Early Readin g

Instruction
What Science Really
Tells Us about How to
Teach Reading

Diane McGuinness
E arl y R e ading Ins t r uc t ion
E arl y R e ading Ins t r uc t ion

What Science Really Tells Us about

How to Teach Reading

Diane McGuinness

A Bradford Book
The MIT Press Cambridge, Massachusetts London, England
( 2004 Diane McGuinness

All rights reserved. No part of this book may be reproduced in any form by
any electronic or mechanical means (including photocopying, recording, or
information storage and retrieval) without permission in writing from the
publisher.

This book was set in Janson and Rotis Semi-sans on 3B2 by Asco Typesetters,
Hong Kong.
Printed and bound in the United States of America.

Library of Congress Cataloging-in-Publication Data

McGuinness, Diane.
Early reading instruction : what science really tells us about how to teach
reading / Diane McGuinness.
p. cm.
‘‘A Bradford book.’’
Includes bibliographical references and index.
ISBN 0-262-13438-1 (alk. paper)
1. Reading (Early childhood) I. Title.
LB1139.5.R43M34 2004
372.4—dc22 2003066585

10 9 8 7 6 5 4 3 2 1
C on t ent s

Preface vii
Introduction xiii
1 Why English-Speaking Children Can’t Read 1

2 On the Nature of Writing Systems 11

3 The Structure of the English Alphabet Code 37
4 How to Teach Reading: Lessons from the Past 73

5 How to Teach Reading: Modern Research 107

6 Phoneme-Awareness Training 153
7 Reading Fluency 189

8 Vocabulary and Comprehension Instruction 211

9 How Does Anyone Learn to Spell? 247
10 The Many-Word Problem: More to Spelling Than
Meets the I 279
11 New Directions for the Twenty-First Century 317
Appendix 1: How Nations Cheat on International
Literacy Studies 349
Appendix 2: Misuse of Statistics 355
Appendix 3: Analysis of Word Lists from Treiman et al.
1995 359

Glossary 363
References 371
Author Index 395

Subject Index 401

 P r e f ac e

Five thousand years ago, Egyptian and Sumerian scholars designed the
first full-fledged writing systems. Though these systems were radically
different in form, the Egyptians marking consonants and whole-word
category clues, and the Sumerians marking syllables, both were complete
and self-contained. Any name, any word, or any word yet to come, could
be immediately assigned the appropriate symbols representing that word’s
phonology.
Schools were established for the sons of the elite—the rulers, priests,
administrators, and wealthy farmers, plus the obviously gifted—and not
much changed in this regard until the nineteenth century, when the
universal-education movement began gathering momentum. Up to this
point, no one kept track of which children were more or less successful in
mastering this extraordinary invention. But with children sorted by age,
and every child in attendance, individual differences in learning rate and
skill were hard to ignore. In most European countries, individual differ-
ences were minor, and when problems did occur, they impacted reading
fluency and reading comprehension. In English-speaking countries, indi-
vidual differences were enormous. Some children were learning to read
quickly but others were not learning to read at all, despite years of teach-
ing. This applied across the board—to decoding, spelling, fluency, and
comprehension. Was this failure due to the teaching method, the nature
of the written code itself, or something inherent in the child?
Answering this question took most of the twentieth century, and now
that the answers are in, there are some huge surprises. Reading and spell-
ing are easy to teach if you know how to do it. Influential theories driving
much of the research over the past 30 years are not supported by the data.
Meanwhile, the volume of research has snowballed to such an extent that
 viii

the quantity of studies has become unmanageable. The huge and formi-
Preface

dable databases on almost every topic related to reading are an impedi-

ment to progress. To get a sense of the actual size of these databases, and
the quality of the studies in them, the National Reading Panel (NRP) de-
cided to keep score. They reported that of the 1,072 studies carried out
over the past 30 years on methods of reading instruction, only 75 survived
a preliminary screening consisting of these criteria: publication in a re-
ferred journal, comparison of at least two methods, random selection of
subjects into comparison groups, and statistical analysis sufficient to com-
pute effect sizes (National Reading Panel, 2000). On further scrutiny, only
38 studies were found to be methodologically sound. It was the same story
for each area of reading instruction. The NRP uncovered a whopping
19,000 papers on the theme that ‘‘reading a lot’’ helps children learn to
read. (It does not, but only 14 studies survived the final screening to prove
it.) The training studies on phoneme awareness, reading fluency, vocabu-
lary instruction, and methods of teaching reading comprehension all suf-
fered a similar fate.
I faced the identical problems when I set out to write a book that
was intended to review the research on reading in the twentieth century.
Trying to squeeze all this material into one volume, while adjudicating
between reliable and unreliable studies, proved impossible. The result was
two complementary, but independent, books. Early Reading Instruction
deals with the historical and scientific research on reading instruction,
including a detailed analysis of the NRP report. The second book (Lan-
guage Development and Learning to Read ) focuses on reading predictors—
whether or not individual differences in certain perceptual, linguistic, or
cognitive skills impact children’s ability to learn to read. The proof (or
lack thereof ) for many of the popular theories in this area of research
lies outside the field, in the mainstream research on language develop-
ment carried out by developmental psychologists, psycholinguists, and
researchers in the speech and hearing sciences, and this adds another level
of complexity to the mix. The table of contents for the second book is set
out following this preface. These books are self-contained and do not have
to be read in any order. However, they reference one another whenever a
greater exposition (or proof ) of a statement or argument is provided in the
other volume.
 ix

A pronunciation key is provided in the accompanying table. It should

Preface
be noted that this key does not conform to the International Phonetic
Alphabet (IPA). Instead, it represents the most common spelling in En-
glish for each phoneme. The IPA is a particularly poor fit to the English
spelling system compared to other European alphabets, which are more
directly tied to the Latin sound-symbol code. As such, the IPA is confus-
ing to people unfamiliar with it. For example, the IPA marks the sound ah
with the letter a. In English, this letter typically stand for the sounds /a/
(cat) or /ae/ (table), while ah is marked with the letter o (hot), which is the
symbol for the sound /oe/ in the IPA. This muddle exists for most vowel
spellings.
A glossary of terms is provided at the end of the book.

English Phonemes and Their Basic Code Spellings

Sounds are indicated by slash marks.
Consonants
Sound As in Basic code spelling
/b/ big b
/d/ dog d
/f/ fun f
/g/ got g
/h/ hot h
/j/ job j
/k/ kid k
/l/ log l
/m/ man m
/n/ not n
/p/ pig p
/r/ red r
/s/ sat s
/t/ top t
/v/ van v
/w/ win w
/z/ zip z
The following are combinations of two consonants with special spellings:
/ch/ chin ch
/ng/ sing ng
 x

/sh/ shop sh
Preface

/th/ thin th
/th/ then th
/zh/ azure —
/ks/ tax x
/kw/ quit qu

Vowels
Sound As in Basic code spelling
/a/ had a
/e/ bed e
/i/ hit i
/o/ dog o
/aw/ law aw
/u/ but u
/ae/ made a-e
/ee/ see ee
/ie/ time i-e
/oe/ home o-e
/ue/ cute u-e
^
/oo/ look oo
/oo/ soon oo
ou out ou
oi oil oi
Vowel þ r
/er/ her er
/ah/–/er/ far ar
/oe/–/er/ for or
/e/–/er/ hair air

There are nine vowel þ r phonemes. All but one (/er/) are diphthongs—
two sounds elided that count as one vowel. Those listed above have special
spellings and need to be specifically taught. The remainder use more
conventional spellings and can be taught as ‘‘two sounds’’: /eer/ /ire/ /ure/
/oor/ /our/ as in deer, fire, cure, poor, our.
L an g u a g e D e v e l o p m e n t a n d L e a rn i n g t o R e a d
C o n t e n ts

Preface
I Does Phonological Awareness Develop?
1 The Origin of the Theory of Phonological Development
2 Development of Receptive Language in the First Year of
Life
3 Speech Perception After One
4 Links: Auditory Analysis, Speech Production, and
Phonological Awareness
5 Young Children’s Explicit Awareness of Language
6 What is Phoneme Awareness and Does It Matter?
II Expressive Language, Reading, and Academic Skill
7 The Development of Expressive Language
8 The Impact of General Language Skills on Reading and
Academic Success
III Direct Tests of the Language-Reading Relationship
9 An Introduction to Reading Research: Some Pitfalls
10 Auditory and Speech Perception and Reading
11 General Language and Reading: Methodological Issues
12 Vocabulary and Reading
13 Verbal Memory and Reading
14 Syntax and Reading
15 Naming Speed and Reading
16 Slow Readers: How Slow Is Slow?
17 Summary: What Do We Know for Sure?
 Int r o duc t ion

It may come as a surprise to many readers that we know how to teach

reading so that every child can succeed. Not only this, but there are highly
successful programs in all categories of reading instruction, including
decoding, spelling, fluency, vocabulary, and comprehension. The knowl-
edge that these programs exist is largely due to the National Reading
Panel. They performed an invaluable service for everyone in the field by
carrying out exhaustive searches of the enormous databases on reading,
and carefully pruning the deadwood. I was able to take advantage of their
screening process, and expand on this in this book to look at the successful
methods in much more depth than they were able to do.
Early Reading Instruction is about much more than current research
on teaching methods. It assesses the most recent discoveries on how writ-
ing systems were designed and the implications for how they should be
taught. It provides an analysis of the efforts to categorize the English
spelling code, and of how our new insight into the structure of this code
affects the process of teaching children to read and spell. The book also
reviews the historical evidence on important breakthroughs in teaching
methods that got lost at the beginning of the twentieth century.
Distilling the lessons from the historical and scientific evidence, I
established a prototype, or template, for what an effective reading program
should and should not contain. The advantage of a prototype is that it is
neutral with respect to specific programs, highlighting the elements that
really matter, as well as those that have no effect or are even detrimental.
When I compared this prototype to the modern research on reading
methods, it precisely mirrored the most successful programs, those that
boost reading scores one or two years above grade level after a year or less
of instruction.
 xiv

This book also addresses a new problem in reading research—what I

Introduction

call the many-word problem. This is the fact that, even if a logical or rea-
sonable instructional method for spelling was introduced into the class-
room, it will never be possible to teach children how to spell every word
in the English language. Because our spelling system is so unpredictable,
what cues or kinds of exposure determine how people learn to spell? The
answers to this question are new and surprising.
Early Reading Instruction is largely an inductive analysis of the histori-
cal evidence and the empirical research on reading instruction. Because
reading methods are complex and the field is controversial, I will be
addressing methodology in more depth than is usually necessary. In part
this is because there is no mechanism in the field for screening out invalid
research. Examples of good and bad research alike can be found in every
journal (even in the top-rated flagship journals) and are very hard to tell
apart. This is just as true of the high-profile studies cited by everyone in
the field, as for research on major breakthroughs that nobody knows
about. This is particularly an issue for chapters 5 through 8, which review
the research screened by the National Reading Panel. To make this
material as accessible as possible, I have provided summaries at the end of
each chapter.

A Few Facts about How to Teach Writing Systems

This book focuses on the research that bears directly on how to teach the
alphabet code, the writing system we inherited from the Greeks and the
one we are stuck with. I will not spend more time than necessary consid-
ering forms of reading instruction that violate the alphabet principle, for
reasons set out below.
As we will see in chapter 2, all writing systems, living or dead, are
based on phonological units of sound below the level of the word. Writing
systems cannot be based on the whole word, because languages have too
many words. The historical evidence shows conclusively that the ability to
memorize sound-symbol pairs closes off at around 2,000 pairs. This is an
ultimate limit requiring years of study. This means no one can learn to
read by memorizing whole words by sight.
There have been four (and only four) phonological units adopted
for the writing systems of the world (McGuinness 1997c, 1998b). Syl-
lables are used for languages with very few syllable types, like Chinese.
 xv

Consonant-vowel units (CV diphones) are used for languages where

Introduction
words are mainly based on those units (CVCVCV ), like the languages of
India and Japan. An English word with this structure is potato. Consonants
only (CCCC) are used for languages where consonant sequences carry
the meaning load and vowels indicate changes in grammar, like Semitic
languages. Phonemes (individual consonants and vowels) are used for all
languages with a highly complex syllable structure and no common pho-
nological patterns, like European languages. History shows these phono-
logical units are never mixed in any writing system.
European languages are written in an alphabet, because they cannot
be written any other way. This is a fact, and there is nothing we can do
about it. The evidence reviewed in this book shows that when you follow
the principles by which writing systems are constructed and teach the
English writing system appropriately, 4-year-olds can easily learn to read
in about 10 to 12 weeks. It makes no sense to continue teaching reading
the way we do.

Reading Skills Vary from One Country to Another

The scientific study of reading is about the mastery of a human invention,
not the study of natural laws like those in chemistry, physics, and biology.
This complicates things. To begin, there is no universal thing called
reading independent of a particular language and a particular solution for
how that language was written down. A reading problem in one country
is not necessarily a reading problem in another. Children in English-
speaking countries have far more difficulty learning to read than children
in many European countries, a fact that is reflected in how reading is
measured. In English-speaking countries, the main test of reading success
is decoding accuracy, the ability to read isolated words one at a time. Yet in a
large number of European countries, decoding accuracy is of little con-
cern, because every child reads accurately. There are no standardized tests of
decoding skill in these countries. Instead, reading is measured by tests of
fluency (reading speed) and comprehension.
These differences are due to the way individual speech sounds called
phonemes are mapped to symbols in the various alphabetic writing systems.
In a highly ‘‘transparent’’ alphabet, like Italian, Spanish, German, and most
Scandinavian writing systems, there is mainly one way to write (spell) each
phoneme in the language, and one way to decode each letter or letter pair
 xvi

called digraphs (sh in ship). Transparent alphabet codes are ‘‘transparent’’

Introduction

in the sense that it is obvious how they work, making them easy to teach
and learn.
Highly opaque alphabet codes, like English, have multiple spellings
for the same phoneme (be, see, sea, scene, field, deceit, radio, marine, lucky,
key), and multiple decodings for the same letter or digraph (soup, soul,
shout, should, rough, cough, journey). Opaque writing systems can be very
hard to teach unless teachers use a structured method that mitigates these
difficulties. Poor teaching of a difficult code creates enormous confusion
and can lead to reading problems and failure.
If success in learning to read is intimately bound to the form of the
script, no universal laws can be applied to it. In other words, reading is not
a natural aptitude—‘‘a property of the child’’—and children who fail to
learn to read do so mainly because of environmental causes, not biological
causes.

An Opportunity Lost
In the early nineteenth century, Isaac Pitman, a self-taught linguist and
author of the famous shorthand method, hit on a solution for how to teach
our complex alphabet code. Spelling reformers had been clamoring for
changes to our spelling system since the sixteenth century, with no suc-
cess. John Hart (An Orthographie, 1569) and Richard Mulcaster (The First
Part of the Elementarie, 1582) advocated sweeping reforms almost to the
point of scrapping everything and starting over. Mulcaster called for a new
phonetic spelling system—a transparent alphabet. Although nothing like
this ever happened, Mulcaster did manage to eliminate some oddities,
such as the unnecessary letter doublings and e’s at the ends of words
(greate, shoppe). He was responsible for standardizing the letter e as a dia-
critic, a symbol that signals a pronunciation, as in the words came, time,
fume, home.
No true spelling reform of the type that Hart and Mulcaster advo-
cated ever succeeded, despite numerous efforts by highly influential peo-
ple over the centuries. And while Samuel Johnson was able to standardize
spelling in his famous dictionary in 1755, he standardized the spellings for
words, but not the spellings for phonemes.
Pitman’s solution was to adapt his shorthand method for use in be-
ginning reading instruction by setting up what I call an artificial transpar-
 xvii

ent alphabet, consisting of one letter or special symbol assigned to each of

Introduction
the 40 English phonemes. He developed this into a classroom program
called Phonotypy with A. J. Ellis in 1847. An artificial transparent alphabet
levels the playing field for English children vis-à-vis their continental
cousins, and provides the novice reader with critical information about
how alphabet codes work:

They are based on the unchanging sounds (phonemes) in our speech.

The number of these speech sounds is finite, providing an endpoint for
managing the code.

Letters are not the units of the code but arbitrary symbols for those

units.

All codes are reversible. Reading and spelling (decoding/encoding) are

mirror images of one another and should be taught together.

By far the most effective classroom program to stem from Pitman’s

insights was designed by Nellie Dale (1898, 1902), a classroom teacher at
the Wimbledon High School for girls in England. Instead of adopting
Pitman’s partially artificial script, Dale set up a basic code using the most
common spelling for each phoneme. This has two advantages: no artificial
symbols have to be unlearned, and spelling alternatives can be pegged
onto the system with no change in logic. Teachers can say: ‘‘We learned
the main way to spell this sound. Now I’ll show you another way to spell
it. Let’s look at the words that use this spelling.’’ Dale’s program incor-
porated the insights of a great teacher with those of a great linguist, and
soon became popular on both sides of the Atlantic.
The rise of the universal-education movement brought these advances
to a halt. By the 1920s, whole-word (sight-word) methods were beginning
to dominate, and by the 1930s they had eclipsed any type of phonics
teaching. Though the breakthroughs of Pitman, Ellis, and Dale have
recently been rediscovered, they could be lost again. It is up to us to see
this does not happen. The sad tale of reading instruction in the twentieth
century, in contrast to the brilliant programs waiting in the wings, is the
main message of this book.
We are in a jam. Reading instruction, if anything, has deteriorated
over the course of the past century. We are stuck with disastrous whole-
word (sight-word) methods that will not go away. To complicate things,
 xviii

reading researchers are constrained by a seriously flawed theory of chil-

Introduction

dren’s language development and cannot think past it. According to this
theory, children gradually develop an awareness of speech sounds in the
order: words, syllables, phonemes—a process alleged to take 7 to 8 years.
This has led to the proliferation of phonological-awareness training pro-
grams, that are, at best, a waste of time (see chapters 5 and 6). The theory
that phonological awareness develops from larger to smaller units of
sound, or even that it develops at all, is contradicted by the evidence from
the mainstream research on early language development (see Language
Development and Learning to Read ).
We have the knowledge to teach every child to read, write, and spell
at an amazingly high level of skill. So far, this knowledge has not been
made available to educators, legislators, parents—or to many researchers.
The vast quantity of invalid and unreliable research clogging the databases
makes it almost impossible for anyone interested in the field to ferret out
what is accurate and important and what is not. The National Reading
Panel has made an important start on this problem, and the primary goal
of Early Reading Instruction is to try to finish the task.1

1. Readers interested in the early history of spelling reform and reading in-
struction should consult Scragg 1974; Morris 1984; Balmuth 1992; McGuin-
ness 1997c, 1998b.
 1
WHY ENGLISH-SPEAKING CHILDREN CAN’T READ

As the universal-education movement began gathering momentum, edu-

cators broke ranks with nineteenth-century traditions. Reading instruction
got so far off track that the twentieth century will go down in history as
the century of the demise of the English alphabet code. The final reck-
oning of an unceasing attempt on its life came in the 1990s. For the first
time, properly conducted national testing, international reading surveys,
cross-cultural studies, and classroom research pointed to the inescapable
conclusion that reading instruction in English-speaking countries is a
disaster. The functional illiteracy rate for American 9-year-olds is 43 per-
cent (Mullis, Campbell, and Farstrup 1993; Campbell et al. 1996).
International reading surveys carried out by Statistics Canada brought
dismal news (Organization for Economic Cooperation and Development,
1995, 1997). In six English-speaking nations, the proportion of function-
ally illiterate/very poor readers among 16- to 65-year-olds ranged from a
low of 42 percent in Canada to a high of 52 percent in the United King-
dom. These figures were in stark contrast to those of many European
nations. The comparable figure for Sweden was 28 percent. Sweden’s
functional illiteracy rate for 16- to 25-year-olds (level 1 of 5 levels) is 3.8
percent. This rate is nearly three times higher in Canada (10.7 percent),
and six times higher in the United States (23.5 percent).
In 1993, an astonishing report came in from Austria. Heinz Wimmer
set out to study poor readers and initiated a citywide search. He asked 60
second- to fourth-grade teachers in Salzburg to refer their worst readers
for special testing. They identified 120 children, about 7–8 percent of the
school population. Imagine Wimmer’s surprise when the worst readers in
the city scored close to 100 percent correct on a test of reading accuracy
and did nearly as well in spelling. Clearly, none of these children had any
 2

difficulty with the German alphabet code. It turned out their problem was
Chapter 1

reading too slowly. But slow is a relative term. How slow is slow?
To find out, Wimmer collaborated with an English researcher
(Wimmer and Goswami 1994) to compare normal 7- and 9-year-olds from
Salzburg and London. The results were startling. The Austrian 7-year-
olds read comparable material as rapidly and fluently as the English 9-
year-olds, while making half as many errors. Yet the Austrian 7-year-olds
had had 1 year of reading instruction, while the English 9-year-olds had
been learning to read for 4 or 5 years. Equal speed and half the errors in
one-quarter of the learning time is an eightfold increase in efficiency!
Wimmer and his colleagues (Landerl, Wimmer, and Frith 1997) got
the same extraordinary results when they compared their worst readers
(incredibly slow) with English children identified as ‘‘dyslexic’’ (incredibly
inaccurate). The children were asked to read text consisting of nonsense
words. The so-called Austrian slow readers were not only more accurate
than the English ‘‘dyslexics,’’ but they read twice as fast. The average
Austrian ‘‘slow reader’’ would be able to read a 500-word passage in about
10 minutes, misreading only 7 percent of the words. The average English
‘‘dyslexic’’ would read only 260 words in this time, and misread 40 percent
of the words. It seems the expression ‘‘worst reader’’ is relative as well.
An even more dramatic study was reported from Italy. Cossu, Rossini,
and Marshall (1993) tested Down’s syndrome children with IQs in the 40s
(100 is average) on three difficult reading tests. They scored around 90
percent correct, breezing through Italian words like sbaliare and funebre.
However, they could not comprehend what they read, and they failed
miserably on tests of phoneme awareness, the skill that is supposed to be
essential to decoding.

What is going on?

The answer is simple. European countries with high literacy rates have
a twofold advantage. First, they have a transparent alphabet code, a
nearly perfect one-to-one correspondence between each individual sound
(phoneme) in the language and a visual symbol—a letter or letter pair
(digraph). For languages with more sounds than letters in the alphabet
(English has 40þ sounds), this problem was handled sensibly. When a
letter or digraph is reused to represent more than one sound, it is marked
by a special symbol (a diacritic) to signal a different pronunciation. In
 3

German, an umlaut distinguishes the vowel sound in Bäume (boimeh)

Why English-Speaking Children Can’t Read

from that of Baum (boum). And while a sound can occasionally be spelled
more than one way, there is never more than one way to read a letter or
digraph. The English spelling system suffers from both afflictions: multi-
ple spellings for the same phoneme, and multiple ways to decode letters
and letter sequences. This is the definition of an ‘‘opaque’’ writing system.
Reading instruction is the second part of the equation. To a great
extent, reading instruction is a function of the complexity of the spelling
code. Teaching a transparent writing system is far easier than teaching an
opaque one, because it is obvious (transparent) how it works. Teaching
can be streamlined and proceeds at a rapid pace. In Austria, children are
taught the sounds of the German language and which letter(s) represents
each sound. Reading and spelling are integrated at every step, which
reinforces the code nature of a writing system—that is, the fact that the
operations are reversible, involving both encoding and decoding. No
clutter or noise clogs the process, such as teaching letter names or lots of
sight words. Because basic reading instruction is fast and pretty well
guaranteed, it can begin late—at age 6 in most countries (age 7 in Scan-
dinavian countries)—and end early (after 1 year or less). Parents sleep
soundly in their beds, safe in the knowledge that their child will be read-
ing and spelling by the end of the first year of school. (This is not to say
that inappropriate teaching methods cannot nullify the advantages of a
transparent alphabet.)
The cross-cultural comparisons reveal that the source of English-
speaking children’s difficulties in learning to read and spell is the English
spelling system and the way it is taught. These comparisons provide ir-
refutable evidence that a biological theory of ‘‘dyslexia,’’ a deficit pre-
sumed to be a property of the child, is untenable, ruling out the popular
‘‘phonological-deficit theory’’ of dyslexia. For a biological theory to be
accurate, dyslexia would have to occur at the same rate in all populations.
Otherwise, some type of genetic abnormality would be specific to people
who learn an English alphabet code and be absent in people who live in
countries with a transparent alphabet, where poor readers are rare. A dis-
order entirely tied to a particular alphabetic writing system is patently
absurd and has no scientific basis. English-speaking children have trou-
ble learning to read and spell because of our complex spelling code and
because of current teaching methods, not because of aberrant genes.
 4

A Century of Whole-Word Methods

Chapter 1

The historical evidence shows that teaching methods most similar to

Dale’s (and the methods used in other European countries) resemble the
programs found to be most effective in recent studies (see chapter 5).
These phonics-type methods were replaced early in the twentieth century
as a consequence of universal education. Self-appointed education gurus
and newly fledged professors of education with little or no knowledge of
how to teach reading decreed that old-fashioned phonics had to go. In-
stead, children should be taught whole words by sight, just as Chinese
people were thought to do, using a method known as ‘‘look-say.’’ The
‘‘whole-word’’ century was launched, and the alphabet code soon vanished
without a trace.
Look-say was replaced early on by a meaning-based sight-word
method. Children were introduced to a few words in each lesson, spent
most of the lesson learning the meanings of these words (words they
already knew), and then read dreary stories where these words were re-
peated endlessly:

‘‘Come, come, John. See me. I can swing. Come and see.’’

Phonics lessons came in late or not at all, and made no sense. This ap-
proach was the platform for ‘‘basal readers’’ (U.S.) or ‘‘reading schemes’’
(U.K.), products of the educational publishing houses. Basal readers
dominated from the 1930s until the late 1970s. In the mid-1960s, a survey
showed that basal readers were used in 95 percent of classrooms in the
United States. Many people still remember Dick and Jane or Janet and
John.
The extreme dullness and repetitiveness of the basal-reader method,
plus other precipitating factors, eventually led to a backlash. Basal readers
were swept away by a third whole-word method that came to be known
as ‘‘whole language’’ (U.S.) or ‘‘real books’’ (U.K.). The theory behind
whole language is that with minimal guidance, children can teach them-
selves to read naturally. They do this by following along as the teacher
reads stories written in natural language, and by reading on their own
while using all their ‘‘cuing systems.’’ These include everything from
guessing words based on context and the illustrations, to sight-word mem-
 5

orization, to attempts to decode by letter names. Children are encouraged

Why English-Speaking Children Can’t Read

to ‘‘invent’’ their own spelling system during creative writing.
The basal-reader approach may have been boring, slow, and wrong,
but at least it was honest. What you saw was what you got (figuratively and
literally). Whole language is based on faith, promising everything and
delivering nothing. Children are passed from grade to grade in the belief
that they will eventually teach themselves to read. And if they do not, it
is their fault. Something is wrong with them. Needless to say, whole
language was not a success. It led to skyrocketing illiteracy rates, quite
beyond anything produced by basal readers. In California, where whole
language was mandated in 1987, the functional illiteracy rate soared to 60
percent, plunging California to last in the nation. The disastrous test
scores dampened the enthusiasm of parents and legislators for whole lan-
guage, but had little or no effect on professors of education, education
publishing houses, curriculum specialists, and many classroom teachers.
Because they control what goes on in the classroom, whole language is
still with us, battered but unbowed, despite lip service to the contrary.

Nouvelle Eclecticism
In the 1990s, reading researchers and directors of research agencies, sup-
ported by state and national politicians, launched a campaign to rescue
children from whole language, claiming they wanted a return to phonics.
But after nearly a century, no one was quite sure what phonics was.
Instead, what they proposed was not phonics, but a new kind of eclecti-
cism. In the past, eclecticism referred to a teacher’s habit of mixing dif-
ferent approaches and materials in the mistaken belief that children have
different learning styles. This form of eclecticism is individualistic and
haphazard.
‘‘New eclecticism’’ is based on the notion of developmental gradualism,
a consequence of the myth that children become more phonologically
aware as they grow older. Children begin by learning whole words by
sight, then move on to syllables (clapping out beats), then to word families
(words with rhyming endings like fight, might, sight), with the goal of be-
ing eased into an awareness of phonemes, a process taking a year or two,
if it is completed at all. This not just a passing whim. It is the method
promoted by people in charge of research funding in the United States.
 6

Full-blown ‘‘nouvelle eclecticism’’ was recently mandated by the British

Chapter 1

government, complete with suitcases of lesson plans, charts, and materials

sent to every elementary school in the country at a cost to the taxpayer of
£56 million.

The Myth of Phonological Development

There is a strange and twisted tale behind this new movement. Inspired by
discoveries in speech perception (A. M. Liberman et al. 1967), Isabelle
Liberman (Liberman et al. 1974) proposed that phonological awareness
‘‘develops’’ throughout childhood and underpins ‘‘reading readiness.’’
Around the same time, paleographers and linguists were launching a new
field of study—the comparative analysis of writing systems. Among a spate
of books on the topic, by far the most influential was by Ignace Gelb
(1963). Gelb proposed that writing systems ‘‘evolve.’’ They begin with
pictograms (recognizable little pictures standing for whole words), gradu-
ate to logograms (abstract signs for whole words), then to syllabaries (syl-
lable signs), and finally to alphabets (phoneme signs). According to Gelb,
this is true of every civilization that invented writing. While Gelb was
highly regarded for his scholarly work (he was instrumental in cracking
the code of the Hittite writing system), his colleagues in paleography were
far less enthusiastic about his evolutionary theory. And as more archaeo-
logical evidence came to light, it became clear that Gelb’s theory was
fatally flawed.
For Liberman and her colleagues, Gelb’s theory was almost too good
to be true. It was assumed that the ‘‘evolutionary’’ order of writing systems
mirrored the developmental sequence of speech perception—moving
from larger to smaller phonological units (whole words, syllables, pho-
nemes). And because children appeared to differ in when and whether
they became ‘‘phonologically aware,’’ the theory provided an explanation
for dyslexia as well.
There is no scientific support for this theory or anything resembling
it (a complete analysis of this issue is provided in Language Development
and Learning to Read ). Instead, the evidence shows that children become
less rather than more phonologically aware as time goes by. Tiny babies
can discriminate between any consonant-vowel contrasts (ba versus pa) in
every language of the world, an aptitude that disappears by 12 months of
 7

age. By 9 months they can tell the difference between legal and illegal

Why English-Speaking Children Can’t Read

phoneme sequences in words (Aslin, Saffran, and Newport 1998; Friederici
and Wessels 1993; Mattys et al. 1999). In English, illegal consonant
sequences commonly form word boundaries (‘‘word boundaries’’). Infants
use these patterns to wrench words out of the speech stream in order to
build a receptive vocabulary. If they could not hear phonemes, it would be
impossible to split phonemes from one another. Chaney (1992) found that
virtually all 3-year-olds can identify a single phoneme error in a spoken
sentence and fix the error. Furthermore, they can blend isolated pho-
nemes into a word and pick that word from a row of pictures with 88
percent accuracy (96 percent scored significantly above chance).
This does not mean that young children know that speech consists of
phoneme sequences or that an alphabetic writing system represents these
phonemes. They will not make this connection unless it is taught. Even
fluent readers are not consciously aware of phonemes until someone
points them out, and there is no reason they should be. The brain carries
out this analysis so rapidly that it operates below the level of conscious
awareness. No one needs to be aware of phonemes unless they have to
learn an alphabetic writing system. Anyone, at any age, who has to learn
an alphabetic writing system must be taught to unravel phonemes in
words to understand how an alphabetic writing system works.
The proposed link between the discoveries in speech perception (its
biological foundations) and Gelb’s theory of the origin of writing systems
(its supposed ‘‘evolutionary’’ foundations) had a powerful impact on read-
ing research in English-speaking countries that has not abated over time.
Yet Gelb’s theory was wrong, and the analogy to speech perception with-
out merit. Writing systems do not evolve.
The comparative analysis of writing systems was in its infancy when
Gelb proposed his theory in 1963. This discipline came of age with the
publication of Florian Coulmas’s Writing Systems of the World (1989), a
synthesis of extraordinary breadth and depth. This was followed in 1996
by Daniels and Bright’s remarkable compendium. Perhaps it is fortuitous
that Coulmas’s book appeared at the precise moment in history when we
finally learned the ghastly truth about the functional illiteracy rates in
English-speaking countries. This is a problem of monumental propor-
tions. It is not merely a question of how to teach our formidable spelling
 8

code, but how to shed 100 years of unsubstantiated beliefs about how to
Chapter 1

teach reading and false theories about why children fail.

Before any real transformation can occur, people need a deeper un-
derstanding of the issues. They need to know what a writing system is and
how it works. They need to know how a particular writing system can and
cannot be taught and which skills are important to success. Coulmas gave
us the first road map to find our way out of this quagmire. There is no
better place to start than with the lessons learned during the 5,000-year
history of the origins of writing systems. A writing system has a central
logic, and this logic is based on how the human mind works. If this logic is
not adhered to, a writing system cannot be taught effectively, if at all.
Before I move on to discuss these new discoveries, I want to present
the commonly held assumptions about writing systems that either directly
or indirectly affect what goes on in the classroom. It may come as a sur-
prise that not one of these assumptions is true.

1. Writing systems evolve from whole-word systems (logographs), to syl-

lable systems (syllabaries), to phoneme systems (alphabets). Logographs
are ‘‘low’’ on the evolutionary scale (poor), and alphabets are ‘‘high’’
(best).
2. Ontogeny recapitulates phylogeny. Children go through the same
stages during development of speech perception, mirroring the evolution
of writing systems: from whole words, to syllables, to phonemes.
3. The evolutionary process (in both cases) is inevitable and goes in one
direction.
4. There is such a thing as a logographic writing system.
5. The Chinese have a logographic (archaic) writing system.
6. The alphabet principle was discovered only once and spread by
diffusion.
7. Alphabets are superior to other writing systems. Nearly all writing
systems today are alphabetic.
8. An alphabetic writing system can be used as a proxy for a logographic
writing system. That is, children can learn to read an alphabetic writing
system by memorizing whole words (random letter strings) by sight.
9. Different sound-based units can (and should) be mixed together in
teaching the alphabet code, including (but not limited to): whole words
 9

(sight words), syllables (ta-ble, droop-ing), rhyming endings or word

Why English-Speaking Children Can’t Read

families (and, band, hand, sand ), consonant blends (br, tw, sl, sts, nt), and
individual phonemes. Children will not be confused by this practice and
will clearly understand how the alphabet code works.
10. Over time, every word is read as a sight word. Ultimately, people read
holistically, and only rare words or unknown words need to be decoded
phonetically.
 2
ON THE NATURE OF WRITING SYSTEMS

The comparative analysis of writing systems sheds considerable light on

what the human mind can or cannot remember, and on how human
memory and language determine the way writing systems are designed.
This knowledge helps us understand how a particular writing system can
and cannot be taught.
Comparative analysis came of age when there was a sufficient body of
evidence to provide a complete succession of forms from protowriting to
full-blown writing systems.
At the beginning of the nineteenth century, these archaic writing sys-
tems still awaited decipherment: Egyptian, Sumerian, Babylonian, Akka-
dian, Hittite, Aramaic, Persian, Mayan, the Cretan scripts of Linear A and
Linear B, and many others. The first breakthrough came in 1823 and the
last (so far) in 1953. Eleven archaic writing systems still defy a solution.
The analysis of an extinct writing system unfolds in a series of steps and
can take well over a hundred years. I will take up these steps briefly, be-
cause each is relevant to understanding how a writing system works.

The Code Breakers

A writing system is a code in which specific elements of a language are mapped
systematically to graphic signs or symbols. Scholars who decipher ancient
writing systems are ‘‘code breakers’’ in exactly the same sense as the
cryptographers who crack the enemy’s coded messages. Codes have an
internal logic and an external system or structure; otherwise they could
not be broken.
Code breakers have a formidable task, and breaking the code is almost
never solved by one person alone. This is why we should never expect a
young child to ‘‘break the code’’ of a writing system unaided. In fact, code
 12

breakers need superb analytic skill in combination with expertise in the

Chapter 2

language in question or a related language. Unless the language is known,

or can be linked to some other known language, the code cannot be
cracked.
The code nature of writing systems is not well understood, in part
because writing systems are imperfect graphic representations of both
semantic (meaning-based) and phonological (sound-based) aspects of a
spoken language. No writing system is purely one or the other (Coulmas
1989). The fact that codes for spoken language are so complex is impor-
tant to bear in mind when thinking about reading instruction. The novice
reader is, in a very real sense, like a code breaker facing an alien script, and
this task can be impossible if instruction is misleading, incomplete, or
absent altogether.
All codes are reversible mapping systems (McGuinness 1997c, 1998b).
What is put into a code (encoded) can be decoded by reversing the pro-
cess. A code is more or less transparent to the degree that it has a one-to-
one correspondence between the original units, whatever they might be,
and their coded form. A completely transparent code is perfectly reversible:
one symbol is associated with only one unit of something, and that unit
has only one symbol. No writing system is completely transparent, but
some alphabets—like those listed earlier—come close.
There are many well-known codes with perfect one-to-one corre-
spondences. There are number codes for quantities, codes for musical
notation that mark pitch, duration, and rhythmic meter and stress, and
computer codes devised to translate binary logic (the zeros and ones of
computer hardware) into octal, then into computer ‘‘languages’’ (and back
again).
A fundamental aspect of codes is that the elements being encoded are
logically (not just perceptually) distinct from the symbols of the code
(McGuinness 1997c, 1998b). To understand what a code is, to master the
system and use it efficiently, a learner should know the direction in which
the code was written. This means being made aware, indirectly or explic-
itly, of what is being encoded and what constitutes the code. Number
symbols code quantity and are not merely transcriptions of number words.
Musical notation codes the music, not a particular instrument or the fin-
gerings on a keyboard or strings.
 13

Similarly, letter symbols in alphabetic writing systems represent pho-

On the Nature of Writing Systems

nemes. Phonemes are the basis for the code, and the letters are the code.
Letters do not ‘‘have’’ or ‘‘make’’ sounds. People have sounds. Unless this
logic is made clear to teachers and used to guide beginning reading in-
struction, the code will lose one of its essential elements: its reversibility.
This is especially critical for learning an opaque writing system. Children
are magical thinkers and can easily believe that letters ‘‘make sounds,’’
leading them to use one logic to read (decode) and a different logic to spell
(encode). For this reason all phonemes will be enclosed in slash marks, and
all spellings (as opposed to readings) will be underlined. It is important to
be able to distinguish between what is being decoded and what constitutes
the code.

The Translators
Translators take over once the code has been cracked and come from a
variety of disciplines. Linguists and grammarians must unravel the pho-
nological, grammatical, and structural properties of the language before a
complete translation is possible. When translators work with the vocabu-
lary of a dead language, such as ancient Egyptian, Sumerian, or Baby-
lonian, they must see the same word in different contexts, in a variety of
texts, to gain any real insight into what the word implies. Once vocabulary
and grammar are tentatively worked out, the translator needs a historian’s
talent for framing a synthesis that accurately represents the customs, the
political system, the economy, and the feelings of the people about all
manner of things, including their sense of themselves.
The historical reconstruction of an ancient civilization is critical for
an understanding of how writing systems develop. Without knowing any-
thing about the Sumerian economy (that it was agricultural on a large
scale, with a temple distribution system and individual ownership of land),
or the Sumerian religious practices (that priests and city administrators
played a joint and important role in economic matters), or the Sumerian
legal system (that it was an effective system of justice), it would have been
difficult to discern how and why the writing system developed the way it
did.
 14

How Writing Systems Work

Chapter 2

One of the major contributions of comparative analysis has been to docu-

ment and classify the critical elements that determine what, in fact, quali-
fies as a writing system.

Function
A writing system codes spoken language into a permanent form so that it
can transcend time and space. But the most important aspect of a writing
system is its purpose. For one thing, it should make life better. A writing
system that has no effect (being too difficult for most people to learn)
or that makes life worse (generating thousands of bureaucratic forms) is
hardly worth the bother.
Writing systems make it possible to permanently record important
things that are hard to remember, such as rules and laws, and case deci-
sions about breaches of those laws. They can record events of critical
importance to everyone, such as migrations, battles, and other historical
events, as well as disasters like floods, drought, and crop failure. They can
record intentions of good faith, as in a business transaction or in mar-
riage vows. And should disagreements arise, family members, magistrates,
or judges do not have to rely on a person’s word, or on the testimony
of witnesses who may or may not remember or be telling the truth.
Recorded accounts of how land and inheritance disputes were settled in
the Middle Ages, when most parties and witnesses were illiterate, make
fascinating reading (Clanchy 1993). A writing system, plain and simple,
makes civilization work, and without a writing system, it cannot work.

Structure
The texts and documents unearthed from civilizations with the earliest
writing systems in the world, Sumer and Egypt, show that writing systems
originated in the same way for the same reasons. They began as systems
for accounting, inventory control, bills of lading, and invoices. This is why
protowriting (which is not a system at all) has so many symbols that are
stylized pictures or icons. Much more is needed to qualify as a writing
system. A true writing system must represent the entire language, and to
do so it has to meet certain fundamental requirements. Coulmas (1989)
specified three: economy, simplicity, and unequivocality. I would add compre-
hensiveness to the list.
 15

Economy means that the number of symbols used for the system must

On the Nature of Writing Systems

not only be complete but must be as small as possible. Economy is essen-
tial to keep the memory load for symbols manageable. Alphabets, for
example, have the fewest symbols and are therefore the most economical.
Simplicity means that the mapping relationships between what is being
encoded (elements of speech) are straightforward, in the sense that there is
only one way to write a particular word, not half a dozen. Mayan and
Egyptian hieroglyphic writing fail the standard of simplicity miserably.
Unequivocality dictates that the meaning of a written expression be
determined by its form. This is a complex way of saying that what you
read must mean one thing, not many things. Writing systems should not
cause confusion by creating ambiguity. Ambiguity poses more of a prob-
lem in writing than in speech, because context, facial expressions, tone
of voice, and knowledge of the personal history of the writer are absent
in written texts. Coulmas used an example of how the written form dis-
ambiguates speech patterns by means of variant spellings, contrasting the
French phrase cette heures (meaning ‘‘this time’’) with sept heures (meaning
‘‘seven o’clock’’). Although Egyptian and Mayan writing systems fail the
simplicity test, they meet this criterion with flying colors. There is almost
no way to mistake meaning, when the same meaning is cued in several
different ways.
Comprehensiveness means that all words, names, and any possible new
words in the future can be represented by the writing system with relative
ease. That is, you would not need to convene a meeting of the National
Writing Board each time a new word appeared in the language, like com-
puter, hardware, software, Internet, website, hacker, nerd.
These examples highlight Coulmas’s central point, that there is always
a trade-off between the elements of the system. Furthermore, one cannot
assume that all will be well by opting for economy alone, which would
lead to the conclusion that every writing system should be an alphabet.
Instead, the syllable structure of a language will dictate which form of the
code is ‘‘simple,’’ and this turns out to be just as important as economy. As
Coulmas notes, the Morse code is extremely economical. Only two values
of duration (long, short) plus silence constitute the entire code, but this is
not a sufficient reason for it to replace the alphabet.
Comparative analysis has revealed that a perfect, isomorphic mapping
between speech and symbol has never been achieved by any culture so far,
 16

and probably would never work. Unfortunately, there is not, and can
Chapter 2

never be, a perfect writing system sufficient to satisfy a linguist’s dream

writing system, because the needs of the user are remote from the ideal
imagined by the linguist. The simple fact that alphabets use different
spellings for homophones (bear/bare; knight/night; pair/pear/pare) shows
that orthographies work better when they are designed to help the user
rather than to satisfy the purist.
In summary, a writing system must meet several constraints in order
to work at all. It must be sufficiently easy to learn so that it does not take a
lifetime to learn it. A reasonable goal would be that it could be mastered
by a child in a short period of time. It must be unambiguous. The printed
script should be the most faithful translation of the true meaning of the
spoken language that it can possibly be. Finally, it should be comprehen-
sive. There should be a way to record every word and every name in the
language, not just some.
Meeting these goals was a formidable challenge. All early writing sys-
tems began with the wrong solution and had to change course, often to
the point of scrapping everything and starting over. There were stopgap
solutions in which special symbols were added to reduce ambiguity. But
the more extra symbols that were pegged onto the system, the harder it
became to learn. The writing systems with the most props, the greatest
amount of redundancy and instability in the size of units for the code,
were the Egyptian and Mayan hieroglyphic writing systems, systems so
difficult to learn that only the scholar-priests could learn them. We will be
looking in more detail at the writing systems that succeeded in all the
functions listed above. Do these successful writing systems have anything
in common, and if so, what? Are there patterns that can be determined
about how and why writing systems differ? Is one form of writing system
better than another?

Writing Systems Do Not Evolve

Comparative analysts have more or less agreed on the types or forms of
writing and the order in which they appear. Coulmas provides a list that
represents the majority view on this issue. However, as he notes, the list is
‘‘abstract’’ in the sense that it is more or less correct, but not specifically
correct. Here is a simplified and slightly revised interpretation of the list.
There are two levels. The first level is whole-word writing. At this level are
 17

the various forms of whole-word symbols: pictograms of either a specific

On the Nature of Writing Systems

person or object (that man, that bird, that ship), or a prototypical form
(a bird, a ship), and logograms, which are completely abstract signs for
whole words.
A writing system in which each word is represented by a unique visual
pattern is unworkable for several reasons. First, it would always be in the
process of becoming and could never arrive. It would forever fail the
standard of comprehensiveness. Second, it fails the criterion of economy
as well. Adult speakers know about 100,000 words, and use over 50,000
words in ordinary conversations. If every word was coded by a unique
symbol, you would need 50,000 symbols just to begin. (The Oxford Com-
panion to the English Language estimates that the English language has
about one million words; see McArthur 1992.) Third, because every sign
must be individually taught, this does not qualify as a code. If the word is
not in your vocabulary, there would be no way to read (decode) the sign.
In short, whole-word writing is not a system. It has no structure. Learning
a logographic writing system would be exactly like memorizing the tele-
phone directory.
For these reasons, the second level—sound-based writing—is inevitable
due to the unworkability of the first. Here, abstract symbols stand for
units of sound in speech. There are far fewer sounds than words in every
language by several orders of magnitude, no matter which sound unit is
chosen, so this fulfills the requirement of economy. The central logic of a
sound-based system is that the sound unit drives the system and meaning does
not. This is true even though sound and meaning can overlap, as they do
in one-syllable words in a syllabary writing system. Another critical factor
is that sound units in a language are fixed in number, whereas words are
never fixed.
Coulmas lists the sound-unit options as syllables or phonemes, but
does not specify a sequence. Gelb proposed that writing systems evolved
in a fixed order: alphabets must follow syllabaries, and syllabaries must
follow logograms. For Gelb, evolution means the inevitability of moving
to higher forms. Pictograms are primitive. Alphabets are sophisticated and
‘‘best.’’
Finally, Coulmas includes a kind of mopping-up period in which
redundancy is eliminated, and the written form is standardized by the
creation of an ‘‘orthography’’ or ‘‘standard form.’’
 18

As we review the development of individual writing systems, it will

Chapter 2

become obvious that Gelb’s position is untenable. Further, I want to take

a slightly different orientation to Coulmas and present a more psycholog-
ical and pragmatic analysis of the similarities and differences between the
various systems. This way, we can include what we know from research on
human memory and perception, and make this relevant to the classroom.
Coulmas is correct that a writing system must be economical due to the
limits of human memory, and should be designed to suit the user. But
these limits need not be vague. There are precise answers, not only from
the psychological research, but from evidence on the analysis of the
structure of writing systems. This means we can discover exactly what a
human learner can and cannot remember, and exactly what a human per-
ceiver can and cannot perceive about elements in the speech stream,
something that Coulmas does not discuss. In other words, there are even
more constraints on the design of writing systems than Coulmas identifies.

The Limits of Human Memory

When early cultures began the transition to a full-fledged writing system,
two things happened. Pictograms disappeared relatively quickly. Logo-
grams expanded in number and soon reached asymptote. A logograph
ceiling was observed in Sumer, China, and Egypt alike. From the histori-
cal evidence, it is obvious that ancient scholars quickly discovered that a
logographic (whole-word, meaning-based) system would not work. If it
had worked, it would not have been abandoned so early. It was not the
comprehensiveness issue that dampened everyone’s enthusiasm for logo-
graphs, but the economy issue. Memorizing thousands of abstract visual
patterns is something human brains simply cannot do. But this is only half
the problem.
In a logographic writing system, each abstract, meaningless visual
pattern is paired with a particular word. Psychologists call this paired-
associate learning. In a typical paired-associate task, people are asked to
memorize a list of random word pairs (tree-mouth), and are cued by one of
the words later in time: tree—? People are particularly bad at this task
when the word pairs are unrelated in meaning, as in this example, and are
hard pressed to memorize more than about 20 such pairs. And these
memories do not survive (Meyer, Schvaneveldt, and Ruddy 1971; Post-
man 1975; Fisher and Craik 1977).
 19

Paired-associate learning is even more difficult if words are paired

On the Nature of Writing Systems

with abstract visual symbols. Even with extensive study over a long period,
memorizing thousands of abstract visual symbols and remembering which
symbol goes with which word may be fine for a computer, but it is a very
bad idea for a human brain. And the difficulty level would go through the
ceiling if the visual symbols consisted of thousands of random sequences
of the same 26 letters. This is what sight-word methods require of a child
learning an alphabetic writing system, and is the reason no one is a sight-
word reader. Nor do people become sight-word readers as adults, as some
researchers claim. Only the profoundly deaf read purely by sight, and they
reach a ceiling at around a fourth-grade reading/spelling level (Aaron et al.
1998; also see chapter 10, this volume).
The evidence from the study of writing systems shows us not only
that sight-word reading is impossible, but exactly when the memory over-
load occurs. In the case of Egyptian hieroglyphics, even the priests, the
keepers of the sacred script who boasted of their prowess in mastering it,
gave up on logographs as the main unit of the system so early that the date
cannot be traced. Already in 3000 b.c., the date of the earliest Egyptian
writing so far discovered, the writing system had three different com-
ponents. There were pictographs/logographs standing for whole words
indeed, but every one of them had a cluster of clues around it. One clue
was a ‘‘category determiner,’’ a symbol that stood for a common semantic
category (man, water, mountain, grain, and so on).
Other clues consisted of a set of phonetic symbols standing for single-,
double-, and triple-consonant sequences. Consonants frame the meaning
of a word (its root) in Hamito-Semitic languages like Egyptian. Vowels
signal changes in grammar and swap in and out, while the root stays in-
tact. Today, these ‘‘consonantal alphabets’’ are the main writing system
for modern Hebrew and Arabic. Sometime before 3000 b.c., the Egyp-
tians had figured this out. Logographs were there all right, but they did
not need to be.
Sumer is the only ancient civilization with an unbroken record of the
writing system dating back to the remote reaches of time. Symbol tokens
carved on clay and inserted into clay balls were used in commerce and
have been dated to as early as 10,000 b.c. (Schmandt-Besserat 1978).
There were symbols for a number system along with pictograms for the
products being conveyed.
 20

A true writing system did not appear until around 3200–3000 b.c.
Chapter 2

The most primitive form of this system used pictograms, abstract symbols
standing for words (logographs), and category determiners for nouns.
This changed rapidly due to the nature of the language. Sumerian was
an ‘‘agglutinative’’ language consisting mainly of one-syllable words with
these syllable types: CV, VC, CVC. In this type of language, one-syllable
words are combined (‘‘glued’’) into longer words and phrases. For exam-
ple, the plural was indicated by duplication: ‘‘I see man-man.’’ Because the
Sumerians used the same symbol for the sound of the word no matter what
role it played in the sentence, the writing system essentially functioned
like a syllabary—a sound-based system.
The shift from meaning to sound is not obvious and has misled
scholars down through the years. Proportional counts of pictograms and
logographs made from the vast number of clay tablets recovered from
archaeological digs show that the logograph count has declined over time.
In part, this is because scholars initially believed all early writing systems
were logographic, leading them to see more logograph symbols than were
actually there. In part, it was because the Sumerians shed logographs as
time went by. For example, Falkenstein (1964) reported that among the
total signs dating back to 3000 b.c. (pictograms, logographs, classifiers),
2,000 were logographs. In subsequent estimates, the logograph tally
was much smaller, and it shrank dramatically over the centuries: 1,200
in 3000 b.c., 800 in the period 2700–2350 b.c., and 500 in 2000 b.c.
(Coulmas 1989). Yet Michalowski (1996), a leading Sumerian scholar,
believes that the Sumerians may have had few, if any, logographs almost
from the beginning. He calculated that the entire corpus of symbols (all
types) was around 800 by 3200 b.c., stating that the system was virtually
complete at this time. A writing system with 800 signs is consistent with a
syllabary.
In 2500 b.c., the Semitic Akkadians adopted Sumerian for their ad-
ministrative language, along with the writing system (much as Europeans
borrowed Latin). Over time, it was adapted to the Akkadian language
and became a syllabary of about 300 signs (Cooper 1996). By the reign of
King Sargon I, in 2350 b.c., any use of logographs had virtually ceased
(Civil 1973).
We can trace a similar development in China. Early Chinese writing
was discovered late in the nineteenth century on a set of oracle bones
 21

dating to the Shang dynasty in roughly 1200 b.c. The characters, which

On the Nature of Writing Systems

numbered around 2,500, were so complex and numerous that Chinese
scholars assume the writing system must have developed much earlier. So
far, no further evidence has been found (Chang 1963; Ho 1976).
The Shang characters are identical in type to modern Chinese char-
acters, though dissimilar in form. Characters are compound symbols. The
main symbol represents a unit of sound (the syllable). A secondary symbol,
which is fused with it, represents a semantic classifier, called a ‘‘radical.’’
This is a clue to the category of the word: man, water, food, mountain,
earth, and so forth. These compound symbols constitute the main writing
system. Karlgren (1923) estimated that 90 percent of all modern Chinese
characters are ‘‘semantic-phonetic compounds.’’ Mair (1996) revised this
figure to 81 percent.
Chinese, like Sumerian, has a small number of syllable types: CV and
CVC. Phonemes combine in these syllable patterns to create 1,277 legal
Chinese syllables. Because there are so few syllables in the Chinese lan-
guage, Chinese (like Sumerian) is riddled with homophones, words that
sound alike but have different meanings. This is less of a problem in nat-
ural speech, because Chinese is a tonal language. Changes in vocal inflec-
tion (pitch) signal a change in meaning. But tones are not marked in
Chinese writing, and the category symbols (radicals) are essential clues to
meaning. They serve the function of unequivocality.
The number of symbols or characters in the modern Chinese writing
system provides a good estimate of the limits of human memory. Children
memorize about 1,260 syllable symbols, plus 214 category classifiers, for a
total of 1,474 basic signs. They must also learn which word (which of the
nine meanings of tang) is represented by which particular syllable-plus-
classifier combination, because this is not always obvious. Mair (1996)
estimated that 90 percent of words in common texts can be written with
1,000 compound characters, and 99 percent with 2,400. He concluded that
‘‘there is a natural upper limit to the number of unique forms that can be
tolerated in a functioning script. For most individuals, this amount seems
to lie in the range of approximately 2,000–2,500’’ (p. 200).
The Japanese writing system is the only writing system that still
retains a large number of logographs (kanji). These symbols were origi-
nally borrowed from China and used for Chinese loanwords. Today, the
Japanese Ministry of Education has set the school requirement for kanji at
 22

about 1,860. Children learn half the kanji symbols in elementary school
Chapter 2

and half in high school, a process taking about 12 years. Mastery of an

additional 2,000–3,000 kanji is considered the mark of a highly educated
person, and memorizing them can take up to a decade or more of adult
life.
Needless to say, 1,860 word signs that take 12 years to learn is
scarcely sufficient to represent the 50,000 or so words required by the
average Japanese man or woman to read a newspaper or magazine. This is
why the lion’s share of the Japanese writing system consists of two (re-
dundant) phonetic scripts (katakana and hiragana) with around 50 symbols
each. These represent the consonant-vowel (CV) diphone units so typical
of the Japanese language (ka-ta-ka-na). The Japanese writing system con-
tains 1,860 þ 50 þ 50 symbols, for a total of just under 2,000, plus the
addition of Roman letters to represent foreign words. Here is a writing
system that fails two out of three of Coulmas’s criteria: economy and
simplicity. By this measure, it does not qualify as a ‘‘successful writing
system,’’ yet it works nonetheless.
The numbers from these tallies tell a story. There is a limit on human
long-term memory for memorizing abstract visual patterns paired with
individual words. This is an ultimate limit, present after years of intense
memorization. Based on an earlier analysis of known writing systems, I
put this limit at approximately 2,000, remarkably close to Mair’s estimate
based on Chinese (McGuinness 1997c).
Logographs do not a writing system make, that much is clear. But
logographs do not even fare well as components of a writing system. It is
obvious that no civilization ever thought they did, or developed anything
remotely approximating a whole-word system. We have a 5,000-year his-
tory to show that writing systems are essentially sound-based systems and
that no whole-word (sight-word) reading method can work. The notion
that children can learn to read whole words purely by sight, or that adults
ultimately become full-fledged sight-word readers, is patently false.
Without exception, in every writing system ever translated, the
memory-overload problem was solved by avoiding the word as the unit of
the system, and switching to phonological units instead. At this point we
reach level 2 of Coulmas’s list, and confront a serious dilemma.
 23

The Sound-Based Level: Syllables, Syllable Parts, Phonemes

On the Nature of Writing Systems

At this level, a major problem was created by a premature classification
system. Put simply, linguists allow words, syllables, and phonemes, but
creators of writing systems did not see it that way. They did not stick to
syllables or phonemes in choosing a sound-based unit. In a sense, linguists
have put a straightjacket on comparative analysis, leaving no wiggle room.
The lack of terminology for writing systems that are neither syl-
labaries nor alphabets creates unnecessary disputes among analysts. The
Egyptian consonant script is a case in point. Gelb claimed it had to be a
syllabary, because the consonant symbols automatically imply vowels, and
also because no writing system can skip a step in Gelb’s evolutionary
sequence. Other paleographers argued that because the symbols do not
mark which vowel to use, or whether it would precede or follow the con-
sonant (VC or CV ), it cannot be a syllabary but must be some kind of al-
phabet ( Jensen 1969). In fact it is neither. It is neither because the reader
does the work filling in vowel sounds that are implied but not marked.
This is because a goodly chunk of the ‘‘writing system’’ is inferred. It has
to be carried in one’s head. Furthermore, a writing system like this can
only work for languages with a particular structure, where consonants
carry the meaning load and stay in a fixed sequence while vowels swap
in and out to signify changes in grammar. These are characteristics of
Hamito-Semitic languages and no other languages. The mere existence of
a consonant-only writing system refutes the notion that there is a univer-
sal model that would apply to the development of all writing systems.
The facts, if you look at them impartially, tell us that the choice of a
sound-based unit has nothing to do with an evolutionary progression. Nor
is the choice an accident, or just ‘‘relatively’’ based on syllable structure, or
borrowed haphazardly via diffusion. The choice is governed by the struc-
ture of the language in terms of both syllable number and phonotactics, the
legal phonetic sequences in a given language. For example, the adjacent
consonant sequence /str/ is legal at the beginning of English words (street),
but not at the end. This is not an academic issue, but a living, breathing,
language issue. Nine-month-old infants are already hard at work analyz-
ing the phonotactic structure of their language (Aslin, Saffran, and New-
port 1998; Friederici and Wessels 1993; Jusczyk 1998; Mattys et al. 1999;
also see Language Development and Learning to Read ).
 24

The idea that the phonotactic structure of a language plays a role in

Chapter 2

the choice of a phonetic unit for a writing system has been suggested by a
number of authors (Mattingly 1985; Coulmas 1989; Katz and Frost 1992;
McGuinness 1997c), but it has not been explored in much depth. I have
proposed (McGuinness 1997c) that there are four types of sound-based
writing systems, not two. These four basic types derive from the syllable
structure and phonotactics of individual languages. The choice of a sound
unit is based on the principle of least effort—a trade-off between two fac-
tors: economy (the ease with which the sound units can be learned), and
perception (the ease of discrimination, or ‘‘naturalness,’’ of the units to be
adopted). This is assuming the other constraints apply as well.
This represents a departure from conventional wisdom on this issue.
Nevertheless, it is worth exploring because it provides answers to some
quite fundamental puzzles, the first being: If Gelb was so certain that a
syllabary is an inevitable step en route to alphabets, where are the syl-
labaries he refers to? So far, we have met only three (Sumerian, Akkadian,
and Chinese). There are the Babylonian and Assyrian syllabaries (lan-
guages that were dialects of Akkadian), the Hittite syllabary, plus the clas-
sical script for Annamese (Vietnamese) borrowed from China. Seven or so
syllabaries in 5,000 years is not an impressive number. Only one, Annam-
ese, was shed for an alphabet, and this was not made official until 1912. A
more compelling argument against Gelb’s linear evolutionary model is
that it cannot explain the many examples where an alphabet was rejected
in favor of something more syllable-like. If Gelb was correct about writing
systems evolving into higher forms, then evolution could go both forward
and backward in time and so would not count as ‘‘evolution.’’
To understand the implications of this new classification scheme, we
need to look at the structure of the four types of writing systems and how
they were designed. In particular, we are going to look at the critical role
that awareness of the phonetic/phonemic level of language played in how
the writing system was set up. This illustrates the fact that no scholars
could have designed a writing system unless they knew the phoneme cor-
pus of their language, and raises the following question: If they knew this,
why didn’t they always opt for an alphabetic writing system?

Syllabary Writing Systems The syllable is the largest phonological unit

below the level of the word, in the sense of containing the most pho-
 25

netic information. The English words straight and I are both one-syllable

On the Nature of Writing Systems

words. Syllabary writing systems only work for languages that have an
extremely limited number of syllable types. Two major civilizations,
Sumer and China, fit this description, and both have a syllabary writing
system.
But did the adoption of a syllabary writing system mean that early
scholars opted for this solution because they were unaware of phonemes,
and could never have discovered that an alphabet was the best solution?
No notes remain from the hundreds of hours it must have taken scholars
to create a syllabary writing system. One can speculate. There are two
solutions. The first would be to work it out orally in a piecemeal fashion.
As each syllable comes to mind—tan, chang, ho, tang, wan, min—a symbol
would be designed for it. This is rather chaotic and dangerous, because
people would forget which symbol went with which syllable as the days
went by.
Another solution is to work out the individual sounds within syllables
(the phonemes) and design symbols for them. There are far fewer pho-
nemes than syllables in any language (Chinese has about 35, and Sumerian
21). If you did this, you could set up a matrix and work it through sys-
tematically. This way, you would be far less likely to leave something out,
and it would be much easier to keep track of what you were doing.
Here is a partial phoneme solution for creating a syllabary in a lan-
guage with the vowels /a/ and /o/ and the consonants /b/ and /t/. The
problem we are addressing is how many syllable symbols we need to rep-
resent every legal syllable (CV, CVC) in a language similar to Chinese.
Beginning with symbols for the 4 phonemes, we would end up with 12
possible legal syllables, as the accompanying table shows. The next step is
to cross off the syllables that are illegal in the language and design a spe-
cial symbol for each syllable that remains. These can be recorded on a
separate chart.

C V C V C
a o a o
b ba bo b bab bob b
t ta to b bat bot t
t tat tot t
t tab tob b
 26

This example is intended to make the point that when your logo-
Chapter 2

graphic solution crashes into the ceiling of human memory, you have to
find a better way, and it must be complete. A partial writing system is no
use. The question is, would it be more efficient to begin by assigning
symbols to 35 phonemes as a memory device, then set up a matrix and
work it through, or to do it randomly, designing each syllable symbol as
you went along? We have no clues from Chinese documents as to what
was done, but we do have clues from Sumer in the thousands of clay
tablets found in temples, palaces, and schools dating back to the fourth
millenium b.c. (Kramer 1963, [1956] 1981; Michalowski 1996).
Michalowski (1996) reported that digs in ancient Uruk revealed an
abrupt transition, which has been dated to between 3200 and 3000 b.c.
The early phase of protowriting had no particular order or structure. At
the point where writing proper begins, suddenly order appears. Having
observed this material, he wrote: ‘‘The structure and logic of the system
indicate that it was invented as a whole and did not develop gradually’’
(p. 35). Michalowski emphasized the fact that from the beginning there
was a concern for the structured transmission of the system.
Kramer reached a similar conclusion in his analysis of Sumerian
tablets written for and by schoolchildren (about 2000 b.c.). The children
were introduced to the syllable signs in a systematic way: all the CV syl-
lables, all the VC syllables, then the CVC syllables, each memorized
independently. Next, they learned lists of words in semantic categories.
Written symbols were not learned in a random order incidentally as part
of reading meaningful text, and it is highly unlikely they were designed
that way.
These examples show that it is just as likely the Chinese and the
Sumerians were aware of the phonemic structure of their language, as it
is they were not. And if they were aware of it, they could have easily
observed that an alphabet was extremely economical. Nevertheless, they
chose not to use it.

CV Diphone Systems When a language has more than three or four sylla-
ble types, a syllabary will not work for the simple reason that it would
breach the limits of human memory. According to the received wisdom
on this issue, the only other option is an alphabet. But this is not remotely
what happened. The most common sound-based unit adopted for all
 27

writing systems, past or present, is the consonant-vowel unit. CV writing

On the Nature of Writing Systems

systems work for the hundreds of languages mainly built on a repetitive
phonological structure like this: CVCVCV. These phonological units ap-
pear universally in the utterances of infants: ba-ba-ba, goo-goo, ba-goo-da.
An English word with this pattern is potato. There are no languages where
VC units commonly form words: VC-VC-VC (edit and idiot are about as
close as it gets).
These writing systems were identified by McGuinness (1997c) and
named diphone systems (two sounds). Up to this time, diphone writing sys-
tems had been classified as alphabets or syllabaries. Coulmas (1989, 225)
referred to one of these systems as ‘‘a Devanagari-type syllabic-alphabet,’’
a good illustration of the missing-terminology problem.
The most important word in the first paragraph of this section, is the
word mainly. All languages written in a diphone writing system have more
syllable types than CV alone. If they did not, they would be one-syllable
languages, and the writing systems would indeed be syllabaries. But these
languages have words that end in a consonant (VC, CVC) and words with
consonant sequences (CCV, CCVC, CVCC). This is especially true of
Indic languages, all of which are written with a diphone system. One of
the features of Indic languages (and all other Indo-European languages) is
the variety of consonant clusters or blends.
This means that diphone writing systems are inherently ambiguous.
There is a problem of how to mute an extra vowel sound when the word
ends in a consonant. Balam is a Mayan word, but in their diphone script it
would read balama. The consonant clusters in krishna, an Indic word,
would read karishana. The solution was to add marks or diacritics indicat-
ing when to drop a vowel, and these can become part of the script itself.
Indic writing systems use ‘‘ligatures’’ (loops that act as connectors) to sig-
nal the reader to drop a vowel between two consonants. This is why a
diphone system is different from a syllabary. The fact that a diphone sys-
tem does not mark all the syllable patterns in the language sets it apart.
It would not be practical to list every writing system that used or uses
the CV diphone as a sound-based unit for the code, but even a partial list
is impressive. The Mayans appear to be unique in designing a compound
system, with pictograms and logograms plus a CV diphone script and
symbols for vowels. The remaining diphone systems are sound-based
only. The earliest so far is Linear B (about 1450 b.c.), found on Crete and
 28

later used in commerce by the Greeks. Next come the writing systems
Chapter 2

based on the Indic Brahmi script, developed sometime after the fifth cen-
tury b.c. This script led to an amazing number of offspring.1
It is assumed that the Indic Pali script, or one like it, led to the
creation of the two Japanese diphone systems, hiragana and katakana, via
Buddhist missionaries. Other diphone systems include Ethiopian (a
Semitic language), circa fourth century a.d., and the Han’gul system
developed during the Korean writing reform in the fifteenth century.
There is also the writing system of the Cherokee Indians, and a most
interesting diphone hybrid from Persia, discussed in more detail below.
The important message is that alphabets may dominate in Western
cultures, but they do so for linguistic (phonotactic) reasons, not because
they are inherently superior to all other forms.
Evidence on how these diphone systems were set up and designed
provides another surprise. At the time the Brahmi script was created, In-
dian scholars had already designed symbols to represent each phoneme in
their language. These ‘‘alphabet symbols’’ were set up in a fixed order and
grouped by place of articulation, showing their sophisticated knowledge of
phonetics. The same ‘‘alphabetical order’’ was used for dictionaries. There
is speculation that this ‘‘alphabet chart’’ was used for novices to learn to
chant mantras with precise pronunciation. But it is highly likely the al-
phabet chart was used to design the Brahmi diphone script, as shown by
how it was constructed. Each consonant plus the vowel /ah/ (the primary
vowel in Indic languages) was assigned a different symbol. Next, these
symbols were systematically modified for each vowel change, as shown in
figure 2.1.

1. The Brahmi offspring include the writing systems for Kusan, Gupta, the
Devanagari script used for Sanskrit, Hindi, and Nepali, plus the scripts for
Siamese, Burmese, Kavi or Sinhalese, Bengali, Assamese, Tibetan, Mongolian,
Kashmiri, Balinese, Madurese, Tamil, Central Indian, Punjabi, and Malaya-
lam, as well as the important Pali scripts designed for the Prakrit languages
associated with Buddhism. These scripts traveled east as Buddhism spread,
giving rise to the diphone systems of Sri Lanka, Burma, Thailand, Cambodia,
and Indonesia.
 29

On the Nature of Writing Systems

Figure 2.1
Brahmi script 5th century B.C. Examples of diphone symbols for 2 consonants and 9 vowels

Judging from the evidence on how these systems were set up, and the
large number of people who adopted and designed them (about 200 of
these scripts have been attested in India alone), it would have been obvious
to ancient scholars that an alphabet could work as a writing system. Yet an
alphabet was never adopted for this purpose.
The Han’gul writing system from Korea is an even clearer example,
because its design is more transparent. First, each vowel and consonant
was assigned a symbol and these were set up in a matrix, 10 vowels across
the top and 14 consonants down the side. At each junction of consonant
and vowel, they needed a symbol for all the CV pairs in the Korean lan-
guage. Instead of designing 140 new symbols like everyone had done
before them, they had a better idea. Rather than wasting the alphabet
symbols, these symbols were fused into pairs, as shown in figure 2.2.
There is one more example from Old Persia, during the reign of
Darius I (522–486 b.c.), the most curious so far. It is written in cuneiform
symbols, no doubt borrowed from the Babylonian script. This is a hybrid
system of 36 symbols. There are 13 consonant-only signs that are used
with 3 vowel signs (an alphabet). In addition, there are 20 different con-
sonant signs that include an inherent vowel and vary according to which
vowel is indicated (CV diphones). Another surprise is that the ‘‘alpha-
betical’’ order of these signs is identical to the one used in India, with
consonants grouped by place of articulation. Our alphabetical order
derives from the West Semitic (Phoenician) system and is quite differ-
ent. Whether this alphabetical order is original to Persia or to India is
unknown, because both writing systems appeared around the same time.
 30
Chapter 2

Figure 2.2
Diphone symbols for combinations of vowels and consonants in Han’gul

Old Persian was an Indo-European language. However, the adminis-

trative language of the Persian empire was Semitic Aramaic. The Aramaic
writing system originated around the ninth century b.c. and marked only
consonants, using simple letter-like characters similar to the Phoenician
script, the ancestor of our alphabet. It seems the Persian script was influ-
enced by two factors: the form of the script (cuneiform) plus the CV
diphone system by the Babylonians, and the concept of consonant-only
symbols featured in the Aramaic writing system. Whatever turns out to be
correct, it is clear that the Persians designed at least a partial alphabet but
 31

failed to use it for a writing system. We now have three clear examples of

On the Nature of Writing Systems

‘‘evolution’’ going backward.

Consonant Cuing Systems Writing systems based on consonant sequences

(no vowels) are unique to Semitic languages. There are scores of conso-
nant cuing systems in the Near East and Middle East, both past and
present, the prototype being Egyptian hieroglyphics (3000 b.c.). In one
sense these systems cannot be classified, because they are so dependent on
the reader knowing all possible words and conventions of the spoken
language before the text can be decoded. One could think of them as
shorthand versions of a diphone system, except this would be inaccurate,
because the script provides no indication of which vowels must be added
to the reading. If the reader saw p t t (CCC), and a diphone rule was
operating, the reader would think: CVCVCV, ‘‘this might be potato.’’ But
the script does not indicate where the vowels go. The sequence could be
read any number of ways: VCVCVC (oputot) or CVVCVCV ( piatuti).
This is not really a writing system. It is more like a sketch or a se-
quence of clues. For example, it would be impossible to read a word that
was not in your vocabulary. Consonant cuing systems are about as eco-
nomical as it gets, so much so that they fail both the unequivocality and
the comprehensiveness tests. Modern consonant systems, like Hebrew
and Arabic, use diacritics to indicate vowels when texts are difficult. This
applies to books for beginning readers, to sacred or classical texts, and to
highly technical books. With the addition of specific vowel markers, these
systems turn into alphabets.

Alphabets When a language has too many syllable types (English has 16,
not counting plurals), when it lacks a basic diphone structure CVCVCV
or any other simple structure, or when it does not have the ‘‘consonant-
root’’ nature of Semitic languages, it must be written some other way. At
this point we are nearly out of sound-unit options, and the only other way
is down—down one more level below the CV unit, to the individual con-
sonants and vowels (phonemes).
All writing systems based on the phoneme are called alpha-bets after
the first two letter names (alpu, beth) in the Old Phoenician consonantal
script. The Greeks borrowed these symbols in the eighth century b.c. to
 32

set up the first alphabetic writing system. They used the 22 Phoenician
Chapter 2

consonant symbols to represent similar sounds in Greek. Any leftover

letters were assigned to vowels, and new letters were created for the re-
maining sounds. When they were finished, there was one letter symbol for
each phoneme in the Greek language, a perfect or transparent alphabetic
writing system. This was certainly not the first alphabet ever designed, but
it was the first one used as a writing system. Everyone who speaks a Euro-
pean language today uses an alphabet derived from the early Greek al-
phabet through the process of diffusion.
A transparent alphabet, like the Greek alphabet, is very efficient and
has no ambiguity. Alphabets have a small set of easily remembered sym-
bols. No props, rules, or clues, like determiners or classifiers, are required
(syllabaries), no vowel sounds have to be dropped (CV diphone systems),
and no vowels have to be supplied by the reader (consonant cuing sys-
tems), a major advantage. Why then aren’t there more alphabets in the
world? Why are CV diphone systems the most common type of writing
system?
Now we are at the heart of the matter. The history of the develop-
ment of writing systems shows that ancient scholars must have been aware
of alphabets. It is extremely unlikely that any writing system could be
designed without one. Examples of alphabet charts, of patterns in the
symbols of the scripts, show that scholars were well aware of phonemes.
Yet they avoided using an alphabet even when one was staring them in
the face. It is hard to escape the conclusion that the phoneme was an un-
popular unit for a writing system.
Diphone systems, on the other hand, have a number of useful fea-
tures. They are well within the memory limit. (A diphone system for
English could scrape by with around 315 symbols, although it would never
work.) They are designed for a highly salient unit of sound, easy to hear in
the speech stream. They fit natural-language patterns in ways that indi-
vidual phonemes absolutely do not. There is more phonetic information
per symbol than in an alphabet, so like syllabaries, they are faster to read
and write (nearly twice as fast).
By contrast, phonemes overlap each other in the speech stream and
slip by so quickly that people are not aware of them. If children are not
made aware of phonemes when they learn to read, they will misunder-
stand how an alphabet works. Because people find it unnatural to analyze
 33

and isolate phoneme sequences in words, this is undoubtedly the reason

On the Nature of Writing Systems

alphabetic writing systems are not found everywhere. It seems that when
an alphabet could be avoided, it was.

Implications for Teaching and Research

The major types of writing systems have been described in some detail to
emphasize the fact that writing systems are inventions. They require time
to design and develop. They have a complex structure. They must be
designed by scholars with sufficient access to power and influence to con-
vince everyone to learn the system. The efforts that went into the design
of early writing systems were monumental. That it was possible to repre-
sent the fleeting sounds of human speech in symbols that mark every word
in a language is nothing short of miraculous. These facts have important
implications for research and for reading instruction.
The assumptions about writing systems held by many educators and
researchers were listed at the end of the last chapter. Here is a recap on
why these assumptions are false.
First, there is no support whatsoever for a fixed evolutionary sequence
in the development of writing systems, nor any evidence that ancient
peoples ‘‘gradually evolved’’ into being able to hear phonemes. What
appears evolutionary (the transition from whole-word to sound-based
systems) is due to historical/logistical factors in trying to use graphic
symbols designed for economic purposes in a full-fledged writing system
for broad cultural ends. During this transition, logographs quickly hit a
ceiling, then decreased as the sound-based system took precedence. At
least this was the pattern in civilizations where writing systems were
created from scratch.
The logograph holdout, in the case of Japan, can also be explained in
terms of historical factors. Early on in the development of their writing
system, the Japanese borrowed Chinese characters and used them as
logographs (kanji) for Chinese loanwords. These logographs were sub-
sequently used for Japanese words with the same or related meanings. A
kanji symbol today has many ‘‘readings,’’ another reason it takes so long to
learn them.
Second, a writing system is an invention and not part of our biological
heritage like spoken language. People must be trained to use inventions.
A reading method must ensure that the nature and logic of the writing
 34

system are transparent to the learner, and that the elements of the system
Chapter 2

are mastered. Methods that ask the child to ‘‘guess’’ how the writing sys-
tem works (whole-language) are utterly irresponsible.
Third, no writing system was ever based on the whole word, nor were
whole-word symbols (logographs) ever more than a minute fraction of any
writing system. Languages have too many words. Human memory for
abstract symbols overloads at about 2,000 symbols, and even achieving this
takes many years. A reading method either totally or partially based on
whole-word memorization of sight words will cause the majority of chil-
dren to fail. These facts also refute the notion proposed by some cognitive
psychologists, that people ultimately read all words by sight. It cannot
be done. (For a description of these theories and what they imply, see
chapter 11.)
Fourth, all writing systems are based on one of four specific mean-
ingless phonological units, specific because phonological units are not mixed.
Teaching methods that train children to be aware of the particular sound
unit for the writing system, and only that unit, and that show them how
these sounds are represented by the symbols, will be effective. Teaching a
potpourri of other sound units (words, syllables, syllable segments, word
families) that have nothing to do with the writing system will lead to con-
fusion and failure for many children.
Fifth, writing systems are designed to fit the phonological structure
(phonotactics) of the languages for which they were written. The choice
of the sound-unit basis of the writing system is not arbitrary. Civilizations
adopted an alphabet solution out of necessity and not from choice, be-
cause no other solution would work. The Greeks, for example, already
had a diphone writing system (Linear B), which they used for commerce.
But it was useless for representing the Greek language, with its complex
syllable structure. All Linear B texts used by the Greeks consist of bills of
lading, inventory lists, and invoices. And it is quite obvious, viewing these
texts, that the CV diphone units of Linear B map extremely badly to the
Greek language (see Robinson 1995).
Finally, history shows that alphabets tend to be avoided if possible.
Scholars from different cultures, separated widely in time and place,
designed alphabets for other reasons, but failed to use them for a writing
system. The record shows that a larger, clearly audible, phonological unit
was the preferred solution, especially if the number of these units was well
 35

below the magic 2,000 limit. People are not normally aware of phonemes,

On the Nature of Writing Systems

and it is unnatural to isolate phonemes from each other. This has impor-
tant implications for instruction. Children who learn an alphabetic writing
system need to be taught to listen at the phoneme level of speech, and
need to learn how to link phonemes to letter symbols. They have to be
trained to listen at the right phonetic level and to look at the right sym-
bolic units (letter(s) in a letter string). Any reading method that teaches
the wrong unit of sound or multiple units of sound (words, syllables, syl-
lable fragments) risks misleading children, causing them to fail.
 3
THE STRUCTURE OF THE ENGLISH
ALPHABET CODE

A writing system is an encoding device for representing units of sounds

in the language by a set of symbols. What we call ‘‘spelling’’ (encoding) is
the fundamental or basic operation, the process of turning sounds into
symbols. What we call ‘‘reading’’ involves decoding those symbols back
into sounds to recover the words. Reading and spelling are reversible
processes, and should be taught in tandem so that this reversibility is ob-
vious. Unfortunately, in English-speaking countries it is common practice
to teach reading and spelling as if they had nothing to do with each other,
using different words, different forms of instruction, taught on different
days. This practice totally obscures the code nature of our writing system,
and makes learning to read and spell far more laborious and confusing
than it needs to be.
Structural reversibility, however, does not mean psychological revers-
ibility. Reading and spelling are structurally reversible, but they draw on
different memory skills. Decoding, or reading, involves recognition memory,
memory with a prompt. The letters remain visible while they are being
decoded. Encoding, or spelling, involves recall memory, memory without
prompts or clues, which is considerably more difficult. To spell a word,
you must first identify each phoneme in sequence in your mind, remem-
ber how each phoneme in that particular word is spelled, and then write it
down. What you can spell, you can easily read (recognize). But what you
can read is not necessarily easy to spell (recall). The fact that spelling is
the primary process and has 100 percent spin-off for reading has been
known for centuries, one reason Noah Webster wrote a ‘‘speller’’ in 1793
and not a ‘‘reader.’’ Maria Montessori strongly advised teachers to teach
children to write (spell) first, and then allow them to discover that they
can read what they have written.
 38

The Sumerians understood this 5,000 years ago. The thousands of

Chapter 3

clay tablets containing stories and exercises discovered in the ruins of

ancient Sumer are a testament to a number of important facts about writ-
ing systems and how to teach them. Based on the historical record, there
is a high probability that the people who designed the Sumerian writing
system played a significant role in setting up the schools to teach it. No
other culture in history has provided such a clear picture of how the
designers of the writing system thought the system should be taught.
It is evident from this material that children were taught each syl-
lable in their language and the symbol that represented it. They learned
this through a series of exercises beginning with the basic syllable forms
(CV, VC, CVC), moving systematically to more complex multiple-syllable
words (CV-CVC-VC-CV ). Encoding (spelling) and decoding (reading)
were connected and learned as a reversible process at every step. The stu-
dents copied syllables, words, and sentences, or wrote them from dicta-
tion. They then read aloud what they had written to their teacher.
Five thousand years ago, the Sumerians bequeathed to us the basic
guidelines for how to teach reading and spelling for any writing system:

1. Make sure the complete structure of the writing system has been
worked out (or thoroughly understood) before a method of instruction is
developed.
2. Teach the specific sound units that are the basis for the code. (Do not
teach other sound units that have nothing to do with the code.)
3. Teach the arbitrary, abstract symbols that represent these sounds.
These symbols constitute the code.
4. Teach the elements of the system in order from simple to complex.
5. Ensure that the student learns that a writing system is a code and that
codes are reversible.
6. Make sure that encoding (spelling) and decoding (reading) are con-
nected at every level of instruction via looking (visual memory), listening
(auditory memory), and writing (kinesthetic memory).

It is hard to imagine a better list of guidelines. Unfortunately, education

practice has strayed so far from its roots that hardly any educators or
classroom teachers adhere to even one of these principles, let alone all six.
 39

We cannot begin to implement reform unless educators (and researchers)

The Structure of the English Alphabet Code

understand how writing systems work in general, and how our writing
system works in particular.
Children must be made aware that specific sound units (phonemes) in
the language are the basis for a written code, and that symbols represent-
ing these sounds constitute the code. It should be made clear that sounds
are ‘‘real,’’ and letters are ‘‘unreal,’’ arbitrary symbols for those sounds.
Letter do not ‘‘have sounds,’’ or ‘‘say’’ anything, as children hear so often
in phonics lessons in the English-speaking world. And they certainly do
not ‘‘say’’ letter names.
The important message is this: a writing system is a code with a
systematic relationship between units of sound and visual symbols. All
codes are reversible. That is the nature of codes. If a code cannot reverse
(through encoding/decoding), it will not function as a code.
There are good codes and bad codes, codes that are obviously revers-
ible (like those in Germany, Italy, Finland, Sweden, and Norway) and
codes that cannot reverse unless they are taught correctly. Purpose-built
codes, like the Greek and the original Anglo-Saxon writing systems, were
designed by scholars familiar with the nature of codes and how writing
systems work. These purpose-built systems were designed in a short
period of time and tend to be ‘‘good codes.’’ A good code is efficient
(entailing a reasonable memory load), comprehensive, and transparent—
only one symbol is used for each sound in the language (whichever unit
of sound is chosen). Transparent codes are transparently reversible. The
student can easily see that the code works in both directions: encoding and
decoding. The sound /d/ is always written d, and the letter d is always read
as /d/. For this reason, transparent codes are much easier than opaque
codes to teach and learn.1
Venezky (1973) discovered just how easy it is to learn a transparent
alphabet in a study on 240 Finnish children in grades 1 through 3. Chil-
dren begin reading instruction in Finland at age 7 (grade 1). A nonsense-
word reading test was designed that included words spelled with every

1. As a reminder, phonemes are enclosed in slash marks and spellings are

underlined.
 40

possible deviation from a one-to-one correspondence in Finnish orthog-

Chapter 3

raphy. At the end of first grade, the children scored 80 percent correct on
this test, a value that remained unchanged through third grade. College
students scored 90 percent correct, failing the words with the most ob-
scure Finnish spellings. In other words, it takes a year or less for Finnish
children to be able to read and spell nearly as well as the average college
student.
As noted in chapter 1, Wimmer (1993) found that the worst readers
in the city of Salzburg, Austria, scored close to 100 percent on a difficult
test of reading accuracy and did nearly as well in spelling. Wimmer and
Landerl (1997) compared Austrian and English children on a spelling test
of English and German words balanced for complexity of spelling pat-
terns. The English children made twice the spelling errors (of all types)
as the Austrian children. More importantly, 90 percent of the errors made
by the Austrian children were legal (phonologically accurate), compared
to only 32 percent for the English children.
Geva and Siegel (2000) reported on 245 Canadian children who were
learning to read and write English and Hebrew at the same time. English
was the first language in most cases. Hebrew is written with a consonant
cuing system where symbols represent consonants, and diacritic marks
are used for vowels when texts are difficult, as for beginning readers. The
consonant symbols and diacritics are nearly 100 percent consistent,
making Hebrew a highly transparent writing system. By the end of first
grade, the children scored 79 percent correct on a Hebrew reading test,
but only 44 percent correct on the English version of the test. Children
did not reach the 80 percent competency level in English until fifth grade,
and by this time they scored 90 percent correct in Hebrew.
Writing systems that evolve over long periods of time tend to be
opaque, lacking consistency between sounds and symbols. This happens
for two reasons. The first has to do with the arbitrary nature of the pro-
cess; the designers are pioneers and there is no prior model of a code or a
writing system. The process proceeds by trial and error, and as insight is
gradually gained, no writing reform occurs to correct early mistakes.
The second reason is historical accident. A country (England) with a
transparent writing system (Anglo-Saxon or Old English) is conquered by
a people who speak a different language and have a different way of writ-
ing (spelling) the same or similar sounds (Norman French). The language
 41

and the spelling system are superimposed on the existing system. Or a

The Structure of the English Alphabet Code

country (England) with a transparent writing system (Anglo-Saxon) adopts
Latin for religious, legal, and academic purposes. The Latin spelling sys-
tem uses different symbols for the same or similar sounds. Over time these
words become part of the vernacular, and Latin words along with their
spellings are pegged onto the existing system. And no writing reform ever
takes place to correct these problems. In addition, spoken language slowly
changes over time. Vowels shift in pronunciation. Consonant sequences
may become illegal (drop out of the language). In the Anglo-Saxon word
for ‘‘know’’ (spelled cnāwan), the /k/, /n/, and /w/ were pronounced
‘‘k-na-wan.’’
By ‘‘reform,’’ I do not mean spelling reform, tinkering with the prob-
lem by standardizing spelling, but real reform in which the one-to-one
correspondence between sound and symbol is restored.
Because the English language and the English writing system were
assaulted by several foreign-language invasions, it is not an exaggeration
to say that the English writing system is one of the most opaque writing
systems in existence. It represents five languages and their spelling systems
superimposed on one another: Anglo-Saxon, Danish, Norman French (a
patois of Danish, Romanz or Old French, and Latin), Classical Latin, and
Greek. The bonus has been one of the most expressive languages in the
world, but the downside is one of the highest functional illiteracy rates
among literate nations. There is no question that the high functional illit-
eracy rate in English-speaking countries is largely a product of our formi-
dable spelling code and the way it is (or is not) taught.
The central argument in this chapter is that a complex, opaque spell-
ing system like the English spelling system needs to be taught with great
care. It cannot function as a code (cannot reverse) if it is taught with the
wrong logic and in the wrong direction: from letter to sound only. Here’s
why. There are 40þ phonemes in the English language, and about 176
common ways to spell them (see below). This number would explode
if every rare spelling was included (Venezky 1970, 1999). If the code is
taught in the direction in which it was written, 40 sounds to multiple
spelling alternatives and back again, the code nature of the writing system
remains intact. The 40 sounds are the only constant in the system.
If the code is taught the wrong way around, from 176 spellings to
200þ ‘‘sounds,’’ there is no way to get to the 40 English phonemes. This
 42

destroys the logic of the writing system. Confusion can start to creep in
Chapter 3

quite early and unravel students long before they get anywhere close to
176 alternative spellings (these spellings would never be taught anyway,
because teachers do not know what they are).
Here is a typical example of a common error. Mrs. Jones is keen to
include phonics principles in her teaching. She works entirely from visual
logic: letter to sound. She teaches the letters k, c, and the digraph ck, but
not at the same time. She says something like: ‘‘The letter see says /k/.’’
(Mrs. Jones is careful not to say ‘‘kuh,’’ which is good.) She writes the
words cat, cup, car, and cow on the board. A week or so later, she says:
‘‘This letter is kay, and it says /k/.’’ She writes the words keg, keep, and kill
on the board. Several weeks later, she says: ‘‘The two letters see-kay say
/k/.’’ She writes the words back, duck, and sick on the board.
If children remember the first lesson (which many will not), they
may think they probably got something mixed up. What they heard a
week ago was not really /k/, but something else they cannot remember.
Other children will assume that the sound /k/ that the letter c says is a
different /k/ from the one the letter k says, and a different /k/ from the
one the letters ck say. Young children make this kind of mistake a lot.
They believe, or actually hear, the same phoneme as sounding differently
depending on where it comes in a word. To some children the /b/ in bat
sounds different from the /b/ in cab. Acoustically, they are different. The
/b/ in bat is considerably more bombastic than the /b/ in cab. These subtle
distinctions are known as allophones (variants on the same sound). But
our writing system is based on phonemes, not allophones. If children are
taught the 40 phonemes in the first place, this kind of confusion could
never occur.
Here is what Mrs. Jones should have said: ‘‘Today we’re going to
learn the sound /k/. There’s more than one way to spell this sound. I am
going to teach you some patterns to help you remember when to use each
spelling.’’
Now the message is clear. The children will not think that they are
losing their minds, or that they are too stupid to learn.
Given the fact that the English spelling code is highly complex, one
would imagine that considerable effort has been expended to work out its
structure by linguists, curriculum designers, and researchers studying how
children learn to read and spell. This has not happened. The nature and
 43

structure of the English spelling code are virtually unknown. As a conse-

The Structure of the English Alphabet Code

quence, research on how children master a writing system is carried out
without any knowledge of its structure, or even of the fact that it is a code.
This is like studying what children do when they are put in a room for
several hours a day over a period of years, with a piano, some musical
scores, a few CDs, and no lessons on how to read music. For the most
part, reading research has largely demonstrated two things: the enormous
ingenuity and versatility of the young mind in the absence of suitable
instruction, plus the miraculous fact that some children actually learn to
read and spell.

A Tower of Babble
Confusion in the field begins with basic terminology like orthography,
spelling rules, and regular spelling. The word orthography appears in almost
every research report on spelling and in much of the literature on reading
as well. Orthography means ‘‘standardized spelling,’’ ortho meaning uni-
form or standard, and graphy, written signs or symbols. We have a stan-
dardized spelling system thanks to Samuel Johnson, but knowing correct
spelling tells us nothing about the structure of the spelling code. Yet
researchers frequently confuse the word orthography with this structure
and with ‘‘spelling rules.’’ Many believe that ‘‘orthographic rules’’ govern
why and how words are spelled in particular ways, despite the fact that no
one is quite certain what these rules are. Yet it is precisely these ‘‘rules’’
that children are supposed to grasp intuitively and internalize as they work
their way through pages of print and spelling tests over the years. Here is
a summary of the terminology problem. When the basic terminology of
a discipline is misused, or never properly defined, the researchers do not
understand what they are studying.

Rules
Clymer (1983) wrote an insightful and entertaining paper on spelling
rules, or as he put it, on ‘‘spelling generalizations.’’ In his words, ‘‘We
were careful not to call the generalizations ‘rules,’ for all our statements
had a number of exceptions. As the class finally formulated a generaliza-
tion regarding the relationships of letters, letter position, and sounds, such
defensive phrasing as ‘most of the time,’ ‘usually,’ and ‘often’ appeared as
protective measures’’ (p. 113).
 44

Clymer stumbled onto the truth about ‘‘generalizations’’ while teach-

Chapter 3

ing an elementary school class. He was advised in a course on reading

instruction to let children discover spelling rules through exercises in cat-
egorizing spelling patterns. While Clymer and his students were busily
engaged in this activity, one student (‘‘Kenneth’’) spent his time pouring
through the dictionary, locating exceptions to the generalizations as fast as
the class could create them. When Clymer explained that the dictionary
was filled with unusual words, Kenneth turned his attention to the basal-
reader word lists instead, with the same results. His interest piqued, Cly-
mer decided to track down phonics spelling rules in various manuals and
programs. Using a corpus of 2,600 words, he worked out how many words
did or did not fit 45 common rules. The only rules that held up 100 per-
cent of the time specified which letters formed digraphs (sh in ship). Every
other rule failed in some way or other. This includes the stalwart: ‘‘c ¼ /s/
before e, i, y’’ (crustacean, suspicion break the rule), and the infamous ‘‘when
two vowels go walking the first one does the talking,’’ a rule that fails 55
percent of the time.
A broken rule is not a rule. If ‘‘washing hands before meals’’ is a
family rule, washing hands ‘‘some of the time’’ breaks this rule. Nor does
it help to fudge matters by calling rules ‘‘generalizations’’: ‘‘Most of the
time we wash our hands before meals in this house.’’ This begs the ques-
tion what ‘‘most of the time’’ means. In any case, even assuming that
rules or generalizations were reliable, they would never work. Children
cannot remember rules, much less apply them, something Noah Webster
observed and commented on in his speller over 200 years ago. Many of
the rules Clymer lists do not even make sense: ‘‘In two- and three-syllable
words, the final e lengthens the vowel in the last syllable’’ (a rule that is
broken 54 percent of the time).
There are no spelling rules. Forget about rules. Forget about
generalizations.

Regular Spelling
The expression ‘‘regular spelling’’ is used all the time in publications on
reading, as well as in descriptions of reading and spelling tests. The casual
use of this expression, and the fact that it is never defined, implies that
everyone is supposed to know what ‘‘regular spelling’’ means. Some authors
refer to ‘‘regular grapheme-phoneme spellings,’’ others to ‘‘grapheme-
 45

phoneme correspondence rules’’ or ‘‘GPCs,’’ apparently in the belief that

The Structure of the English Alphabet Code

there are rules governing which decodings are ‘‘regular’’ and ‘‘irregular.’’
But there are no such rules.
Nevertheless, some spellings are obviously more common than
others, and in the absence of any definition for ‘‘regular spelling,’’ I will
provide one here. ‘‘Regular spelling’’ is the only spelling for a particular
phoneme (including letter doubling), or the most probable or least am-
biguous spelling. For example, the sound /b/ is always spelled b or bb.
The spellings ee and ea are the most probable (common) spellings for the
sound /ee/, and represent this sound equally often. However, ea is am-
biguous (bead, head ) and ee is not.
Note that ‘‘regular spelling’’ means regular spelling, not to be confused
with regular decoding, which is often how this term is used. For instance,
the digraph oa is supposed to be ‘‘regular’’ because it is usually decoded as
/oe/ (broad and oasis are exceptions). But oa is not the most common (reg-
ular) spelling for the sound /oe/, one of the many problems that arise when
the code is analyzed the wrong way around. (The most common spelling
of /oe/ is o-e, as in home.)

Orthography
Wagner and Barker (1994) unearthed 11 definitions of orthography in
publications by leading reading researchers. Each definition was different.
Three were circular, using the word orthographic or orthography to define
orthography. According to Szeszulski and Manis (1990, 182), for example,
‘‘Orthographic coding allows direct access to a mental lexicon for familiar
words based on their unique orthography.’’ Other authors stressed the
function and structure of the spelling code, the ‘‘general attributes of the
writing system,’’ which included things like ‘‘structural redundancies’’
(Vellutino, Scanlon, and Tanzman 1994) or similar structural features
(Leslie and Thimke 1986; Jordan 1986). Some incorporated phonological
processing into the definition (Foorman and Liberman 1989; Ehri 1980;
Goswami 1990; Olson et al. 1994), while others took pains to exclude it,
stressing that orthography was different from ‘‘phonological mediation’’
(Stanovich and West 1989). Perfetti (1984, 47) came closest to the real
meaning, defining orthographic ability as ‘‘the knowledge a reader has about
permissible letter patterns.’’
 46

I will reserve orthography for its true meaning (standardized spelling)

Chapter 3

to avoid further confusion. The goal of this chapter is to establish a set of

criteria that define the structure of the spelling code. One of these criteria,
as Vellutino and colleagues correctly pointed out, is ‘‘structural redun-
dancy’’ or probability. But a ‘‘probability structure’’ is not an orthography.
Probability refers to the fact that some alternative spellings for a particular
sound are more or less likely than others. This is a statistical likelihood,
most definitely not a rule or a generalization. Statistical likelihood can be
estimated reliably, whereas rules and generalizations are unreliable, as we
have seen.
In an opaque writing system, the only factors that make it a code and
not chaos are reversibility and statistical likelihood, the incidence of pre-
dictable structural patterns.
The complexity of our spelling code has led most people to believe
that the English spelling system goes on to infinity with no end in sight.
This is why classroom teachers, curriculum designers, and researchers find
such solace in ‘‘word families’’ (damp, lamp, clamp, stamp), those little
islands in an endless sea of chaos, the few discernible patterns in the sys-
tem. The view that English spelling is untamable, and thus unteachable,
has been reinforced by three well-known attempts to fathom the con-
straints or limits of the system. All were unsuccessful, though for quite
different reasons. Before outlining why these attempts failed, let’s look at
the peculiar complexities of our spelling code. Here is a laundry list of the
woes of the English spelling system:

1. There are not enough letters in the alphabet for the 40 phonemes in
the English language. To solve this problem, the Anglo-Saxons followed
the Romans’ lead by reusing letters in combination (digraphs) to stand for
a single phoneme (sh in ship).
2. There are only 6 vowel letters for approximately 23 English vowel
sounds (15 vowels, and 8 vowel þ r vowels). The multiplicity of vowel
digraphs and phonograms (igh in high) make vowels particularly difficult
to read and spell.
3. There are multiple ways to spell the same sound: spelling alternatives.
4. There are multiple ways to ‘‘read’’ the same letter(s): code overlaps.
5. The connection between spelling alternatives and code overlaps is not
straightforward. This creates the false appearance of two independent
 47

systems, one specific to encoding (spelling) and one specific to decoding

The Structure of the English Alphabet Code

(reading).

As an example of points 3–5 above, the sound /ee/ can be encoded

(spelled) ten ways: ee (as in green), ea (mean), e (be), ie (siege), ei (deceive),
e-e (serene), ey (key), y ( folly), i (radio), and i-e (marine). The spelling ee is
always decoded /ee/, but the remaining spellings can be decoded in a
multiplicity of other ways. The spelling ea can be decoded as /e/ (head )
or /ae/ (break), the spelling e as /e/ (bed) or / / (the), and the spelling ie as /
e
ie/ (die), /ue/ (view), /e/ ( friend ), or /i/ (sieve), and so forth.
Given this complexity and the disorganized manner in which reading
and spelling are taught, it is easy to see that, for a child (as for teachers
and reading researchers), the spelling code appears to lack any discernible
structure. But brains are pattern analyzers and cannot code or retrieve
randomness. They actively resonate with recurring regularities in the in-
put, and automatically keep score of the probabilities of recurring pat-
terns. By 9 months of age, infants can learn the statistical redundancies
of four alien syllable sequences in spoken words in about 2 minutes of
listening time (Aslin, Saffran, and Newport 1998). True randomness,
unpredictable relationships between sound and symbol, is unlearnable, in
the same way that the telephone directory is unlearnable. The fact that
the majority of people learn to spell tells us the spelling code cannot be
random. It must have a probability structure in which recurring patterns
are more or less predictable. And if there are patterns, these can be clas-
sified so that they can be taught.

Attempts to Classify the Spelling Code

Webster
The first attempt to classify the English spelling code, as well as to set out
a sequenced method for reading and spelling instruction, was carried out
by Noah Webster and published in 1783. The book became known as the
American Speller or the Blue-Backed Speller. Webster was well aware that
the spelling code works from sound to print and back again from print to
sound, and that the spelling system is the code: ‘‘Spelling is the foundation
of reading.’’ But, perhaps because ‘‘sounds’’ are hard to represent in print,
he set up the speller from print to sound only. Here are his opening
remarks on how to use the speller: ‘‘Let a child be taught first the Roman
 48

letters, both small and great . . . then the Italics . . . then the sounds of
Chapter 3

the vowels; not pronouncing the double letters a and u separately, but
only the sound that those letters united express . . . then the double letters.
All this a child should know before he leaves the Alphabet and begins
to spell’’ (Webster 1783, 28). The speller consists mainly of word lists
organized by syllable length (up to seven syllables) and by types of suffixes.
The only phonological structure imposed on these lists was the use of
word families (rhyming endings). There was no attempt to provide the
common patterns of sound-to-symbol relationships in our spelling code.
No change to this basic format occurred in the six editions of the speller
over 100 years. (For a thorough discussion of Webster’s analysis of the
problem, see McGuinness 1997c, 1998b.)

Venezky
Since that time, surprisingly few attempts to assess the structural elements
and limits of the code have been made. Venezky (1970, 1995, 1999), also
using a letter-to-sound approach, analyzed the incidence and variety of
code overlaps (point 4 on the list above). As he put it, ‘‘Orthography con-
cerns letters and spellings, the representation of speech in writing. That’s
not exactly what this book is about, however. Here only one direction in
the speech-writing relationship—that from writing to speech—is stressed’’
(Venezky 1999, 3).
Venezky set himself the task of discovering all possible ways a partic-
ular letter or letter combination can be decoded, using a corpus of 20,000
words. For example, he identified 17 different ways to decode the letter o.
There are 48 ways to decode the five (single) vowel letters. However,
Venezky tends to give equal weight to all options, which only serves to
highlight the number of exceptions. He did not write, for example: ‘‘Most
of the time, the letter o is decoded three ways—/o/, /oe/, and /u/, as in
hot, told, among.’’ Instead, common and weird spellings alike are set forth
(sapphire, catarrh), and while these curiosities are interesting, they lead the
reader to believe that the English spelling system is beyond redemption.
Venezy did not work out the probability structure of these code over-
laps, but he did offer two classification systems. In the first system, letters
were categorized functionally as relational units, markers, or silent. Rela-
tional units signal a particular decoding (a phoneme); markers refer to let-
ters used as diacritics; silent letters serve no function. However, letters are
 49

not confined to just one category (i.e., are not mutually exclusive), so it is

The Structure of the English Alphabet Code

not clear what function this classification scheme serves. For example, the
letter e is a relational unit for three decodings (be, bet, the), it is a marker or
diacritic for both vowel and consonant pronunciations (can/cane; bulge, not
bulg), and it is silent in axe.
In the second classification scheme, Venezky labeled spelling patterns
as invariant, variant but predictable, or unpredictable, meaning that the
decoding is either fixed (b is always read /b/), or limited to a set of alter-
natives, or unknown without instruction. But Venezky did not specify
which letter patterns fit which of his classification schemes, and did not
clarify the connection between them. Can a letter or digraph be a ‘‘variant
but predictable’’ ‘‘relational unit’’? Because few examples were provided
to illustrate these categories, no overall structure can be developed from
them. One gets a sense of hopelessness about our spelling code, and it
seems Venezky (1995, 29) did as well: ‘‘If orthographic patterns were sym-
metrical, pronouncing words from their spellings and spelling words from
their pronunciations would be relatively similar tasks. . . . Such symmetry
would allow a collapsing of the teaching of reading and spelling, as well
as simplifying the description of orthographic patterns. But alas, English
orthography is painfully asymmetrical.’’
This is a surprising statement on two levels. First, it is a confession
that there is no symmetry if the code is analyzed backward, solely from
print to sound (the way he did it). Second, it is tantamount to a claim that
our spelling system is not reversible. But if it is not reversible, it cannot be
a code. And if it is not a code, no one can learn it. Venezky’s statement
was followed by three pages of illustrations and descriptions of the star-
tling complexity created by multiple spelling alternatives for the sounds
/k/ and /z/, and of how these, in turn, can be decoded in a multitude of
unrelated ways. These examples have the effect of shocking the reader into
recognizing the utter futility of trying to fathom the limits of our spelling
system.
The American Way of Spelling (Venezky 1999) is the most systematic of
Venezky’s writings and is devoted to an analysis of our spelling system.
The title belies the contents, because the purpose of the book is to set out
the ‘‘decodings’’ of the various spelling patterns, not how the 40 pho-
nemes in our language can be spelled. The book begins with an in-depth
account of the many spelling patterns in our system and ends with a
 50

chapter on phonics instruction, setting out a more simplified list of which

Chapter 3

spelling patterns should be taught and in what order. Despite including

a chapter on the phonetic distinctions in English, the book provides
virtually no analysis of how the spelling patterns are related to these dis-
tinctions, or an explanation of the relevance of this phonetic structure to
the nature of the code or the teaching of reading and spelling.
Venezky’s primary goal in this book is to provide some type of order,
a set of categories or ‘‘rules’’ for how our spelling system works, although
he recognizes that no one can learn to read by trying to memorize or
apply these rules. His search for rules is actually a search for logic, as if
our spelling system was idiosyncratic by design. Thus, he makes many
observations like ‘‘Visual identity of meaningful word parts takes prece-
dence over letter-sound simplicity’’—this to explain why the plural s sig-
nifies two pronunciations (as in cats and dogz). He searches for patterns in
homophone spellings, such as the use of ‘‘silent letters’’ ( plum, plumb; rain,
reign; him, hymn), as if their use was intentional. Some of his observations
are correct and quite general (useful), such as the consistent way simple
vowels govern the spelling of final consonants: final /k/ is spelled ck after
simple vowels (sick, sock, sack), but not after digraphs (weack, seeck) or con-
sonants (milck, sharck). Final consonants usually double after simple vowels
(well, stuff ) but not after digraphs or consonants (real, leaf, girl, self ).
Venezky attempts to look at ‘‘regularity’’ in two ways. One has to do
with pseudowords or pronounceable nonwords. As he points out, ‘‘No full
set of rules for generating legal spellings for pseudowords has ever been
published.’’ He tries to set up some rules here, such as ‘‘all one or two
letter words are irregular by strict criterion,’’ listing ebb, odd, and egg as
examples of why the final letter is doubled. (But there are many regular
spellings that do not follow this rule, such as in, is, it, at, am, an, as.)
The second way is to identify every consonant and vowel spelling,
and establish a frequency count of how often each spelling is used in the
overall corpus of words and in the initial, medial, and final positions
in these words. A table is provided for consonants, but not for vowels.
One problem with this endeavor is that the 20,000-word corpus contains
foreign words and proper names, which make the tallies misleading.
Examples include Ghent, khaki, khan, Slav, Yugoslav, Chekhov, leitmotiv,
Himalayas, Maya, Sawyer, Agassiz, and Suez. The consonant spelling rrh
for /r/ appears in six Greek words for diseases (cirrhosis, diarrhea).
 51

From this exercise we learn that the most common initial consonant

The Structure of the English Alphabet Code

letter is c, the most common medial letter is r, and the most common final
letter is m. But this does not tell us which sounds ( phonemes) are common
in these positions or how these sounds should be spelled. Nor does it tell
us how these letters are decoded. For example, knowing that c is most
common in initial position does not tell us how often it represents /k/ or
/s/ (cut, cent).
The chapter on consonants traces the history of 38 consonant spell-
ings, such as where the letter(s) itself came from, the regular and irregular
types of spellings or words it can appear in, and the sound or sounds into
which it can be decoded. The letter b is usually decoded /b/, can double in
certain words (bb), and is subject to being ‘‘silent’’ in debt, doubt, and sub-
tle, is subject to ‘‘pre or postjunctural deletion’’ (silent) in bdellium, bomb,
and so on, and is subject to ‘‘medial cluster leveling’’ (silent) in subpoena.
The 38 consonant spellings are listed in alphabetical order and include
many alternative spellings, including: ck, dg, gh, gn, kh, ph, pph, rh, rrh,
sch, tch, u (u stands for /w/ in a words like quiz, anguish, persuade). As
a reminder, there are only 23 consonant phonemes in English, including
/zh/ (vision) plus /kw/ (qu) and /ks/ (x).
The chapter on vowel spellings lacks a table of spelling patterns and
frequencies (i.e., no probability analysis for vowel spellings). Instead, the
vowels are set out in a categorical scheme. This goes as follows:

Major primary patterns: a, e, i, o, u, y. These letters signify vowels that are

‘‘checked,’’ as in cat, bet, sit, hot, cut, gym, or ‘‘free,’’ as in cake, mete, kite,
home, cute, why.
Primary subpatterns. Final e pattern: corresponds to spellings as above:
a-e, e-e, i-e, o-e, u-e.
Geminate consonant pattern. Special decoding for vowels in conjunction
with double consonants: mamma, albatross, coffee, gross, doll, pudding.
Consonant influences: Vowel þ r Venezky sets out 15 vowel þ r patterns,
plus 12 exception patterns. There are too many overlaps in both spellings
and pronunciation for this to be valid. Linguists usually recognize about 5
to 6 vowel þ r phonemes. I could only find 9, as listed in the preface.
Before final l or ll. This refers to so-called ‘‘l-controlled’’ vowel pronun-
ciation in words like ball, doll, almond, calm.
w-controlled a: As in: wand, swan, where the w signals the vowel /o/.
 52

i before nd, ld, gn.: As in: kind, mild, sign.

Chapter 3

Suffix Patterns: Syllable stress and schwa vowel.

Exceptions to decoding words with final e spellings: have, morale.
Major Secondary Patterns: ai, ay, au, aw, ea, ee, ei, ey, eu, ew, ie, oa, oi,
oy, ou, ow, ui, and uy.
Minor secondary patterns are low-frequency spellings: aa, ae, eau, eo, ieu,
iew, oe, ue, ye

In the final chapter, Venezky states that 100 ‘‘common graphemes’’

should be taught initially in beginning phonics instruction, a process he
estimates will take two to three years to complete, and this does not in-
clude the additional time for learning prefix and suffix spelling patterns.
These graphemes consist of 58 consonant spellings and 38 vowel spellings,
plus 4 spelling rules (for a total of 100). However, the chapter lists only 34
vowel spellings and 23 consonant spellings, plus one rule, and there is no
explanation for why the remaining spelling patterns are omitted. The list
excludes 32 common spelling alternatives in my system (see page 55) but
includes many spellings that children will not need at this level, such as:
t for /sh/ in nation, c for /sh/ in ocean, s for /zh/ in measure (the phoneme
/zh/ is not mentioned anywhere in the book), ch for /k/ in chord, h is
silent in honor, and b is silent in bomb. (The phoneme representations are
mine.)
Also provided is a list of 37 ‘‘exception words’’ (sight words), plus
32 prefixes and suffixes. Sixteen of these words have perfectly regular
spellings and are in no way exceptional. There are no guidelines for
how these spelling patterns, rules, exception words, and affixes should be
taught. Only a few words are provided for each spelling, and rare spellings
are just as likely as common ones. In short, it is difficult to see how
teachers are supposed to use this material, and difficult to understand
how reading can be taught without reference to the 40 phonemes in our
language.
Venezky suggests that spelling should be taught along with reading,
but he provides no information on how this should be done. He often uses
a logic that is contradictory to the way the code works, one which blocks
teaching spelling in any meaningful way. For example, he urges teachers
to reinforce multiple ways to decode the same spelling: ‘‘Students should
be asked repeatedly ‘what other sounds could that letter (digraph) have?’
 53

and ‘do you know a word in which it has a different sound?’ ’’ (p. 232,

The Structure of the English Alphabet Code

italics mine).
Letters do not have sounds. People do.

Hanna, Hanna, Hodges, and Rudorf

In the third well-known attempt to come to grips with the spelling code,
the code was analyzed properly from print to sound, the way it was
designed. This project was much more ambitious than Venezky’s. Using
a corpus of 17,310 English words, Hanna et al. (1966) worked out all
the spelling alternatives for sounds in the English language, apparently
the first attempt to do so. Their goal was to classify the various spelling
alternatives according to the ‘‘frequency’’ with which particular spellings
appeared within this 17,310-word corpus.
I want to distinguish here between frequency in terms of probability
within a corpus of words (how often a particular spelling appears in all
words), and frequency in print (how often words with particular spellings
appear in text), which is not what Hanna et al. were interested in. For
example, ee and ea are the most probable spellings for the sound /ee/. But
e is the most frequent spelling because words with this spelling appear
frequently in print (be, he, she, me, we), and this spelling also appears in
common prefixes (de-, e-, be-, re-, pre-). I will use the word probability for
the former and frequency for the latter. Both are important in working
out the structure of the code and in setting up a curriculum for how to
teach it.
Hanna et al.’s work is a monumental effort in every sense of the word,
and it is unfortunate that a series of major blunders and other problems
led to a largely unusable outcome. The classification was based on 52
English phonemes, about 10 too many. These include 15 primary vowels:
^ /oo/, /ou/, and /oi/,
/a/, /e/, /i/, /o/, /u/, /ae/, /ee/, /ie/, /oe/, /ue/, /oo/,
plus schwa ( ), an unaccented /uh/ sound. The vowel sound /aw/ was
e
omitted. Also included were five of the eight vowel þ r vowels: /ar/, /er/,
/or/, /air/, and /eer/. Omitted were /ire/, /ure/, and /our/.2

2. Vowel þ r vowels are known as ‘‘rhotacized’’ vowels (‘‘r-colored’’). Because

/r/ by itself is a vowel (/er/ in her), a vowel þ r is effectively a digraph that
represents a diphthong. This is like other vowel-vowel combinations that
 54

Some vowel sounds were counted twice: /er/ (her) and /ur/ ( fur); /o/
Chapter 3

(soft) and /o/ (odd ). There were numerous errors in classifying vowel
spelling alternatives. The most serious stemmed from misunderstanding
the role of e as a diacritic marker for consonants in words like juice, dense,
live, siege, judge, and soothe. The e was erroneously coded with the vowel,
producing several nonexistent spelling alternatives like oo-e (choose) and
ui-e ( juice), plus hundreds of misclassified words where the e works with
the consonant but was coded as working with the preceding vowel: e-e
(license), o-e (dodge), and i-e (massive).
There are 23 consonants in English (25 if one includes the letter
symbols for consonant clusters: x /ks/ and qu /kw/). Hanna et al. listed 31.
These included /ks/ and /kw/, plus silent h (honest) and several phonemes
that do not exist, marked with a glottal stop /'l/, /'m/, and /'n/, as in table,
chasm, and pardon. These are two-syllable words. Every syllable, by defi-
nition, must have a vowel, not a glottal stop. These words are pronounced
ta-bP l, kazzP m, and pardP n, with a schwa vowel in the second syllable.
Hanna et al. uncovered a total of 174 spellings for 52 phonemes (93
for consonants and 81 for vowels), including the spellings for the Latin
and Greek layers of the language. The spelling alternatives were classified
in turn according to whether they appeared in a stressed or unstressed
syllable. However, English spelling is largely unrelated to syllable stress.
The schwa vowel always appears in an unstressed syllable, but knowing
this does not tell you which of its six spellings to use. Next, Hanna and
colleagues classified multisyllable words according to whether a spelling
appeared in the initial, medial, or final syllable, another exercise in futility.
Spelling patterns are affected by phoneme position within a syllable, not
between syllables in multisyllable words. All this greatly increased the level
of complexity of their results.
There were errors in data entry. People were trained to classify words
by phoneme and by spelling alternative. Initially, there was a high success
rate with a trial corpus of 565 words. This level of accuracy was not

count as one vowel: /ou/ ¼ ah-oo (out); /oi/ ¼ oh-ee (oil ). Vowel þ r vowels are
most affected by dialect, and in some dialects the /r/ portion is not sounded.
There is considerable disagreement among linguists as to how many vowel þ r
vowels there are.
 55

maintained, however, and Hanna et al.’s study is replete with hundreds of

The Structure of the English Alphabet Code

data-entry errors. The most glaring mistake was putting the words with
the vowel sound /ue/ (cue) into the lists of words with the vowel sound
/oo/ (coo), an error that does not appear in the vowel tables at the front of
the book.
But there was a far more serious problem, essentially dooming the
whole endeavor. They did not separate the Latin (or Greek) layers of
the language from common English words (Anglo-Saxon and Norman
French). This needs to be done, because there are major structural differ-
ences in the spelling code for Latin words, and there are special spelling
patterns for words of Greek derivation.
The inaccurate set of spelling alternatives, the complex and unneces-
sary syllable-classification scheme, plus the inclusion of an abundance of
affixed forms of the same root word (idea, ideal, idealism, idealistic, idealize,
ideally), led to an unwieldy tome of 1,716 pages, taking up the space of
four large books on the library shelf. The awesome complexity of this
document along with its bulk only confirmed in another way that our
spelling system is too wild to tame.
These efforts had the effect of convincing researchers that spelling
could not be taught. The fact that children learned to spell at all was taken
as evidence that children go through ‘‘developmental stages’’ (see chapter
9). The extreme form of this notion is tantamount to a belief that English
brains evolved to learn to spell the English writing system. This specificity
is a requirement of the model, because the spelling errors English children
make as they progress through these so-called spelling stages do not ap-
pear for children learning transparent orthographies who receive appro-
priate instruction (Venezky 1973; Wimmer and Landerl 1997; Geva and
Siegel 2000).

McGuinness
The most recent attempt to classify the spelling code is my own
(McGuinness 1992, 1997a, 1997c, 1998a, 1998b). This classification is
based on a strict sound-to-print orientation like that of Hanna et al. But
the goals and the procedure were different, in that the purpose of my
classification was similar to Webster’s—to systematize and illuminate
the spelling code so that it could be taught. The process began with the
phonotactic structure of the English language (legal phoneme sequences
 56

in words). All phonotactically legal syllables were determined. If a syl-

Chapter 3

lable constituted a real word, or was part of a real word, its spelling was
recorded. Based on this analysis, it was estimated that there are over
55,000 legal English syllables. Contrast this to the 1,277 legal syllables in
Chinese, and you have the reason our writing system is an alphabet and
not a syllabary.

Basic Code The first step in any classification process is to establish limits
or endpoints. Endpoints create a frame or boundary. This is essential for
codes, because they must have a pivot point, around which the code can
reverse. This endpoint or pivot is the finite number of phonemes in the
language. Once the student knows 40 phonemes, there are no more to
learn.
The next step is to set up a basic code (an artificial, transparent
alphabet) using the most probable spelling for each phoneme, as defined
in the section ‘‘Regular Spelling’’ above. A basic code can easily be taught
to 4- or 5-year-olds in a relatively short period of time ( Johnston and
Watson 1997, forthcoming; also see chapter 5, this volume), though this
should not be mistaken for the complete code. A basic code makes it pos-
sible to read and write a large number of common one- and two-syllable
words. It is transparently reversible, so children can see and experience the
logic of a writing system. The idea of a basic code is not new (Ellis 1870;
Dale 1898), and it is common to several phonics programs today, particu-
larly in the United Kingdom. But nearly all programs stop here, or at best,
teach only a fraction of the remainder of the code.

The Advanced Spelling Code The major hurdle in our writing system is
mastering the multiple spellings for each phoneme. This is the reason
English-speaking children have so much difficulty learning to read and
spell. It is the ‘‘advanced code’’ that causes the major problems for poor
readers, and teaching this turns out to be far more important during
remediation than training phoneme-awareness skills (C. McGuinness, D.
McGuinness, and G. McGuinness 1996).
Because of its complexity, the advanced code itself needs a classifi-
cation scheme with limits and boundaries. All classification systems are
somewhat arbitrary, and whether this one is best for the purpose remains
 57

an empirical question. The first hurdle is to find out which spelling alter-

The Structure of the English Alphabet Code

natives are common and which are rare, and the number of words suffi-
cient to establish a bona fide spelling alternative. The number of spelling
alternatives needs to be kept within reasonable bounds, while at the same
time avoiding hundreds of special exceptions. If an improbable (uncom-
mon) spelling of a phoneme occurs in a few words or less, what should be
done with these words? The solution was to create four more categories at
the advanced-code level.
The first category comprises the major spelling alternatives beyond the
basic-code level. Spellings that did not qualify as major were classified in
one of three ways: as a special group (few words, improbable spelling, high
frequency in print), as sight words (singular spelling, high frequency in
print), or omitted (singular spelling, low frequency in print).
In making these distinctions, probability (how many times a spelling
alternative is used in a large corpus of words) was a key determiner of a
major spelling alternative, whereas frequency in print was more important
for inclusion as a ‘‘special group’’ or a ‘‘sight word.’’ Special groups like
could, would, should, and door, floor, poor, and break, great, steak are impor-
tant high-frequency words with improbable spellings.
Another decision was to avoid ‘‘silent letters’’ by merging them into
digraphs or multiletter phonograms whenever possible. Thus igh is one of
four main spelling alternatives for the sound /ie/ (high). This helps teach-
ers avoid making nonsensical statements like this one: ‘‘In the word high,
the I says its name, and the gee aitch is silent.’’ This solution works well
except in rare cases. For words like honor, hour, and honest, there is no
choice but to tell the children: ‘‘The first letter isn’t sounded in these
words.’’
Every effort was made to avoid classifying a word as a sight word.
Most reading programs produced by the major publishing houses include
a large list of sight words, many using ‘‘regular’’ spellings. It is a bad idea
to teach sight words to children learning any writing system, as we saw
in chapter 1. But there is more at stake. Teaching whole words by sight
promotes a faulty decoding strategy. This happens because memorizing
whole words seems logical and is relatively easy initially, leading to a
false sense of security. But a whole-word strategy will inevitably collapse,
depending on the child’s vocabulary and visual-memory skills. Meanwhile,
 58

this strategy can harden into a habit that can be difficult to break
Chapter 3

(McGuinness 1997b).
For these reasons, the sight-word category was reserved for common
words where one or more phonemes have a unique spelling that is hard
to decode without direct instruction. There are almost no words where
every phoneme has an unpredictable spelling. By this criterion, there are
remarkably few true sight words. The following sight words and special-
group words did not fit a major spelling category in a large corpus of words
of English/French origin. There are approximately 100 sight words:

/a/ aunt, laugh, plaid

/e/ friend, leopard
/i/ been, busy, sieve, pretty, women
/o/ abroad, broad, cough, father, gone, trough, yacht
/u/ a, because, does, blood, flood, of, once, one, the, was, what
/ae/ straight, they Group: ea break, great, steak
/ee/ people, ski
/ie/ aisle, choir, I, height, sleight
/oe/ sew
/ue/ beauty, feud, queue
/oo/ move, prove, shoe, deuce, through Group: o do, to, who, whom,
whose
^
/oo/ Group: oul could, would, should
/ar/ are, heart, hearth Group: orr borrow, tomorrow, sorrow, sorry
/er/ acre, glamour, journey, syrup, were Group: ure leisure, measure,
pleasure, treasure
/or/ drawer, laurel Group: oor door, floor, poor
/air/ bury, heron, scarce, their, there, they’re, very, where

Final /k/—c or ch arc, tic, ache, stomach Group: /k/—lk baulk, caulk,
chalk, stalk, talk, walk
/t/—bt Group: debt, doubt, subtle
Final /th/—th smooth (final voiced /th/ is usually spelled the, as in breathe,
clothe)
Final /v/—f of
Initial h is not sounded: honest, honor, hour
Initial /h/—wh who, whom, whose, whole
 59

The ‘‘omitted’’ category eliminates nuisance sight words—those rare,

The Structure of the English Alphabet Code

low-frequency words with rogue spellings that only clog up the system,
like sapphire and catarrh. These words are usually easy to decode and only
hard to spell. They need to be seen a lot or looked up in a dictionary until
they are memorized.
So far, this analysis covers words derived from Anglo-Saxon and Nor-
man French. There are 132 major spelling alternatives for these words, 55
for consonant spellings and 77 for vowels (including 4 vowel þ r spell-
ings). Once this structure was worked out, it was set up in an easily acces-
sible form as a ‘‘spelling dictionary’’ organized by phonemes. All words
containing a phoneme with a particular spelling were listed under the
appropriate heading and alphabetized in two dimensions down the page.
Spelling alternatives were listed in a row across the top of the page from
most to least probable, with words spelled in basic code on the left. Thus,
a corpus of over 3,000 common English words could be presented in only
75 pages, providing a clear visual display of the probability structure of the
spelling code (McGuinness 1997a). This structure is obvious, even to a
child. A page from the dictionary is shown in table 3.1.
Very little active memorization is necessary when learning is based on
exposure to predictable patterns (structural redundancies). Our brains do
the work for us. The only active memorization required is to learn the 40
phonemes in the language and their basic-code spellings. Spelling alter-
natives for each phoneme can be mastered through controlled exposure
and varied repetition, via dictionary activities, fail-safe worksheets, spe-
cially designed stories, copying, and creative writing.
Once the structure of the spelling code is set up visually, numerous
features and patterns come to light. Use of these patterns can dramatically
speed up learning while reducing the memory load, and is essential for
setting up a sequence of instruction. Here are some of the most important
patterns.

Structural Features of the Code

1. Some phonemes have no, or rare, spelling alternatives. Apart from
consonant doubling, there is no way to misread or misspell them:

/b/, /d/, /l/, /p/, /th/, /a/, /ar/

Table 3.1
Sound: /ae/ Key word: came
Spelling alternatives
a-e ai a ay/ey ei/eigh

h hail
hale halo
haste
hasty
hate hatred hay
haze hey
i inflate
j jade jail jay
l label
labor
lace ladle
lady
laid
lain
lake
lame
lane
late
lathe lazy lay
m mace
made maid
mail
maim
main
maize major
make
male
mane manger
maple
mate may
maze
n nail
name
nape nasal
native
nature
naval
navel
navy neighbor
o obey
Source: McGuinness 1997a.
 61

2. Because some phonemes have one or two spelling alternatives used in

The Structure of the English Alphabet Code

a very small number of words, it is possible to teach by exclusion: ‘‘This
sound is spelled in basic code except in these words.’’

/g/, /h/, /m/, /n/, /ng/, /t/, /i/

3. Some consonant spellings are determined by whether the consonant is

in the initial or final position in a syllable.

/j/ jinx barge bridge

/th/ this bathe
/v/ vain groove

4. Some consonant spellings in final position are determined by whether

they are preceded by one of the five ‘‘checked’’ vowels: /a/, /e/, /i/, /o/,
and /u/. (There are exceptions.)

/ch/ lunch beach catch

/j/ siege barge hedge
/k/ beak milk sick
/l/ girl fool dull

5. There are 76 legal consonant blends (adjacent consonants) in the En-

glish language, and only 3 appear in both initial and final position in a
syllable. These are sk, sp, and st (skip/ask; spoon/gasp; stop/fast). Nearly all
consonant blends are always spelled in basic code, something very useful
to know. If children can segment them, they will always be able to read
and spell them correctly.
6. Patterns within patterns (statistical regularities) make learning multiple
spelling alternatives more manageable. Once these are known, they can be
taught directly. Exposure (practice) is the key, not rules or memoriza-
tion. For example, /ee/ has the most spelling alternatives of any phoneme.
These can be broken down in several ways. Multisyllable words ending in
/ee/ are spelled y or ey most of the time (exceptions: cookie, collie, coffee,
toffee), and these spellings rarely represent /ee/ in other slots. Children
should learn this by exposure—by reading, writing, and spelling a range
of multisyllable words ending in the sound /ee/. Teaching these patterns
 62

chips away at the complexity, by accounting for when and where specific
Chapter 3

spelling alternatives are used. By a process of exclusion, the complexity is

reduced from 10 spellings for /ee/ to 8, and can be reduced to 7 by mem-
orizing the few words and prefixes where /ee/ is spelled e (be, he, me, she,
we, be-, de-, pre-, re-, tre-), and so forth.
7. The classification reveals danger zones, such as spelling alternatives
with no obvious or memorable patterns, or those with conflicting code
overlaps, signaling the need for more exposure. This can be illustrated
with another example from the /ee/ spelling complex. The two main ways
to spell /ee/ (ee and ea) are equally probable in common English words:
bean, clean, green, mean, queen, seen, screen, teen, wean / beat, beet, bleat, feat,
feet, fleet, greet, heat, meet, neat, peat, seat, sleet, street, sweet, treat, wheat. I
have set these up in the een and eet families to make the point that ‘‘word
families’’ do not help you spell. Because there are no clues for which of
these spellings to use, the only solution is to see and write these words
often.

The Latin Layer of the Code A different type of classification is

required for the Latin layer of the spelling code, one that relates to the
morpheme level of language (McGuinness 1998a). A morpheme is the
smallest unit of sound that signifies meaning. English, Norman French,
Latin, and Greek are compounding languages, in which root words can
combine with other root words or affixes to alter parts of speech, change
verb tense, or create entirely new words like deadline, firefly, household,
hotdog, and railroad. Teaching spelling alternatives in Latin words in the
context of this compounding structure greatly simplifies the nature and
type of spelling alternatives that need to be taught. It considerably
enhances multisyllable decoding and spelling, while building vocabulary at
the same time.
The compounding structure of English, French, Latin, and Greek
words, along with special issues like adjacent vowels, homophones, and the
schwa, can be fully addressed with an additional corpus of 3,000 words.
We will skip over the details of English compounding and the trans-
formations that occur when English prefixes and suffixes are added to root
words. These are reasonably well known and sufficiently orderly (see
McGuinness 1998a). For example, if a root word ends in e, this must be
dropped prior to adding a suffix that begins with a vowel, such as -ing
 63

or -ed. When adding a suffix to CVC-type words with ‘‘simple’’ vowels,

The Structure of the English Alphabet Code

if the suffix starts with a vowel (-ing, -ed, -er, -y), the final consonant
is doubled to preserve the vowel sound. Examples involving bat include
batting, batted, batter, and batty (not bating, bated, bater, and baty). This is
a fairly stable convention, coming about as close as it gets to a rule in our
spelling code. Adding prefixes poses no problem, because no transforma-
tions are necessary.
The Latin layer marches to a different drummer. In the first place, it
is rare that a Latin root word is an English word ( fact, duct, and port are
exceptions that come to mind). This means that the core of Latin/English
words is bereft of meaning, a definite disadvantage. English/Latin words
are compounded by prefixes, suffixes, or both. As an example, the root
word struct (‘‘to build’’) is not an English word, but is ‘‘English’’ in a
number of common affixed forms: instruct, construct, instructor, instruction,
instructed, construction, constructed, destruction, indestructible, structure, ob-
struction, and so forth. (The prefix in- means ‘‘in,’’ con- means ‘‘with,’’
de- means ‘‘undo’’ or ‘‘opposite to,’’ and ob- means ‘‘in the way of’’ or
‘‘against.’’ The suffix -or means ‘‘a person,’’ and ‘‘shun’’ (-tion) means the
word is a noun.)
Fortunately, Latin roots and prefixes are usually spelled in basic code,
which is a bonus. But Latin suffixes are not and pose particular problems if
they are taught as a sequence of individual phonemes. This would produce
a dazzling array of spelling alternatives and code overlaps. The phoneme
/sh/ can be spelled ti (nation), si (tension), ci (magician), ce (ocean), shi
(cushion), sci (conscious), or xi (anxious). It is tidier to keep Latin suffixes in-
tact as multiphoneme units. This is a radical departure from the approach
used for the English layer of the language.
A good example is the suffix ‘‘shun,’’ the most common Latin suffix. It
has three main spelling alternatives and four rare ones. Far and away the
most common is -tion, which appears in tens of thousands of English
words. The next most common is -sion, followed by -cian. The rare
spellings are -tian (dietitian, gentian, Martian, titian), -cion (coercion, suspi-
cion), -cean (crustacean, ocean), and -shion (cushion, fashion). There are lots
of useful clues for these ‘‘shun’’ spellings:

When in doubt, spell -tion.

If an occupation or person, spell -cian.
 64

The suffix -sion usually attaches to word stems ending in /s/: access,
Chapter 3

compress, concuss, confess, convulse, depress, digress, discuss, and so on, though
other word stems take this spelling as well: admit/admission, ascend/
ascension.

Memorize the words that use the -tian, -cion, -cean, and -shion

spellings.

What is more, -tion and -sion get around. They move as a unit, another
reason to teach them as a unit. They pop up as spellings for ‘‘zhun’’
(equation, vision), and for ‘‘chun’’ (digestion, question, suggestion). In all, there
are 38 Latin and 2 Old French suffix spellings that need to be taught. This
brings the total of spelling alternatives to 172, plus 4 extra ‘‘Greek’’ con-
sonant spellings at this level, a total of 176. This is surprisingly close to the
tally of 174 reached by Hanna et al., but for quite different reasons.
Teaching these Latin suffixes requires a different instructional ap-
proach, and keeping the suffixes intact has a number of spin-offs. One is
that it makes long, scary words like advantageous and unconscious easy to
read and spell. Knowing that the formidable -geous or -scious spellings
are merely ‘‘jus’’ or ‘‘shus’’ in disguise makes them far less daunting. Once
these suffixes are demystified, they can be identified first, making the rest
of the word easy to decode and spell: ad-van-ta-/geous, un-con/scious.
The front ends of Latin-derived words are remarkably well behaved and
usually spelled in basic code, or with a highly probable spelling alternative.
A second spin-off is that suffixes attach to root words in predictable
ways. When -tion attaches to a root word, the most common form
involves adding the letter a to the root: inform-information, limit-limitation.
When word stems ends in e, the e is dropped: agitate-agitation, create-
creation. A large family of -ate words follow this pattern, another reason to
group words when teaching these suffixes.
Greek words came into the language via philosophy, medicine, and
science, originally in Greek, and later as transliterations with special
‘‘Greek’’ spellings. If one steers clear of specialist or technical words,
the Greek invasion is remarkably less intrusive than people believe. Very
few common words use these spellings, and only eight of these spellings
appear in familiar words. These words can be listed on a single page. The
spellings include ch for /k/ (chorus), y for medial /i/ (myth) or /ie/ (cycle),
 65

ph for /f/ (dolphin), and rarer spellings like pn for /n/ ( pneumatic), rh for

The Structure of the English Alphabet Code

/r/ (rhapsody), ps for /s/ ( psalm), and x for /z/ (xylophone).
This is not to say that everything is solved by classification alone.
Other elements of our spelling system present different challenges, and
some make heavy demands on visual memory. Words with adjacent vow-
els are a problem because they tend to look like common digraphs and
sometimes overlap with them. This group includes words like poet, briar,
ruin, create, fluent, denial, radio, annual, oasis, alien, and idiot. Children need
lessons on how to break these words into syllables.
The biggest spelling headaches, those that never go away, are when to
double consonants, and how to spell a schwa. These are the spelling errors
most responsible for the wavy red lines produced by your computer spell-
check feature. The schwa, an unaccented ‘‘uh’’ sound, appears in tens of
thousands of multisyllable words. Though initial and final spellings were
standardized to a (among, agenda), there is no rhyme or reason for how to
spell the schwa anywhere else in the word: hesitate, benefit, reliant, gratify,
impolite, cultivate, economy, important, adjacent, and the list goes on. Letter
doubling in Latin-based words is supposed to be resolved by word-
derivation rules. But because nobody knows Latin any more, this is no
help at all. Which of the following spellings is correct: recomendation,
reccomendation, recommendation, reccommendation?

Research Based on Spelling-Classification Systems

The ultimate question is, does classifying the spelling code so that it can
be taught make it easier for children to learn to read and spell? There is
almost no research on spelling (or reading for that matter) based on an
informed understanding of the spelling code, so little, in fact, that while
the National Reading Panel’s report (2000) has sections on phoneme
awareness, reading, fluency, vocabulary, and comprehension, it has no
section on spelling. I know of no applied research that relies on Hanna
et al.’s efforts to systematize the code or on their subsequent recommen-
dations (Hodges 1981, 1982). A few studies use the spelling levels pro-
posed by Venezky (1995), and at least one study uses my classification
system. These studies shed considerable light on the difference in out-
come between teaching the code from letter to sound (visual strategy)
versus from sound to letter (phoneme strategy).
 66

Berninger et al. (1998) created a set of 48 words derived from eight

Chapter 3

levels or sequences proposed by Venezky. These levels are letter driven,

so it is important to point out that they are decoding levels, not spelling
levels (the code taught backward). The levels are as follows:

0. CVC words: bat nut men mad hit wet

1. mixed words: flag back last ask drop club
2. words with a-e, i-e, o-e spellings
3. words with consonant and vowel digraph spellings: sh ch ng oy ou aw
4. words with vowel digraph spellings: ai ay ee ea oa ow
5. words spelled with single or double l
6. words with vowel þ r: ar er ir ur or
7. words spelled with wr kn igh mb tch dge

Altogether, there are 33 phonemes here represented by 47 spellings.

Missing from the list are seven phonemes (/ue/, /oo/, ^ /oo/, /th/, /th/, /v/,
and /z/), 5 vowel þ r vowels, and the spellings qu and x. This arbitrary,
visual approach is typical of many phonics programs, and I want to unpack
it to illustrate the problems created by a letter-to-sound approach like this.
Bear with me, because if this confuses you, think how it will confuse a
child.
The first two levels introduce words with letters that stand for these
phonemes:

Level 0: /b/, /d/, /t/, /m/, /n/, /h/, /w/, /a/, /e/, and /i/
Level 1: /f/, /l/, /g/, /k/, /s/, /t/, /p/, /o/, and /u/

Level 1 words (see above) include three spellings for the sound /k/: c,
k, and ck, as seen in the words club, ask, and back. Do the children learn
these as three spellings for the same sound, or as three different sounds?
At level 2, children learn to decode words with the vowel spellings a-e
(lane), i-e ( fine), and o-e (tone). At level 4 they learn alternative spellings
for two of these vowels (/ae/ and /oe/): ai (hail ) and ay (day); oa (soap) and
ow (snow). The ow spelling is also a code overlap for the phoneme /ou/
(cow), so it is not clear which sound is meant. The question is, do children
learn the level 4 vowels as new sounds, or do they learn that ai, ay, oa, and
 67

ow are alternative spellings for two sounds they learned at level 2? And if

The Structure of the English Alphabet Code

this is the case, where are the spelling alternatives for the sound /ie/ that
was also introduced at level 2? These alternatives are i ( find ), y (cry), and
igh (high).
Level 5 is devoted to teaching words spelled with single and double
l’s. Why this is important is unclear. There is no ‘‘level’’ for the equally
tricky single and double f spellings (elf, staff ) or any other spelling where
final consonant doubling is an issue: bag, egg; gas, dress; quiz, buzz.
At level 6, children learn ar and or, as well as three spelling alter-
natives for the sound /er/: er, ir, and ur. Again, do children learn er,
ir, and ur as different sounds, or as three spellings for the same sound?
If the latter, why aren’t the main spellings for /or/—ore (bore), oar
(soar), and our ( your)—listed here as well? (/ar/ has no common spelling
alternatives).
At level 7, children see a mix of common and rare digraphs and pho-
nograms. There is no logical or structural reason why these are here. Five
are spelling alternatives for phonemes previously taught, but it is highly
unlikely the children will ever know this. One assumes that igh is taught as
a ‘‘new’’ sound, bearing no relationship to the /ie/ sound taught five levels
earlier.
Level 7 plays havoc with the structure of the code by divorcing spell-
ing alternatives from the basic code. The basic-code spelling for /ch/ is ch
(taught at level 3). Do children learn that tch (level 7) stands for the same
sound? Are they told which is used only in word-final position, and when?
The spelling dge for the sound /j/ (also a word-final spelling) is introduced
here as well. But the basic-code spelling j was never taught. How then
should children spell words like just, June, and jelly? Are they to be written
dgeust, dgeune, and dgeelly?
This approach and its haphazard organization are notable for the total
disregard of the sounds in the language. This makes it impossible for
children to understand that our writing system maps 40 speech sounds to
a set of spellings, because these sounds are never taught.
Using Venezky’s system, Berninger et al. trained a large number of
third and fourth graders identified as ‘‘poor spellers.’’ The children were
divided into several groups that received different kinds of training, plus
a control group. They were taught to spell 48 words over a period of
 68

4 months (24 sessions of 10 minutes each, totaling about 4 hours). The

Chapter 3

control group was trained in phonological awareness and letter-name

knowledge and did not see these words.
After training, the children took a spelling-dictation test on ‘‘transfer
words,’’ which differed in some minor way from the training words. Most
(75 percent) involved a simple change of the initial consonant. The inves-
tigators found that children in the training groups spelled the 48 transfer
words more accurately than the control group (40.5 correct versus 30).
There was no difference as a result of different types of training. It did
not matter whether children memorized the words as sight words, or via
phoneme segmenting, ‘‘onset-rime’’ analysis, or any combination thereof.
In effect, these results show that if you study the same 48 words on 24
occasions, you will be more likely to spell nearly identical words correctly
than if you have never seem them before.3
The critical test is whether the spelling instruction transferred to
standardized measures of spelling. It did not. The authors reported that
they ‘‘failed to find any reliable differences between the treatments as a
group or singly and the control groups, or among the different treatments
on the standardized normed measures that contained different words than
the words that were taught’’ (p. 596).
In short, teaching words based on Venezky’s sequence of spelling
levels had no greater effect on spelling skill measured by a standardized
spelling test than teaching nothing, and this was true no matter what train-
ing emphasis (whole-word, phonemes, onset-rime) was employed.
Henry (1989) added a multisyllable component for the Latin and
Greek layers of the language to a phonics program developed by her and
Calfee (Calfee and Henry 1985). The phonics program, Project Read, had
been ongoing in 14 classrooms prior to this study. This is a letter-driven

3. The authors reported ‘‘no significant differences’’ between the different

instructional approaches, but this was not the message in the abstract or
the closing section. Abstract: ‘‘Combining whole word and onset-rime training
is most effective in achieving transfer of the alphabet principle across word
context.’’ Final summary: ‘‘Explicit training in whole word and onset-rime
connections led to greater transfer of the alphabet principle to untrained
monosyllabic words’’ (p. 603).
 69

program very similar to Venezky’s. Eight classrooms of third and fifth

The Structure of the English Alphabet Code

graders used Henry’s new addition (READ PLUS) along with the Project
Read lessons. A control group was taught with a basal-reader method.
READ PLUS consisted of 5 lessons 30–45 minutes long, spread out over
5 weeks (2.5 to 4 hours). The lessons included Latin root words, prefixes,
and suffixes, grouped by common spellings, with several examples of each.
Unfortunately, the results of this study are unclear. The reading and
spelling tests were not standardized tests, and reliabilities were only ‘‘fair.’’
Also, because the reading and spelling test scores were combined in the
data analysis, the separate impact of the training on reading and spelling
cannot be determined.
The gain scores (raw scores) for ‘‘reading/spelling’’ over the school
year were 6.2 points (controls), 6.7 (Project Read), and 9.0 (READ
PLUS). READ PLUS children had significantly higher gains ( p < :05).
However, when scores on the reading/spelling test were converted to per-
cent improvement over the year, gains were not impressive and groups did
not differ. Furthermore, the Project Read children scored identically to
the children in the basal-reader group, suggesting this phonics program is
not particularly effective.
A study using a spelling program based on my analysis (McGuinness
1997a, 1998a) was carried out in New Zealand (Smith 1999) on teenagers
13–14 years old. All had extreme reading and spelling delays as measured
by standardized tests. They were between 2:6 and 5 years below chrono-
logical age on word recognition, and 5 to 8 years below on spelling dicta-
tion. Everyone was below the 25th percentile on reading comprehension.
The students received 12 hours of one-on-one instruction in spelling up
through the Latin layer of the spelling code using a corpus of over 6,000
words.4
Students with such extreme profiles are hard to shift, especially in
only 12 hours, and results were very encouraging. Average standard-score
gains on the Woodcock Reading Mastery subtests and the Test of Written

4. The instruction involves lessons designed around an English and a Latin

spelling dictionary, fail-safe worksheets, stories featuring a phoneme and its
multiple spellings, and spelling dictation.
 70

Spelling–3 (TWS-3; Larson and Hammill 1976) were as follows: word

Chapter 3

recognition improved by 10.6 standard-score points to a final score of 88,

word attack by 18.6 points to 100, word comprehension by 14.8 to 104,
passage comprehension by 10.4 to 96, reading comprehension by 12.8 to
100, and spelling by 9 to 90 for regular words and by 7 to 78 for irregular
words. Gains of this magnitude for this age range are equivalent to an
average increase (all tests combined) of 27 percentile points, bringing the
students up to the mid-average range (50th percentile) overall.
One of the most important results of this study is that a program
designed specifically to teach the advanced spelling code had an even
greater effect on decoding and reading comprehension than on spelling.
This is proof (though tentative due to the small sample size) that teaching
the complete spelling code the right way around, from sound to print,
allows the code to reverse. What people can spell, they can automati-
cally read. The program does not directly teach multiple decodings (code
overlaps) of the same spelling (soup, out, soul ), yet children learned them
anyway. This means that Venezky’s concern over the lack of symmetry
between decoding and encoding is unfounded. The code is asymmetric
only if you teach it the wrong way around. These results support and
extend the findings of Ehri and Wilce (1987) and Uhry and Shepherd
(1993) that children score higher on reading tests when they are taught to
spell than when they are taught to read!
Spelling scores did not improve at the same rate as reading and com-
prehension scores. As noted earlier, spelling relies on recall memory
(no clues or prompts available), while reading requires only recognition
memory. It will take longer than 12 hours for spelling patterns to leave
a stable trace in long-term memory. However, spelling scores are rarely
impacted even in long-term interventions of this type (National Reading
Panel, 2000; see chapter 5, this volume), so a gain of 9 standard-score
points is remarkable.
There are a number of problems with these studies. Much more re-
search is required to assess the value of spelling/reading instruction based
on an analysis of the spelling code. Smith’s study used a small sample and
there was no control group. Gains in standard scores were compared to
pretest scores and to norms on the test, but there is always the chance that
these norms may not be truly representative of the children in the study.
 71

Berninger et al. used a large sample, well-matched groups, plus a con-

The Structure of the English Alphabet Code

trol group. While transfer of training did occur for words nearly identical
to those that had been taught, it did not generalize beyond these words
to standardized tests. Furthermore, children taught a whole-word (sight-
word) strategy did just as well as everyone else, suggesting the results
were due to exposure (visual memory) and not to any particular method.
Henry’s results for READ PLUS were encouraging, but reliable (and
separate) reading and spelling tests are essential to allow comparisons to
other research.
We return to this topic in chapter 9, which reviews the evidence on
how children learn to spell using the traditional spelling methods that have
nothing to do with the structure of the spelling code. In these methods,
children are given lists of spelling words chosen at random, or that feature
structural patterns like word families that are irrelevant. Given the fact
that our spelling system is highly complex, how children learn to spell
under these circumstances is somewhat of a mystery. Recent studies have
provided some surprising information about how the brain is able to code
the probability structure of our spelling system nonetheless.
At this point we move on to reading instruction.
 4
HOW TO TEACH READING:
LESSONS FROM THE PAST

In 2000, several committees of scholars, convened as the National Read-

ing Panel (NRP), released a report reflecting an extensive review and
analysis of the research on reading instruction. The topics were reading
programs, phonological-awareness training programs, and instructional
methods designed to improve reading fluency, vocabulary, and compre-
hension. This formidable undertaking provides an enormous service to
educators and to the research community by screening a database of
thousands of papers for scientific merit, and by assessing the major out-
comes of the better studies.
My goal in the next two chapters is to compare the lessons from the
past with the research findings over the last 40 years. To begin, I want to
summarize what we have learned from the comparative analysis of writing
systems and the structural analysis of the English alphabet code, and work
forward in time. This will allow us to establish the basic parameters for an
effective reading program—a prototype—and to match these parameters to
the most successful programs identified in the NRP survey.

Lessons from the Past

We begin with what we have learned from the attempts to design a func-
tional writing system.

Lesson 1 The ancient scholars who designed the first writing systems be-
gan by using the same logic people had used to set up accounting systems:
one symbol for each word. This attempt failed quickly, irrevocably, and
universally. Scholars were forced to abandon the word for a sound unit
below the level of the word due to the extreme limitations of human
memory for mastering sound-symbol pairs. The average person has an
 74

upper memory limit of about 2,000 of these pairs, no matter which sound
Chapter 4

unit is chosen. This is an ultimate limit, a memory ceiling, which does not
improve with further training. Thus, this type of paired-associate learning
obeys the law of diminishing returns.
We have abundant evidence, dating back over 5,000 years, that a
whole-word, meaning-based writing system does not work, never did
work, and never will work.
The evidence from the NRP report provides incontrovertible support
for this conclusion. Whole-word teaching methods lead to consistently
lower reading test scores than methods that emphasize phoneme-
grapheme correspondences.
There is another problem with whole-word methods: they are highly
misleading. Children and adults alike are strongly biased in favor of
linguistic meaning. A whole-word (sight-word) reading method is very
appealing to children, especially because memorizing letter sequences and
‘‘word shapes’’ is quite easy early on. This gives children the false im-
pression that they are learning to read. But it is just a matter of time be-
fore this strategy begins to implode. Whole-word memorization starts to
fail toward the end of first or second grade, depending on the children’s
vocabulary and visual-memory skills, and unless they figure out a better
strategy, their reading will not improve (McGuinness 1997b).

Lesson 2 Only four types of sound units have been adopted for the writ-
ing systems of the world: the syllable, the CV diphone, consonants only,
and the phoneme. Which unit is chosen depends on the phonotactic or
phonetic structure of the language. These sound units are never mixed. If
more than one unit was adopted, this would make the writing system
highly ambiguous and extremely difficult to learn.1
A reading method must teach the sound for which the writing system
was designed, and no other unit. This rules out ‘‘eclectic’’ or ‘‘balanced’’
reading methods that teach whole words, syllables, syllable fragments like
rhyming endings, and phonemes. This is tantamount to teaching four

1. Japan is an exception since they added the Roman alphabet to their two CV
diphone scripts.
 75

writing systems simultaneously. The NRP report confirms this also.

How to Teach Reading: Lessons from the Past

Methods that introduce a variety of phonological units, or the wrong
phonological units, are much less successful than methods that stick to
phonemes.

Lesson 3 Ancient scholars avoided adopting the phoneme as the basis for
a writing system. Yet there is abundant evidence they were well aware of
the phonemic structure of their language and used this to set up the writ-
ing system and to design dictionaries. The fact that phonemes are harder
to isolate or segment than larger phonological units appears to be the
primary reason, perhaps the only reason, why every civilization today does
not have an alphabetic writing system. For this reason, it makes sense to
teach children to segment (and blend) phonemes if they have an alpha-
betic writing system. A method that includes this type of instruction ought
to be more effective than one that does not. This too is confirmed by the
NRP report. Reading methods that include phoneme-analysis tasks are far
more successful than methods that do not.
Here is the message so far: If you have an alphabetic writing system, you
must teach an alphabetic writing system. There is no use pretending you have
something else.

Lesson 4 The English alphabet code is highly opaque. There are two
ways to mitigate this problem and help children manage this complexity.
The first is to ensure that children understand the direction in which the
code is written, from each sound in speech to its spelling. For an opaque
writing system to function as a code, it must be anchored in the finite
number of sounds of the language and not in the letters or letter combi-
nations of the spelling patterns. Unless this is done, the code nature of the
writing system is obscured, and the code cannot reverse. A code that can-
not reverse will not function as a code.
By contrast, transparent codes are relatively easy to learn and to teach.
The second way to help children master an opaque alphabet code is to set
up a temporary ‘‘artificial transparent alphabet’’ or basic code. This reveals
the nature or logic of an alphabetic writing system, making it ‘‘transpar-
ent’’ or accessible to a child. It also provides a platform, a foundation,
from which the code can expand, and spelling alternatives can be pegged
 76

onto the system without changing the logic. Reading programs based on
Chapter 4

these two principles ought to work better than programs that are not. So
far, no analysis, including the NRP review, has focused on these possibil-
ities: a sound-to-print orientation, and teaching via an artificial transpar-
ent alphabet.
We begin this analysis here, and it continues in the following chapter.
It is of considerable interest to follow the history of these new ideas, and
track the programs that meet these guidelines from the nineteenth century
to the present time. This way we can evaluate how a prototype reading
program based on these principles fares in well-conducted research.

The Origin of the Artificial Transparent Alphabet

Scragg (1974) and Morris (1984) provide an interesting history of reading
instruction in Great Britain and of the attempts over the centuries to im-
prove the spelling code. As noted in the introduction, the major break-
through in how to teach our opaque system derived from Isaac Pitman’s
shorthand, which he designed in 1834 using common letters plus invented
letters for the leftover phonemes. He collaborated with a colleague, A. J.
Ellis, to develop a classroom reading program based on this new alphabet.
This was subsequently redesigned by Ellis in a program called Glossic in
1870. The final transformation occurred when Nellie Dale, a classroom
teacher at Wimbledon School for Girls, wrote a more practical and user-
friendly version, using the most common spelling for each of the pho-
nemes in English and no special characters—in other words, what I call a
‘‘basic code.’’ This is the first true classroom program based on a basic
code and taught from sound to print.
Dale’s program was set out in a teachers’ manual titled On the Teach-
ing of English Reading in 1898, and was expanded in 1902 (Further Notes on
the Teaching of English Reading). She described 83 innovative lessons that
could be taught to the whole class, and that ‘‘worked well for classes of up
to 70 children.’’ Sounds of the language were the first emphasis. Children
were taught to listen for a target phoneme in the initial, middle, and final
position in words, learned how phonemes were produced, and found out
what articulatory features they had in common. For example, consonants
with the same place of articulation (/b/, /p/) were identified as ‘‘brothers.’’
Children checked which one was voiced (vocal cords vibrating) by touch-
ing their windpipe.
 77

Once several phonemes had been taught, and children practiced

How to Teach Reading: Lessons from the Past

combining them, they were shown large, cutout wooden letters. As each
sound-letter correspondence was taught, a letter was hung on a frame at
the front of the room according to its articulatory features. A space was
reserved at the bottom of the frame for word building. Children copied
each letter on a blackboard inside the lids of their desks, saying the sound
as they wrote the letter. In other exercises, children moved about the
classroom ‘‘becoming’’ a phoneme or a letter. Children made words by
standing in a row (left to right) according to the order of their sounds (/k/
/u/ /p/), or their letters (c-u-p), while the class blended the phonemes into
a word.
There were no standardized reading tests and no statistics in the early
twentieth century, so the success of Dale’s program can never be known.
Based on what we know today, her program would have been highly suc-
cessful. Teachers certainly thought so at the time, and it sold well on both
sides of the Atlantic.
These innovations were soon to be eclipsed by the expansion of
universal education in the early twentieth century. Reading instruction
was hijacked by newly minted education gurus in collaboration with
the fledgling educational publishing houses. Dale’s program, and other
phonics-type programs, began disappearing from classrooms during the
1920s, to be replaced by whole-word (sight-word) methods of the ‘‘look-
say’’ or basal-reader variety.
While a few phonics-type programs survived, lingering on in some
parochial and private schools, the basic code did not. It resurfaced 40 years
later in three different reading programs. One owed its origin to family
ties. Pitman’s grandson, Sir James Pitman, designed the initial teaching al-
phabet or i.t.a. (Pitman and St. John 1969). Pitman believed that digraphs
(sh in ship) were confusing for beginning readers. For this reason, i.t.a.
used an artificial (temporary) script with special symbols for digraphs.
The two other programs were American, though one had obvious ties to
Dale, linking phonemes to an analysis of their speech patterns and mouth
movements. This method, Auditory Discrimination in Depth, was designed
as a remedial program by Pat and Charles Lindamood in the late 1960s
(Lindamood and Lindamood 1969). The third method had some name
changes in its history. It began life as Hay-Wingo, named after its authors
Julie Hay and Charles Wingo. This was the program used by Rudolf
 78

Flesch in Why Johnny Can’t Read (Flesch [1955] 1985). It was later revised
Chapter 4

and expanded to include stories and exercises by McCracken and Walcutt

(1963), and since then it has become known as the Lippincott program,
after the publisher. We will look at the success of these programs in the
following section.
At this point, I need to say a few words about phonics. By the 1960s,
the term phonics could mean any method that had to do with teaching let-
ters, letter names, and/or letter-sound correspondences. It could refer to
such disparate forms of reading instruction as teaching the 26 names and
sounds of the letters of the alphabet, or a linguistic-phonics program as
sophisticated as Dale’s. Phonics came to have quite different connotations
in different countries. In the United Kingdom, for example, phonics is
more likely to mean something similar to Dale’s program. In the United
States, phonics commonly refers to letter-driven methods, like the one set
up by Webster (see McGuinness 1997c, 1998b). For this reason, the so-
called methods wars, which began in earnest in the 1960s, cannot really be
characterized as a debate between phonics versus whole-word methods.
Framing the problem in these simplistic terms is not helpful.
Early in the twentieth century, decisions about what should go into
a reading method and how it should be structured came to be based on
partiality to the alphabet principle or partiality to the logograph principle,
and subsequently sparked the reading ‘‘wars.’’ Prior to the 1960s, there
was no way to determine whether one type of reading method worked
better than another. Logic alone could not resolve this issue. There was
little understanding of how writing systems work. There were no reliable
tests to measure the success or failure of a teaching method, and no sci-
entific or statistical tools. Despite the rhetoric and the rancor, there was
not a scrap of evidence that a particular phonics method or whole-word
method worked better in the classroom or anywhere else.
We needed solid research on the content of reading programs and on
which components were most effective for young children. We had to wait
a very long time. The first major efforts began in the mid-1960s, and we
turn next to the monumental studies from that decade.

Doing Research on Reading Methods Is Not as Easy as You Think

Prior to the 1960s, educators assumed that a reading method was au-
tonomous, so much so it would override everything else: the school, the
 79

children, the classroom, the teacher, the parents, and so forth. Thus, it

How to Teach Reading: Lessons from the Past

seemed reasonable to compare classrooms using method X to classrooms
using method Y. If, on the whole, children taught with method X had
better reading test scores, then everyone could rejoice and switch to X.
Several problems came to light as these studies accumulated.
The first problem was that results from one study to the next were
contradictory. Method X would work well in one school but not in
another. It would work well in one classroom but not in another in the
same school. As people searched for reasons, they became aware that the
children, the teacher, the principal, the parents, and the reading tests
mattered more than they thought. People realized that baseline measures
had to be taken before methods X and Y were compared, because children
enter school with widely varying skills.
The measures themselves had to be valid and reliable. The tests had
to accurately reflect the types of performance you wanted to measure, and
provide information on a broad range of aptitudes. For example, reading
tests should include measures of decoding, spelling, comprehension, and
fluency. The tests should be normed and standardized. They should be
reliable, so that when they are given at different times (on alternative
forms), students get similar scores. These elements, if uncontrolled, can
add up to a multitude of ‘‘confounding variables,’’ which is a fancy way
of saying that the results are not a consequence of the method but of an
unknown factor or factors.
A persistent question dogged many thoughtful people. If method X is
‘‘better,’’ then how much better is ‘‘better’’? From a practical standpoint,
is an average reading score of 26 compared to an average score of 19 on a
reading test sufficient for the entire school district to scrap everything, train
teachers, and purchase new materials? Is an improvement of two months
above age or grade-level norms worth the bother and expense? Until the
1960s, data were reported in simple means or average scores, because sta-
tistical tools were not well known. The first statistics textbook for the be-
havioral sciences, Psychological Statistics by Quinn McNemar, was published
in 1949. McNemar commented in the preface that he wrote the book for
his students, because there was nothing else available. Of course, statistics
is only part of the answer to the question ‘‘how much better is better?’’
There is another problem. Reading programs are not ‘‘pure.’’ They
are sequenced differently, take various lengths of time, and contain a
 80

variety of activities. If method X was found to be better than method Y

Chapter 4

(even highly significant on powerful statistical tests), what exactly was it

about method X that made the difference? Was it the emphasis on mem-
orizing sight words, the encouragement to engage in shared reading, the
time spent chanting letter names, a focus on the sounds of English, the
stress placed on the code nature of a writing system, or the emphasis on
guessing words in context? This is still the key issue in reading research,
though we have some important answers (see the next chapter).
For all these reasons, research on reading made little headway until
the 1960s. In recounting this history, Graves and Dykstra (1997) thought
the major catalyst for a more rigorous approach to research was the pub-
lication of Why Johnny Can’t Read. It certainly stirred the pot. But Flesch’s
book appeared in 1955, and the pot had long since stopped boiling before
anything happened. Some rumblings were heard at a meeting of invited
participants at the National Conference on Research in English in 1959.
And from one person’s animated concern, a famous study emerged.

Jeanne Chall’s Road Map

In 1961, Jeanne Chall began an intensive quest to provide a thorough
analysis of reading programs and teachers’ performance in the classroom,
and to do a survey of research on reading. The project took three years
and resulted in the publication of the book Learning to Read: The Great
Debate in 1967. The investigation involved interviews with 25 authors and
editors of reading programs, an analysis of 22 reading programs, over 300
hours of classroom observations in the United States, England, and Scot-
land, and a review of the research literature.
Chall was trying to connect the opinions of authors and curriculum
developers with the actual program content in the manuals and materials,
with what teachers did in the classroom, and hopefully with children’s
reading achievement. At this time, about 95 percent of U.S. classrooms
were using basal-reading programs. Basal-reader authors tended to imitate
one another, even holding the same contradictory set of assumptions. On
the one hand, they agreed that children should learn to read with words in
their vocabulary, memorizing these words by sight. On the other hand,
the bulk of the lesson was devoted to exploring the meaning of the new
words in the lesson, words the children already understood. These authors
believed that reading should precede writing and viewed reading and
 81

writing as entirely separate processes. This meant that for most beginning

How to Teach Reading: Lessons from the Past

readers, from the 1920s to the time of this study, the encoding aspects of
the writing system were never mentioned. Letters and words were neither
traced nor copied, spelling was not taught, and there was no creative
writing to be seen. Most programs did not introduce writing or spelling
until the second or third grade.
When basal-reader authors were asked why they thought some chil-
dren failed to learn to read, they provided a laundry list of amorphous
causes unsubstantiated by any facts: dysfunctional families, an unsup-
portive culture, substandard schooling, poor teachers, lack of funding,
overcrowded classrooms, the children’s emotional and psychological well-
being, physical disabilities, and so forth. Of course, the authors whose
methods were out of favor said ‘‘methods’’ were the primary problem.
Spelling reformers focused on the spelling code, claiming that children
would continue to fail until the spelling code was revised.

A Program Analysis
Chall did an analysis of program content and sequence, plus an in-depth
account of three programs. The basal-reader programs as a group featured
very few words in their stories and ‘‘readers.’’ Neither spelling regularity
nor word length was considered important. Letter-sound correspondences
were introduced late, and children learned slowly, gradually being exposed
to the word from whole to part, a method known as analytic or intrinsic
phonics. These lessons do not begin until around second grade and con-
tinue for several years. Details on two basal-reader programs are pre-
sented in the following section.
The programs described as synthetic phonics were much more variable.
Chall described these methods as having reading vocabularies based on
spelling regularity and word length. Early words were simple and short,
gradually increasing in complexity. The emphasis was on mastering letter-
sound correspondences, and new words were introduced rapidly. To this
end, children were taught to blend and segment sounds in words and to
connect sounds to letters.
A third group, the linguistic programs, had several features in com-
mon. First, they were written by linguists. There was little emphasis on
meaning for obvious reasons, and the focus was on the alphabet, especially
letter names. The reading vocabulary consisted of predictable, short words
 82

set out in lists that abounded with word families: cat, rat, sat, mat. Chil-
Chapter 4

dren were expected to spontaneously detect the sound-symbol regularities

from these spelling patterns and read words as wholes, not to ‘‘sound them
out.’’ Oddly, the linguistic aspect of the code was the most ignored and
most distorted. Sounds of the language were not taught. No segmenting
or blending was allowed. If children encountered a new word, they were
to spell it aloud with letter names, then read it as a whole. Illustrations
were minimal or absent. Visually, these programs look remarkably like no-
nonsense spellers, and if Noah Webster’s speller had been among them, it
would have fit in perfectly. It is difficult to understand what is ‘‘linguistic’’
about these programs.
Two programs in the synthetic group fit the prototype—a sound-to-
print orientation and a basic code. These were the Initial Teaching Al-
phabet i.t.a. and the Lippincott program. As noted above, the i.t.a. uses a
special script to represent 44 phonemes of English, and children are
obliged to transition to conventional letters and spelling. The i.t.a. pro-
gram has a strong emphasis on writing. Writing begins as soon as possible,
and creative writing is especially encouraged. The Lippincott program will
be discussed below.
Chall studied three programs in depth from the prereader phase (kin-
dergarten) through third grade. Two were basal-reader programs: Scott-
Foresman and Ginn, the most popular programs in the United States.
The Scott-Foresman program will be familiar to anyone who remembers
Dick and Jane. The third program was the Lippincott program.
The basal-reader programs were very much alike. The stories gradu-
ally increased in length through repetition. New words were introduced
slowly at a constant rate of 1 to 2 words per 100 running words in a story
across all grades. By the end of third grade, children had seen about 1,500
words. Teachers were instructed to focus on meaning as well as on the
visual elements of the word—its length and its shape, or pattern of
ascenders (b, d, f, h, l, and t) and descenders ( g, j, p, q, and y). The Ginn
program actually had lines around new words to emphasize their shape.
Illustrations were numerous and colorful, especially in the early books,
with 6.5 pictures per 100 words.
The word count of what the teacher was supposed to say in the par-
tially scripted lessons, versus what the children were allowed to contribute,
was always overwhelmingly in favor of the teacher. Despite this verbal
 83

barrage, Chall could not find a single statement about the fact that the

How to Teach Reading: Lessons from the Past

alphabet is a code in pages and pages of teacher instructions in either
program. Comments on sound-symbol relationships, or segmenting and
blending, were absent entirely until the end of first grade/beginning of
second grade.
In excerpts from the so-called phonics lessons in the basal-reader
manuals, it was obvious that the authors had no idea how the alphabet
code works. Elements were introduced at random, and the statements in
the lesson make no sense. In a grade 2 example, children were supposed to
tell the teacher ‘‘which i and which a’’ (letter names) they heard in the
words sit, night, bird, hand, rain, car, and ball. This assumes that children
‘‘get it’’ that letter names stand for the letters, and that letters, in turn,
represent sounds.
The Lippincott program was an extreme contrast. Each lesson intro-
duced a phoneme and its most common spelling, plus a story with the
target sound appearing in all positions in the word (no word families).
Lessons began with the five ‘‘simple’’ (short) vowels and moved on to
consonants. Consonant blends came next, followed by vowel and conso-
nant digraphs. Toward the end of first grade, vowel-spelling alternatives
were taught, including phonograms (ough in bought). Alternative spellings
and code overlaps (out, soup) were sometimes introduced together to il-
lustrate certain problems with the code.
There was a strong emphasis on teaching the relationship between
sounds and letters, and connecting this immediately to words. Twenty-
two percent of the lesson time was devoted to writing, compared to zero
in the basal programs. Children copied words and were given spelling
dictation. Reading aloud was emphasized.
This approach led to a rapidly increasing reading vocabulary. By the
end of first grade the children had been introduced to over 2,000 words, ten
times as many words as the basal-reader children had seen at this stage,
and 500 more words than they will see by the end of third grade. Because
the Lippincott children were taught to decode, whereas basal-reader chil-
dren were not, their skills transferred easily to decoding unfamiliar words.

Classroom Observations
The third phase of Chall’s investigation included visits to over 300 class-
rooms in kindergarten through third grade. The observations consisted of
 84

subjective impressions of elusive qualities like ‘‘mood of the classroom,’’

Chapter 4

‘‘restlessness,’’ and ‘‘interest’’ on the part of the children, as well as ‘‘mo-

mentum,’’ ‘‘support,’’ and ‘‘expectations’’ on the part of the teacher. Chall
opted for variety (300 classrooms in several months) over stability (fewer
classrooms studied repeatedly), to study the impact of lots of different
reading methods on the children’s enthusiasm for learning them, and on
their success in learning to read. However, the second goal turned out to
be a fruitless quest. In some cases, school records and test scores were not
made available. In other cases, records were incomplete, or the tests used
by different schools measured different things, or children were tested at
different ages. In short, testing was haphazard and differed to such a de-
gree that test scores could not be compared.
What remained was Chall’s impressions and insights. She observed
that a good teacher can make children enthusiastic about a lesson that an
adult would think was dull, and a poor teacher can make children listless
or fidgety during a lesson that should have been interesting. Many of
Chall’s comments were essentially descriptions of good teaching. Good
teachers have control of the class, get the children excited about the les-
son, push them to the limits of their confidence and no further, move the
lesson along at a good pace, confront and solve problems quickly and
fairly, and involve as many children as possible in the lesson. No surprise
here. But one wonders, are these qualities sufficient to make a poor read-
ing program succeed, and their absence sufficient to make a good reading
program fail? Without objective data on the children’s reading perfor-
mance, these questions cannot be answered.
One of Chall’s most important discoveries was that teachers tend
to be eclectic. If teachers are asked, or decide, to change to a new pro-
gram, they do not abandon old activities and lessons from programs they
enjoyed teaching or felt were important. Nor do they abandon their phi-
losophies. This can create a situation where elements from different pro-
grams with contradictory logics cancel each other out, such as an emphasis
on decoding and an emphasis on memorizing the shapes of words. This
has profound implications for classroom research, because it means that
there will always be an overlap of different methods, depending on the
teachers’ training and on how many different methods they have been
asked to teach. It seems highly unlikely a particular method will ever be
 85

taught ‘‘pure,’’ unless a teacher really commits to it, or the program is so

How to Teach Reading: Lessons from the Past

effective it succeeds in spite of the odds.

A Research Summary
The last component of Chall’s review was an analysis of classroom re-
search on reading methods. All the comparison studies between look-say
and phonics were published between 1912 and 1940, and will not be dis-
cussed further. The findings from the studies were summarized in a series
of tables. The method that produced the best outcome on tests of word
recognition, comprehension, spelling, and so on was indicated by its
initials. Thus, if look-say did ‘‘better’’ on a particular test, an LS was
reported. (Synthetic phonics ¼ SP; intrinsic or ‘‘basal’’ phonics ¼ IP.)
Words and phrases like better, higher, and ‘‘had an advantage’’ were used
to describe these outcomes. Better was not defined and no numerical data
were provided.
What kind of data did Chall rely on? The majority of the studies
reported outcomes in average scores. It appears that no statistical analysis
of the data occurred before the 1960s, as shown by the fact that ‘‘no dif-
ference’’ was indicated by E (for equal) in all studies prior to the 1960s,
but as NS (not significant) from the 1960s on. The tables are peppered
with SP’s, showing a huge ‘‘advantage’’ for synthetic phonics. In the
comparisons listed from first through sixth grade, SP was ‘‘better’’ 68
times, IP only 11, and there were 34 draws (29 E’s and 5 NS’s), giving the
impression that synthetic phonics was the overwhelming winner. This is
very misleading.
The absence of numerical data, the reliance on average scores to de-
termine better, and the failure to define what better means, make it impos-
sible to draw any conclusions from this review. Chall leaves us with a basic
contradiction. If synthetic phonics is vastly superior, as her tables seemed
to show, then the quality of teaching is far less important than the
method. In fact, teaching skill pales into insignificance compared to the
overwhelming superiority of synthetic phonics according to the tables. Yet
Chall’s classroom observations indicated that the teacher’s ability can
override the method, at least in making the lesson exciting and stimulat-
ing. On the one hand, the message from the classroom is that the teacher
matters as much as or more than the method. On the other hand, Chall’s
 86

presentation of the research seemed to show that the method matters far
Chapter 4

more than the teacher. Which conclusion is true?

Chall mapped out the important issues and asked the fundamental
questions, many of which have still not been answered and need to be. But
the form her research took and her conclusions from it failed to persuade
people that we had been doing something terribly wrong and needed to
change our ways. As Chall began her study, another project was underway
that had precisely these goals.

The Cooperative Research Program

In 1963 a dedicated group of scholars, including Chall, Bond, Clymer, and
Durrell, convinced the U.S. Department of Education to fund an ambi-
tious project that would settle the basal-reader/phonics debate once and
for all. This project was called the Cooperative Research Program in First-
Grade Reading Instruction, and the project directors were Guy Bond and
Robert Dykstra from the University of Minnesota. The project was
funded in 1964. The first published report appeared three years later
(Bond and Dykstra 1967) and was reprinted in Reading Research Quarterly
(Bond and Dykstra 1997). There is no better place to begin an in-depth
analysis of quantitative research on reading than with this pioneering
study and all that it implies.
Project proposals were requested by the funding agency on the topic
of reading methods and their efficacy, and had to conform to one re-
quirement: since basal readers were used in 95 percent of classrooms,
every project had to compare a basal-reader program to something else.
The rationale went like this: if ‘‘Johnny can’t read,’’ is this the fault of
those ubiquitous basal-reading programs? There was actually no hard
evidence that ‘‘Johnny’’ could or could not read, and neither Flesch nor
Bond and Dykstra had a word to say about student test scores. The issue
was more subjective. It had to do with the war between the whole-word
enthusiasts and the phonics enthusiasts.
A large number of proposals were received, and 27 projects were
funded. Probably nothing on this scale will ever be seen again. The key to
the success or failure of the entire enterprise was the degree to which it
could be ensured that all project directors used the same measures to the
same standards. The study proceeded as follows. First the world was di-
vided up by types of reading programs. Everything that anyone could
 87

Table 4.1

How to Teach Reading: Lessons from the Past

Comparison of five methods: Number of Projects, subjects, Classrooms, and means
Group testing
Individual
No. of No. of testing
Methods Projects N classes means N
Basal 1,038 49 98 149
vs. 5
i.t.a. 1,055 48 96 163
Basal 722 33 66 161
vs. 4
Basal þ Phonics 1,002 42 84 204
Basal 1,523 61 122 138
vs. 3
Language Experience 1,431 60 120 134
Basal 597 25 50 120
vs. 3
Linguistic 760 31 62 146
Basal 525 24 48 97
vs. 3
Lippincott 488 23 46 94

Total N 9,141

think of that might affect learning to read was measured as precisely as

possible. The clock was started and time ran on without interference for
140 classroom days (7 months) of first-grade instruction. Several reading
tests were given. Table 4.1 sets out the overall structure of this study.
This study is highly complex. Classrooms are nested within Projects,
Projects within Methods, and Methods within the Study as a whole.
(I capitalize the terms ‘‘Study,’’ ‘‘Method,’’ ‘‘Project,’’ and ‘‘Classroom,’’
which represent the variables in the research design, to distinguish them
from their generic form.) Here is a glossary to help keep these compo-
nents straight.

Study. Refers to the Study as a whole, including all Methods and Projects.
Method. The particular Method being compared to basal-reader in-
struction. Five different Methods were investigated.
 88

Projects. Each Method comparison included several Projects run by

Chapter 4

different people in different geographic locations.

Classrooms. Each Project included a large number of Classrooms.

Data Collection
Baseline Measures Demographic data were collected on the children, the
teachers, the community, and the school. For the children, these were age,
sex, months of preschool experience, and days absent. Teacher variables
included sex, age, degrees, certification, years of experience, marital status,
number of children, attitude toward teaching reading, days absent, and
supervisor-rated teacher effectiveness. Community-/school-based infor-
mation included median education of adults, median income, population,
type of community (urban, rural, and so on), classroom size, length of
school day/year, per-pupil costs, and so forth.
These data can be dismissed rather quickly. None of the community
and school variables were more than modestly correlated with the chil-
dren’s reading test scores. ‘‘Teacher experience’’ was the only marginally
relevant teacher variable (r ¼ :30). Sex was found to be a strong predictor,
and for this reason, it was included as a variable in all further statistical
analyses.
The children took a battery of tests thought to be predictive of sub-
sequent reading skills. These included auditory- and visual-discrimination
tasks, tests of letter-name knowledge, reading readiness, and IQ. When
this testing was completed at each school, the 140-day clock started
running.

Outcome Measures At the end of the 140 days, children were tested once
more on various reading tests. They were given the five subtests of the
Stanford Reading Test, which had just been renormed. This is a group-
administered test. The subtests are word reading (word recognition), para-
graph meaning (comprehension), vocabulary (receptive vocabulary), spelling
(spelling dictation), and word-study skills (tests of auditory perception as
well as decoding skill).
In addition to the Stanford Reading Test, four individually adminis-
tered tests were given to a subsample of children, randomly selected from
every classroom. The total number of children in this sample was 1,330.
The tests were the Gilmore Tests of Accuracy and Rate, the Fry Word
 89

List (decodable, regularly spelled words), and the Gates Word List (based

How to Teach Reading: Lessons from the Past

on word frequency, including many irregularly spelled words).

Correlational Analyses
The first question was whether scores on the baseline tests correlated with
(predicted) scores on the Stanford Reading Test seven months later.
There were complete data on 8,500 children. Correlations were computed
separately for each Method, and also for Basal-reader Classrooms com-
bined across all Projects and Methods.

Group-Administered Tests The best predictors of performance on the

Stanford tests were the Murphy-Durrell Letter-Name and Phoneme Dis-
crimination tests, and the Pitner-Cunningham IQ. In the phoneme test,
the child makes same/different judgments between two spoken phonemes
(/b/ versus /b/ or /v/ versus /t/). Correlations for the six Methods ranged
from .48–.55 for letter-name knowledge to .37–.48 for phoneme dis-
crimination and .32–.50 for IQ. However, these measures were correlated
with each other: letter-name knowledge correlated with the phoneme-
discrimination test at .50. IQ was correlated with letter-name knowledge
at .43, and with the phoneme test at .46. Had IQ had been statisti-
cally subtracted, it is unlikely the letter-name and phoneme tests would
have had much predictive power. The remaining baseline tests (Metro-
politan Reading Readiness, Thurstone visual discrimination tests) were
not strong predictors of reading skill. The correlations were positive but
low (.20–.35).
The subtests of the Stanford Reading Test were found to be strongly
correlated with each other in every case. Apart from vocabulary, these
ranged from .63 to .76 (X ¼ :67). This shows that, for first graders, all
aspects of reading skill—decoding, spelling, comprehension, and phonics
skills—are strongly related, and doing well on one predicts doing well
on the others. Vocabulary was correlated with the reading measures at
around .50.

Individually Administered Tests No correlations were carried out between

the baseline tests and scores on the individual tests (Gilmore, Fry, and
Gates). However, the individual and group reading tests were correlated
with each other. Apart from the Gilmore Rate of Reading test (correlations
 90

from .45 to .62), all other correlations were very high indeed, ranging
Chapter 4

from .61 to .86. Clearly, the group tests and the individual tests were
measuring the same skills. This shows that both the individual data and
the group data were valid and reliable measures of reading skill.
A particularly important result was the almost-perfect correlation
(.86) between the Fry Word test (a test of decoding regularly spelled
words) and the Gates Word test (a test of visual recognition of mainly
irregularly spelled words or sight words). This was surprising, because
people assumed, as many still do, that decoding and sight-word memori-
zation require different skills. Children trained to memorize sight words
(basal-reader groups) were expected to be good on the Gates and worse
on the Fry. Conversely, children taught to decode regularly spelled words
(phonics only/few sight words) were expected to do the opposite. These
patterns would produce low to moderate correlations. Instead, perfor-
mance on one test almost perfectly predicted performance on the other.
There is only one interpretation: children who are good decoders are
good decoders, no matter how words are spelled. Children who rely
mainly on visual memory (nondecoders) are not only poor decoders but
fare just as badly on sight words.

Comparison of Reading Methods

The major focus of the study was on the comparison of the different types
of reading programs to the basal programs. The Projects were reduced to
15 from the original 27, so that at least three Projects were using the same
method. The Methods were the Initial Teaching Alphabet (i.t.a.; five Pro-
jects), Basal þ phonics (four Projects), Language experience (four Projects),
Linguistic (three Projects), and Lippincott (three Projects).
The basal-reader programs were described earlier. The most fre-
quently used basal in this study was the Scott-Foresman series (Dick
and Jane). Details on the other Methods are provided in the following
paragraphs.
The i.t.a. uses standard and special letters to provide separate symbols
for 44 phonemes in English. There is a one-to-one correspondence be-
tween each phoneme and its symbol in all materials, which included
readers. Children are encouraged to do lots of writing and reading. Later,
children transition to conventional spelling. It was noted that many chil-
dren had not made this transition by the end of the first year.
 91

The Basal þ phonics condition added a separate phonics component to

How to Teach Reading: Lessons from the Past

the Basal-reader curriculum in half of the classrooms. The add-on phonics
programs were either Speech to Print, Phonic Word Power, or Phonic Exer-
cises, all of which are code-emphasis programs. Descriptions of these pro-
grams were not provided.
The Language experience program involves highly individualized one-
on-one interactions between teacher and child. The idea is to develop
reading and spelling ability directly from each child’s vocabulary. The
process begins by the child telling a story, which is copied by the teacher.
The child reads what the teacher has written, copies the words, and grad-
ually develops enough skill to write on her own, using whatever spellings
she can. Teachers listen to children read their own work, correct spelling,
and so forth. Later, children move on to books of their choosing.
The Linguistic programs were described earlier as being the least
‘‘linguistic’’ of any phonics program. These were the McGraw-Hill pro-
gram, and programs authored by Fries and by Bloomfield and Barnhart.
Both Fries and Bloomfield were linguists.
The Lippincott program was used in three Projects in this group. A
detailed analysis of the Lippincott program was presented in the previous
section.

Results from the Stanford Reading Test (Group Administered) At this point,
in this otherwise impeccably controlled research, several statistical blun-
ders took place that essentially voided most of the results. The authors
decided to use mean scores from each classroom instead of individual
scores, one for boys and one for girls. This reduces the data from 20–30
children in a classroom to a prototypic boy and girl. Next, the data were
analyzed with analysis-of-variance (ANOVA) statistics for the five Meth-
ods separately. Reducing data to means invalidates the use of ANOVA
statistics, because the mathematics is based on variances derived from nor-
mal distributions of individual scores, not on group means—hence its
name.2

2. There is currently a theory that the unit of instruction must also be the unit
of analysis. If the ‘‘treatment’’ is the whole class, then the measure must be the
 92

By substituting means for the children’s individual test scores, two

Chapter 4

things happened. First, the focus of the study changed. Now all the vari-
ance (variability in each outcome measure) was due to the variability be-
tween classrooms, instead of variability between children within classrooms.
This turns a study comparing children learning different Methods into a
study comparing Classrooms (teaching skill perhaps?). Second, this repre-
sents a huge loss of statistical power. Table 4.1 illustrates the contrast be-
tween the actual number of children in the study and the number of scores
used for the data analysis, a tenfold reduction (100 to 10).3
I will examine the Basal-reader-versus-i.t.a. comparison to illustrate
the kinds of problems this created. There were large main effects of sex
(girls’ better p < :001), effects of Method (i.t.a. better on the Word sub-
test, Basal readers better on Spelling), and effects of projects (some did
better than others). However, there were also highly significant Project 
Method (treatment) interactions on every test except Vocabulary. Some-
times Basal Classrooms did better and sometimes i.t.a. Classrooms did
better, depending on which Project they were in. This is the ultimate
muddle in methods research, the experimenter’s worst nightmare: ‘‘Now
you see it here, now you see it there, now you don’t see it anywhere.’’
The investigators tried to reduce or eliminate these interactions by
covariance analyses (ANCOVA) on the assumption that the erratic re-
sults occurred because children started out with different levels of skill.
ANCOVA helps to level the playing field by equalizing scores on the

whole class as well (mean scores). This is a doubtful practice, and, as far as I
am aware, unknown to Bond and Dykstra at that time. A full analysis of this
problem is presented in appendix 1.
3. The authors provide no explanation for why they reduced the data to class-
room means and analyzed the methods comparisons one at a time. When this
project was carried out, computers were physically big and computationally
small. The data and the program-code information went onto punch cards.
To compare 15 projects in one analysis requires a four-factor design with 300
cells! It is unlikely anything this complex could have been handled by com-
puters at that time. Mean scores may have been used for the same reason.
ANOVAs are simple to compute by hand (by calculator) if the data set is small
(a few hundred), but nearly impossible when numbers run into the thousands.
 93

baseline tests (letter-name knowledge, phoneme awareness, IQ, and so on)

How to Teach Reading: Lessons from the Past

and making corresponding adjustments to the Stanford Reading tests. If a
child did too well on the baseline tests, reading scores are adjusted down,
and vice versa.
However, covariance analysis must meet stringent requirements.
McNemar (1949) cautioned that covariance should be used only when
groups being compared cannot be randomly assigned, or if they have
something that will cause differences in performance between the two
groups. For example, differences in IQ will impact the ability to learn new
skills.
The Bond and Dykstra study in no way met these criteria:

1. The only skill that would directly ‘‘cause’’ subsequent reading perfor-
mance is initial reading performance, and this was not measured.
2. Correlates of reading scores cannot be used to infer causality, and in
any case the baseline measures are correlated with each other, so it is im-
possible to know what is causing what.
3. One valid covariate in this study was IQ, but IQ was not used as a sep-
arate covariate in the ANCOVA analyses.

The authors do not report what kind of data were used for the covariance
analyses, but the tables reveal that mean scores were used again, which
voids the use of ANCOVA statistics. (You cannot do covariance analyses
with mean scores.) The tables also reveal that grossly inflated degrees of
freedom were used on all analyses. (These problems are discussed more
fully in appendix 1.)
Despite these statistical manipulations, the Project  Methods inter-
actions did not go away. Something else was causing this effect. The re-
searchers decided to reanalyze the data for each Project separately (see
their table 23). Two Projects had strong across-the-board results favoring
i.t.a. over the Basal-reader groups. Three Projects essentially found no
differences between the two methods, except for spelling. Basal-reader
classes had significantly higher spelling scores. On closer inspection, the
explanation was obvious. Many of the children in i.t.a. classes were still
using nonstandard script, and scoring did not allow for this. However, it
was not at all clear why two Projects produced a strong i.t.a. advantage on
the remaining tests, and three Projects did not. The authors could not
 94

explain this result. Using means instead of individual data in the analysis
Chapter 4

could have produced this effect, or alternatively, it could be a real effect.

There is no way to tell.
Data from the other Methods comparisons (Basal þ phonics, Lin-
guistic, Language experience, Lippincott) were analyzed in the same way,
producing the same erratic mix for the same reasons. Because of the prob-
lems of data analysis raised above, I will not review these statistics further.
Instead, I have reassessed the data and will present the results later.

Individually Administered Tests Analyses of the individually administered

tests (Gilmore, Fry, and Gates) did use individual data instead of means.
This was good news, especially because there were more individual scores
than mean scores (see table 4.1, right column).4
Here are the results for the most discriminating of the individual tests,
the Fry and the Gates. The children using the Lippincott program were
superior on both the Fry and the Gates to the Basal-reader controls in all
three Projects (6 out of 6 comparisons: 100 percent at p < :01). i.t.a.
classess also did well. They were superior in 7 out of 10 comparisons (70
percent). Basal þ phonics were superior in 5 out of 8 comparisons (62
percent), Language experience in 2 out of 6 (33 percent), Linguistic in
none. Basal-reader groups were not superior on any of the 36 comparisons,
despite the fact that the Gates is a sight-word test and contains many of
the same words these children were taught to memorize.

A New Look at Old Data There are 92 tables in this report but no table
summarizing the entire project. In view of the inappropriate use of the
data in the statistical analyses, a summary of combined data from these
more than 9,000 children is in order. It will be far more informative and
more valid than its predecessor. I calculated the grand means across all
Classrooms within a particular Method. These results are shown in tables
4.2 and 4.3. The tables illustrate the number of children who contributed

4. The degrees of freedom were incorrect here as well. However, when I

compared their probability values to statistical tables using the correct degrees of
freedom, the probability values were accurate.
Table 4.2
Mean scores and grade-level equivalents on the Stanford Reading subtests
Grade- Para- Grade- Grade- Grade- Grade-
level graph level level level level
Word equiv- Mean- equiv- Vocab- equiv- equiv- Word equiv-
Group Reading alent ing alent ulary alent Spelling alent Study alent
Basal (1,028) 20.1 1.7 19.6 1.7 22.0 1.9 11.4 1.9 35.9 1.9
i.t.a. (1,055) 23.3 1.9 20.9 1.8 21.9 1.9 10.8 1.9 38.6 2.0
Basal (722) 19.0 1.7 16.7 1.7 20.2 1.7 8.7 1.7 32.8 1.7
95

Basal þ phonics (1,002) 20.9 1.8 20.5 1.8 21.1 1.8 10.8 1.9 35.3 1.8
Basal (1,523) 20.0 1.7 20.7 1.8 21.2 1.8 12.1 2.0 36.6 1.9
Language experience (1,431) 21.5 1.8 21.1 1.8 22.1 1.9 12.3 2.0 37.3 1.9
Basal (597) 19.1 1.7 19.2 1.7 21.5 1.9 10.8 1.9 36.3 1.9
Linguistic (760) 19.0 1.7 15.8 1.6 19.6 1.7 9.3 1.7 33.8 1.8
Basal (525) 19.6 1.7 19.6 1.7 22.2 1.9 10.8 1.9 36.1 1.9
Lippincott (488) 26.6 2.2 24.4 1.9 23.7 2.2 14.1 2.2 41.4 2.2
Basal total (4,405) 19.6 1.7 19.2 1.7 21.4 1.8 10.8 1.9 35.6 1.9

How to Teach Reading: Lessons from the Past

 96

Table 4.3
Chapter 4

Mean scores on individual tests

Gilmore Gilmore Fry Gates
Group N Accuracy Rate Word Word

Basal 149 23.3 59.0 7.4 13.3

i.t.a. 163 26.0 60.0 17.2 19.3
Basal 161 21.6 59.2 6.2 12.1
Basal þ phonics 204 23.5 59.9 9.9 14.5
Basal 138 18.9 52.2 5.9 12.1
Language experience 134 21.8 53.0 9.1 13.8
Basal 120 23.3 59.1 6.5 12.1
Linguistic 146 17.9 43.8 7.8 10.5
Basal 97 24.4 56.2 6.0 12.3
Lippincott 94 29.5 62.4 18.4 20.5

Basal grand means 22.3 57.2 6.4 12.4

to these means. With numbers this large, statistical tests are not necessary.
One can confidently assume a normal distribution, and the mean is the
most accurate measure of this distribution if standard deviations are low.
Fortunately, table 75 in the Bond and Dykstra paper provided the stan-
dard deviations for each subtest on the Stanford Reading Test for every
Method, and for all Basal-reader groups combined. These standard devi-
ations were computed correctly on the children’s individual test scores. All
were low and extremely consistent, indicating the excellent psychometric
properties of the Stanford tests as well as outstanding testing and data
collection. Thus we know we are dealing with normal distributions, and
combining means across Classrooms is a valid indication of what hap-
pened in this study. The table values represent only the nontransformed
(noncovaried) data.
Table 4.2 provides the grand means for the Stanford subtests, plus
grade-level conversions (in decimals, not months). The expected grade
level for these children at the end of the Project was 1.7 (first grade, eighth
month). The Basal-reader groups consistently scored at or near this level
across all measures. Because the Stanford tests had recently been normed
 97

on American children, 95 percent of whom received Basal-reader instruc-

How to Teach Reading: Lessons from the Past

tion, the match in table 4.2 between grade-level norms and Basal-reader
children’s test scores is no surprise. But it gives one a good deal of confi-
dence that the data are real. Table 4.3 contains the grand means on the
individually administered tests. Unfortunately, no conversion table was
provided for these raw scores, so it is not possible to translate them to
grade level or age.
The tables illustrate that, overall, the phonics-type programs worked
somewhat better than Basal-reader instruction, but not much better. The
main result is the clear and unassailable evidence that one method (Lip-
pincott) is consistently superior to the rest, something not reported in the
article. This method produced the highest scores across the board, in both
group and individual test results. The value 2.2 translates to six months
above grade level. This is in striking contrast to the small or nonexistent
effects of the other programs. There is something about the Lippincott
program that worked in the classroom in a way that the Linguistic pro-
grams and other phonics programs did not.
The obvious next step would have been to look in detail at the Lip-
pincott program. What was it about this program, and what was going
on in Lippincott classrooms, that worked better compared to other pro-
grams like the Linguistic programs, which were surprisingly unsuccess-
ful. Could this lack of success have been due to the emphasis on letter
names in the Linguistic programs to the exclusion of phoneme analysis
and phoneme-to-letter correspondences? Could it be due to the fact that
they were designed by linguists with no experience in the classroom? Did
the teachers or the children find any elements in these programs that were
especially good or especially confusing? Surprisingly, issues like these
were not addressed as a consequence of this report, and the Lippincott
program never received the attention it deserved.

What Really Happened in This Study?

Bond and Dykstra’s summary of the Methods comparisons tended to be
biased toward the statistical results that favored phonics or linguistic
methods. Comparing the Basal-reader and i.t.a. methods, they wrote:
‘‘The i.t.a. treatment produced superior word recognition abilities as
measured by the Word Reading subtest of the Stanford and the Fry and
Gates word lists’’ (1997, 414). This is true, but they did not report that
 98

this was the only Stanford test on which i.t.a. children excelled, nor did
Chapter 4

they mention that this effect did not occur in three out of the five Projects.
Rather than pursue these issues further, especially in view of the many
problems with data handling and statistical analysis, the following is a
summary of what this study demonstrated.

The Study Design and Measurement The investigators made an excellent

choice of baseline and outcome measures. The study also employed a
thoughtful (useful) set of demographic variables, which helped to rule out
a wide range of factors as contributing to reading performance. Finally,
there was superb control over this study from the top down to every
Project director and every test site. This led to extremely reliable test-
ing and highly robust data for such a vast undertaking. (The authors
commented on the excellent spirit of cooperation among the Project
directors.)

The Results Sex differences favoring girls were very large and consistent
and appeared on every test in every comparison.
Correlations between the baseline measures (IQ, phoneme discrimi-
nation, letter-name knowledge) and the Stanford Reading tests were
modest, with the best predictors accounting for about 25 percent of the
variance. However, there was a high degree of overlap (shared variance)
between these baseline measures that was not controlled.
The fact that phoneme discrimination predicted 20–25 percent of the
variance in subsequent reading scores was a new finding, and needed to be
followed up.
The correlations between group and individual reading tests were
very high, indicating good test administration and good validity.
There was unequivocal evidence that good decoders can decode both
regularly and irregularly spelled words with the same facility, and that
sight-word memorization does not work (see the scores of the Basal-
reader children on the Fry decoding test in table 4.3).
Classroom means were used in the ANOVAs for the Stanford tests,
and incorrect degrees of freedom were used on about 75 percent of the
analyses. As a consequence, grand means have greater validity. These
showed that children who were taught with the Lippincott method scored
six months above grade on nearly all the Stanford tests, and were superior
 99

to all other method groups on all tests, and there were large effect sizes on all

How to Teach Reading: Lessons from the Past

measures except vocabulary. These conclusions were strengthened by low
and consistent standard deviations on the test scores.
Bond and Dykstra provided 15 numbered paragraphs in their final
conclusions on the Methods’ comparison data. Many comments are ad hoc
and unwarranted by the findings, even if the statistics had been reliable.
However, paragraph 15 stands out as the most prescient statement in their
summary:

It is likely that improvement would result from adopting certain elements

from each of the approaches used in this study. . . . For example, the i.t.a. and
the Phonic/Linguistic [Lippincott] programs, both of which were relatively
effective, have in common a vocabulary controlled on sound-symbol regular-
ity, introduction of a relatively large reading vocabulary, and emphasis on
writing symbols as a means of learning them. It would be interesting to know
which of these elements, if any, are primarily responsible for the effectiveness
of the program. (1997, 416)

In paragraph 13, Bond and Dykstra wrote that these two programs
‘‘encourage pupils to write symbols as they learn to recognize them and to
associate them with sounds’’ (1997, 416).
This is the first time in the report that the reader has some idea of the
specific characteristics of the Lippincott program, and it is the first time
that Bond and Dykstra give any indication that they are aware of what
happened in this study. From what we know today, had these suggestions
been followed up with appropriate research, we might be one or two de-
cades ahead of where we are now, and we might have prevented the
unnecessary suffering of hundreds of thousands of children who have
struggled to learn to read. But these suggestions were not followed up.
One cannot help wondering why.

Missing Variables The problem with doing pioneering research is that

you miss things. Hardly any of the demographic, school, or teacher vari-
ables the authors thought to measure ahead of time were found to corre-
late with reading performance. Of course, this information is just as
valuable as finding out what does correlate. In this project, and in follow-
up publications (Dykstra 1967, 1968b), the authors pondered what they
 100

might have missed. In the conclusion to their study, they wrote: ‘‘Evi-
Chapter 4

dently, reading achievement is influenced by factors peculiar to school

systems over and above differences in measured prereading capabilities of
pupils’’ (Bond and Dykstra 1997, 415). They did not speculate on what
these factors might be. They suggested that teachers needed better train-
ing, though they did not explain why they thought this, or what it implied.
Throughout their 1967 paper, Bond and Dykstra commented that the
results were unstable because of large Project effects, and possibly because
of a novelty effect, arguing that new programs ‘‘undoubtedly influence’’
and were ‘‘likely’’ to have contributed to the better performance of chil-
dren in the classrooms using novel programs. Chall also held this view,
and added the insight that the teacher’s personality may contribute to
these results. Yet ‘‘novelty’’ cannot explain the dismal showing of the
Linguistic programs both in this study and in Chall’s observations, or
the absence of any strong effects for the Basal þ phonics approach or for
the Language experience program.

The Importance of the Teacher In a commentary on the reprint of Bond

and Dykstra’s 1967 article, David Pearson (1997) related a story told to
him by Dykstra when the original project was ongoing. One day, Dykstra
popped into a project teacher’s classroom unannounced. This teacher,
assigned to the Language experience method (individualized teaching),
was happily doing a phonics lessons with the whole class. When asked
about this, she explained that children needed phonics to be able to learn
to read. She obviously thought, or had ascertained, that her students were
not learning to read, at least to her standard, with the other approach.
This experience prompted Dykstra to raise the issue of ‘‘treatment
fidelity’’ in his article on the second-grade extension of this study (Dykstra
1968b). Were the teachers teaching the methods they were assigned to
teach to the same standard in the same way? Were they mixing methods
or leaving out elements? Dykstra wrote: ‘‘One of the most important
implications of this study is that future research should center on teaching
and learning situation characteristics, rather than method and materials’’
(p. 66). This echoes Chall’s observations on the success or otherwise of
different teachers using the same program, and underscores the fact that
teachers tend to be eclectic. However, research approaches emphasizing
 101

teachers and methods are not mutually exclusive. Both could be studied at

How to Teach Reading: Lessons from the Past

the same time.
The story about the wayward teacher is supported by Chall and
Feldman’s (1966) classroom observations that teachers’ reports of their
activities did not tally with their observational notes. Teachers do not
always do what you think they are doing, nor do they always tell you what
they are doing, so asking them to fill out forms or keep a record is no use
either. Also, teachers may not always know or remember what they did.
There are interruptions; a lesson that is going well will be extended;
something the teacher intended to do does not get done. There is, re-
grettably, only one solution: sit in the classroom and record what the
teacher does and for how long.
The teacher’s behavior in Pearson’s story also highlights a real ethical
dilemma. If a teacher sincerely believes that what she has been asked to do
in a research project is not working, she is entirely within her rights to
stop doing it. She is responsible for human lives. Nonreaders and poor
readers do not do well in our society. Anyone who has ever tutored poor
readers can attest to the amount of suffering they and their families have
endured. The teacher’s dilemma is similar to what occurs in medical/
clinical trials when some patients are allocated to the miracle-drug group,
and others to the placebo group.

The Impact of the Study

This study appeared to have a strong negative impact on future research.
Pearson (1997) wrote that the value of a scientific project is determined
by how much research it generates. He observed: ‘‘By that standard, the
First-grade Studies were a dismal failure . . . they, in conjunction with
Chall’s book, marked the end of the methodological comparisons in re-
search on beginning reading (at least until the 1990s)’’ (p. 431).
Of course, a study like this almost shouts out: ‘‘Now top this!’’ Per-
haps because this study was so complex, with its 92 statistical tables, which
only seemed to highlight the unpredictable outcomes, other researchers
were intimidated by it. This was also the last decade where research
funding flowed like water from a tap, and the 1970s saw a severe cutback
in support for research of all kinds. No study like this one ever appeared
again.
 102

Nevertheless, there are so many tantalizing leads here that it is sur-

Chapter 4

prising that no one was inspired to follow them up. Here are some of them:

Phoneme discrimination may be important in learning to read.

Being taught how to decode is more beneficial than being trained
to memorize sight words. Basal-reader children had serious decoding
problems.

A reading method with the following characteristics produced far larger

gains than all other methods in this study: sound-symbol regularity, a

controlled but large reading vocabulary introduced systematically, copying
letters, words, and phrases, saying the sound the symbol stands for, and
reading stories that target a particular phoneme.

What teachers do in classrooms can be unpredictable, matters a lot, and

needs to be investigated quantitatively in tandem with classroom research

on methods.

What We Have Learned about Research

It is easy to look back to 1967 and say: ‘‘Well, we don’t do things like that
any more.’’ We have powerful computers to handle complex data sets like
these, and all is well. This perception may be comforting, except it is not
accurate. In some ways the problem is worse because we have become
complacent. We assume that in every scientific report, the correct statis-
tics were used on the appropriate data for the right reasons, and that the
interpretation of the findings in the study is an accurate reflection of what
really happened. We assume that all the flaws will be picked up by the
reviewers, that we can rely on the reviewing process totally, and that
we can accept the findings and the author’s conclusions at face value.
Unfortunately, these are dangerous assumptions, as witnessed by the fact
that the Bond and Dykstra article was reprinted in 1997 without any
comment about the problems with data handling and statistics in the
original study. The editors were either not aware of this or chose not to
comment. The NRP report (see the next four chapters) is witness to the
serious methodological problems that plague this field.

A Postscript: Follow-On Studies

In 1968, Dykstra (1968a, 1968b) published two brief reports on follow-on
data for 960 children who had completed second grade. They came from
 103

ten Projects: two each for Basal reader, i.t.a., Language experience, Lin-

How to Teach Reading: Lessons from the Past

guistic, and Lippincott. No Basal þ phonics groups were included. In addi-
tion, 250 students were followed up on the individually administered tests.
Correlational data revealed much the same results as those found
in first grade. The same three baseline tests—letter-name knowledge,
phoneme-discrimination, and IQ—correlated with the Stanford Reading
subtests: Vocabulary and Comprehension (the only two tests reported).
Each of the baseline measures accounted for approximately 16–25 percent
of the variance on the Stanford scores.
Reading achievement measured at the end of second grade correlated
with reading at the end of first grade at values ranging from .60 or higher,
showing that reading achievement is the best predictor of reading achieve-
ment. This led Dykstra (1968b, 60) to conclude:

The best predictor of success in learning a task is prior success in learning a

similar task. . . . There is little indication that any of the readiness subtest skills
uniquely related to success in the various types of programs. . . . Results indi-
cate that it is not feasible to place pupils differentially in instructional pro-
grams on the basis of a profile of readiness tests administered early in the first
grade.

In the Method comparisons, i.t.a. and Lippincott were superior to the

other Methods on most comparisons, and Basal readers superior on none.
However, no tables or numerical data were provided to support these
statements. Sex differences favoring girls were still in evidence. Dykstra
concluded: ‘‘On the average boys cannot be expected to achieve at the
same level as girls under the current methods.’’ There was no indication
from these reports that the statistics problems had been solved, or that
individual scores had been used in these analyses. All Method comparisons
are still subject to the same criticisms raised earlier.
There was one other large-scale project from this period. This was
Project Follow Through, involving research on how to prevent children
in Head Start programs from losing gains after preschool. Altogether,
22 instructional approaches were examined. These fell roughly into three
groups: those that focused mainly on academic skills, those that empha-
sized cognitive development, and those that emphasized affective or emo-
tional development. A survey of the outcomes of the various methods
 104

conducted by Abt Associates and Stanford Research Institute (Stebbins

Chapter 4

et al. 1977) proved difficult. Data had not been collected on the same
measures, and testing was not synchronized in time nor reported in a
similar fashion.
This research lacked the advance planning and uniform controls
imposed by Bond and Dykstra in their study, and only the vaguest con-
clusions could be drawn. There was a tendency for the academic approach
to be most successful, and within this group the direct-instruction method
known as DISTAR was most effective. DISTAR was designed at the
University of Oregon by Engelmann and Bruner (1969) and was intended
for disadvantaged children in small-group instructional settings. It re-
quires minimal teacher training, because lessons are scripted and teachers
are strongly encouraged not to deviate from the script.
DISTAR includes a variety of components, including math and lan-
guage. The reading program is phonics oriented and features a modified
script. Letters vary in size depending on their perceived importance in
decoding, or are marked to indicate pronunciation. Children do not learn
to segment but to read the ‘‘slow way’’ by stretching the sounds. The
long-term effects of the complete DISTAR program were reviewed by
Becker and Gersten in 1982. They provided an analysis of the follow-on
data for fifth and sixth graders collected in 1975. These children had been
in a three-year DISTAR program for reading, language, and mathematics,
beginning in first grade in 1969–70. When DISTAR training ended at
third grade, children had gained over 20 standard-score points on the
Wide Range Achievement Test (WRAT) for reading, 9 points in spelling,
and 7.5 points in arithmetic. Percentile scores were 67, 40, and 45 (50 is
the national average). These are good results.
There was a further follow-on at fifth and sixth grade after the DIS-
TAR children had been transferred to schools with normal instruction.
Compared to a matched control group, they were significantly ahead in
WRAT reading in 23 out of 31 comparisons. At fifth grade, DISTAR
children were superior in one-third of the comparisons on the Metropol-
itan Achievement Test (MAT) battery, and equal to the controls on the
remaining two-thirds. By sixth grade this advantage had dwindled to
around 10–20 percent of comparisons.
However, when compared to national norms, these children did not
do well. When Becker and Gersten plotted the average percentiles and
 105

standard scores for the children across six grades, the DISTAR children

How to Teach Reading: Lessons from the Past

showed a strong peak (near norms or well above norms) at both second
and third grade (during the period of DISTAR instruction), followed by a
sharp decline by fifth and sixth grade. WRAT reading tumbled into the
30th percentile and MAT reading into the teens, along with math scores.
Despite these declines, the DISTAR children were still ahead of the
matched control groups, and kept this advantage at a ninth-grade follow-
on (.8 of a grade equivalent).
The failure of these children to maintain their hard-won gains is both
puzzling and troubling. Becker and Gersten believed that they suffered
from a lack of challenging work, and that once the DISTAR program
ended, ‘‘without effective instruction which continues to build on these
skills in the intermediate grades, the children are likely to lose ground
against their middle-income peers’’ (p. 89). The results also suggest that
teachers did not assess these children’s skills appropriately when they
transferred to normal classes at age 9. This study was important for
showing what can be accomplished with disadvantaged youngsters, and for
illustrating the critical importance of longitudinal data.
As these monumental studies faded from memory, the insights they
provided were lost. A proper reading of this work could have saved much
time and impelled us toward studies that did not appear for another 30
years.
 5
HOW TO TEACH READING: MODERN RESEARCH

It is the ultimate irony that Chall’s book and Bond and Dykstra’s study
had the opposite effect from the one they intended. Bond and Dykstra’s
analysis of the data failed to reveal what really happened (that is, the su-
periority of the Lippincott program on all measures), and seemed to show
that the success of a reading method was completely unpredictable. This
message undermined Chall’s assessment of the research (that synthetic
phonics was the clear and unassailable winner) and gave more weight to
her observations on the impact of the teacher. Taken together, the pro-
jects pointed to the same conclusion: teachers matter more than the method.
Dykstra (1968b) even suggested this in a follow-up report, expressing
concern that teachers needed to be monitored in future research.1
The net effect of Bond and Dykstra’s project was to virtually shut
down applied research on classroom reading methods. If the teacher mat-
ters more than the method, the method is largely irrelevant. If basal-
reader programs are no worse than anything else, there is no reason to
change. And unless some way can be found to neutralize or stabilize the
teacher’s input into the process, any future research on methods is a waste
of time. There was no point in funding this kind of research when a study
of over 9,000 children failed to show anything definitive. The fact that
only 38 research reports on reading methods passed the final screening in
the recent National Reading Panel survey says it all. And of these 38
reports, half are tutoring programs for older readers. Twenty valid studies

1. Much later, Chall offered a more chilling explanation at a conference

I attended, stating that research on reading methods in the classroom is
impossible.
 108

on classroom reading programs in 30 years is two studies every three

Chapter 5

years, one reason we have made so little progress in establishing solid sci-
entific evidence on how to teach children to read.
The 1960s projects had even more ripple effects, opening the door for
the whole-language movement. Chall’s book (in company with Flesch’s
book in 1955) unmasked the basal programs. The facts were there for all
to see, complete with virtual pages from those boring readers. By the end
of third grade, children had been exposed to a reading vocabulary of a
mere 1,500 words. They did not learn to spell. They did no writing of any
kind until second grade.
This was the antithesis of what should happen, according to the
founders of the whole-language movement, who believed that learning to
read was as ‘‘natural’’ as learning to talk. Children should learn to read by
reading and use stories written in natural language, not in the stilted and
repetitive style of the Dick and Jane readers. This way, children could
apply all their linguistic skills (vocabulary, syntax, sensitivity to context) to
understand what they read. According to Goodman (1967), reading is a
‘‘psycholinguistic guessing game,’’ where the main goal is to follow the
gist of the story. Accuracy is largely irrelevant.
Of the face of it, the ‘‘natural-language’’ approach sounded like a
better solution, and had the added bonus of being fun—fun for the
teacher whose main task was to read interesting stories from real chil-
dren’s literature out of Big Books, and fun for children who got to listen
and ‘‘read along’’ in little books. It was fun for teachers to encourage
creative writing and watch in admiration as children invented their own
spelling system, and fun for children to ‘‘write stories’’ regardless of
whether anything they wrote could be deciphered. It was motivating for
children to believe they were learning to read, write, and spell, despite the
fact they were not. The whole-language movement was the third (and
final) whole-word method of the twentieth century, and it took the
English-speaking world by storm—with catastrophic consequences (see
chapter 1).
The founders of whole language were not alone in believing that
natural-language development had something to do with learning to read.
This has been a major theme in reading research as well. Most of the
research on spelling derives from the belief that children go through
 109

developmental spelling stages (see chapter 9). As noted in chapter 1, the

How to Teach Reading: Modern Research

dominant theory in reading research for the past 30 years has been that
‘‘phonological awareness develops,’’ a theory unsupported by any data.
This gave rise to a proliferation of phonological-awareness training pro-
grams, the topic of the following chapter.
Since the early 1970s, the phonological/language-reading connection
dominated the landscape in scientific studies on reading. At a rough esti-
mate, approximately 90 percent of the scientific research over the past 30
years has been descriptive or correlational, addressing the following ques-
tion: What phonological or linguistic skills do good readers possess that
poor readers lack? (See Language Development and Learning to Read.) The
remaining 10 percent of published studies is applied and deals with more
fundamental issues, like pinning down what matters most for teaching
children to read.
Only a few studies in this 30-year period have addressed the critical
problem that Chall and Dykstra identified. What is the relative contribu-
tion of the teacher and the method to children’s reading skill? This is the
most important question of all, because unless we can answer it, methods
research is largely a waste of time.

Methods and Teachers Matter

In their observational study, Chall and Feldman (1966) discovered that
what classroom teachers said they did bore only a vague resemblance to
what was recorded on their observational checklists. Perhaps this is not
surprising. It is difficult to monitor your own behavior when you are part
of the action. Classrooms are highly dynamic, and what happens minute
by minute is unpredictable. Lessons get interrupted. A lesson that is going
well may be extended. A lesson that was planned never gets done.
This means that asking teachers to keep diaries or fill out checklists
on what they did in the classroom that day or that week is not the most
productive way to find out what is really going in the classroom. It is cer-
tainly not the way to find out which of the activities students engage in
really matter for learning to read. The only way to do this is to sit in the
classroom and record what the children are doing minute by minute. Few
researchers have had the fortitude to do this, and the first rigorous study
did not appear for 20 years.
 110

Canadian psychologists Evans and Carr (1985) carried out observa-

Chapter 5

tions on first-, second-, and third-grade classrooms. Ten classrooms fol-

lowed the language experience (LE) format described in the preceding
chapter. Lessons are developed around the child’s vocabulary. The child
tells a story to the teacher. She writes it down and copies the words onto
cards. These become the child’s ‘‘word bank,’’ which he is supposed to
memorize by sight. This phase ends when the bank reaches around 150
words. After this, the child works with published materials, following
the tenets of whole language. In another ten classrooms, the children
were using a program based on ‘‘traditional phonics.’’ Children memo-
rized sight words, learned to link letters and sounds (decoding), filled in
phonics workbooks, practiced writing letters, words, and sentences, and
read phonics-type readers.
Each of the 400 children in the study was observed on 50 different
occasions for 10-second periods. Observations covered the entire school
day. Behaviors were coded into categories, and the total ‘‘time on task’’ for
each category was recorded.
The two types of classroom were noticeably (and significantly) dif-
ferent in how time was allocated. Children in the LE classes worked
independently about two-thirds of the time, while children in the phonics-
oriented classes spent over half the time (57 percent) in whole-class,
teacher-led activities. They spent far more time on word analysis and
decoding—three times longer in groups, and six times longer when work-
ing alone. They also spent more time writing letters and words, more time
reading silently, and less time reading out loud.
On direct comparisons of several standard tests of reading compre-
hension at the end of the year (the only reading skill measured), children
in the phonics classes were significantly ahead. There were no differences
between the two groups on any tests of cognitive and language skills.
The critical aspect of this work was the relationship between the time
spent on the various activities (time on task) and reading test scores, irre-
spective of the reading method (all classes combined). There were several key
findings. Time spent memorizing sight words was negatively correlated
with every reading test. This was true in both group and independent
settings (median scores: .32 group, and .63 independent). The negative
correlation means that the more time the children spent memorizing
 111

sight words, the lower their reading scores. Time spent reading aloud (to

How to Teach Reading: Modern Research

teacher or class) was uncorrelated with reading skill. Interestingly, lessons
on comprehension and on how to use context to understand the story
were unrelated to reading comprehension (correlations at zero). All classes
benefited from group silent reading (correlations strongly positive) but not
from independent silent reading (no correlation). Apparently, it is easier to
focus on reading when everyone in the class is silent than when only you
are silent.
Other findings were more specific to the types of instruction. Chil-
dren in the phonics classes spent far more time copying words and in
various writing activities. These activities were positively correlated with
reading comprehension scores (range r ¼ :37 to .75). There was a modest
relationship between reading and time spent on phonics-type activities
whether done by the whole class or independently (range r ¼ :04 to .41).
The children in the LE classrooms engaged in far more oral-language
activities led by the teacher (lessons on vocabulary and grammar, listen-
ing to a story). Time spent in these activities was strongly negatively cor-
related with reading skill at surprisingly high values (range r ¼ :70 to
.80).
The authors commented on the negative outcome for several of the
independent-learning tasks, and suggested that children ‘‘may be at risk
for degenerating into almost random learning which may detract from or
interfere with more systematic practice of reading skills’’ (Evans and Carr
1985, 344). The one exception was writing practice, which was highly
predictive of reading skill and rarely took place in a group setting. Talking
about the meaning of words and listening to the teacher read stories takes
time away from more productive activities. Time spent on these activities
had negative consequences for reading comprehension.
Sumbler (1999) in Toronto used a similar approach. Ten kinder-
garten classrooms were using a linguistic phonics program called Jolly
Phonics ( JP), a program close to the prototype. The alphabet principle
is taught from sound to print using a basic code. Children get extensive
practice in blending and segmenting sounds and letters in real words, both
orally and in writing. We will look more closely at this program later in
the chapter. The other ten classes were using a ‘‘balanced’’ approach, which
included a wide range of activities chosen at the discretion of the teacher.
 112

The children were attending half-day sessions in senior kindergarten.

Chapter 5

The average age was six years at midyear when the training and the ob-
servations began. Observations followed the format developed by Evans
and Carr. Every child in every classroom was monitored for a brief time
period on numerous occasions, and the observations continued for several
months.
Overall activity patterns were similar in the two types of classrooms.
Children went out to play for the same amount of time, spent the same
amount of time transitioning between activities, and were interrupted
equally often. However, children in the ‘‘balanced’’ classes spent a sig-
nificantly larger proportion of the time in non-literacy-related activities
(35 percent versus 28 percent). Of the ten literacy activities coded, there
were significant differences on four, with JP children participating sig-
nificantly more often in each one. JP children spent 10 percent of the
language period on phonics-related activities (explained below), compared
to 2 percent for the other children. They spent more time on auditory
phoneme-awareness tasks (7 percent versus 3.7 percent), and more time
memorizing sight words and learning about grammar, though the last two
activities were infrequent in all classrooms. There were no differences be-
tween the two types of classroom in the time spent on the remaining cat-
egories: learning concepts of print, like the order of words on the page
and order of pages, reading aloud or silently, ‘‘pretend writing’’ sentences,
copying letters/words/sentences and writing them from memory, learning
letter names, and vocabulary lessons. Vocabulary work took up by far the
most time in both types of classrooms (18 percent and 20 percent).
The children took five standardized reading and spelling tests at the
end of the school year. Children in the JP classrooms were significantly
ahead on every test. When scores on these tests were correlated with time
on task for the various activities, only two activities were significantly (and
positively) related to reading and spelling scores. These were phonics
activities (range r ¼ :48 to .62), and copying/writing letters and words
(range r ¼ :50 to .55). The correlations between time spent on auditory
phoneme analysis (no letters) were essentially zero for all five reading
measures. Although no other values were significant, the following activ-
ities were consistently negatively correlated with all five reading and spell-
ing tests, with values for r ranging from .20 to .31: learning letter
 113

names, vocabulary lessons, and ‘‘nonliteracy’’ activities (15 negative cor-

How to Teach Reading: Modern Research

relations out of 15 would not occur by chance).2
The phonics category consisted of five subcategories: oral spelling:
spelling out loud with letter names, visual spelling: saying letter names of
words, learning sound-letter correspondences, word analysis: sounding
out and blending sounds in words using letters, and Jolly Phonics ‘‘ac-
tions.’’ These actions only occurred in the JP classes. Because they always
overlapped with learning sound-to-letter correspondences (nonindepen-
dent categories), they will not be discussed further.
Children spent little time doing either oral or visual spelling, and
comparisons were not significant. JP children spent 8 minutes a day learn-
ing sound-to-letter correspondences and 8 additional minutes on word
analysis (segmenting, blending). The children in the ‘‘balanced’’ programs
averaged less than a minute per day on either activity. Time spent on these
two activities was significantly correlated with reading and spelling (cor-
relations ranged from r ¼ :35 to .55 across the reading and spelling tests).
The negative impact of story time on children’s reading skill was also
shown by Meyer et al. (1994). Two large cohorts of children in three dif-
ferent school districts were followed from kindergarten to the end of first
grade. At kindergarten, the more time the teacher spent reading to the
class, the lower the children’s scores on a variety of reading tests, espe-
cially tests of decoding (r ¼ :44 to .71). Reading to children in either
kindergarten or first grade had zero impact on first-grade reading scores.
However, time spent teaching decoding skills was strongly correlated with
reading skill on five tests (r ¼ :44 to .62).
These findings are remarkably consistent. Time devoted to learning
sound-to-letter correspondences, segmenting and blending sounds in
words, and writing letters and words predicts subsequent reading and
spelling skill in both beginning readers and more advanced readers. Most
other ‘‘literacy activities’’ have no effect, while memorizing sight words,
doing vocabulary lessons, and listening to stories have a consistently neg-
ative effect.

2. It should be noted that the ‘‘subjects’’ in this study were ‘‘classrooms,’’

which sharply reduced N and statistical power.
 114

The importance of writing for locking sound-symbol correspon-

Chapter 5

dences into memory has been investigated by Hulme and his colleagues
(Hulme 1981; Hulme and Bradley 1984; Hulme, Monk, and Ives 1987).
They compared learning speed for mastering phoneme-grapheme cor-
respondences for copying, using alphabet cards, or using letter tiles.
Children learned much faster when they wrote the letters. Hulme and
colleagues (1987) concluded that motor activity promotes memory, and this
assists children in learning to read. But this is only part of the answer.
Copying letters forces you to look carefully and hold this image in mind
while you are writing. This, plus the act of forming the strokes, makes it
clear how letters differ. (See McGuinness 1997c for an analysis of how
copying assists memory.)
Cunningham and Stanovich (1990b) reported the same effect for
spelling accuracy. First graders memorized spelling lists using three dif-
ferent methods: copying by hand, utilizing letter tiles, and typing the lists
on a computer keyboard. Children spelled twice as many words correctly
when they learned by copying the letters than with letter tiles or typing.
They also found that saying letter names while writing letters had no im-
pact on spelling performance.
The most surprising result to emerge from the observational studies
was the large number of activities that were either nonproductive (zero
relationship to reading) or actually detrimental (negative correlations).
Negative correlations can mean either of two things: time wasted (a trade-
off between learning one thing at the expense of another), or a negative
outcome, like the creation of a maladaptive strategy. Correlational re-
search can never prove causality, but it is hard to imagine that vocabulary
activities and listening to stories are ‘‘bad’’ for children. Here, the time
trade-off argument makes sense, especially because vocabulary work and
listening to stories took up more time than any other literacy activity in
most classrooms. (The big question is whether time spent on vocabulary
work actually improves vocabulary. We will come back to this question in
chapter 8.)
On the other hand, we know that time spent memorizing sight words
can cause a negative outcome by promoting a strategy of ‘‘whole-word
guessing.’’ This is where children decode the first letter phonemically and
guess the rest of the word based on its length and shape. This strategy
is highly predictive of subsequent reading failure (McGuinness 1997b). It is
 115

well known that program emphasis (sight word, context-based guessing,

How to Teach Reading: Modern Research

part-word analysis, phonemic decoding) strongly affects the child’s decod-
ing strategy, and that this strategy quickly becomes entrenched (Barr 1972,
1974/1975; Vellutino and Scanlon 1987; McGuinness 1997b, 1997c).
Boronat and Logan (1997) showed convincingly that what you pay
attention to is automatically encoded by the brain and automatically cued
in memory. As they put it, ‘‘What one pays attention to acts as a retrieval
cue that draws associations out of memory’’ (p. 45). What you ignore,
even though it is physically adjacent to what you are looking at, is not
encoded at all. The more a child focuses on the wrong patterns and com-
binations of letter sequences in words, the more automatic (habitual) this
becomes.
In the observational studies, time spent memorizing sight words was
negatively and weakly correlated with reading scores for the kinder-
garteners, but negative and strongly correlated for the older children (6 to
8 years). A sight-word strategy begins to overload between 7 and 8 years. I
found that children who adopted a whole-word strategy by the end of first
grade had not improved their performance when they were followed up
in third grade (McGuinness 1997b). These children often made the same
decoding errors on the same words they had made two years earlier, and
they were uniformly the worst readers in the class.
The observational studies are incredibly powerful, because they do
something that research on classroom methods cannot do. They cut
through the noise of curriculum details and teacher variability and get to
the heart of the matter. They make it possible to link success in learning
to read and spell to what is actually being taught. Training in listening
for the sound units of the writing system, learning the correspondences
between these units and their symbols, and grasping how the units link
together to make words are the essential ingredients of a successful read-
ing method. The more these skills are practiced by eye, by ear, and by
hand, the better.
After 5,000 years, the Sumerians have been vindicated by science.

Spelling Helps Reading, But Seeing Misspelled Words Is Bad for Spelling
There were other important discoveries during this 30-year period. Two
sets of studies, in particular, bear directly on the importance of writing
and spelling practice. One group of studies looked at the impact of spelling
 116

practice on reading skill. The second group of studies stemmed from

Chapter 5

concern about teaching methods that advocate misspelling words as a

strategy for learning to spell, or that tolerate the writing of misspelled
words by children.
In chapter 3, I pointed out that learning to spell (correctly) impacts
reading, because spelling requires recall memory, a substantially more com-
prehensive (deeper) form of memory than what reading entails. Reading
relies on recognition memory (memory with a prompt), a more superficial
type of memory. This is why learning to read is less likely to impact
spelling than the reverse. Smith (1999) found that very poor readers
scored normally on tests of word recognition and reading comprehension
after 12 hours of structured lessons on the advanced spelling code. Other
scholars have investigated this more directly.

Learning to Read by Learning to Spell

Ehri and Wilce (1987) taught one group of 5-year-olds to spell nonsense
words with letter tiles, and another group to read the same words, using
a slightly modified alphabet of 10 letters. After training, the spelling
group had higher reading scores than the reading group on a transfer task
(same letters/different words). In a more thorough investigation, Uhry and
Shepherd (1993) trained two groups of first graders matched for reading
skill for 40 minutes a week for 28 weeks. One group received segmenting
and spelling training, while the other group received a mirror-image ver-
sion consisting of blending and decoding (reading) training. At 32 weeks,
the children took a battery of tests. The segmenting/spelling group was
significantly ahead on all tests of reading, spelling, and phoneme aware-
ness (including blending). These children used letter tiles or typed on a
computer keyboard to spell words. Based on the evidence that writing
letters speeds up learning, one would expect this to make an even greater
difference.

Invented Spelling
There is little research on the consequences of invented spelling, because
spelling is assumed to follow natural developmental stages. When children
are encouraged to ‘‘just write’’ and ‘‘invent’’ their own spelling system, the
most common pattern that emerges (other than complete randomness) is
letter-name spelling (so, for example, the word far would be spelled fr), as
 117

noted in chapter 1. This is not a ‘‘stage of development,’’ as researchers

How to Teach Reading: Modern Research

claim. Letter names are all the children have been taught. (If they had
been taught the sounds the letters stand for, correctly spelled words would
appear a lot earlier in spelling development.)
We know that chanting letter names as you spell has no effect from
Uhri and Shepherd’s research, and we know that saying the sounds does
from the Lippincott results. It turns out that using letter names to spell is
actually detrimental. Treiman and Tincoff (1997) discovered that learning
letter names focuses children’s attention on the syllable instead of the
phoneme (es, em, ef, kay, pee, are), blocking their conceptual understanding
of how an alphabet works, to say nothing of the fact that they will not
learn how to spell.
The argument for teaching letter names to young children has always
been that the names contain the sound the letter stands for. But this is
another myth. Try this little game with a friend. Write out the following
letter-name sequences on a piece of paper and ask a friend what they spell:
sea-oh-double-you, aitch-ee-ay-tee, and ef-are-oh-gee.
It is obvious that invented spelling, by definition, will impede, halt, or
otherwise delay correct spelling. Does correct spelling eventually emerge
from this mess, as educators believe? So far, there are no studies on chil-
dren that shed light on this issue, but there are studies on adults.

The Negative Impact of Seeing Misspelled Words

Any teacher or college professor knows that after hours of marking stu-
dent papers, one’s sense of the spelling code begins to falter. What was
once easy (spotting spelling errors effortlessly) now becomes tentative. Is it
possible that a sense of the ‘‘wrongness’’ or ‘‘rightness’’ of particular spell-
ings (learned over decades) could start to evaporate in such a brief period?
This phenomenon was first documented by Pintner, Rinsland, and
Zubin (1929) and by Nisbet (1939). They discovered that if students took
a spelling test, then saw some of the same words misspelled, when they
took the spelling test again the spelling errors increased by about 15 per-
cent. More recent studies have supported these results. Brown (1988) gave
college students a spelling-dictation test of intermediate difficulty. Half
the students saw half the words again and had to generate two possible
phonetic spellings (misspelled) for each word. The other students did an
unrelated task. The original spelling test was given again. The spelling
 118

errors increased by 17 percent for the experimental group versus 7 percent

Chapter 5

for the controls, a highly significant result.

In a second experiment, Brown looked at the effect of seeing mis-
spelled words rather than creating them. The experiment was run on
a computer. The format was similar: a spelling-dictation test, then a
spelling-recognition test (or not), followed by either a dictation or recog-
nition test. The spelling-recognition test consisted of several spellings of
the same word (tramendous, tremendus, tremendous, tremmendus), and the
task was to choose the correct spelling. Students also had to rate each
misspelled word according to how closely it resembled the correct spell-
ing. The control group did a jigsaw puzzle. On the final dictation test,
spelling errors were twice as high for the experimental group.
One of the most intriguing and complex studies on this topic was
carried out by Jacoby and Hollingshead (1990). They were interested in
whether merely seeing misspelled words was sufficient to depress spelling
skill, as compared to active processing, which requires decisions or writing
out the words. College students saw hard-to-spell words one at a time on a
computer: 20 spelled correctly, 20 incorrectly. Students were assigned to
one of three conditions: read the words out loud, type the words exactly
as spelled, or print the words exactly as spelled. All students were told
that some of the words were misspelled. After this, they took a spelling-
dictation test on the same 40 words plus 20 new words of equal difficulty.
Spelling accuracy (the dictation test) was strongly affected by previous
exposure to the correct and incorrect spellings alike. The results were the
same whether the students had previously read, typed, or printed the
words. Scores were as follows: old correctly spelled words (93 percent),
new correctly spelled words (87 percent), incorrectly spelled words (83
percent). Seeing misspelled words significantly depressed spelling scores,
and the probability of using the same misspelling was high (.76). This
shows a ‘‘priming’’ effect in that recent exposure to correctly spelled words
improves spelling (93 percent) for those words, as compared to correctly
spelled words not seen recently (87 percent).
While spelling accuracy did not vary with the three types of exposure,
typing speed did. The groups that had typed or printed in the exposure
phase, typed faster in the dictation test than the group that just read. This
suggests that motor involvement has more of an impact on efficiency than
 119

on accuracy. In a recognition-memory test (‘‘Was this word on the list?’’),

How to Teach Reading: Modern Research

performance was also superior for the groups that typed or printed.
In a second experiment, students saw the same word lists. This time,
one group copied each word while it remained in view. A second group
studied the word for 3 seconds, then did a distracting task for 10 seconds.
Both groups took the same spelling-dictation and spelling-recognition
tests. The two groups did not differ in accuracy on the spelling tests, and
scores mirrored the first experiment: correctly spelled words (92 percent),
new words (86 percent), incorrectly spelled words (80 percent). This
shows that an interfering task does not erase the negative impact of seeing
misspelled words or the priming effect of seeing correctly spelled words.
Not only this, but recognition memory was higher for the distraction
group (85 percent correct versus 65 percent).
Jacoby and Hollingshead explored several possible theoretical expla-
nations for these results and dismissed most of them. For example, no
‘‘template’’ theory, in which every spelling has a special memory trace, or
a system prompted by ‘‘recollection’’ or ‘‘spelling rules,’’ can explain these
findings. Unconscious (automatic) processing is at work here. We will
come back to what these results mean shortly.
A similar study was carried out with undergraduates in the United
Kingdom (Dixon and Kaminska 1997). Good and poor spellers were
selected on the basis of their performance on words from The Awful
Speller’s Dictionary (Krevisky and Linfield 1990), which contains difficult-
to-spell words. They first took a spelling-dictation test consisting of 60
words from the above dictionary, then did one of three intervening tasks:
read misspelled words from the same test, read correctly spelled words
from the same test, or read correctly spelled words not on the test. A final
spelling test was given immediately and again one week later. The delay
had no effect. Exposure to misspelled words depressed the spelling scores,
and they did not rebound at all over the week. Seeing correctly spelled
words twice had no more beneficial effect than seeing them once. Good
and poor spellers were not differentially affected by any of these manipu-
lations. That is, good and poor spellers alike were equally prone to the
impact of misspelled words immediately and after a delay.
The authors of all three studies strongly cautioned educators and re-
searchers to not use spelling-recognition tests, either for school testing or
 120

for spelling research. This type of test appears in well-known test batteries
Chapter 5

like the California Achievement Test, Comprehensive Test for Basic

Skills, Metropolitan Achievement Test, and the McGraw-Hill Basic Study
Skills. There are two reasons for concern. First, a spelling-recognition test
is an invalid method of testing spelling ability. It is a test of recognition
memory (‘‘choose the spelling that looks ‘familiar’ ’’), not a test of recall
memory. Second, merely taking the test can create spelling amnesia for
the misspelled words. Jacoby and Hollingshead reported on the experi-
ence of the second author who ran the experiment many times, exposing
her to the misspelled words again and again. This seriously eroded her
confidence or sense of what correct spellings looked like, not only for
these words, but for other words as well.
These studies provide considerable insight into the process by which
the brain codes and remembers spellings. The fact that a single exposure
can erase or degrade a particular spelling shows that memory traces, es-
pecially for complex spellings seen infrequently, are extremely unstable
and can easily be overwritten by brief exposure to something ‘‘almost cor-
rect.’’ The brain is a statistical pattern analyzer par excellence. Incoming
patterns are instantly matched to prior representations of the same or
similar patterns. If a small deviation (a spelling change) in a relatively un-
familiar word exists, statistical probability will shift in the direction of the
new form. That we have stable memories of spellings at all, given the id-
iosyncrasies of the English spelling code, seems miraculous. The instruc-
tional implications are enormous. First, a regime of ‘‘invented spelling’’
makes it impossible to form stable memories of correctly spelled words.
Second, the spelling techniques and tricks widely used by classroom
teachers where children are told to write out all the ways a word can be
spelled to see what looks right, are fatally flawed.

truble, trubble, trubbel, troble, trobble, trobbel, trouble, troubble, troubbel

The argument is that children should do this, because adults commonly

try out different spellings when they forget how to spell a word, or when
a spelling ‘‘looks funny.’’ But adults try out likely spellings, not a list of
highly improbable, intentionally misspelled words. And adults know how
to spell; otherwise a misspelled word would not ‘‘look funny’’ in the first
place, prompting the need to try another spelling. Children do not know
 121

how to spell, and so writing out lists of all possible spellings, carefully

How to Teach Reading: Modern Research

scrutinizing each one, then crossing it off, is a sure-fire ‘‘antispelling
method.’’ Children will see far more incorrect spellings than correct
spellings, making it impossible for the brain to do its job. Consistent pat-
terns lead to accessible memories. The brain cannot code randomness. As
Boronat and Logan’s research showed, what you are forced to pay atten-
tion to is what you remember. Rightness or wrongness makes no differ-
ence to this automatic process.

The Prototype: An Update

Before moving on to an analysis of reading programs in light of this new
information, here is what a successful reading/spelling program should or
should not contain based on our discussion so far:

No sight words (except high-frequency words with rare spellings).

No letter names.

Sound-to-print orientation. Phonemes, not letters, are the basis for the

code.

Teach phonemes only—no other sound units.

Begin with basic code (a one-to-one correspondence between 40 pho-

nemes and their most common spelling).

Teach children to identify and sequence sounds in real words by seg-

menting and blending, using letters.

Teach children how to write each letter. Integrate writing into every

lesson.

Link writing (spelling) and reading to ensure children learn that the al-

phabet is a code and that codes are reversible: encoding/decoding.

Spelling should be accurate or, at a minimum, phonetically accurate (all

things within reason).

Lessons should move on to include the advanced spelling code (the 136

remaining, common spellings).

Reading Instruction: The National Reading Panel

Results
This section reviews the evidence from the National Reading Panel’s
(NRP) survey on instructional methods for reading. My goal is to look
more closely at the programs in the NRP database that have the greatest
 122

and most consistent success. How well do these programs align with the
Chapter 5

prototype program I have been assembling in the last two chapters?

In the introduction to the project, the NRP reported that a general
screening of the databases unearthed approximately 100,000 research
studies on reading published since 1966. To handle the volume of this
material, the NRP divided ‘‘reading’’ into five topic areas. I will discuss
one of these areas—reading instruction—in this chapter. The remaining
areas (phoneme-awareness training, methods for improving fluency, in-
struction in vocabulary, and instruction in comprehension) will be taken
up in the following chapters. Spelling instruction was not included in the
NRP survey, no doubt because so little research is available on this topic.
The panel held hearings for parents, teachers, researchers, and other
interested parties who were the intended audience for this report. Here is
one of the seven themes they identified from this input: ‘‘The need for
clear, objective, and scientifically based information on the effectiveness of
different types of reading instruction and the need to have such research
inform policy and practice’’ (National Reading Panel, 2000, 1-2).
From these themes they developed more specific questions to guide
their search. The general statement above was narrowed considerably,
producing questions like; ‘‘Does phonics instruction improve reading
achievement? If so, how is this instruction best provided?’’ (p. 1-3).
At no point do they explain why they limited the focus to phonics,
specifically to ‘‘systematic phonics,’’ as you will see below. Part of the
reason appears to reflect the panel’s preference, though the main reason
may lie in the fact that whole-language advocates rarely conduct research
on their approach. The same can be said of publishers of basal-reader
programs.
This is a very curious state of affairs. The dominant reading method is
accepted universally without a shred of evidence attesting to its efficacy,
while phonics advocates continually have to prove that phonics programs
work. One would imagine that there would be voluminous research on
whole language, a method used in 90 percent of classrooms for 30 years,
and that research on other programs, like phonics, would be hard to find.
Yet the opposite is true, not only in terms of volume, but in terms of
quality. This says a lot about why our schools are in trouble.
Of the original 1,072 studies on reading instruction screened by the
NRP, only 75 passed the first screening based on these guidelines:
 123

An experimental or quasi-experimental design with a control group.

How to Teach Reading: Modern Research

Published in a refereed journal after 1970.

Must provide data testing the hypothesis that systematic phonics in-

struction improves reading performance more than alternative phonics

instruction or no phonics instruction (i.e., other programs).

Reading must have been measured as an outcome.

Adequate reporting of statistics sufficient to compute effect sizes.

Other problems came to light as the NRP reviewed the 75 studies in more
depth, such as missing control groups, too limited a focus or time span,
incorrect or inadequate statistical analysis, inappropriate outcome mea-
sures, and duplicate studies (or data) found in another publication. As a
result, only 38 research reports (a total of 66 individual comparisons)
passed the final screening—a sad state of affairs.
Yet even this dismal showing is better than the research database on
whole language. In 1989, Stahl and Miller attempted to review the whole-
language/language experience research and provide a quantitative synthe-
sis of the research findings. The search was exhaustive and dated back to
1960. It included all the obvious databases, dissertation abstracts, and bib-
liographies. Personal letters were sent to the major figures in the field,
asking for information and help. Apart from Bond and Dykstra’s research
on the language experience method, only 46 studies were found, and only
17 had sufficient statistical data to compute effect sizes. These 46 studies
constituted the total pool of research on the whole-language and language
experience methods. Contrast this with the 1,027 studies found by the
NRP. These studies compared whole language/language experience to
various basal-reader programs.
Because Stahl and Miller located so few studies, none could be ex-
cluded, regardless of the numerous methodological problems they de-
tected and the fact that most of these studies had never passed peer review.
They were obliged to report the outcomes in a table similar to the one
used by Chall. The table showed that 58 comparisons were not significant,
26 favored whole language/language experience, and 16 favored basal
readers. On closer inspection, the advantage to whole language was en-
tirely on nonreading tests, like ‘‘concepts of print,’’ and only in the
kindergarten classes. Of the 17 programs with sufficient statistical infor-
mation to compute effect sizes, only 4 were published papers. Of these 4,
 124

whole language had marginally higher effect sizes on ‘‘print concepts’’ and
Chapter 5

‘‘readiness.’’ Otherwise the results were either nonsignificant or in favor of

the basal programs.
In contrast, all studies in the final NRP data pool were published. All
had data sufficient to compute effect sizes and to provide a meta-analysis
of the results. Before reviewing this evidence, I need to discuss what a
meta-analysis is, and what it can and cannot do. Readers who are knowl-
edgeable about meta-analysis may want to skip to the next section.

Meta-Analysis Made Simple

The NRP converted each statistical comparison between reading methods
to an effect size (ES). These ES values are used in a meta-analysis, which is
essentially a grand average of all effect sizes for a large number of studies.
To make sense of the NRP findings, one has to understand what this
means.
An effect size provides a common metric for comparing two reading
methods on any test, by transforming means and standard deviations into
standard deviation units. An effect size is computed by taking the difference
between the two means of the methods being compared on a given mea-
sure, and dividing this value by the average of their standard deviations
(m1  m2/s.d.1 þ s.d.2  .5). This provides an estimate of the magnitude
(effect size) of the difference. An ES of 1.0 represents a one-standard-
deviation difference between the two groups, very large indeed. This
would reflect 15 standard-score points on a standardized reading test, the
difference between scoring at the 50th percentile (score of 100) or the
15th percentile (score of 85).
A meta-analysis pools the effect sizes from a large number of studies
and adjusts values according to the number of participants in each study.
After this, studies can be sorted into baskets and compared in different
ways. This provides very useful information. For example, training pho-
neme awareness in tandem with letters has a much greater impact on
reading test scores than when the training is purely auditory.
Despite its usefulness, meta-analysis is not a magic bullet. There are a
number of constraints and difficulties with this technique. A meta-analysis
requires a large number of studies to make reliable generalizations. Studies
need to be as alike as possible. Children should be the same age and grade.
The general content or teaching time within categories of reading meth-
 125

ods (phonics, whole language) should be similar. Because of the enor-

How to Teach Reading: Modern Research

mous variety of reading programs, and the variability with which they are
taught, this constraint is impossible to meet. For this reason, I will be
looking at the studies in much more depth than the NRP was able to do.
Another difficulty with meta-analysis is that an effect size is entirely
dependent on the control group or contrasting group. Applied research on
methods typically contrasts two different programs with the goal of find-
ing out which is best. If one program is truly better than another, the ES
value will be large. But not all research designs are alike. There are several
studies in the NRP database where ‘‘at-risk’’ children were compared to
‘‘normal’’ children who got no special training. The purpose of the train-
ing was to get the at-risk children caught up. If this effort has been suc-
cessful, the two groups will not differ, and the ES values will be close to
zero. This means an effect size cannot tell you anything absolute. It can
only tell you something relative to something else. When a large number
of studies with different research designs or other methodological varia-
tions are combined in a meta-analysis, they will partially cancel each other
out, giving a false impression of what is true. A meta-analysis is rather like
taking a census by counting houses instead of counting the people in the
houses.
I also need to add that I computed effect sizes using the formula pro-
vided by the NRP. For the most part my values were identical or compa-
rable to those reported by the panel. However, in some cases differences
were very large, and I will comment on this as we go. The most likely ex-
planation for these gross discrepancies are mathematical errors in the
NRP report.

A Quick Overview
With these reservations in mind, we will explore what the NRP’s meta-
analysis showed about the global effectiveness of the different types of
reading programs. For the most part, the studies in the database compared
a phonics-type program with something else. It was rare to find two
phonics programs being compared. Taking all cases (66 ES values), the
mean ES after training was .41 for combined reading scores. At follow up
(62 cases), it was .44 (see table 5.1). The positive value shows an advan-
tage of phonics methods over contrasting methods. These effect sizes are
a composite value that represents every reading measure, on every age
 126

Table 5.1
Chapter 5

National Reading Panel: Reading-instruction effect sizes for phonics versus other
N Read- N Spell- N Compre-
cases ing cases ing cases hension

All studies
Immediate testing 65 .41
End-of-year testing 62 .44
K and 1st grade only 30 .55 29 .67 20 .29
2nd to 6th grade 32 .27 13 .09 11 .12

group, on normal and special populations alike, and for all types of re-
search designs. It is interesting to see what happens when a subset of
studies is subtracted from the mix. A general summary of ES values for the
studies as a whole is shown in table 5.1.
I separated the studies on beginning readers (kindergartners and first
graders) from the studies on older readers (children receiving remedial
tutoring). For beginning readers, the ES value increases for phonics
methods (30 cases, ES ¼ .55). For older, poor readers, it is reduced sub-
stantially (32 cases, ES ¼ .27). The first value is moderately large and sig-
nificant, and the second is neither. Does this mean that phonics-type
programs do not work for older, poor readers? Well, no, it does not. But
it does mean that a large proportion of the remedial programs in this
particular database were unsuccessful. For instance, programs based on the
Orton-Gillingham model were particularly ineffective (10 cases produced
an ES value of only .23).
Because my analysis will focus on beginning reading instruction, we
can unravel this further by examining what the ES value of .55 represents.
The 30 individual comparisons that contributed to this value tell us, by
and large, that a phonics-type program produces a 0.5-standard-deviation
advantage over a non-phonics-type program. Not all phonics programs
were equally successful. Particularly weak were programs described as rime
analogy. These programs teach larger sound units, like word families, and
encourage children to make analogies to other words by swapping word
parts (cr-own, fr-own, d-own). The average effect size was .28 for this group
of studies, not significantly different from the comparison programs.
 127

Studies on Reading Mastery, or Direct Instruction (DISTAR), pro-

How to Teach Reading: Modern Research

duced highly inconsistent results. It is informative to compare the NRP’s
pool of these studies to a recent meta-analysis on DISTAR carried out by
Adams and Engelmann (1996). They located 350 studies, reduced by
careful screening to 37. To meet criteria, the study had to include a com-
parison group, pretest scores, means and standard deviations, and substan-
tial training (no short-term studies were allowed). Moreover, DISTAR
could not be confounded with some other method. Studies did not have to
be published, however, which was a requirement in the NRP’s protocol.
Of the 37 studies, only 13 were specific to reading or spelling. After
excluding studies on adults and one on DISTAR’s advanced spelling pro-
gram (seventh grade), this left 6 studies on special-education populations
and 4 on classroom reading instruction (normal children). The effect size
for children in special education was solid (ES ¼ .74), but DISTAR was
less impressive in the normal classroom (ES ¼ .40).
Because only one study appeared in both databases, I will assess the
studies in the NRP database separately. I eliminated three. Gersten, Darch,
and Gleason (1988) had an unusual study design, comparing children who
received two years of DISTAR, but who either did or did not partici-
pate in the DISTAR kindergarten program. Effect sizes were close to zero.
In another study, first-grade ‘‘at-risk’’ children were individually tutored,
but children in the control group were not, producing an overly large
effect size. The third study involved seven different teaching approaches
(tutoring, computers, small groups, and so on), some of which combined
DISTAR with other methods. The cumulative effect size was zero.
After eliminating these studies, the total effect sizes for the remainder
were higher than those reported by Adams and Engelmann: ES ¼ 1.1 for
reading and .45 for spelling. Nevertheless, the variability between the
studies was still large for no obvious reason. Because of this, DISTAR will
not be included in any further analysis. I also excluded the one-of-a-kind
studies using an unknown and unpublished reading program. It is impos-
sible to generalize from a single study, especially when authors fail to de-
scribe the program adequately.
There are several reading programs in the NRP database that closely
resemble the prototype. I will take them up in turn to see how they com-
pare with the total effect sizes reported in table 5.1. Before I do this,
however, I need to provide a glossary of common terms.
 128

A Glossary
Chapter 5

Basal readers or traditional basal is jargon for a complete reading curriculum

of any type published for the schools by an educational publisher. The
word basal means ‘‘basic’’ and ‘‘comprehensive.’’ In the United Kingdom,
basal readers are called reading schemes, a more descriptive term. These
programs contain all the necessary components for the classroom teacher:
teacher manuals, lesson plans, suggestions for activities, workbooks, over-
heads, graded readers, and recommended-reading lists.
Basal programs tend to be alike (publishers copy one another). Most
hedge their bets and include all possible ways to teach reading: sight-
word memorization, part-word decoding (word families), and specialized
readers, along with a sprinkling of phonics lessons and worksheets. It is
typical for the content and logic of the phonics lessons to mismatch the
readers, and for everything to mismatch the spelling lessons. There are
differences in emphasis; some programs are more oriented to whole lan-
guage and others more oriented to phonics. By and large, phonics in this
context does not conform to most people’s notion of phonics.
Phonics is a problematic word. In most research reports, the descrip-
tion of what passes for phonics is exceedingly vague. The phrase ‘‘letter
sounds were taught’’ could refer to the 26 sounds of the letters of the al-
phabet, or to the 40þ sounds of the English language and their common
spellings, or neither. In the United Kingdom, the term letter sound is
nearly synonymous with phoneme, a practice that is causing a great deal of
confusion. On rare occasions, the reader might be able to decipher what is
meant from other clues. Griffith, Klesius, and Kromrey (1992, 86) wrote
this about a traditional basal-phonics unit: ‘‘Instructional focus is primarily
upon the acquisition of letter-sound correspondence information and less
upon the mapping of spoken language onto written language. In fact,
during first grade children are exposed to approximately 90 different in-
dividual sounds. It is felt that through direct instruction on individual
letter-sound correspondences children will learn to both decode and spell
words.’’
Griffith and colleagues have no problem with the statement that chil-
dren learn 90 different sounds, despite the fact that there are only 40
sounds in English! Nevertheless, this statement tips off the reader that
the phonics component of this program is entirely letter driven and far
removed from the prototype.
 129

Some basal-phonics lessons resemble a guessing game or a cognitive

How to Teach Reading: Modern Research

puzzle rather than anything to do with instruction. It is the student’s job
to decipher what the teacher is talking about, discover the purpose of the
game, and then solve the problem. A teacher may ask a question like
‘‘Is the short sound of ‘a’ in these words?’’ and point to a set of pictures.
Note that she uses a letter name, not the sound /a/, and assumes the child
can make the connection between letter names and sounds without any
knowledge of how to do this. The child has been asked to ponder some-
thing that does not exist—the ‘‘short sound’’ of a vowel letter name.
We need much better descriptors for the different types of phonics.
The NRP reading group has made one attempt, but they do not opera-
tionalize their classification in terms of exactly what is taught and how it is
taught. In the introduction to the NRP report on phonics instruction,
phonics is defined as a means of teaching the alphabetic system by explicit
and systematic instruction in letter-sound correspondences and spelling
patterns. The report goes on to specify variations among phonics pro-
grams that fit this general category, as follows:

Synthetic phonics programs teach children to convert letters into sounds or

phonemes and then blend the sounds to form recognizable words. Analytic
phonics avoids having children pronounce sounds in isolation to figure out
words. Rather children are taught to analyze letter-sound relations once the
word is identified. Phonics-through-spelling programs teach children to trans-
form sounds into letters to write words. Phonics-in-context approaches teach
children to use sound-letter correspondences along with context cues to iden-
tify unfamiliar words they encounter in text. Analogy phonics programs teach
children to use parts of written words they already know to identify new
words. (National Reading Panel, 2000, 2-89)

This classification is unsatisfactory because it does not identify the critical

difference in logic between programs that teach the code backward from
print to sound, and those that teach it forward from sound to print (lin-
guistic phonics). They describe ‘‘phonics-through-spelling’’ as a sound-to-
print method, but this appears to refer to a strategy for teaching spelling,
not a method for teaching both decoding and spelling. There is also the
problem that the United Kingdom, in particular, and some other English-
 130

speaking countries, use phonics (specifically synthetic phonics) to refer to

Chapter 5

reading instruction based on a ‘‘basic code.’’ This kind of phonics is

anchored in the 40 phonemes and their major spellings. This is not what
Americans mean by phonics or by synthetic, as can be seen above.
To avoid this kind of confusion, I have set out a new classification
system:

Junk Phonics The practice of teaching aspects or elements of the al-

phabet code in a chaotic, nonsensical, and unstructured fashion. Most
basal phonics comes under this heading.

Visual Phonics The short version (a) teaches the 26 ‘‘sounds’’ of the
26 letters of the alphabet. The long version (b) teaches 40 to 256þ
‘‘sounds’’ of the letters, digraphs, and phonograms.

Whole-to-Part Phonics (also known as embedded, analytic, intrinsic).

The practice of easing children into phonemes by starting with whole
words, then word parts (word families, blends), then individual phonemes,
either overtly or covertly (embedded). This is tantamount to teaching
three different writing systems one after the other, each canceling out the
one before.
(Whole-to-part phonics is not the same as an eclectic reading pro-
gram, which contains elements from any of the above, plus sight-word
memorizing, plus whole language.)

Multisound Phonics Same as above, but the different-size sound units

are mixed up and taught in a random fashion.

Linguistic Phonics (Not to be confused with ‘‘linguistic programs’’

circa 1960s.)
a. Incomplete. (Called synthetic in the United Kingdom.) Teaches from the
sound to the letter. Teaches the 40þ phonemes of English and their main
spellings (basic code), plus some spelling alternatives.
b. Complete. Includes (a) above plus 136 spelling alternatives.

The Prototype Fits (b) above, plus the other components of the pro-
totype listed earlier. There are no programs that fit all the elements of the
prototype in the NRP database, though some come close.
 131

Programs Predicted to Shine

How to Teach Reading: Modern Research

The programs outlined in this section are those most representative of the
prototype.

Lippincott and i.t.a. The Bond and Dykstra 1967 study included two
reading programs close to the prototype: the initial teaching alphabet
(i.t.a.) and the Lippincott program. (See chapter 4 for descriptions of these
programs.) Bond and Dykstra’s study was not included in the NRP data-
base because of their 1970 cut-off date. I am adding it here, because the
effect sizes represent over 3,000 children, providing the most accurate
estimate of the power of these programs. I computed the effect sizes for
i.t.a. and Lippincott on the Stanford Reading Test subtests. (‘‘Word
study’’ is a measure of phonics knowledge.) The comparison groups were
the basal-reader classrooms (see chapter 4).
Table 5.2 contrasts the effectiveness of these two programs. The
Lippincott program has much higher effect sizes than the i.t.a. program
and is superior to the basal programs in all respects (ES ¼ 1:12 for read-
ing, .61 for spelling, .62 for phonics knowledge, and .57 for reading com-
prehension). Perhaps having to learn, then unlearn, a special script (i.t.a.)
wastes time and/or causes confusion. More recent studies on Lippincott
that are in the NRP data pool are shown in table 5.2. Effect sizes are
smaller but generally confirm the Bond and Dykstra results, which are far
and away the most accurate reflection of the strength of this program.

Lindamood: Auditory Discrimination in Depth Another program that par-

tially fits the prototype (basic code, sound-to-print orientation) is the
Lindamood program. This program was originally designed to remediate
poor readers. Like Dale’s program, it focuses on how speech sounds are
produced and includes a number of exercises for classifying and analyzing
phonemes. Lessons were designed around various materials, such as pic-
tures of mouth postures, letters and digraphs printed on cards, felts, and
tiles, plus colored blocks and special charts. There is no writing compo-
nent in this program. A movable alphabet is used instead. The program
adheres largely to a one-to-one correspondence between phonemes and
their basic-code spellings. A major departure is that word work tends to be
confined to nonsense words. Some alternative spellings are taught (about
 132

Table 5.2
Chapter 5

Initial teaching alphabet, Lippincott program, effect sizes

Word Phonics
recog- knowl- De- Compre-
nition edge coding Spelling hension
Bond and Dykstra 1967
(all comparisons to basal readers)
i.t.a.
N ¼ 2,100
Grade 1 .49 .31 .11 .03
Lippincott
N ¼ 1,000
Grade 1 1.12 .62 .61 .57
Lippincott
(comparisons to whole language or basal)
Brown and Felton 1990
N ¼ 48
Grade 1 .02 .94 .51
Grade 2 .52 .68 .38
Fulwiler and Groff 1980
N ¼ 147
Grade 1 .81 1.65 .76
Silberberg, Iversen, and Goins 1973
N ¼ 69
Grade 3 .70 .36
Tests: Stanford Achievement Test, Woodcock Reading Mastery, Canadian TBS,
TWS Spelling, DST Decoding.

16) but no common sight words are included. Spelling activities are indi-
rect in the sense that children build and alter nonsense words using letter
tiles. The program takes, on average, 80 hours for poor readers to reach
normal reading levels.
Because of the population for which it was written, the Lindamood
program requires some modification and flexibility to teach beginning
readers. The exercises on phoneme analysis need pruning, and lessons
on the 40þ phonemes and their basic-code spelling need to be speeded
 133

up. With these modifications, the program can be highly effective in

How to Teach Reading: Modern Research

the classroom when taught in small groups. This was shown in the work
of Howard (1982, 1986) and in an unpublished study by Lindamood
(1991)—studies that did not appear in the NRP study pool.
Two studies using Lindamood were in the NRP database. These are
McGuinness, McGuinness, and Donohue 1995, a study on normal first
graders, and Torgesen et al. 1999, a study on at-risk kindergartners. In the
study I and my colleagues carried out, two classrooms were taught the
Lindamood program in small groups of five to eight children. The control
classroom was also taught in small groups for the same amount of time,
but received a combination of whole-language instruction, invented spell-
ing, and a sprinkling of phonics. The training took 40 minutes a day for
one school year, about 100 hours per group, roughly 15 to 20 hours per
child.
In the Torgesen et al. study, children were tutored individually for 80
minutes a week starting in the middle of kindergarten and continuing to
the end of second grade (if necessary), for an average of 88 hours. There
were three control groups, two of which were also tutored. One was
taught an embedded-phonics program designed by a local teacher, which
included sight-word memorization, letter-sound training, and reading
text. The second group was tutored on whatever was being taught in the
classroom, and the third control group had no tutoring. The results of the
two studies are shown in table 5.3. Results from the McGuinness,
McGuinness, and Donohue 1995 study include both my own analysis and
that of the NRP. I computed the effect sizes for the standardized tests only
from the Torgesen et al. study. I omitted the kindergarten data due to
large floor effects (too many zero scores). Because the Lindamood training
relies on nonsense words almost exclusively, test scores on ‘‘word attack’’
(nonword decoding) always exceed ‘‘word-recognition’’ scores.
The two sets of results are difficult to compare, given the different
formats. In one school year, the normal children taught in groups by a
single classroom teacher made at least the same gains as the at-risk chil-
dren with 2.5 years of individual tutoring plus their regular classroom
program. One would expect individually tutored children to do much
better, and in a shorter space of time. There are several possible reasons
why they did not:
 134

Table 5.3
Chapter 5

Lindamood effect sizes

Word Word Compre-
ID attack Spelling hension

McGuinness, McGuinness,
and Donohue 1995
N ¼ 42
Grade 1 .30 1.66
Torgesen et al. 1999
N ¼ 180
K to grade 1 .32 .71 .25
Grade 2 .48 .89 .43 .53
N ¼ 138
Tests: Woodcock Reading Mastery (reading, comprehension), WRAT (spelling).

1. There was a clash of methods in the tutoring study, with children

getting one method during tutoring and a different method (unknown) in
the classroom.
2. The Lindamood program is not easy to teach. In our study, I and
my colleagues used highly experienced classroom teachers who were
trained for one full week prior to the study, plus refresher workshops. In
the Torgesen study, 9 lead tutors received 18 hours of training initially,
plus extra bimonthly sessions. However, the ‘‘instructor aides’’ who appear
to have done the bulk of the tutoring received only 2.5 hours at the outset,
followed by 2.5 hours per month after this. The Lindamood clinic rec-
ommends 2 weeks of training before teaching this program.
Training is important with this method, because the speed through
the program is based on the child’s input moment by moment. Unless the
tutor/teacher knows how to present an activity properly, help children
correct errors quickly and effectively, and judge when an activity has be-
come counterproductive, they can get stuck reteaching the same exercise
over and over again. A typical example is phoneme-manipulation training
using colored blocks. The child is asked to alter a row of colored blocks to
match a sequence or chain of nonsense words: ‘‘If that is ip, show me pip.’’
(If /i/ is ‘‘blue’’ and /p/ is ‘‘red,’’ the child must add another red block
on the left.) This task has a high cognitive load, and is significantly corre-
 135

lated with age, vocabulary, auditory and visual short-term memory, and

How to Teach Reading: Modern Research

oral comprehension, with all values exceeding r ¼ :40 (McGuinness, Mc-
Guinness, and Donohue 1995).
3. The third explanation is related to the second. If children are
young, developmentally delayed, or have low verbal IQs (all of which was
true in the tutoring study), the program will be harder to teach, and tutors
need to be even more skilled. In our small-group study, the children were
older, came from upper-middle-class families, had high verbal ability, and
had the guidance of well-trained and highly experienced teachers. These
children had few problems with any of the lessons.
Nevertheless, we know from Lindamood’s own study in the Santa
Maria school district that with proper teaching, even sons and daughters
of migrant farmworkers can be taught effectively. In this study, first
graders were taught by the classroom teacher trained in the method plus
an experienced tutor from the Lindamood clinic who was in the classroom
each day. At the end of first grade, average reading scores were three years
above age norms. No child did worse than one year above national norms.
These children have been tracked for several years by the local school
district and continue to score well above norms. Because this study was
never published, we have no knowledge of how or whether the lessons
were modified to suit these children. The format in the study by Mc-
Guinness and associates was similar to the one in Santa Maria, minus the
expert aide. Our results were far more modest, and the obvious conclusion
is that teaching expertise is critical to the effectiveness of this program.

Open Court There is only one study in the NRP database on the Open
Court program. This is unfortunate, because this program has received a
good deal of national attention, and because it is a good fit to the proto-
type, at least according to the description provided by Foorman et al.
(1997, 67):

Components of the first grade program include: 1) Phonemic awareness

activities during the first 30 lessons (10–15 minutes daily); 2) Forty-two
sound/spellings are introduced in the first 100 lessons, one per day in lessons
11–30, and at a slower pace thereafter; phonics principles are reinforced
through sound/spelling cards, alliterative stories, and practice stories whose
vocabulary is tightly controlled for the sound/spelling just taught; 3) Blending
 136

is regarded as the key strategy for applying the alphabetic principle, and,
Chapter 5

therefore, 8–10 new words are blended daily; 4) Dictation activities move
from letter cards to writing words sound by sound, to whole words (by lesson
17), to whole sentences (by lesson 27); 5) Shared reading of Big Books; 6)
Text anthologies (with uncontrolled vocabulary), plus workbooks are intro-
duced in the middle of first grade, when all sound/spellings have been intro-
duced; and 7) Writing workshop activities are available in individual and small
group formats.

There are some peculiarities in this program, such as the extensive use of
color-coded text (both background and print) to mark consonants and
three types of vowel spellings. There is little empirical support for the use
of color-coded text, and there is a risk that children will come to rely on
the colors to the exclusion of noticing the specific print features. Trans-
ferring to normal text may cause difficulties. The program includes some
aspects of whole language, which can muddy the waters.
The second-grade program starts with a review of the sound-to-letter
correspondences, includes more blending exercises, and adds two new
anthologies. There is no mention of how spelling is taught after the basic
code has been mastered.
The study by Foorman et al. (1997, 1998) was ambitious, involv-
ing 375 children (1997), reduced to 285 (1998), who received tutoring
through Title 1 in addition to a classroom program. There were three
control programs: whole language with no special teacher training, whole
language with teacher training, and an embedded-phonics program with
teacher training. Open Court teachers (and tutors) were taught by Open
Court trainers. The embedded-phonics program was developed locally,
and was a visually driven method based on letter patterns and featuring an
‘‘onset-rime-analogy’’ type of instruction.
These children were scattered among 70 classrooms in all (70
teachers). The three methods were taught to the whole class, but only the
tutored children participated in the study, about 3 to 8 children per classroom.
Classroom time for literacy activities was 90 minutes daily, and tutoring
was provided for 30 minutes per day. Whether these periods overlapped is
unknown. This was a complex study design in which the tutorial method
sometimes matched the classroom program and sometimes did not. To
complicate matters further, the tutors had been trained the previous year
 137

Table 5.4

How to Teach Reading: Modern Research

Open court
Compre-
Reading Spelling hension

Foorman et al. 1998

N ¼ 285
Grades 1 and 2 combined scores
OC vs. embedded phonics .58 .36 .37
OC vs. whole language .48 .38 .31
(trained teachers)

Test: Wide Range Achievement.

Note: The NRP had access to data not in the published report and broke out ES
values by grade. However, their values—first-grade reading ES ¼ 1.63 and spell-
ing ES ¼ .56, as well as second-grade reading ES ¼ .32 and spelling ES ¼ .19
—do not tally with the published data. The NRP values are too large.

in Reading Recovery (a whole-language-type intervention program for

children who fall behind), and had to be retrained in one of the two new
methods. Half of the tutors had to deliver two different tutorial methods.
The results from the complete study, which lasted from early grade 1
to the end of grade 2, are in Foorman et al. 1998. I have computed the ES
values from the results presented in that article, which combines first- and
second-grade test scores. I also provide the ES values from the NRP re-
port (see table 5.4). The NRP appears to have had access to data not
available to the public, and broke out the first- and second-grade scores.
However, these values do not tally with the published data (the ES values
are too large).
This study did not provide a fair assessment of the Open Court pro-
gram, and better research is called for. So far, there is no published study
on Open Court in the classroom. The research design was overly complex
and there were other methodological problems. The authors stated that
the matched versus mismatched tutorials had ‘‘no effect,’’ and for that
reason ‘‘tutoring was ignored in subsequent analysis.’’ It is hard to under-
stand why extra one-on-one help for 2.5 hours per week for one school
year had no effect, especially when the lessons matched what was ongoing
in the classroom.
 138

Jolly Phonics Jolly Phonics was developed by Sue Lloyd (1992), a class-
Chapter 5

room teacher in England who devoted many years to perfecting this pro-
gram. It takes its name from the publisher, Chris Jolly. Jolly Phonics
meets nearly all the requirements of the prototype, and goes beyond it in
some respects. Jolly Phonics is the product of what can happen when
popular myths of how to teach reading are challenged.
First to go was the myth that reading is hard to teach. Second to go
was the notion that a linguistic-phonics program cannot be taught to
the whole class at the same time. Third to go was the age barrier. Jolly
Phonics is taught to 4-year-olds. Fourth to go was the belief that young
children cannot pay attention for more than about 10–15 minutes at a
time. Fifth to go was the related belief that if young children are kept at a
task for longer than about 15 minutes, they become bored and frustrated
and are unable to learn. Sixth to go was the idea that teachers need ex-
tensive training to teach the alphabet code properly.
Lloyd’s initial goal was to reduce the lessons to the essential elements
and present them at an optimum rate, as quickly and as in depth as pos-
sible. Undoubtedly, her greatest insight was in figuring out what these
elements are. Certainly nothing in teacher training provides any useful
information on this issue. The next questions were how these elements
should be taught and how early and quickly they could be taught, given
the appropriate sequence and format. She discovered that young chil-
dren forget what they have learned when lessons are spaced too far apart.
This necessitates constant reteaching and review, wasting an extraordinary
amount of time.
Lloyd discovered that a comprehensive reading program can be
taught to young children in a whole-class format if three conditions are
met. First, the lessons should be fun and stimulating and engage all the
children. Second, sufficient backup materials for individual work have to
be available to support what is taught in the lessons. Third, parents need
to be involved enough to understand the program and know how to sup-
port their child at home. When lessons are enjoyable and when children
see that they and their classmates are actually learning to read, they have
no trouble paying attention for up to an hour.
Lloyd found clever and ingenious ways to engage the whole class and
keep them interested. She invented simple action patterns to accompany
learning each phoneme. Children say each phoneme aloud accompanied
 139

by the appropriate action. An undulating hand movement accompanies

How to Teach Reading: Modern Research

the sound /s/, a finger to the lips represents the sound /sh/, and so forth.
Apart from being fun for the children, the action patterns fulfill a number
of functions. They help anchor the speech sounds in memory. Because the
actions are visible to everyone, including the teacher, they ensure that all
the children are engaged (no daydreamers allowed). In this sense, they
function as gentle peer support for everyone to get on board and learn
quickly. A similar group activity is used when writing training begins.
However, it is possible that these actions are not essential, and research is
needed to sort this out.
Jolly Phonics proceeds rapidly. Children learn one phoneme per day,
along with the accompanying action and letter symbol. They get hand-
writing training almost from the beginning, and are soon able to write the
basic-code spellings for the phonemes taught so far. The basic code is
taught in 9 to 10 weeks, taking about 50–60 hours of direct instruction.
After this, children move on to simple phonics books, are given spelling
dictation for words and sentences, and begin to learn spelling alternatives
(22 spelling alternatives are taught). Little teacher training is necessary.
There is a simple handbook with brief, clear instructions, and an excellent
video.
One study using the JP program (Stuart 1999) was included in the
NRP database, but there are several other studies in the literature from
the United Kingdom and Canada. Some were ongoing when the NRP
completed its search. I have calculated the effect sizes for these studies
(unless provided by the authors). Table 5.5 sets out the average scores on
standardized tests to illustrate the extraordinary gains. Table 5.6 provides
the effect sizes for each study.
None of these studies were exactly alike. The study truest to Lloyd’s
intentions was the study by Johnston and Watson (1997), which was car-
ried out at Lloyd’s school. The children were matched on a wide range of
skills (IQ, phoneme awareness, and so on) with a control group in Scot-
land that was learning the traditional Scottish method, known as analytic
phonics (similar to whole-to-part phonics). Children also learn to identify
phonemes in initial, final, then middle position, but these are taught in
isolation over an extended period of time.
Johnston and Watson’s second study (2003) used a variant of the JP
program designed by Watson, called: Fast Phonics First (FPF). The study
 140

Table 5.5
Chapter 5

Jolly Phonics standardized tests in age-equivalent or standard scores

Initial Read- Compre-
N age Training ing Spelling hension

Johnston and Watson 10 weeks

1997 50 hours
JP 25 4:8 age at test 5:9
AP 29 4:9 5:1 5:0
JP age at test 6:7
AP 5:4 5:3
JP age at test 8:3
AP 7:6 7:5

Johnston and Watson 16 weeks

2003 26 hours
JP/FPF 117 5:0 age at test 6:0 6:0
AP only 109 5:0 5:4 5:4 5:2
AP þ Phon 78 5:0 5:4 5:3
JP/FPF age at test 7:7 7:8 7:3
AP þ Phon þ FPF 6:7 7:5 7:5 7:0

Stuart 1999 12 weeks

60 hours
JP 55 5:0 age at test 7:1 6:8
WL 57 5:0 6:6 6:3 5:8
Sumbler 1999 20 weeks Word Word Spelling
33 hours ID attack
JP 145 6:0 age at test 107 100 104
Eclectic 120 6:0 6:5 101 86 98
JP ¼ Jolly Phonics, FPF ¼ Fast Phonics First, AP ¼ Analytic Phonics, Phon ¼
phoneme awareness, WL ¼ whole language.
 141

included all beginning readers (337) from Clackmannanshire, the smallest

How to Teach Reading: Modern Research

county in Scotland (nicknamed ‘‘Wee County’’). Children from the towns
with the most free lunches and clothing allowances were in the FPF
classrooms. The more affluent children constituted the control groups.
There were several modifications to Lloyd’s format. The action patterns
were dropped. Phonemes were introduced in the same order as in JP, but
at a slightly slower rate (just under one per day). The lesson periods were
shortened and spaced out over 16 weeks, totaling about 26 hours overall.
Magnetic letters were used to teach blending and spelling simple words
from the beginning. Otherwise, the content and sequence were the same,
with the exception that no consonant blends or spelling alternatives were
taught.
There were two control groups. Both received the standard analytic-
phonics fare, but were taught the same phonemes in the same sequence as
the FPF children. The children spent the same amount of time on the
lessons, but only 16 sound-letter correspondences could be taught in this
time frame. One of the control groups split the time with a phonological-
training component (no letters). This involved onset rime and phoneme
segmenting and blending. As can be seen in table 5.6, the phonological
training added something above analytic phonics alone. The effect sizes
for FPF versus AP þ Phon are smaller than for FPF versus AP only. Effect
sizes were large across the board in favor of the FPF program, close to 1.0
for reading and over 1.0 for spelling.
Stuart’s famous ‘‘docklands’ study’’ (Stuart 1999) was carried out in
London’s east end, in the impoverished docklands area. Children had little
spoken English, and 53 percent knew no English words whatsoever. The
JP program followed Lloyd’s format closely, and children had about 60
hours of instruction. Despite the fact that these children had such impov-
erished English-language skills, the results are remarkably like those of
Johnston and Watson’s in almost every respect. Furthermore, they held
up well over time.
In Sumbler’s observational study discussed earlier, half the chil-
dren were taught with JP. The duration of the lessons was considerably
shortened, and they were extended over the entire school year. The results
provide scientific support for Lloyd’s assertion that learning should be fast
and intense for maximum effect. The problem was not in the study design,
but in the reluctance of the kindergarten teachers to teach at this pace.
 Chapter 5

Table 5.6
Jolly Phonics effect sizes
Phoneme
segmen- BAS Young Nonword Compre-
tation reading reading decoding Spelling hension
Johnston and Watson 1997
N ¼ 53
JP vs. analytic phonics
Immediate (5:1) 1.52 .90
1st posttest (5:4) 3.27 1.0
142

JP group 2 (7:6) 1.1

Johnston and Watson 2003
N ¼ 304
FPF vs. analytic phonics (5:4) 1.46 .91 1.45
FPF vs. APþPA (5:4) .73 .85 1.17
Stuart 1999
N ¼ 112
JP vs. WL
1st posttest (5:6) 1.80 1.18 .63 1.49 1.39
2nd posttest (6:6) 1.13 .71 .62 .74 .86 .37
Sumbler 1999 Word ID Word Spelling
N ¼ 265 attack
JP vs. eclectic (6:5) .52 .68 .44
GRAND MEANS Phoneme All reading Nonword Spelling Compre-
segmenting decoding hension
143

1.65 .84 .89 1.22 .73

UK tests: Yopp-Singer segmenting; British Ability Scales: reading, spelling; Young Reading; Primary Reading Test; Neale Comprehen-
sion; Schonell Spelling; Willows Spelling.
Canadian tests: Woodcock Reading Mastery.

How to Teach Reading: Modern Research

 144

They did not believe that young children could learn a phoneme and let-
Chapter 5

ter per day, or endure hour-long sessions. There was a compromise. Les-
sons were reduced to 20 minutes or less, and total recommended hours
were nearly cut in half. As noted in the discussion of the observational
study, the JP program was taught along with a mix of other language
activities, reducing time still further. We can see the impact of the slower
delivery of the program, plus the lack of focus, in the lower effect sizes
(table 5.6). Sumbler commented that after working with the program,
some teachers realized that children could learn this material much more
quickly and had begun speeding up the training.
There were three follow-on studies among this group. Johnston and
Watson (1997) followed the children for an additional year, to 7.5 years
old. JP children maintained their gains and were about one year ahead of
the control group and national norms. Furthermore, one-third of the
control children scored more than one standard deviation below the mean,
but only 9 percent of the JP children scored this low. Stuart’s one-year
follow-up produced much the same results.
Johnston and Watson (2003) followed up the ‘‘Wee County’’ children
for an additional year. However, because of the initial success of the pro-
gram, school officials insisted that the entire county be switched to FPF,
and group differences were eliminated by the end of the next year. Now
all the children in the Wee County scored one year above national norms.
Further follow-on studies (see Johnston and Watson 2003) showed that
the advantage for all the children increased over time to two years above
national norms by age 9.5 on reading (decoding), and one year ahead in
spelling. Reading comprehension was only marginally above norms (see
table 5.7).
None of the prototype programs include anything other than the
most rudimentary attempt to teach the advanced spelling code. Yet spell-
ing scores were surprisingly high, certainly much higher than the na-
tional norms. Because norms are based on the current status quo, this
tells us what a parlous state spelling instruction is in. Merely teaching the
basic code the right way around, getting the logic straight, and adding a
dozen or so of the 136 remaining spelling alternatives makes an enormous
difference.
Several of the studies in this group measured phoneme awareness in
addition to the reading and spelling tests. Table 5.6 shows the enormous
 145

Table 5.7

How to Teach Reading: Modern Research

Fast Phonics First: Clackmannanshire follow-on results. Johnston and Watson
(2003)
Age-equivalent Years:months
scores Above national
Age in years:months norms
7:9 Reading 9:4 1:7
Spelling 8:6 :9
Comprehension 8:0 :3
8:9 Reading 10:6 1:9
Spelling 9:6 :9
Comprehension 9:2 :5
9:7 Reading 11:9 2:2
Spelling 10:9 1:2
Comprehension 10:0 :3
U.K. tests were: British Ability Scales (reading); Schonell Spelling Test, Primary
Reading Test (comprehension).
Source: Scottish Executive Education Department. Internet publication: Insight
www.scotland.gov.uk/sight/

impact of the JP/FPF programs on phoneme-awareness scores. Johnston

and Watson used the Yopp-Singer segmenting test, a reliable test that is
highly correlated with decoding skills. This raises the question of whether
special (additional) phonological-awareness training programs are really
necessary—the topic of the next chapter.
What is most remarkable about the Jolly Phonics research is the con-
sistency of the results. This is in striking contrast to most of the studies in
the NRP database. The grand means are provided in the final row of table
5.6, and these values act as a benchmark for what can be achieved in
whole-class beginning reading instruction.

Potpourri
Before we leave the assessment of reading methods, I want to report on
some of the other programs that made it into the NRP database. A brief
synopsis of these studies provides a startling glimpse of the confusion and
variability that abound in research on reading methods, and in the use of
the term phonics.
 146

A major source of this confusion is the lack of specificity of program

Chapter 5

content. As a general rule, reading researchers with a background in ex-

perimental psychology ignore what is being taught in the classroom. They
craft clever training programs and carefully match all the groups in the
study on important variables, like age, IQ, and so forth. The experimental
groups then get special training (described in great detail) while the con-
trol group participates in the ‘‘usual classroom program,’’ whatever this
might be. The actual program content is presumed to have so little impact
(positive or negative) on the outcome of the study that the program itself
is not even identified. But the classroom program could be complemen-
tary or contradictory to one or more of the experimental treatments. We
saw this problem in Torgesen’s study above, where children were indi-
vidually tutored in a highly structured linguistic-phonics program, while
they participated for at least 5 hours a week in some unidentified ‘‘class-
room program,’’ probably whole language.
Researchers with a background in education take more care to pro-
vide information on the classroom programs, but these descriptions are
often vague and misleading. As noted earlier, descriptions of ‘‘basal phon-
ics’’ make so little sense that they strain the imagination.

More Phonics A new phonics program designed by Blachman et al. (1999)

was included in the NRP’s database for reading instruction. This program
was an extension of a kindergarten phoneme-awareness training program
designed by Ball and Blachman (1991). The study design (kindergarten
through grade 2) does not allow for an independent assessment of the two
programs. For this reason, the Blachman program will be considered in
the following chapter on phoneme-awareness training.
Two published programs were tested in citywide or countywide
studies using large numbers of students. The first was a reading program
designed by Beck, called the New Primary Grades Reading System (NRS)
(Leinhardt and Engel 1981). It was tested over a two-year period on first
and second graders. It was described as ‘‘an eclectic code-breaking ap-
proach as opposed to a whole word approach,’’ with explicit instruction
in letter-sound correspondences along with extensive blending practice.
After initial learning (levels 1 and 2) children read books and listened
to audiotapes. The content of these tapes was not described. The con-
 147

trast programs were basals, either Scott-Foresman, Harper and Row, or

How to Teach Reading: Modern Research

Houghton-Mifflin.
At the end of the school year, the first-grade children were tested on
the Stanford Achievement Tests (SAT) reading total, and the second-
grade children on the California Test of Basic Skills (CTBS) reading total.
The NRS program worked well in first grade but did not hold up in sec-
ond grade. The ES values for the SAT for three large first-grade studies
were .76, .67, and .44. The ES values for the second-grade children were
essentially zero, reasons unknown. Unfortunately, there was a major con-
found in this study. NRS teachers assigned about 100 more minutes of
reading instruction per week than the basal-reader teachers, which essen-
tially invalidates this study.
Bond et al. (1995–1996) looked at the impact of a published program
called Sing, Spell, Read, and Write (SSRW ). This was a partial test of the
method, because the district agreed to allow only 20 lessons. A feature of
the program is singing, as the title suggests. There are several audiotapes
of songs about ‘‘phonics generalizations’’ (term unexplained). Otherwise,
this is visual phonics, strongly letter driven. Early lessons involve memo-
rizing the ‘‘sound’’ of each letter of the alphabet. ‘‘Phonetic Storybooks’’
are introduced after a few letter sounds are learned. Spelling and writing
activities are included in the program. The program was tested on three
age groups: kindergarten, first grade, and second grade, using nearly 900
children. Effect sizes for the Woodcock letter-word identification test
were low across the three grades (.39, .24, .44), but much higher for ‘‘word
attack’’ (first grade .60, second grade .56). Most kindergartners scored
zero on this test. There was no special advantage of the SSRW program
for oral reading (first grade .02, second grade .33).

Whole Language There were three studies specifically oriented to whole

language. Klesius, Griffith, and Zielonka (1991) compared whole language
to a ‘‘traditional basal’’ program that emphasized ‘‘explicit phonics.’’ The
experimental question was whether first graders in a whole-language en-
vironment, who were ‘‘submerged in print’’ and doing a ‘‘range of reading
and writing activities,’’ could learn the alphabet principle ‘‘implicitly’’
without direct instruction. (This is the general claim of whole-language
advocates.) There were 112 children from 6 classrooms in the study.
 148

Teachers received additional training and continuing help from the re-
Chapter 5

searchers. The whole-language activities were typical and involved listen-

ing to the teacher read children’s books, poems, and rhymes, while the
children read along. The children did a lot of writing, far more than the
children in the basal classroom. However, they had little, if any, formal
spelling instruction.
The basal program was not well described, other than that it included
‘‘isolated skill instruction,’’ including ‘‘letter-sound correspondences,’’
plus directed reading from basal readers. There were weekly spelling tests.
The amount of writing varied extensively from one classroom to the next.
Children also worked independently on phonics workbooks, doing exer-
cises like circling words or objects that started with the same sound.
The basal-phonics program produced generally higher scores on a
battery of reading, spelling, and writing tests. Effect sizes were low for
comprehension (.10), marginal for word recognition, spelling, and writing
(range .30 to .34), moderate for phoneme awareness (.56), and large for
nonword decoding (.71). This, of course, does not mean that this ‘‘tradi-
tional basal’’ program is a good program, only that whole language was
worse. (The NRP ES values were highly discrepant from the values
shown above.)
In a second study on the same two reading methods by the same
team (Griffith, Klesius, and Kromrey 1992), children were selected on the
basis of extreme scores (high, low) on a phoneme-awareness test. The
goal was to find out whether children with high phoneme awareness did
better overall, and whether the type of reading program was influenced
by phoneme-awareness skills. Unfortunately, when children are sorted
into extreme groups, other factors come into play. IQ is a major factor
on some phoneme-awareness tests, and this was not controlled. Nor was
home environment. The children with high phoneme-awareness skills
may have been taught to read at home. In any case, children with high
phoneme awareness did better in both types of classrooms. For the most
part, the type of reading instruction made no difference (results were not
significant). This study had too many uncontrolled variables to be valid.
Also, several reading test scores were at ceiling for the high-phoneme-
awareness children. (The NRP reported very large effect sizes in favor of
the basal-reader group, which is clearly incorrect and misrepresents what
happened in this study.)
 149

The final study (Eldredge 1991) was an attempt to improve whole

How to Teach Reading: Modern Research

language by adding a phonics component for 15 minutes each day. Three
classrooms of ‘‘modified whole language’’ were compared to three class-
rooms using basal-reader programs. Children were all ‘‘low achievers’’ in
first grade. From the description of the phonics supplement, this was
partly visual phonics (‘‘20 vowel-team graphemes’’), plus a variety of
phoneme-analysis activities: identifying sounds in initial, middle, and final
positions in a word, lessons in segmenting and blending, and so on. Chil-
dren learned spelling patterns for certain syllable types. The basal pro-
gram (Houghton-Mifflin) also appeared to be whole language oriented.
Most of the activities were devoted to reading, vocabulary training, and
sentence building. Some letter-sound relationships were taught, but no
segmenting or blending.
Results showed a consistent superiority of whole language plus
phonics over the basal programs. The effect sizes were .44 for phonics
knowledge, .55 for word recognition, .73 for reading total, and .83 for
comprehension (the last three tests from the Gates-MacGinitie battery).
This is the only study to show a substantially higher effect size for com-
prehension than for basic tests of word recognition and word attack.
However, this study was comparing apples and oranges, and the evidence
is not convincing that whole language plus phonics causes the advantage,
or simply phonics alone.

Conclusions
How valid and reliable is a meta-analysis for research of this complexity
where nearly every study uses a different method, involves different types
and ages of children, employs different research designs, and utilizes dif-
ferent measures of reading and spelling competence? I should point out
that nearly all reading and spelling measures in these studies were prop-
erly constructed, normed, standardized tests. This does at least ensure
that the effect sizes (when computed accurately) are statistically valid and
reliably represent group differences on these measures. This is a bonus,
because standardized tests are very much the exception in the phoneme-
awareness training studies reviewed in the next chapter. Nevertheless, effect
sizes will vary for other reasons, such as study design and extraneous factors.
I have taken special care to analyze each study separately in detail and
to compute effect sizes myself (many times). This revealed that the com-
 150

putations in the NRP report were not always accurate. Sometimes this was
Chapter 5

due to misunderstanding the study design. For example, one study in the
database (Martinussen and Kirby 1998) had nothing to do with comparing
reading methods. Instead, it focused on whether teaching different strat-
egies for learning the same reading program made any difference. Results
were not significant for reading, and there were huge floor effects on the
Woodcock reading tests with standard deviations three times higher than
the means. Yet effect sizes were computed anyway.
Besides noting anomalies like this, I was concerned about the fact that
my values sometimes deviated significantly from those in the NRP’s
tables, especially since we were using the identical formula. In part, this
was because I was more interested in individual test scores, whereas the
NRP was more likely to collate the data from several tests. However, in
other cases, there were gross computational errors. For example, Griffith,
Klesius, and Kromrey (1992) (see previous section) reported no significant
differences between two contrasting reading methods on every reading
and spelling test except one. ‘‘No significant differences’’ translates into
low or zero effect sizes, which is what I found. The NRP, however, re-
ported large effect sizes in every case (some higher than 1.0) in favor of
whole language. Errors like these bias the general meta-analysis results (in
this case away from phonics and toward whole language).
In view of the nature of meta-analysis as a statistical tool, and the
problems outlined above, the overall summary of the panel was disap-
pointing. They relied exclusively on the summary tables and argued from
global effects to conclusions. There were sweeping generalizations like the
following: ‘‘Phonics instruction failed to exert a significant impact on the
reading performance of low-achieving readers in second through sixth
grades’’ (National Reading Panel, 2000, 2-133). This is a dangerous
statement. It implies that no phonics instruction works for poor readers or
that none of the programs in the NRP database was effective for this pop-
ulation, neither of which is true. The statement also gives the false im-
pression that all remedial-phonics programs found their way into their
data pool, when this was far from the case. In fact, the most successful
remedial-reading programs today were missing (see D. McGuinness
1997c; C. McGuinness, D. McGuinness, and G. McGuinness 1996).
As for beginning reading instruction, the panel concluded that the
meta-analysis value (MS ¼ .44), the one we started with at the beginning
 151

of this section, ‘‘provided solid support for systematic phonics’’ (p. 2-132),

How to Teach Reading: Modern Research

when this value is marginal and was the outcome of a composite of every
program comparison.
In a question-and-answer section (on pages 2-132 to 2-136), they
asked the question ‘‘Are some types of phonics instruction more effective
than others?’’ They answered it by loosely grouping the studies (66 cases)
into three types. They found no statistical differences between the effec-
tiveness of these types and concluded that the type of phonics program did
not matter. Once more, this statement is misleading. As we have seen, the
type of phonics program matters enormously. The real issue here is how
you classify them.
The term synthetic phonics is particularly troublesome. It was defined as
‘‘teaching students to convert letters (graphemes) into sounds (phonemes)
and then to blend the sounds to form recognizable words.’’ This definition
suggests synthetic phonics is synonymous with what I call visual phonics.
The panel does not seem to recognize the distinction between visual
phonics (letter driven) and linguistic phonics (phoneme driven), which is
clear from the way the summary tables are coded. Visual-phonics pro-
grams and linguistic-phonics programs were both coded as ‘‘synthetic
phonics.’’
The problems created by (visual-phonics) logic appear in the
question-and-answer section. The panel asked this question: ‘‘How many
letter-sound relations should be taught and how many different ways of
using these relations to read and write words should be practiced?’’ Later,
they posed the same question again and answered it: ‘‘It is clear that the
major letter-sound correspondences, including short and long vowels and
digraphs, need to be taught.’’
But what is clear? And what is ‘‘major’’?
In linguistic-phonics logic, neither the question above, nor the an-
swer, makes sense. Children need to be taught the 40 sounds in the
English language and their most common (probable) spelling. Everything
follows from this. The real question is: When and in what order should
the remaining 136 spelling alternatives be taught?
When the NRP addressed guidelines for future research, many ques-
tions the panel raised were already answered by studies in their data pool,
or by the research literature reviewed in this chapter. These were ques-
tions like the following: ‘‘How many months or years should a phonics
 152

program continue?’’ (It takes one semester to teach the basic code.)
Chapter 5

‘‘What are the active ingredients of effective systematic phonics pro-

grams?’’ (The key components include sound-to-letter logic, a basic code
marking all 40 phonemes, the integration of reading and spelling, the
teaching of segmenting and blending, and lots of writing.) ‘‘How does the
use of decodable text as early reading material contribute to the effective-
ness of phonics programs?’’ (Certainly, text that is too difficult to decode
contributes nothing.)
As for my own conclusions, I think the news is exciting. The
linguistic-phonics programs—those most like the prototype—were the
clear winners, nearly all producing very large effect sizes, despite the
enormous variation in subject populations (including children who had no
spoken English). Moreover, these effect sizes are sustained or even in-
crease over several years. We do know how to teach reading efficiently
and effectively. What we need is more research on these good programs to
pin down the details. This would include an exploration of materials,
activities, sequence, and pace, plus an assessment of the value of special
program features like the Jolly Phonics ‘‘actions,’’ or the analysis of artic-
ulatory motor patterns used in the Lindamood program. Do they really
help? We need to compare the various linguistic-phonics programs with
one another and stop doing research comparing ‘‘phonics’’ to whole lan-
guage and basals. We know that whole-word methods do not work. We
have abundant evidence that whole language is a dismal failure, and that
the majority of basal-reader programs are not far behind.
In chapter 11, I provide more information on important research
questions for the twenty-first century. Meanwhile, we will look at
phoneme-awareness training programs. Are they really necessary?
 6
PHONEME-AWARENESS TRAINING

The importance of phonological- or phoneme-awareness training for

mastering an alphabetic writing system has been the dominant theme
in reading research for 30 years ( phonological refers to all sound units,
including words, syllables, and phonemes). The NRP located nearly 2,000
studies on phoneme awareness. There is no question that performance on
certain phonological tasks correlates strongly with reading and spelling
skills. What is in question is the nature of this correlation, the type of
phonological skills that are important, and their precise role in reading
and spelling.
According to the theory that ‘‘phonological awareness develops’’ over
childhood, children gradually become aware of words, syllables, syllable
fragments, and phonemes, in that order. Development is long and slow,
with ‘‘explicit awareness’’ of phonemes emerging around age 6 or 7
(Liberman et al. 1974; Adams 1990). This theory has had a profound im-
pact on the design of phonological training programs, with important
consequences for when phonological awareness is taught, what is included
in the lessons, and in what order.
Research on speech perception and language development, as well as
the training studies themselves, do not support this theory. Awareness
of phoneme contrasts (ba-da) is present at birth, and phoneme analysis
becomes operational by about 6 months of age, when infants begin to
extract words from the speech stream (spoken phrases) and build a recep-
tive vocabulary. At the babbling stage, infants learn how spoken language
is produced mechanically, and check on how they are doing through au-
ditory feedback (Vihman 1993; also see Language Development and Learn-
ing to Read ). That children are truly aware of individual sounds is evident
from the fact that they begin to self-correct mispronounced phonemes in
 154

their speech before age 3 (Chaney 1992). (These studies are reviewed in
Chapter 6

depth in the companion book cited above.)

Two-year-olds are generally incapable of demonstrating that they
know what they know, but 3-year-olds can. Chaney found that 93 percent
of 3-year-olds (average age 44 months) could listen to a sequence of iso-
lated phonemes, blend them into a word, and select this word from a set
of pictures with accuracy well above chance, scoring 88 percent correct.
This was the second-easiest task in a battery of 22 language tasks. The
easiest task was saying a ‘‘word’’ (‘‘Can you say a word for me?’’), showing
that young children are well aware of what a ‘‘word’’ is, despite another
myth to the contrary. Another easy task was detecting a mispronounced
phoneme in a phrase and fixing the error by saying the words in the
phrase correctly.
On the other hand, children had enormous difficulty with tasks that,
according to the phonological-development theory, are ‘‘developmentally
early,’’ like word play, rhyming, and alliteration. Few children could judge
whether a word ‘‘sounded like’’ a series of spoken rhyming words, and
almost no one could provide a matching rhyming word. They had even
more difficulty with alliteration (matching initial sounds). These findings
show two things: children’s natural sensitivity is to words and phonemes,
and children have difficulty making abstract judgements such as whether
fragments of words ‘‘sound alike.’’
This is not to say that explicit awareness of phonemes is instantly on
tap when children have to learn an alphabetic writing system. As percep-
tion and production of speech increase in efficiency, they begin to run on
automatic pilot, operating below the level of conscious awareness, the exact
opposite of the developmental theory. Adults are not aware of individual pho-
nemes as they listen or speak; their attention is focused on meaning. Yet
most people can detect a phonemic error and correct the speaker without
a moment’s hesitation, though they may be uncertain where the error
occurred. The brain performs these functions for us, and we do not need
to monitor them.
The only reason anyone would need to be aware of phonemes is if
they had to learn an alphabetic writing system. But once the necessary
skills are mastered, phonemes ‘‘go unconscious’’ once more. Ask a family
member or friend how many sounds are in the word straight, and you will
get a number between 4 and 8. (Try it yourself.) Because phoneme anal-
 155

ysis occurs below the level of conscious awareness, children (or illiterate

Phoneme-Awareness Training
adults) must be taught to pay attention to the phonemic level of speech to
learn an alphabetic writing system. This does not necessarily require ex-
plicit awareness (‘‘awareness that’’), as Cossu, Rossini, and Marshall’s (1993)
research on children with Down’s syndrome has shown. These children
were able to learn the sound-symbol correspondences of the Italian alpha-
bet code by simple matching and repetition, and could read (decode)
real and nonsense words at fairly high levels of skill. Yet they could not
comprehend what they read and failed dismally on phoneme-awareness
tests.
Another powerful component of the myth is that phoneme awareness
is difficult to teach. This is said to be because phonemes are coarticulated
and hard to tease apart, and because consonants cannot be produced in
isolation from a vowel. Thus the best rendering of a segmented word
like cat is ‘‘kuh-aa-tuh.’’ Neither statement is true. Only five of the 40þ
English phonemes are hard to produce in isolation. These are the voiced
consonants: /b/, /d/, /g/, /j/, and /l/. But even they can be managed by
keeping the vowel extremely brief. No other consonants need to be pro-
duced with a vowel, and certainly not /k/ and /t/.
Because the evidence shows that children have good phoneme sensi-
tivity, and isolating phonemes in words is not nearly as difficult as re-
searchers seem to believe, do we really need special phoneme-awareness
training programs? More to the point, children do need to be made aware
of phonemes to use an alphabetic writing system and to learn to match
each phoneme to its letter symbols. But this is what a linguistic-phonics
program teaches. What special benefit, then, does a phoneme-awareness
training program confer?

Are Special Phoneme-Awareness Training Programs Really Necessary?

Answering this question will take up most of this chapter. To help with
this comparison, let’s review how a linguistic-phonics program works in
practice.
Lessons begin by drawing the children’s attention to a particular
phoneme by producing it in isolation (/sss/), and/or having the children
discover it by listening to a little story that features the sound over and
over. For example, here is the opening section of a story for introducing
the sound /p/ in a program I designed (McGuinness, forthcoming).
 156

Pretty Penny is happy.

Chapter 6

She has a pig named Poppy.

Poppy is a pet.
The best pet yet.

Most pigs are sloppy.

Not Poppy.

Poppy is perfectly pink and pale.

From the top of her head to the tip of her tail.

Notice that the sound /p/ appears in all positions in a word. After the
children learn to hear this sound, and to say the sound in first, middle, and
last positions, they see the letter p and are told that this letter represents
the sound /p/. They practice tracing, copying, and writing each new letter
until its shape is firmly embedded in memory. As they write, they say the
sound the letter stands for.
Once a few sounds are introduced, they are combined to make real
words (not nonsense words). This way children get to ‘‘meaning’’ as soon
as possible and understand the value and purpose of the exercise. Real
words can be written down, sound by sound, and retrieved (decoded) from
print. Phonemes and their basic-code spellings are introduced as quickly
as children can learn them, every lesson building on the one before, until
all 40 phonemes and their spellings are mastered. Spelling alternatives
follow next.
Children learn to segment and blend phonemes by seeing and writing
letters. They learn that phonemes occur one after the other over time, and
that letters are sequenced one after the other over space (left to right).
Reading and writing (spelling) are integrated in the lessons so that chil-
dren learn the code nature of our writing system.
These are the basic building blocks of a good linguistic-phonics
program.
The central question is whether a phonological-training component
adds anything to what I have just described. Will it enhance phoneme
analysis, reading, and spelling skills beyond the programs that fit the pro-
totype outlined in the previous chapter? To prove that it does, it would
 157

have to produce still larger effect sizes for reading, spelling, and phoneme

Phoneme-Awareness Training
awareness than the successful phonics programs alone, with ES values well
above 1.0.
To answer this question, I will draw on the efforts of the National
Reading Panel once again. Their focus was on phoneme awareness. To
meet the selection criteria, the study had to include training in phoneme
awareness. The panel unearthed 1,962 articles referenced by this term,
and they were screened by these criteria:

Had an experimental or quasi-experimental design with a control group.

Appeared in a refereed journal.

Tested the hypothesis that phoneme-awareness training improves read-

ing beyond some alternative form of training or no training.

Phoneme-awareness training was not confounded with other types of

training.

Statistics had to be adequate to compute effect sizes.

Only 78 studies passed this screening, and careful reading eliminated 26

more, leaving 52 studies in the NRP database. These 52 studies included a
total of 96 comparisons or ‘‘cases.’’ Using the NRP values, I prepared a
simplified, summary table of the important findings (see table 6.1). Bear in
mind that these studies are just as variable in their own way as the studies
on phonics instruction. The table sets out the most informative compar-
isons. First, we see that teaching phoneme awareness improves perfor-
mance on phoneme-awareness tests, compared to alternative training or
no training, proving, at least, that phoneme awareness is easy to teach.
However, being ‘‘phonemically aware’’ has little functional utility
unless there is some causal impact on learning to read and spell. A major
tenet of the ‘‘phonological-awareness-develops’’ theory is that children
should learn to read after they become aware of phonemes, with the added
proviso that if phoneme awareness is not ‘‘developing’’ in a timely fashion,
children need additional help. Nevertheless, the important question is
whether phoneme-awareness training per se has an impact on learning to
read.
The simple answer to this question is no. Here is the evidence. For
the combined data (72 cases), the meta-analysis showed that phoneme-
awareness training has a moderate impact on reading (ES ¼ .53) and
 158

Table 6.1
Chapter 6

Phoneme-awareness training effect sizes by category based on NRP analysis

Pho-
neme
N aware- N Read- N Spell-
Contrast cases ness cases ing cases ing
Immediate 72 .86 90 .53 39 .59
Follow-up 14 .73 35 .45 17 .37
K only 39 .95 40 .48 15 .97
1st grade only 15 .48 25 .49 16 .52
English 61 .99 72 .63 32 .60
Foreign language 11 .65 18 .36 7 .55
Letters used 39 .89 48 .67 27 .61
No letters 33 .82 42 .38 12 .34
Teacher/trainer 64 .89 82 .55 33 .74
Computer 8 .66 8 .33 6 .09
In-house test 58 .61 24 .75
Standardized test 39 .33 20 .41
Composite* eliminating: 14 1.10 45 .74 24 1.01
3rd grade or higher
Foreign language
Computers
N ¼ <20
Hours ¼ <3
* Not corrected for letters/no letters or for in-house vs. standardized tests.

spelling (ES ¼ .59). But when the training is purely auditory (subtracting
all studies where letters were used), the impact on reading and spelling is
substantially reduced (ES values shrink to .38 and .34). Also, if reading is
measured by a standardized test instead of an experimenter-designed test,
the impact of phoneme-awareness training on reading and spelling shrinks
as well (ES ¼ .33 and .41). In-house tests inflate effect sizes, because they
measure what was taught.
Another meta-analysis of phonological-awareness training studies was
provided by the Dutch team of Bus and Van IJzendoorn (1999), using a
 159

smaller database (36 studies). They focused mainly on phoneme aware-

Phoneme-Awareness Training
ness, excluding training on rhyme and syllable segmenting. They reported
an effect size of 1.04 for the impact of phoneme-awareness training on
phoneme tests, and .44 for reading. However, follow-on studies showed
the impact of phoneme-awareness training on reading was nil (ES ¼ .16).
Bus and Van IJzendoorn commented on the enormous variability be-
tween these studies. In an attempt to reduce the ‘‘noise’’ in the data, they
recomputed the meta-analysis using only the studies from the United
States where children were randomly assigned to groups or matched.
Surprisingly, the effect size for the impact of phoneme training on pho-
neme awareness declined (ES ¼ .73), while it rose considerably for read-
ing (ES ¼ .70). Bus and IJzendoorn were overly optimistic about this
result: ‘‘The training studies settle the issue of the causal role of phono-
logical awareness in learning to read: Phonological training reliably en-
hances phonological and reading skills. About 500 studies with null results
in the file drawers of disappointed researchers would be needed to turn
the current results into nonsignificance’’ (p. 411).
We will see shortly that a more careful reading of these U.S. studies
will turn these results into ‘‘nonsignificance.’’
The NRP’s data pool reflected a similarly diverse group of studies
with different types of training, different tests measuring different things,
and different populations in different countries. Eleven studies were
carried out in non-English-speaking countries, including Israel, which
does not even have an alphabetic writing system. Some of these countries
have transparent alphabets (Finland, Germany, Spain, Sweden, Norway),
and some do not (Denmark). In countries with a transparent alphabet,
standardized reading tests measure reading speed (fluency), not accuracy,
while the reverse is true for countries with opaque alphabets. Subtract-
ing foreign-language studies increases the connection between phoneme
awareness and reading for studies in English-speaking countries (English:
ES ¼ .63; foreign: ES ¼ .36). Eliminating the computer-training studies,
which were singularly unsuccessful, had the same impact: (teachers: ES ¼
.55; computers: ES ¼ .32).
The NRP summary tables illustrate shifts in effect sizes when one
variable at a time is subtracted. However, tables were also provided for
each study, listing individual effect sizes along with information on im-
portant descriptors, like the number of children in the study, the country
 160

where it was conducted, the number of hours the training lasted, and so
Chapter 6

forth. This makes it possible to subtract studies that obscure a true picture
of the findings. To this end, I recomputed the average ES value for read-
ing and spelling (see the bottom row of table 6.1) after eliminating studies
with these characteristics:

Foreign-language studies

Computer-training studies

Total N less than 20 children

Training lasting less than 3 hours

Studies on older poor readers (grade 3 or higher)

The new ES values are now much larger: ES ¼ .74 for reading and
ES ¼ 1.01 for spelling. These results are more in line with the results
from the good phonics programs, and in better agreement with Bus and
IJzendoorn’s stripped-down meta-analysis of U.S. studies. This seems to
show that phoneme-awareness training has a strong impact on reading
and spelling. But does it?
Two pieces of information were missing from the NRP’s tables on
the individual studies. The first is whether letters were included in the
training, and the second, which reading tests were employed. After careful
reading of the remaining studies in the data pool, and recomputing many
of the effect sizes, I again found that the NRPs effect sizes were not always
accurate, and I will comment on any discrepancies as we go along.
NRP’s effect sizes were often based on the combined scores of in-
house and standardized tests. Reviewing the individual studies, I found
that in-house tests lacked reliability estimates and were highly specific to
what was taught. This will grossly inflate effect sizes, giving a false picture
of what these training studies accomplished. By contrast, the effect sizes
for reading instruction were based on standardized tests (see the previous
chapter). If one wants to know the true impact of a phoneme-awareness
program on reading, the same rigorous measures of reading and spelling
must apply. To examine this problem more closely, we have to turn to the
studies themselves.
In the remainder of this chapter, I will be reviewing those studies
left in the data pool after the studies in the categories listed above
 161

were eliminated. I will be looking at studies that compared phoneme-

Phoneme-Awareness Training
awareness training with or without letters, and examine the outcomes for
experimenter-designed tests versus standardized tests. I will also examine
the impact of various types of training, such as programs using larger or
smaller phonological units, and how, precisely, phoneme awareness is
taught. To this end, I have included a few well-designed foreign-language
studies that investigated neutral factors such as the impact of teaching
phoneme awareness with or without letters.

A Glossary
Before we move on to the analysis of the individual studies, I want to clear
up some terminology problems. There are several types of phoneme-
awareness skills. Suffice it to say that there are no standardized tests of
these skills, and few tests have norms. The descriptive terms for these
skills vary as well. Here is a glossary of how I intend to use terms.

Discrimination. Telling phonemes apart. ‘‘Are the sounds /v/ and /v/ the
same or different?’’

Identification. Being able to identify (say the sound) when asked for a
phoneme in a particular location in a word (first, middle, final position).
‘‘What is the first sound in cat?’’

Sequencing. Being able to say phonemes in the order in which they appear
in a word. There are three teaching methods:

Stretching phonemes. mmmmmaaaaannnn for man.

Segmenting. Isolating each phoneme in sequence: /m/ /a/ /n/.

Blending. Being able to join up isolated phonemes to form the word.

Manipulation. Mentally delete, add, or reverse individual sounds in words,

close up what remains, and say the word: ‘‘Say stand without the /t/.’’

There is considerable confusion about the terms segmenting and blend-

ing and how these relate to reading and spelling. Many people believe that
reading involves blending, and spelling involves segmenting. According to
 162

this view, when we read, we rapidly translate letters into phonemes and
Chapter 6

blend them into the word. When we spell, we say the word, segment the
sounds, translate each sound into a letter or letters, and write it down.
This is not what happens.
Children see an unfamiliar word: sting.
To read the word, they sound out each phoneme (segmenting): /s/ /t/
/i/ /ng/. Then they blend the sounds into the word and check the out-
come. It is quite common for beginning readers and poor readers to seg-
ment correctly and blend incorrectly: /s/ /t/ /i/ /ng/—sing.
To spell the word sting the children say the word, hear each segment
in sequence, and blend the segments into the word as they write. Seg-
menting and blending are intimately connected in both reading and
spelling. Even when reading and spelling are efficient, and it seems like
processing is instantaneous, it is not. The same sound-by-sound analysis
continues, only at a phenomenal speed. This is what is meant by automa-
ticity. Our brains operate in the realm of milliseconds, while conscious re-
flection operates in the realm of seconds.
We become aware of the interplay of these two processes when we see
a word we cannot read or start to write a word we cannot spell. My spell-
ing occasionally falters when I write technical words on the board, an in-
teresting phenomenon caused by the 90 degree rotation from the normal
plane and by the wider visual angle. To solve this I have to consciously
slow down, segment each sound, then deliberately blend the sounds vo-
cally as I write.
As I review the individual studies in the next section, the litmus test
for evaluating the relevance of a phoneme-awareness training program
will be the evidence from the Jolly Phonics research that 4-year-old chil-
dren can learn the 40 sounds of English and their basic-code spellings in
about 11 weeks of whole-class teaching. At the end of the school year,
these children were one year advanced on standardized reading and spell-
ing tests compared to the control groups, and compared to national
norms. The average effect sizes for the JP studies were 1.60 for phoneme
segmenting, 1.0 for reading, and 1.42 for spelling. Follow-on data showed
that an effect size of 1.0 for reading held up for 3 years, the longest time
measured. To make the case for a separate phoneme-awareness training
regime, it would have to produce results better than this.
 163

An Analysis of the Studies in the Data Pool

Phoneme-Awareness Training
Phonological-awareness programs differ in a number of ways, including
the size and the type of phonological units taught, how much of the focus
is on larger versus smaller units, how many phonemes are taught, and how
long it takes to teach them. Many programs in my selected data pool teach
larger units of sound as well as phonemes. What works best?

Big and Little Phonological Units

Lundberg, Frost, and Petersen (1988) designed a training program that
adhered to the phonological-development theory. This Swedish team
tested their program on 400 Danish kindergartners. Kindergarten begins
at age 6 in Denmark. Lessons included segmenting sentences and phrases
into words, segmenting syllables, and playing rhyming games, which took
up 14 weeks of an 18-week program. Matching initial phonemes began in
week 15, and phoneme-segmenting and phoneme-blending tasks did not
appear until week 17. This program had no impact on reading measured
later in time compared to the control group that got no training (play is
the main activity in Danish kindergartens). The effect size for reading was
ES ¼ .19. Results were much better for spelling (ES ¼ .60).
In an earlier correlational study on Swedish children (Lundberg,
Olofsson, and Wall 1980), performance on rhyming and syllable tasks was
found to be weakly correlated with tests of reading and spelling, while
correlations between phoneme tasks and reading and spelling were mod-
erate to high, with r ranging from .41 to .78 for reading, and from .49 to
.58 for spelling. These results indicate that phoneme-analysis skills are
useful in learning to read, and rhyming and syllable segmenting are not. In
view of these results, it is odd that Lundberg’s training program focused
so extensively on skills that were found to be unrelated to reading and
spelling.
A similar result was reported by Williams (1980) in a study on older
poor readers (grades 2 and up). She made two attempts to design and
implement a remedial-reading program with a phonological-awareness
component. In the first study, a good deal of time was devoted to word
and syllable exercises, and the results were not encouraging. The program
was revised, the word and syllable work was dropped, the phoneme train-
ing was increased, and lessons in segmenting and blending CVC words
 164

were added. Now there was a strong impact on decoding words and non-
Chapter 6

words made up of the phonemes that were taught. Though standardized

tests were not used, this study does show, at least, that word and syllable
training is simply a waste of time.
One of the most famous studies in the literature, and frequently cited
as proof that the phonological-development theory is correct, was carried
out in the United Kingdom by Bradley and Bryant. A preliminary re-
port appeared in the journal Nature (1983) and the full report appeared as
a book (1985). The study was based on the premise that children are
more sensitive to rhyme (vowel plus final consonant(s)) and alliteration
(first sound) early in development, and this leads ultimately to phoneme
awareness and to reading skill.
The study was set up in two parts. The first part measured whether
early skill in rhyming and alliteration was correlated with subsequent
reading test scores. The second part was a training study. The ‘‘Sound
Categorization’’ task was designed to measure phonological awareness. In
this task, the child must identify the ‘‘odd one out’’ among a set of words
and point to one of three or four pictures. The ‘‘odd one’’ can be a pho-
neme in the initial position (first sound), as in the set bud, bun, rug, bus, or
in the middle (red, lid, fed, bed ), or in final position ( pin, win, fin, sit). The
middle and final sound matches are part of the rhyme (ed versus id and
in versus it). The word sets were heavily biased toward rhyme, and the
initial training phase strongly emphasized rhymes. Bradley and Bryant
found that children had less difficulty judging middle- and final-sound
contrasts (part of the rhyme) than initial sounds. This was interpreted as
rhymes being more ‘‘salient’’ or developmentally ‘‘early’’ than alliteration.
A simpler explanation is that it is easier to discriminate between vowel-
consonant pairs (VC) than between initial consonants (C).
There were two age groups in the study. One group had just turned 5,
and the other group was 5.5. The Sound Categorization task was sim-
plified to three words per set for the younger children. In tasks where
people choose among a small number of items, a certain proportion will
be correct by chance. To correct for guessing, a binomial test must be
carried out. This estimates how many answers need to be correct to score
better than chance at p ¼ :05. This estimate is based on the number of
choices and the number of items on the test. After computing this value, I
found that the younger children did not score above chance.
 165

However, scoring was well above chance for the 264 older children.

Phoneme-Awareness Training
With age, IQ, and memory controlled, rhyme scores (middle and final
sounds) did not predict performance on standardized reading and spelling
tests. Alliteration (first sound) was a significant predictor, accounting for 5
to 8 percent of the variance. But there was a problem. Alliteration pre-
dicted math test scores even better (10 percent of the variance)! Whatever
the alliteration test is measuring, it is not specific to reading. In short,
nothing was found in this study. The ability to detect a contrasting pho-
neme (the ‘‘odd one’’) in initial, middle, and final position in three-sound
words does not predict reading and spelling test scores.
Problems with the Sound Categorization test were reported by
Schatschneider et al. (1999), who were strongly critical of tests where
guessing plays a major role. In their normative study on 945 children in
kindergarten through second grade, they investigated the overlap (redun-
dancy) of seven phoneme-awareness tasks, one of which was the Sound
Categorization task. Six phoneme tasks were highly correlated with each
other. The Sound Categorization was an outlier in this analysis. Nor did it
‘‘load’’ on a general factor of phoneme awareness at high levels like the
other tasks. It was also extremely unreliable, producing a wide range of
scores (high standard deviations). This led the authors to conclude that
‘‘an inspection of the difficulty parameters for the Sound Categorization
subtest revealed that item difficult is highly dependent on where the target
word is placed in the string of words. This dependence, coupled with the
low discrimination parameters associated with these items, indicates that
this subtest is a relatively poor indicator of phonological awareness’’
(Schatschneider et al. 1999, 448). Despite Bradley and Bryant’s disap-
pointing results, they did not abandon their hypothesis. Instead, the data
were interpreted to fit the hypothesis, both in the conclusions to this study
and in the introduction to their training study, which opens with this line:
‘‘Children who are backward in reading are strikingly insensitive to rhyme
and alliteration’’ (Bradley and Bryant 1983, 419).
The training study is presented in the next section, because it primarily
demonstrates the benefit of using letters during phonological training.
Overall, these studies show that teaching larger phonological units
has little impact on reading and spelling skill. There is no evidence,
either from correlational studies or from training studies, that children
need to be eased into phonemes from larger units of sound. Large-scale
 166

correlational studies like those of Lundberg, Olofsson, and Wall (1980)

Chapter 6

and others (Share et al. 1984; Yopp 1988; McGuinness, McGuinness, and
Donohue 1995) show that phoneme-analysis skills are strongly correlated
with reading and spelling, while syllable segmenting and rhyming are not.
Williams found that children in the control group could segment and
manipulate syllables just as easily as the children who had spent weeks
practicing this.

Phoneme Analysis with and without Letters

One of the most consistent findings in the literature, and evident in the
NRP’s meta-analysis, is that when phoneme-awareness training is meshed
with teaching letter-sound correspondences, this has a much stronger im-
pact on reading and spelling than training in the auditory mode alone.
Overall, ES values for reading were .67 with letters versus .38 without, and
for spelling, .61 versus .34.
Studies on phoneme training with and without letters go back over 25
years. Haddock (1976) reported that blending training with letters was far
more effective than auditory training alone or training on letter-sound
correspondences alone. Bradley and Bryant’s training study is in this tra-
dition. Children in the age range 5.5 to 7.5 years were selected on the
basis of their poor performance on the Sound Categorization task, and
divided into four groups. Each child received 40 individual lessons over a
period of 2 years. There was no information on how long the lessons
lasted or how often they occurred. Bradley reported later at a conference
that the total time per child was less than 10 hours. This would be about
30 minutes per month in a school year.
Bradley and Bryant’s use of the terms categorizing, rhyme, and allitera-
tion is misleading in the context of their training study, and obscure what
was actually taught. Instead, I will classify the tasks in keeping with the
glossary above. Groups I and II got phonological-awareness training using
picture sets like those in the Sound Categorization test. The word sets
were unsystematic, using 15 consonants and the five simple vowels, plus a
sprinkling of vowel digraphs. Children named the pictures and had to
discover and say the common sound: ‘‘What is the common sound in mix,
dip, swim, sick?’’ This is a phoneme-identification task. Next, they learned
how to find the odd one out among items in a set (phoneme discrimi-
nation and identification). In the final exercises they worked with sets of
 167

rhyming words. There were no phoneme-sequencing activities (no blend-

Phoneme-Awareness Training
ing or segmenting).
Group I worked exclusively in the auditory mode. Group II did the
auditory tasks and worked with plastic letters, as described below:

In the second half of the training sessions Group II was taught with the help
of plastic alphabet letters as well. Whenever a new sound category was intro-
duced, it was demonstrated first with the help of the picture cards in the nor-
mal way. But the child then made each word in the set with plastic letters. (All
of the children had by this time attended school for at least two years, and
they were quite familiar with the alphabet.) (Bradley and Bryant 1985, 88)

Group III (control) had an equal number of lessons with the same picture
cards, but sorted them into semantic categories, like animals, furniture,
and so forth. A final control group (Group IV ) had no special training.
If these English children had been in school for ‘‘at least two years’’
and ‘‘were quite familiar with the alphabet,’’ they would be reading at
some level. The groups should have been matched for reading skill, and
they were not.
The training was not how it was portrayed. Children did not ‘‘cate-
gorize’’ sounds, they did phoneme-identity tasks. Nor were children
trained in phonological awareness prior to learning to read. We have no
idea what the children were taught in the classroom. This study cannot
test the authors’ central hypothesis that training in phonological aware-
ness ‘‘causes’’ reading, because reading was well underway, and children’s
reading skill was not controlled.
Furthermore, plastic letters were not simply added to the mix as sup-
port for phoneme identification. They were used primarily for spelling
dictation. After the children had identified each phoneme in a word set,
they were asked to spell the words with plastic letters, which they selected
from a box. The authors commented that many children learned to ‘‘with-
hold certain letters’’ (not put them back in the box) when the word sets
shared phonemes in common: sand, band, land.
After 2 years of training (and 4 years of classroom reading instruc-
tion), the children were tested on standardized reading and spelling tests.
Group II (phoneme identification plus spelling with plastic letters) was the
only group to score in the normal age range (8 years). They were significantly
 168

ahead of the two control groups on reading (a 1-year advantage), ahead

Chapter 6

of all groups in spelling, and had a 2-year advantage over the untreated
control group. Group I (phoneme awareness only) was superior to the
untreated control group (6 months ahead in reading, and 1 year ahead in
spelling) but not to the semantic-categorizing group. The two control groups
did not differ from each other.
The NRP used a special formula to compute effect sizes for this study
because standard deviations were not provided. These effect sizes were
unusually high, especially for Group II (phoneme þ letters): ES ¼ 1.17
for reading compared to the semantic-categorizing group, and ES ¼ 1.53
compared to the no-treatment controls. Effect sizes for spelling were even
higher: ES ¼ 1.59 and 2.18. It strains credulity that 10 hours of lessons
spread over 2 years would generate such large effects.
A number of factors challenge the validity of these results and/or the
NRPs calculations. First, there was the failure to control for initial reading
skill. Second, there was the small sample size (13 children per group).
Third, the effect sizes computed by the NRP were far too large. Even in
the comparison between Group I (no letters) and the no-treatment con-
trol group, ES values were .86 and 1.0. Recall that Group I was trained
with pictures, saw no printed words or letters, identified phonemes in the
auditory mode only, did no phoneme sequencing of any type, and never
read or spelled anything. In the NRPs’ meta-analysis (see the previous
chapter), the ES values for this type of training were .38 for reading and
.34 for spelling compared to no-treatment control groups. Thus the ma-
jority of studies are distinctly at odds with this result.
Fourth, interpretation of the results is problematic, even if they were
accurate. The main effect appears to be a spelling effect, not a phoneme-
awareness-plus-letters effect. If phoneme-identification training was really
effective, then the phoneme-only group would have been superior to the
semantic-categorizing group, but it was not. Nowhere in the description
of these exercises did children ever read any words. Even for Group II, the
only reading was indirect, a consequence of spelling with plastic letters. If
these data are valid, this study has demonstrated the impact of spelling
practice on reading skill, supporting findings by Ehri and Wilce (1987)
and Uhry and Shepherd (1993).
Unless this study is replicated and these issues addressed, Bradley and
Bryant’s results cannot be relied on. So far, this has not occurred.
 169

The NRP report provides convincing evidence that learning pho-

Phoneme-Awareness Training
nemes in conjunction with letters is better than learning with none. But
there are two ways to train phoneme awareness without letters. One is to
do this in the auditory mode alone, and the other is to use unmarked
tokens or counters to represent each sound. The token method does seem
to have some success.
Cunningham (1990) divided kindergartners and first graders into
three groups. She provided a program similar to Williams’s program,
with a range of activities to teach phoneme analysis, segmenting, and
blending. Tokens, but no letters, were used. Children were taught in
groups for 10–15 minutes twice a week, for 10 weeks (5 hours). One group
was taught with a matter-of-fact ‘‘skill-and-drill’’ approach, and a second
group with a more cognitive ‘‘metalevel’’ approach,’’ in which explana-
tions and goals were provided. A third group got no special training.
The results were puzzling. In kindergarten, the ‘‘meta’’ and the
‘‘skills’’ groups scored much higher than the controls on three phoneme-
awareness tasks (ES values all above 1.0), and did better on the Metro-
politan Reading Test as well (ES ¼ .57 and .43). The ‘‘meta’’ and ‘‘skills’’
groups were very much alike. They did not differ in reading, or on the
Lindamood Auditory Conceptualization (LAC) test of phoneme discrimi-
nation/manipulation, or on the Sound Categorization test, although the
‘‘meta’’ group was superior on the difficult phoneme-deletion test
(ES ¼ .81).
The first-grade results were different. Although the ‘‘skills’’ group far
outshone the controls on all the phoneme-awareness tasks (ES range ¼
.76 to 1.46), this apparently had no impact on reading (ES ¼ .09). The
‘‘meta’’ group did even better compared to controls on the phoneme tasks
(ES range ¼ .83 to 2.08), and this did impact reading (ES ¼ .53).
Something external to the study may be responsible for the strange
result, in which a high degree of expertise in phoneme awareness confers
no benefit for reading in one group of children but does in another.
Perhaps reading test scores were nonnormally distributed for these two
groups, or something was going on in the classroom that affected these
results. Otherwise, one would have to argue, despite the vast amount of
evidence to the contrary, that phoneme awareness taught in first grade has
no effect on reading unless you receive a ‘‘metalevel’’ explanation for why
you are learning it! This seemed to be Cunningham’s argument. She
 170

stated that a metalevel explanation was ‘‘not an important factor in

Chapter 6

kindergarten-age children’s transfer of phonemic awareness to a measure

of reading achievement’’—whereas in first grade it was ‘‘a more effective
method of instruction than a skill and drill approach’’ (p. 441). It is hard
to understand why this should be the case.
Brady et al. (1994) also reported a beneficial effect of a phoneme
training program that used blank tokens. One group of kindergarten chil-
dren got weekly phonological-awareness activities for about 4.5 months
(20 hours), and a matched control group participated in the usual whole-
language instruction. The first 4 weeks were spent on rhyming activities
and syllable segmenting, the next 6 weeks on phoneme-isolation tasks, and
the final 7 weeks on segmenting and blending phonemes using tokens.
Elements of the Lindamood program were used, including exercises on
monitoring mouth postures for each phoneme. At the end of the program,
when the children were 6 years old, there was no effect on reading or
spelling measured by standardized tests due to floor effects. One year
later, they were able to locate 42 of the original 96 children, who were
tested on the Woodcock Word ID and Word Attack subtests. The trained
group were superior to the controls on Word ID (ES ¼ .68) and Word
Attack (ES ¼ .71). These children scored about 5 to 8 months above age
norms as well.
These results beg the question that if blank tokens are effective,
wouldn’t letters be even more effective?
In a direct test of the letters-versus-blank-tokens question, Hohn and
Ehri (1983) found that letters were more powerful aids to segmenting
than blank disks used in the same exercises. The segmenting-plus-letters
group was superior to the blank-disks group and to an untrained control
group on tests of phoneme segmentation, phoneme deletion, and on sim-
ple decoding tests. In comparison to the control group, ES values were .20
for blank disks versus .68 for letters. The fact that letters (symbols for
specific sounds) are more beneficial to phoneme segmenting suggests that
letters provide a memory aid for remembering which phonemes appear in
the sequence. Hohn and Ehri observed that they helped the learner dis-
tinguish the correct size of the sound to be segmented.
The same effect was demonstrated in another way by Ball and Blach-
man (1988, 1991), who looked at learning letter-sound correspondences
with and without phoneme-awareness training. The children were 5.5
 171

years old and nonreaders. They were divided into three groups and given

Phoneme-Awareness Training
20-minute lessons four times a week for 7 weeks (about 9 hours). The first
group did phonological exercises, learned letter names and sounds for nine
letters (a, m, t, i, s, r, f, u, and b), practiced segmenting and blending, and
did ‘‘word assembly’’ with tokens and letter tiles. Children in this group
never used more than two letter tiles (plus blank tiles) in any one session,
and so never saw the words spelled in full. The second group focused
on general language activities (vocabulary, categorizing), but also learned
the nine letter names and sounds. There was no phonological training
for this group. The third group participated in their ordinary classroom
lessons.
An in-house reading test was designed with words composed of the
letters that were taught. Recall that none of the groups had seen these
words, though group 1 had seen portions of them. Group 1 read 10.9
words correctly, group 2, 3.9 words, and the control group, 2.2 words.
The effect sizes comparing group 1 and the other groups were substantial
(ES ¼ .71 and .98). However, this did not transfer to a standardized
reading test. This was reported as a ‘‘significant difference,’’ yet there were
large floor effects on the reading test, which puts statistical analysis off
limits.
The definitive study on this issue was carried out in Germany by
Schneider and his colleagues (2000). It is notable for the extensive and
informative test batteries that were employed. These provide a clear pic-
ture of which phoneme skills are easy or hard to teach (a direct test of the
developmental theory). And, by adding an extra experimental group, they
were able to pin down the value of teaching phoneme awareness alone,
letter-sound knowledge alone, or a combination of the two. This was a
complex study, and before I get into the details, I need to alert the reader
to certain factors that influence how these findings can be interpreted.
First, in Germany, where kindergarten originated, it was intended to
be precisely that: a ‘‘children’s garden,’’ where children play and interact
socially. From a research standpoint, this is useful, because children in the
control group are really taught nothing about how to read, and parents are
discouraged from teaching reading at home. Second, parents and teachers
in Germany have strong feelings about the value of this practice. The
authors reported that many parents and teachers were opposed to intro-
ducing any type of training at the kindergarten level, and obtaining
 172

parents’ permission was difficult. Third, the German alphabet code is

Chapter 6

nearly transparent. If the German writing system is taught properly, chil-

dren learn to read and spell in the first year of school (Wimmer 1993). All
that remains is mastering a few quirks of the spelling code. There are no
quirks of the reading code (no code overlaps). If you know the code and
can segment and blend phonemes, there is no way to misread a word.
Fourth, because everyone reads accurately, standardized reading tests
measure speed and comprehension, not accuracy. All these factors play a
role in how the results are interpreted.
At the start of the study, over 700 kindergartners were tested on a
normed ‘‘reading-readiness’’ test, which measured phoneme awareness,
phonological memory, verbal-processing speed, and visual attention.
Parent permission was obtained for 138 of the kindergartners who had
low scores on the test. These at-risk children were divided into three
matched groups, with each group receiving different training. The control
group consisted of 115 children who scored normally on the test.
The goal of the training regime was to ensure that the at-risk children
caught up to the normal controls on reading and spelling skills measured
later. If phoneme training and/or learning letter-sound correspondences
was important, then effects sizes on reading and spelling tests should be
close to zero, with no difference between the at-risk children and the normal
control group.
The three types of training were phoneme awareness only (PA),
letter-sound training only (LS), or both (PAþLS). The hypothesis was
that the group that got both types of training would be most successful.
The lessons were taught to the whole class at the same time for 10–15
minutes each day. This continued for around 5 months, for a total of ap-
proximately 20 hours.
A phonological training regime was based on Lundberg’s program
but dramatically speeded up. For the phonological-awareness group,
large phonological units were taught in the first month (words, syllables,
rhymes), and phoneme-analysis tasks were introduced in the second
month. This group had 20 weeks of training. The letter-sound group
learned 12 sounds and 12 letters, which took 10 weeks. A phoneme was
introduced in a story and in various games, then children were shown the
letter for that sound. Games and activities included alphabet cards, iden-
 173

tifying initial phonemes in matching sets of pictures, and so forth. There

Phoneme-Awareness Training
was no writing in this component. The combination group that had both
types of training learned fewer letters and spent less time on the phono-
logical tasks. This group had 20 weeks of training.
Special tests were designed to measure phonological-awareness skills,
along with verbal short-term memory, naming speed for colors and pic-
tures of objects, and measures of early literacy such as letter knowledge
and word recognition. These tests were given prior to the start of training
and after training was completed. Pretraining scores on the phonological
tasks were poor for everyone, and when children had to choose among
alternatives, performance was no better than chance. Test scores were
reliable after training was completed in July.
It is informative to look at the order of difficulty of these tasks. This
order was nearly identical for all groups regardless of the type of training.
From easy to hard, the order was: (1) Identify (say) an initial phoneme in a
word. (2) Blend isolated phonemes into a word. (3) Segment phonemes in
a word. (4) Identify the odd one out in a set of rhyming words that varied
in the final phoneme (part of the rhyme). (5) Delete phonemes. Say the
word that remains after the initial phoneme is deleted. (6) Carry out the
‘‘alliteration’’ portion of the odd-one-out task. Children did not score
above chance on this test.
These results support Chaney’s findings reported earlier (Chaney
1992). In her study, blending phonemes was the second-easiest phono-
logical task for 3-year-olds, and the alliteration tasks were the most diffi-
cult out of 22 tasks. The results also support Schatschneider et al.’s (1999)
finding that the Sound Categorization task is one of the most difficult
among a variety of phoneme-awareness tests. As seen above, even the
phoneme-deletion test, the most difficult of the phoneme tests, was easier
than the alliteration task.
Schneider et al. found a large impact of training on phoneme-
awareness skill for both the PA group and the PAþLS group. Both groups
did significantly better than the normal control group, with the PA group
having a decided advantage. However, none of the groups had much suc-
cess reading simple words.
In the fall of the same year, children entered first grade and were
taught to read in the usual way. The authors do not describe the reading
 174

Table 6.2
Chapter 6

Effect sizes from Schneider, Roth, and Ennemoser 2000: kindergartners at risk vs.
normal controls, three types of training
Phoneme Phoneme Initial
Kindergarten synthesis analysis phoneme
Pretest
PAþLS vs. controls .45 1.0 1.12

Posttest
PAþLS vs. controls .34 .61 .90
PA only vs. controls .55 .96 1.18
Reading
speed Spelling Comprehension
End grade 1
PAþLS vs. PA only .17 .27
PAþLS vs. LS only .22 .49
PAþLS vs. controls .38 .17

End grade 2
PAþLS vs. PA only .31 .30 .19
PAþLS vs. LS only .48 .48 .20
PAþLS vs. controls .30 .32 .50
Note: Tests used: German standardized tests on reading fluency, spelling, and
comprehension. PA ¼ phoneme-awareness training. LS ¼ letter-sound training.

method, but the German writing system is typically taught with a

linguistic-phonics approach (Wimmer 1993). Reading and spelling tests
were administered at the end of grade 1 and again the following year.
Intelligence tests were also administered at the end of grade 1. Because the
control groups had significantly higher IQs, IQ was controlled (covaried)
in all statistical analysis. These results are described below. The effect sizes
(IQ not controlled) are shown in table 6.2.
With IQ controlled, the control group was superior to the PA and the
LS groups, but not to the PAþLS group at grade 1 and grade 2 on tests
of decoding speed and reading comprehension. Results were similar for
 175

spelling, except that the PAþLS group was also statistically superior to the

Phoneme-Awareness Training
LS group, especially at the second testing. In other words, the PAþLS
combined training had largely eliminated the difference between these at-risk
children and the normal children. Training in phoneme awareness alone, or
learning 12 letter sounds alone, did not erase this gap.
I calculated the ES values for the most successful group (PAþLS)
versus all other groups. We can see that this group is ahead of the PA
group, and especially the LS group, in reading, spelling, and comprehen-
sion. Although the ES values are consistently negative in the comparisons
to the normal control group, most values are not statistically significant. It
would be interesting to see what happens in the future. Differences in IQ
may come to matter more over time, especially because the training did
not eliminate the difference between the normal children and the at-risk
(low-IQ) children on the comprehension test.
This study showed that phoneme-awareness exercises alone, or
sound-symbol associations alone, have little impact on learning to read. At
least this is the case in countries with a transparent alphabet, where
teaching is appropriate. But we are left wondering what would have hap-
pened if all the phonemes had been taught (and their letters) and not just
12 of them, and if the letter-sound group had had the same 20 weeks
training that the other two groups received.
These at-risk children, apart from having lower IQs, started off at
a distinct disadvantage compared to the normal children on several of
the readiness measures. They knew half as many letter-sound corre-
spondences as the normal children, and had lower phoneme-identity and
phoneme-segmenting scores. These things have to be taught. It may be
the case that the ‘‘normal’’ children entering kindergarten had more of a
boost at home than their parents let on. (Wimmer reported that, in Aus-
tria, some first graders know all the letter-sound correspondences and
some know none, even though parents are told to teach nothing.)
Although programs combining phoneme and letter training always
produce a greater impact on reading test scores than either alone, if we ask
whether any of these combination programs does a better job than a good
linguistic-phonics program, the answer is still no. The only exception here
is the study above, which measured fluency and comprehension for chil-
dren learning a transparent alphabet.
 176

Contrasting Types of Phoneme-Awareness Training

Chapter 6

Earlier I provided a glossary of the different types or measures of pho-

neme awareness. Two training studies focused specifically on types of
phoneme training.

Phoneme Identity and Letter-Sound Training versus Normal Classroom In-

struction Byrne and Fielding-Barnsley (1989, 1990) began a quest to
determine what preschool children, ages 4 and 5 years, found easy or
difficult to learn. They investigated the children’s ability to make sound-
symbol associations. The training focused on phoneme identity for pho-
nemes in initial and final position, and for segmenting initial phonemes.
They discovered that identity training generalized to words made up of
the same phonemes in new combinations and concluded as follows:

Preschool children can be taught to recognize the identity of phonemic seg-

ments, both consonants and vowels, across words, whether the phonemes are
singletons or members of a consonant cluster, and the position of the target
phonemes in the word appears to make no difference. . . . Successful instruc-
tion in phoneme identity can, when combined with relevant letter-sound
knowledge, promote acquisition of the alphabetic principle. (Byrne and
Fielding-Barnsley 1990, 810)

In 1991, they reported a study on a preschool program based on this ini-

tial research called Sound Foundations. Two groups of 4.5-year-olds were
matched on vocabulary scores and received either the training program or
regular kindergarten lessons. The training program focused on learning
nine consonant phonemes and one vowel. Training took 30 minutes once
a week for 12 weeks (6 hours). The trained children had larger gains on a
phoneme-identity test for sounds in both initial and final position. They
also could generalize this knowledge to phoneme sequences in spoken
words that had not been directly taught.
The children were followed up at the end of kindergarten when they
were 6 years old (Byrne and Fielding-Barnsley 1993). They were given
phoneme-identity and phoneme-blending tasks, a test of letter-sound
knowledge, and reading and spelling tests. The children who had the train-
ing were superior on the phoneme-identity task for the final sound only.
No significant differences were found for blending or letter-sound knowl-
 177

edge, both groups doing well (24 correct). Likewise, both groups did well

Phoneme-Awareness Training
on the Woodcock Word ID subtest, and phoneme-identity training played
no role (ES ¼ .21), nor did the training affect an in-house spelling test
(ES ¼ .09). However, the trained group did outshine the controls in
decoding on an in-house nonword test (ES ¼ 1.2). This result suggests
that the training had oriented the children toward the alphabet principle.
At this point in the data analysis, the authors chose to re-sort the
children into two new groups (regardless of the prior group assignment)
based on whether they passed or failed the phoneme-identity test at the
end of kindergarten. (The pass/fail cutoff was 66 percent correct.) The
new groups were compared statistically on various reading and spelling
tests. When groups no longer reflect who did or did not get trained, there
is no way to measure the outcome of training! Furthermore, this brings
uncontrolled factors into play, raising the following question: Why would
some children trained on ‘‘phoneme identity’’ fail a phoneme-identity test,
and some of the untrained children pass? IQ and home environment be-
come important, and neither variable was controlled. The new results
were highly significant, which was not the case previously.
The reassignment of children to groups based on their ability explains
the extremely high effect sizes reported by the NRP for this group of
studies. The NRP averaged the results together with those of the original
groups, enormously inflating effect sizes for this study as a whole. (ES ¼
1.61 for reading and 3.14 for phoneme awareness, by far the largest values
we have seen.) This grossly exaggerates the impact of this training pro-
gram, as well as the meta-analysis values as a whole. We know these values
are false by a table of first-order correlations from the 1993 publication,
which showed that the strongest contributors to reading and spelling test
scores were blending skills and letter-sound knowledge, not phoneme-
identity skills.
In subsequent follow-on reports on these same children, similar re-
sults were found at the end of grades 1 and 2, and at grade 5 (Byrne and
Fielding-Barnsley 1995, 2000). In-house tests were used to measure read-
ing and spelling in the early grades. The two groups did not differ on
these tests, with the exception of a small but significant effect for nonword
decoding. At the fifth-grade level, most tests were standardized tests. The
trained group, now 11 years old, outscored the control group on the
Woodcock Word Attack test ( p < :04), though the effect size was small
 178

(ES ¼ .34). No other comparisons using normed, standardized tests were

Chapter 6

significant. (It should be noted that pass-fail groups were reconstituted in

this report also, with the same exaggerated impact on effect sizes in the
NRP report.)
It appears that a short (6-hour) training program on phoneme iden-
tity, using nine consonants and one vowel, had a small but lasting effect on
nonword decoding. Why this is so is not clear, because the same training
had no impact on other reading and spelling skills at any time. Finally, it
should be noted that the NRP effect sizes for this series of studies are far
too large and need to be revised.

Phoneme Identity versus Phoneme Sequencing A study in Norway (Lie

1991) provided 208 first graders (age 7) with one of two phoneme-
awareness training programs. Children learn to read at age 7 in Norway,
and Norwegian is written in a transparent alphabet. The training took 15
minutes per day for 9 weeks (11 hours) and was delivered by the classroom
teacher as part of normal reading instruction. The two programs had an
emphasis on articulation, listening for a phoneme in sound-targeting
stories, and learning to recognize individual phonemes in words. Children
learned 27 phonemes at a rate of 3 per week. Overall training was the
same for both groups with the exception that the phoneme-identity group
learned phonemes in isolated positions in the word, and the phoneme-
sequencing group learned phonemes in the correct sequence. A third
group had lessons in concept categorization. No letters were used in this
training, but children were learning letter-sound correspondences in their
regular lessons.
At the end of the school year, the sequencing group was consistently
superior to the phoneme-identity group and to the controls in reading
(ES ¼ .42 and .62) and in spelling (ES ¼ .42 and .68). However, this effect
did not persist. At second grade there was no difference between the
identity and sequencing groups, though both groups were marginally
ahead of controls (ES values ranging from .33 to .53). This shows that
phoneme-awareness training can contribute to reading and spelling skills
above and beyond a linguistic-phonics program. The sequencing approach
is more effective, at least in the short term. It is interesting that this pro-
gram made a difference even in a country with a transparent alphabet
where reading is properly taught.
 179

On the whole, the evidence is not compelling that an independent

Phoneme-Awareness Training
phoneme-awareness training regime has a special benefit over and above a
good linguistic-phonics program. The really effective reading programs
reviewed in the previous chapter have far more impact than anything we
have seen here. Surprisingly, the strongest support for a positive impact of
special phoneme-awareness training comes from the European studies
(Schneider and colleagues; Lie) in countries with transparent alphabets,
and where traditional classroom instruction is likely to be similar to lin-
guistic phonics. However, neither Schneider and associates nor Lie reveal
what specifically was being taught in the classroom.

Studies Comparing Phoneme Training with Classroom Programs

Several studies have combined or compared phoneme-awareness train-
ing with training in specific classroom reading programs. For various
methodological reasons, none provide a direct test of how or whether a
phoneme-awareness program is necessary.
Blachman (Blachman et al. 1999) used the findings from her earlier
research (Ball and Blachman 1991) to develop a kindergarten phoneme
training program and mesh this with a new first-grade phonics program.
There were 159 at-risk (inner-city) children in the study. The hypothesis
was that early intervention in kindergarten can help at-risk children, es-
pecially when followed up by a good phonics program that links these
phoneme-analysis skills to reading instruction.
Half the at-risk children had phoneme-awareness training in kinder-
garten for 11 weeks. This consisted of 15–20 minutes of group lessons 4
days a week, for a total of about 13 hours. Training was limited to 8 letters
(a, m, t, i, s, r, f , and b). The children were taught to segment words orally
and move blank disks to represent each phoneme while saying the sound
(‘‘say it and move it’’). Later, letters were pasted on the disks. Children
were taught the letter names and sounds for the 8 letters. There was
phoneme-identity training: matching pictures with first sounds in com-
mon. The remaining at-risk children (controls) participated in their usual
kindergarten program (unspecified).
At the end of kindergarten, the children were tested on phoneme
segmenting, letter-sound knowledge, and two reading tests. One was an
in-house reading test consisting of words made up of the 8 sounds/letters
that had been taught. This, not surprisingly, showed a strong advantage
 180

for the special kindergarten program (ES ¼ 1.08). However, this knowl-
Chapter 6

edge did not transfer to a standardized reading measure. (The ES value

for the Woodcock Word ID test was .17.) It goes without saying that
scores on phoneme-segmenting and letter-sound knowledge were higher
for the children who had been taught these things than for children who
were not.
The experimental group was followed into first grade and given a
phonics program that completely replaced the usual program. The kin-
dergarten activities were continued and letter-sound knowledge was grad-
ually expanded to ‘‘all letter sounds.’’ Whether this means the ‘‘sounds’’ of
all the letters of the alphabet, or all the sounds in the English language,
is not clear. Children were taught some vowel digraphs, but how this
was done was not specified. Children did workbook exercises connecting
letters and sounds, learned some common sight words, and practiced
sequencing using the DISTAR phoneme-stretching technique (‘‘mmmm-
aaaaannnn’’), instead of segmenting. (Blachman subscribes to the view that
segmenting consonants in isolation from a vowel is difficult.) Slingerland’s
pocket-chart technique was used to sequence letter cards to make words.
Children practiced with flash cards to improve decoding speed. They
read simple phonics books with ‘‘regular’’ spellings, such as the Primary
Phonics series, and wrote words and sentences to dictation. This looks like
a sensible phonics program, but there was insufficient detail to make a
valid judgment.
Toward the end of training, children were introduced to six sylla-
ble types: ‘‘closed,’’ ‘‘open,’’ ‘‘final e,’’ ‘‘vowel team’’ (vowel digraphs),
‘‘vowel þ r,’’ and ‘‘consonant þ le.’’ This classification scheme is a feature
of Orton-Gillingham remedial programs. English spelling patterns are not
specific to syllable types. Furthermore, these types are not mutually ex-
clusive, and therefore do not constitute independent categories.1 Teaching
this only adds unnecessary complexity and will create confusion.

1. The same spelling for the same sound can occur in more than one of these
syllable types. The spelling o for the sound /oe/ occurs in both ‘‘open’’ and
‘‘closed’’ syllables: go, most. Ditto ee as in fee and feed; ie as in die and died; ue as
in cue and fuel; oo as in too and food. The same word fits more than one syllable
type: goat (‘‘vowel team’’ and ‘‘closed’’), care (‘‘final e,’’ ‘‘vowel þ r’’), soar
 181

The control group followed the Scott-Foresman basal series. This

Phoneme-Awareness Training
is a sight-word, plus basal-reader, plus disconnected-phonics curriculum.
Teachers also used parts of the DISTAR program as well. Children read
trade books from classroom libraries. This is a typical eclectic mix.
Both groups of children were taught in a whole-class format. And
both groups received whole-class spelling instruction using the Scott-
Foresman spelling program, which is vaguely phonics oriented (‘‘regular’’
CVC words first, followed by the ‘‘long-vowel’’ words, blends, and
digraphs, gradually increasing in length and complexity).
The teachers in the experimental program received 13 hours of train-
ing. No special training was given to the teachers of the control group.
The comparison between the two programs showed that the experi-
mental group had much higher phoneme-segmenting and letter-sound
knowledge scores (ES values ¼ 1.0), as would be expected. However, they
were only marginally better on standardized reading and spelling tests
(ES ¼ .35 for word recognition and .38 for spelling). The children were
followed up at the end of second grade, and standardized reading and
spelling tests were given again. Effect sizes for reading improved slightly, in
favor of the experimental group: ES ¼ .44 for word-recognition tests and
.46 for word attack. However, the ES value for spelling was zero. These
values are not particularly remarkable for 2 years of special teaching.
This study had a major design flaw if one of the goals was to evaluate
the impact of the kindergarten phoneme-awareness training. It is missing
a critical control group, and possibly two. This would be a group of chil-
dren who did not receive the kindergarten phoneme-awareness program
but did receive the first-grade phonics program. To be absolutely certain
that the kindergarten program was relevant, a fourth control group is also
needed. This group would receive the kindergarten program but be
switched to Scott-Foresman at first grade. Instead, the control group got
neither the kindergarten program nor the new reading program. In view
of this, one would expect to find a much more substantial advantage for
the experimental group, especially if both the kindergarten program and

(‘‘vowel team,’’ ‘‘vowel þ r’’). Furthermore, this classification scheme does not
fit multisyllable words at all, which make up about 80 percent of the words in
English.
 182

the first-grade program were effective. Instead, effects sizes were modest,
Chapter 6

suggesting that neither program is as effective as it could be, and/or the

teachers were not committed to them.
Hatcher, Hulme, and Ellis (1994), in England, compared a phoneme-
awareness program to a remedial-reading program known as Reading
Recovery, based on work by Marie Clay (1985). Hatcher and colleagues
trained 125 second-grade children (7.5 years) who had been identified
as poor readers. The training consisted of individual tutoring taught by
classroom teachers over a period of 25 weeks (two half-hour sessions per
week) for a total of 20 hours. After this, the children were retested, and
they were followed up 1 year later and tested again. The groups were
matched for age, IQ, and initial reading test scores.
The children were divided into three experimental groups and a con-
trol group. A phonological training group received a program similar to
the one designed by Lundberg and colleagues. It involved identifying and
producing words, syllables, rhymes, and a range of phoneme-analysis
tasks. These included lessons on segmenting, blending, omission, substi-
tution, and transposition of phonemes in words.
A reading training group received lessons based on Reading Recovery
(see Clay 1985). Teaching was visually oriented (reading and rereading the
same books), included some strategies for decoding (unspecified), frequent
diagnostic tests, plus writing activities (invented spelling) involving letters,
words, and stories. Children learned letter names but no letter-sound
correspondences.
The phonological-plus-reading group received both types of instruc-
tion. There was a trade-off due to time constraints, and only half the
phonological tasks were completed. This group also got new activities
that linked reading and phonological awareness. These included learn-
ing letter-sound correspondences, using plastic letters to spell words, and
writing words while attending to sound-to-letter correspondences. These
‘‘linking’’ activities took up 10 minutes of every session.
The ordinary classroom instruction (received by everyone) was de-
scribed as ‘‘phonics’’ minus phonological-awareness training. There was
no description of what ‘‘phonics’’ meant or which program was used. The
control group received only this ‘‘phonics’’ instruction, and some children
from this group had outside remedial help as well (program and time
unspecified).
 183

From this description, it appears that the children in the combined

Phoneme-Awareness Training
group (phonological awareness þ reading þ linking activities) got a Mul-
ligan stew of reading methods: a variant of whole language in Reading
Recovery lessons, some type of phonics in the classroom, multisound-unit
phonological tasks, a variety of phoneme-analysis tasks, plus exercises to
‘‘link’’ phonological knowledge and the alphabet code.
Perhaps this mix was too confusing, because neither the combined
group (predicted to be superior to all groups) nor any of the other groups
did particularly well. No differences between the groups were found on an
in-house word-recognition test or on the British Ability Scales, a standard-
ized test of word recognition. There was a small but consistent advantage
for the combined PA þ read group on the remaining standardized tests, as
shown in table 6.3. The other groups did not differ from one another.

Table 6.3
Effect sizes: Data from Hatcher, Hulme, and Ellis 1994
Neale
Neale Compre- Schonell
Accuracy hension Spelling
Testing age 8:1
Read þ PA vs. read only .45 .41 .31
Read þ PA vs. phonemes only .40 .52 .14
Read þ PA vs. controls .52 .61 .33

Testing age 8:10

Read þ PA vs. read only .46 .47 .35
Read þ PA vs. phoneme only .35 .44 .22
Read þ PA vs. controls .38 .57 .30
Read þ phonological Testing: age 8:1 Testing: age 8:10
group only
Age- Years Age- Years
equivalent below equivalent below
score age score age
Neale Accuracy 6:1 2:0 6:8 2:2
Neale Comprehension 6:4 1:9 7:0 1:10
Schonell Spelling 6:8 1:5 7:2 1:8
 184

The comparisons between the combined group and the other three
Chapter 6

groups produced similar effect sizes (see table 6.3), with ES values in a
modest .30–.45 range. This is not much to show for 20 hours of one-on-one
help. In other words, neither Reading Recovery nor phonological-aware-
ness training on its own provided any greater benefit to these poor readers
than what was taught in the phonics classroom, which was not much.
More revealing were the combined PA þ read group’s age-equivalent
scores on the standardized tests at beginning and end of training, and at
follow-up 9 months later. These scores are shown at the bottom of table
6.3. The children began the study scoring 1.5 years below age norms on
all tests. The table lists the years:months these children lag behind age
norms for reading and spelling at two testing times (ages 8:1 and 8:8
years). The discrepancies between these scores and national age norms
actually increased over time.
For reasons unknown—the mix of methods, the training programs
themselves, poor teacher training or monitoring—these programs were
not effective in getting the children caught up. Other intervention pro-
grams work much better, and in a much shorter space of time (see D.
McGuinness 1997c, 1998b; C. McGuinness, D. McGuinness, and G.
McGuinness 1996). The rule of thumb for successful remediation is to
remain faithful to the prototype, and to avoid teaching skills that have
nothing to do with an alphabetic writing system.
Finally, Brennan and Ireson (1997) carried out a study that, despite
methodological problems, is the most direct test of whether phonological-
awareness training provides any particular advantage over a good phonics
program. They adapted a Lundberg-type program for kindergartners (age
range 4:10 to 6:1 years) attending an American school in England. The
children spent 3 months on phonological units above the level of the
phoneme. These included listening to nonverbal sounds, playing rhyming
games, clapping out syllable beats, and using markers to represent syllables
in multisyllable words. In the middle of the third month, children began
learning phonemes in initial position, and subsequently in all positions
over the course of the school year. Lessons were mainly in the auditory
mode only (no letters). However, there was a major confound in this
study. The phonological program took up 15–20 minutes of a 2-hour
language-arts period, during which children learned letter names and
sounds, copied letters, and wrote ‘‘stories’’ using invented spelling.
 185

The phonological training group was contrasted with two other

Phoneme-Awareness Training
groups. One group was using a published phonics program (Success in
Kindergarten) designed to integrate learning sounds and letters. Copying
and writing words was a particular emphasis. The third group was taught
‘‘Letterland’’ characters (letters drawn to look like animals) to teach letter
sounds. This is a visually driven approach that focuses on letter shapes and
‘‘the sounds the letters make.’’ Lessons include hands-on activities to do
with memorizing letter forms, such as tracing letters in the sand tray and
making letters out of play dough. Children recite rhymes related to the
target letter.
Both the phonological training group and the Success in Kinder-
garten group were considerably advanced on standardized reading and
spelling tests compared to the Letterland group at the end of the year.
However, the phonological-awareness group was not superior to the Suc-
cess in Kindergarten group. When I compared the two programs, effect
sizes were in favor of Success in Kindergarten: ES ¼ .38 for Schonell
reading, .56 for ‘‘high-frequency words,’’ and .23 for Schonell spelling.
(These values are different from the NRP’s effect sizes.)
Brennan and Ireson do not report which phonemes were taught for
any group, or how they were taught, and there is the fact that about 90
minutes of other reading activities were ongoing in the classrooms. In
contrast to the situation in most other European countries, children in the
United Kingdom are taught to read in kindergarten. The only clear result
was the extremely poor showing of the Letterland program. There is no
way to tell whether the differences between the other two programs had
anything to do with specific characteristics of these programs.

Conclusions
As a general observation, one of the most consistent findings to
emerge from these studies is that phoneme-identification and phoneme-
sequencing (segmenting/blending) training are the only phoneme-analysis
skills that consistently impact reading test scores. This confirms the evi-
dence from the correlational research. Helfgott (1976) was among the first
to discover that segmenting skill for CVC words was the highest correlate to
reading 1 year later (r ¼ :72) among a variety of phoneme-awareness skills.
The definitive study was that of Yopp (1988), who measured
kindergartners’ performance on 11 phonological-awareness tests. She
 186

investigated the statistical overlap between these tests and correlated each
Chapter 6

test score with the time to learn to read novel words. Yopp’s findings
mirror the training studies. The highest correlates of learning rate were
‘‘sound isolation’’ (phoneme identification) (r ¼ :72), and phoneme-
sequencing tasks (blending and segmenting) (average r ¼ :67). Auditory
discrimination (r ¼ :27) and rhyming skills (r ¼ :47) did not predict
learning rate. Phoneme-deletion tests were too difficult for this age group.
For those knowledgeable about factor analysis, phoneme-
identification and phoneme-sequencing tasks loaded on the same factor
(factor I) at values ranging from .76 to .89. Auditory-discrimination and
rhyming skill loaded on none, and phoneme-deletion tests loaded on a
separate factor (the Rosner test loaded on factor II at .94).
It should be noted that phoneme identification and sequencing are
precisely the skills that are trained in a linguistic-phonics program as a
matter of course.
Despite the overall agreement between these studies, most studies
reviewed in this chapter have a number of design flaws that make it diffi-
cult to know exactly how or whether a phonological training program
impacts reading and spelling. There is little attempt to discern which of
the many phonological tasks are truly necessary. Often there is no de-
scription of which phonemes and letters (if used) are taught. And when
this information is provided, too few sound-symbol correspondences are
included in the program, and far too much time spent teaching them. On
the whole, the majority of programs grossly underestimate what 5-year-
old children (or even younger children) can learn. One is struck by the
fact that the balance is exactly backward between the number of phonemes
taught (and their spellings), and the time spent on larger phonological
units. The unwritten assumption seems to be that teaching phonemes and
letters is hard for young children, but teaching lots of ‘‘unnatural’’ and
conflicting phonological tasks is not. This appears to be one of the many
legacies of the phonological-development myth.
The evidence is not convincing that special programs for teaching
phoneme awareness and letter-sound correspondences instead of using a
good linguistic-phonics program at the outset provides any additional
benefit.
This was not the conclusion of the NRP, which seemed to take it for
granted that phoneme-awareness skills are so hard to learn that separate
 187

phoneme-awareness training is a necessity. The panel commented in the

Phoneme-Awareness Training
introduction to the report that Texas and California have prescribed the
inclusion of phoneme-awareness training as part of early reading instruc-
tion in the context of praising these two states for their forward-looking
policies. (We have already seen what happened the last time California
mandated a curriculum.)
If special phoneme-awareness training is essential, precisely which
of the many training programs reviewed here should teachers adopt?
And what is the evidence that any of these programs confers a benefit
beyond what a good linguistic-phonics program confers? So far there is
none.
The summary to the NRP report on phoneme awareness (‘‘Implica-
tions for Reading Instruction’’) shows that the authors have not considered
the overlap between phoneme-awareness training and a linguistic-phonics
program. For example, in a question-and-answer section, the NRP ad-
dressed the question of whether phoneme-awareness training helps chil-
dren learn to read and spell. The panel stated that teaching children to
manipulate phonemes ‘‘transfers and helps them learn to read and spell.
PA training benefits not only word reading but also reading comprehen-
sion. PA training contributes to children’s ability to read and spell for
months, if not years, after the training has ended’’ (p. 2-40).
As to which phoneme-awareness method has the greatest impact on
learning to read, they wrote: ‘‘Teaching students to segment and blend
benefits reading more than a multiskilled approach. Teaching students to
manipulate phonemes with letters yields larger effects than teaching stu-
dents without letters. . . . Teaching children to blend the phonemes repre-
sented by letters is the equivalent of decoding instruction’’ (p. 2-41). No,
it is identical to decoding (reading) instruction.
As for spelling: ‘‘Teaching children to segment phonemes in words
and represent them with letters is the equivalent of invented spelling in-
struction’’ (p. 2-41). No, it is identical to proper spelling instruction.
Spelling should never be ‘‘invented.’’
The authors of the NRP report appeared to be satisfied with the
quality of the studies and the validity of the large effect sizes reported in
their analysis—effect sizes that, as we have seen, are grossly inflated due to
invalid in-house tests and other methodological anomalies. Nowhere is
this more evident than in a statement that misrepresents the fact that
 188

effect sizes based on standardized tests are marginal (i.e., ES ¼ .33 for
Chapter 6

reading, ES ¼ .41 for spelling).

According to the report, ‘‘The NRP analyses shows that the evidence
rests solidly on well-designed studies. Significant effect sizes were appar-
ent on standardized tests as well as experimenter designed tests’’ (p. 2-42).
As noted in the previous chapter, the effect sizes in the NRP report for
the individual studies themselves are not always accurate. The problem is
compounded when these inaccurate values are averaged together in a
meta-analysis. As we have seen from this review, many studies were not
‘‘well-designed,’’ and many test instruments and much data manipulation
were invalid. I cannot recall more than one study where reliability esti-
mates were computed for an in-house test.
Of course, it is difficult to measure the impact of a phoneme-
awareness training program on reading and spelling before children are
taught to read and spell, which is the case in most kindergarten studies.
Researchers are justified in trying to ascertain whether the training im-
pacted decoding and spelling without having to develop their own reading
and spelling tests from scratch, complete with norms and standard scores.
However, there are other criteria for a valid test. Test reliability is cru-
cial, as is item analysis. Items may be too difficult or too easy. Some items
may not measure the construct (construct validity). Schatschneider et al.
(1999), for example, discovered in their analysis of phoneme-awareness
tests that items in the same test often varied enormously in difficulty in an
idiosyncratic fashion.
As a final comment, let’s return to the third criterion used in the ini-
tial screening of these studies: ‘‘Test the hypothesis that phoneme aware-
ness training improves reading beyond some alternative form of training
or no training’’ (p. 2-15). By ‘‘alternative form of training,’’ the panel
meant something similar or parallel to phoneme-awareness training. But
what if the ‘‘alternative form of training’’ was a linguistic-phonics program
like Lippincott or Jolly Phonics? In only one case in this group of studies
was this hypothesis even marginally tested, and until this is done, the
conclusion must be that separate phoneme-awareness training programs
like those reported here do not come close to ‘‘improving reading’’ com-
pared to a good linguistic-phonics program. Phoneme awareness programs
do not even produce comparable results on tests of phoneme awareness.
 7
READING FLUENCY

The goal of reading instruction is to ensure that children learn to read

accurately and fluently. But while researchers in English-speaking coun-
tries carefully measure accuracy in all its forms, fluency is rarely measured.
In the eight slots for effect sizes on outcome measures in the NRP report
on reading instruction, there was no slot for reading fluency or speed.
Very few standardized test batteries include such a measure.
For the most part, researchers and educators believe that fluency is a
by-product of accurate decoding. It is certainly the case that inaccurate
decoders are halting, dysfluent readers. But the reverse is not true. A slow
reader is not necessarily inaccurate. The slow but accurate reader is a
common phenomenon in countries with a transparent writing system. As
we saw in the last chapter, reading tests in these countries measure reading
speed and comprehension, not accuracy. When Wimmer (1993) tested
Salzburg’s poorest readers, no one had decoding or spelling difficulties but
nearly everyone read very slowly.
This means there are two types of slow readers, children who read
slowly because they cannot decode (inaccurate), and children who read
slowly despite good decoding skills. The first type of reader needs most
help with decoding skills (perhaps followed by fluency training), and the
second type needs fluency training only. How fluency can best be trained
is the topic of this chapter.
Before we look at the training studies, we need to understand why a
reader would read accurately but slowly, and address the question of
whether it matters if a child is a slow but accurate reader. The standard
answer has been that a ponderous decoding speed prevents children from
comprehending what they read. However, the evidence to support this
theory is largely anecdotal. For the most part, the correlational studies
 190

have been carried out on English-speaking children who read slowly and
Chapter 7

inaccurately, so when reading speed and comprehension are correlated, we

cannot know why. And other factors like age, sex, and IQ have a major
impact on processing speed but are rarely controlled in the correlational
studies. Wimmer found that 80 to 85 percent of slow readers at first
and second grade were boys, and that verbal IQ scores for these slow
readers were significantly below those of normal readers by a wide margin
( p < :001).
The only way to answer the question raised above is to control both
fluency and comprehension at the same time. If slow reading has no effect
on comprehension, there is no reason to be concerned. We also need a
better conceptual grasp of potential causes. The children may be lagging
developmentally, or may not have been taught letters and sounds prior to
going to school. They may have been victims of poor instruction, causing
them to have an inefficient (slow) decoding rate.

Speaking Rate and Brain Speed

What is the optimum reading rate? One would imagine that reading speed
ought to be close to speaking rate. Speaking rate is extremely rapid due to
the fact that phonemes are coarticulated (physically overlap). Each pho-
neme in a word modifies the one preceding it. This is a kind of ‘‘backward
propagation’’ in which the last phoneme in a word controls the speech
patterns of the one before it, and so on, backward through the entire
word. When we begin speaking (forward in time), the brain has already
made this calculation before we utter the first phoneme. The /g/ in the
word dog modifies the vowel /o/, which in turn modifies how the /d/ is
physically produced. As a result, the /d/ in dog and the /d/ in dip have
different acoustic profiles, quite easy to distinguish on a paper printout
of spectral patterns, even with the naked eye. Coarticulation speeds up
speech enormously, making the transfer of information from one brain to
another more efficient.
Articulation speed (speaking rate) needs to be optimal for the human
brain to extract meaning from speech utterances. Speak too quickly, and it
sounds like gibberish; speak too slowly, and the sentence falls apart. Lis-
tening to slowed speech is such hard work for the listener that interest
wanes with every word. (The same applies to music played too quickly or
too slowly.) Speaking rate is surprisingly uniform across different lan-
 191

guages, considering the variations in their phonetic structure. This rate is

Reading Fluency
measured by the number of words uttered per unit of time in normal
conversations. The natural speaking rate for English is about 250 to 300
words per minute (wpm).
Because this rapid speaking rate makes conversation possible, one
would imagine that for readers to process meaning at the brain’s preferred
rate (to comprehend what they read), they should be able to decode at the
same rate as people speak. What might this optimal reading rate be? In his
review of eye-movement research, Rayner (1998), described a study with
this goal. College students were identified by their excellent performance
on a reading-comprehension test. They were asked to read passages at
their optimum reading rate. The fastest reader was clocked at 380 wpm
and the slowest at 230 wpm. The average for this group of expert readers
was 308 wpm. This is about the same as the optimal speaking rate, and
even this takes years to accomplish. The normal second-grade reader
reads about 90 wpm. This jumps to 150 wpm at fourth grade. By sixth
grade, the average reader is closing in on the rate of the slowest college
reader (200 wpm).
The variability across age, and the individual differences in reading
speed, mean that defining a ‘‘slow reader’’ is not going to be easy. The
problem is compounded when English children are compared to children
who learn a transparent alphabet. As noted earlier, Wimmer, in collabo-
ration with English colleagues (Wimmer and Goswami 1994; Landerl,
Wimmer, and Frith 1997), compared normal children from Salzburg and
London. The Austrian 7-year-olds with 1 year of school read as fast as
English 9-year-olds with 4 years of school, making half the number of
errors, an eightfold increase in efficiency. When the worst readers in
Salzburg (very slow) were pitted against the worst readers in London (very
inaccurate), the Salzburg children read the same material twice as fast while
misreading only 7 percent of the words. The English children not only
read more slowly, but misread 40 percent of the words. ‘‘Slowness,’’ it
seems, is a function of the writing system, not a property of the child.
If ‘‘slow reading’’ is relative, tied to a particular writing system and
method of instruction rather than to age or innate ability, there are no
guidelines for determining an optimum reading rate other than anecdotal
reports of the teachers or complaints of individual children. To complicate
matters, the content of the reading material also determines reading rate.
 192

Difficult material is read more slowly, with many more regressive fixa-
Chapter 7

tions, than simple material.

For the time being, I will adopt the average second-grade reading
speed of 90 wpm as a base rate, below which comprehension may become
difficult. This rate applies to text with simple vocabulary used to describe
content that is familiar or conventional. Anything less than 90 wpm will
not be adequate to process meaning.
For children whose reading speed falls below this value, what kind of
training might help? One clue comes from the eye-movement research.
Rayner (1986) found that adults and children utilize different amounts of
information in their peripheral visual field as they read. This is known as a
perceptual span. The span is larger on the right than the left due to the left-
right direction of our writing system. It takes about a year of reading
practice for this asymmetry to be established.
The average perceptual span for adults is around 14–15 characters to
the right of fixation, and adults use all this information. The span is similar
for children, but Rayner (1986) found that younger children’s functional
spans are smaller, about 11 characters or less. Children rely more on cen-
tral (foveal) vision, and do not take advantage of the information to the
right of fixation until they are around 12 years old. Perhaps slow readers
make less use of this information than normal readers. Their plodding,
word-by-word-by-word decoding style suggests they are focusing on one
word at a time.
Despite the fact that humans cannot see detail in their peripheral
visual field, they can see something. Everyone with normal vision, includ-
ing slow readers, can make out the first two letters of the following word
quite well. Beyond this, the remaining letters provide a blurry, global im-
pression of the word’s length and shape, its pattern of ascenders (b, d, f, h,
l, and t) and descenders (g, j, p, q, and y). For this information to be ac-
cessible it must be coordinated ‘‘online’’ (in real time) with meaning and
sentence structure. Readers have to be aware of syntax and anticipate what
part of speech is due next: ‘‘The little boy [verb goes here].’’
Readers also have to be aware of context—what is going on in the story
(the boy just tripped over a log).
These four clues (initial letters, word length and shape, syntax, con-
text) narrow the choice of words that go in the next slot in the sentence.
The brain, given half a chance, will automatically toss out a few suitable
 193

words that begin with those particular letters, have that shape, fit that slot,

Reading Fluency
and relate to the meaning of the story. Human brains are especially good
at associative pattern matching, and superb at anticipating meaning, a
phenomenon first reported by William James. Today, this is known as
analysis by synthesis or top-down processing. Listeners continually anticipate
words that are coming up next in a speaker’s utterance. This is the rea-
son that puns, other forms of word play, and sudden or odd shifts of
context are surprising and amusing, because certain words mismatch our
expectations.
Of course, people usually do not do this consciously. No one reading
this book is aware that they see print they are not looking at. Nor can we
monitor the fact that our brain is busily putting this ‘‘nonseeing’’ to good
use. How then can we get slow readers to make use of this peripheral
information to speed reading along? One way would be to train this fac-
ulty directly, by setting up a series of exercises for slow readers to practice
making peripheral glances or to fixate straight ahead while trying to iden-
tify blurred shapes on the right. However, it turns out that there is a much
simpler solution, one that produces the same results without the need to
make an unconscious process conscious.

The NRP’s Analysis of Fluency-Training Studies

In the introduction to the NRP’s analysis of this topic, the authors pointed
out that the old notion of ‘‘automaticity’’ as a natural outcome of accurate
decoding has to be modified. In particular, they discussed the importance
of syntax—‘‘the ability to group words into meaningful grammatical
units’’—plus comprehension. It is not merely that fluency frees up ‘‘cog-
nitive resources’’ for interpretation, but that fluency is implicated in the
process of comprehension because it allows for ‘‘preliminary interpretive
steps.’’ This anticipatory analysis is similar to the notion of top-down
processing discussed above. They link this to ‘‘parallel processing,’’ the
ability to perform multiple perceptual and cognitive tasks at the same
time.
They also bring up another important point, that automaticity is not
simply present or absent, but develops gradually with practice over a
period of time. Reading speed is on a continuum, and, as we have seen
from the developmental norms, there is no definitive measure of reading
fluency. The reader is referred to the panel’s excellent analysis of the
 194

problem of defining automaticity and fluency (National Reading Panel,

Chapter 7

2000, 3-5 to 3-11). The introduction to the section on fluency is one

of the highlights of the NRP report.
There are various approaches to training fluency. The NRP found
studies on only two of these approaches: the process of ‘‘encouraging
students to read more’’ and the technique of ‘‘rereading.’’ We can dis-
pense with the first at the outset after discussing some extraordinary facts.
There is, perhaps, no more entrenched belief than that ‘‘reading a lot’’
plays a causal role in turning children into accurate and fluent readers
who comprehend what they read. ‘‘Reading a lot’’ is supposed to increase
vocabulary as well. One would imagine that this long-standing belief has
been put to rigorous tests a number of times and been upheld, but this is
far from being the case.
Nothing in the field of reading has spawned a more bountiful crop of
articles and papers than the belief that ‘‘reading a lot’’ makes a difference
with respect to reading skill, attitudes toward reading, and a host of other
language and cognitive skills. The NRP search revealed 30 different ap-
proaches to how to increase children’s reading volume. Using these 30
terms as search terms in one database (PsychINFO), they unearthed
over 19,000 papers! Combining these terms with the primary areas of
interest—reading speed, reading comprehension, reading skills, and so
on—this number reduced to just over 1,000 (900 after foreign-language
articles were eliminated). This was repeated with the ERIC database and
overlapping articles eliminated.
Limiting the studies to those that appeared after 1991, there was a
final set of 603 articles. The NRP investigators looked for studies that met
the following criteria: research studies only, English K–12 reading edu-
cation, and publication in a refereed journal. This eliminated all but 92
studies, of which only 79 could be located. After a close reading of these
articles, only 9 could be retained, because most of this research was
correlational. They extended the search to the bibliographies of the 79
studies. This produced 10 more likely candidates, of which only half could
be retained. They now had 14 bona fide experimental investigations
involving the scientific study of the impact of ‘‘reading a lot.’’ Unfortu-
nately, most of these studies had such a weak research design or other
methodological problems that no meta-analysis could be applied to them.
This is an appalling state of affairs.
 195

The 14 studies were reviewed briefly. Most of these studies measured

Reading Fluency
the impact of ‘‘sustained silent reading’’ or a similar approach. There was
no evidence to support the idea that having children read for a fixed
period of time, inside or outside the classroom, made any difference to
vocabulary, reading comprehension, reading attitude, word recognition, or
performance on general achievement tests and standardized tests. The
NRP authors concluded as follows: ‘‘None of these studies attempted to
measure the effect of increased reading on fluency. Instead, most of these
studies considered the impact of encouraging more reading on overall
reading achievement as measured by standardized and informal tests. It
would be difficult to interpret this collection of studies as representing
clear evidence that encouraging students to read more actually improves
reading achievement’’ (p. 3-26).
They went on to stress that the poor quality of this research provides
no definitive proof, one way or the other, that a regime of scheduled or
controlled silent reading helps reading achievement. Given the fact that
the outcome measures included the whole spectrum of reading compe-
tence (except fluency), this is a scathing indictment. This pattern has been
observed in every topic area so far. It is clear that we urgently need a
proper database for scientific research on reading. ERIC has certainly
never fulfilled that function, and now, it appears, neither does Psych-
INFO. The NRP’s discovery that only three of the studies in their review
were methodologically sound is also an indictment of the research journals
themselves and their editorial boards. Teachers need our help. Teachers
should be able to find out how or whether time spent reading makes a
difference to reading skills.

Rereading
Teachers have known for a century or more that rereading text will in-
crease reading speed. Slow readers read faster after they have read the
same story or passage many times. This technique has such an ancient
history that E. B. Huey described it in his famous book on reading in
1908. However, the central problem with this technique is whether
improved speed on one story will transfer to another story. If fluency is
specific to only one passage or story, the rereading technique is worthless.
The NRP data search on rereading was much like the previous one,
covering a wide range of programs with fanciful names like ‘‘echo reading’’
 196

and ‘‘neurological impress.’’ The same tireless search was carried out, with
Chapter 7

the result that 98 articles met preliminary screening requirements and

were carefully coded. From this analysis, they found that there were a
variety of different experimental approaches, which they classified into
four categories. Only one category met the requirements for inclusion in
a meta-analysis. This was a group of 16 studies (1970–1996) that were
properly designed with pretest and posttest measures of reading, plus a
control group. The other categories not included were the following: 14
studies on the immediate impact of repeated reading (no tests for trans-
fer); 8 studies comparing different rereading methods and approaches, but
lacking a true control group; and single-subject designs only.
Due to the differences in populations (younger/older children; good
versus poor readers), only the most basic effect sizes could be com-
puted. The first outcome measure was a global reading score based on
whatever type of measure was used. The overall effect size for rereading
training was moderately positive (ES ¼ .48). Looking at separate measures
of reading, the greatest impact appeared to be for word recognition
(ES ¼ .55), followed by fluency (ES ¼ .44), and then by comprehension
(ES ¼ .35).
When normal versus poor readers were separated, and the data con-
trolled for sample size, the effect size was larger for normal readers (ES ¼
.50) than for poor readers (ES ¼ .33). But, as the authors pointed out, this
was confounded with the amount of time the rereading program lasted.
Studies on normal readers tended to last longer.
I want to look in detail at the programs with the greatest success and
take a more historical perspective. Little research appeared on the effects
of rereading until the 1970s, when S. J. Samuels (the senior member of the
NRP fluency panel) began to explore the aptitudes and strategies of effi-
cient fluent readers compared to poor readers. He and his colleagues
looked at how or whether using context enhanced reading speed. They
investigated the impact of training sight-word recognition speed com-
pared to rereading. Samuels (1979) reported that rereading with a target
goal of 100 wpm led to increased speed and accuracy, with savings (trans-
fer) from one story passage to the next, an effect that increased in effi-
ciency with each of five story passages. By contrast, sight-word training
(visual memory) was not effective.
 197

Samuels’s colleague, Dahl (1979), compared three methods (context,

Reading Fluency
sight word, rereading) in a completely crossed research design, in which
second-grade children were trained in one, or none, or any possible com-
bination of these three methods. Dahl found that context-based training
and the method of rereading were equally likely to enhance reading accu-
racy, but that rereading was most likely to increase reading speed. Sight-
word practice on the 800 isolated words that constituted the text had no
effect on either accuracy or speed.
While these studies were ongoing, Chomsky (1976) developed a
rereading technique in which children read along with an audiotape. She
reported that children improved in reading speed and accuracy, but there
was not any hard evidence to back up this claim.
Since this early work, a number of important issues have come to
light, and there has been increasing scrutiny of the assumptions about
what rereading actually achieves. Because so many factors are critical in
optimizing the effects of rereading, I will outline them here before dis-
cussing the more recent research.
Here are the questions that need to be addressed in doing research on
rereading or in designing an effective rereading program:

1. Does reading get faster with or without a target goal (criterion of

words per minute)?
2. Do children do better reading alone or with an audiotape?
3. Is rereading more effective than reading the same number of different
stories?
4. What is the optimum practice time necessary to produce lasting
changes?
5. How should this time be allocated: concentrated into a few days, or
spread out over weeks or months?
6. How many different stories should be reread to ensure a lasting effect?
7. Does the difficulty level of the text make a difference?
8. Does overlap in the words from one story to the next make a difference?
9. Does overlap in story content/context make a difference?
10. Does prosody (reading expression) improve along with speed?
11. What kinds of transfer effects are there and how should they be
measured?
 198

Five main goals or outcomes need to be measured for rereading to be

Chapter 7

deemed a success. First and second, reading speed should increase with no
loss of accuracy (or an increase in accuracy) as readers read faster. Third,
oral rereading should produce increasingly appropriate phrasal boundaries
and inflection (prosody). Fourth, comprehension should improve. Finally,
there should be transfer effects. Reading speed should increase from
one story to the next. If a criterion reading speed is set, this should be
achieved more quickly with each new story. Accuracy and comprehension
should improve with each new story as well. Getting children to read
‘‘fast’’ achieves nothing if they are inaccurate, fail to comprehend what
they read, and do not stay ‘‘fast’’ from one story to the next.
Sorting out these issues turned out to be more difficult than people
imagined. It is easy to get slow readers, accurate or inaccurate, to read
much faster in a short space of time (within an hour). It is not easy to im-
prove comprehension of that same passage or to show transfer effects. We
may not have all the answers to the questions raised above, because indi-
vidual researchers study different problems, and the children in the studies
vary in age and reading skill, but we are getting close.
Setting a target criterion seems, on the face of it, a better approach
than using an arbitrary number of repetitions—in other words, letting
children reread without a goal. Children report they like rereading and
enjoy having a goal. Because the studies vary, there is no direct proof of
this assertion. What is important is that if a goal is set, children are capa-
ble of meeting it. This raises the question of where to set the goal, because
rereading stops when the goal is met.

Training Studies on Improving Children’s Reading Speed

Herman (1985) did a training study on slow readers in grades 4 through 6
whose reading speed ranged from 35 to 40 wpm. This is exceedingly slow.
The slow readers were selected from children attending a reading lab in
the school. These children were not only slow but inaccurate, scoring
below the 17th percentile on a reading test.
The children read five different stories suitable for their reading level,
about 2 years below their grade. A target speed of 85 wpm was set. The
initial speed on story 1 was 47 wpm and the final speed 93 wpm, slightly
above the target. At this point, the children chose another story and re-
peated the process. Students worked on this task for 10 minutes per ses-
 199

sion, twice a week, for total of about 21 days (210 minutes) spaced out

Reading Fluency
over 3 months. The initial speed on the fifth and final story was 70 wpm, a
clear improvement over story 1 (transfer effect). The final speed on story 5
was the same as on story 1 (92 wpm) because rereading stopped when the
children met the target goal. (Would they have improved further if the
target goal had been shifted higher for each story?) Accuracy showed
excellent transfer. The average error score was 11 on the first reading of
story 1, and this dropped to 7.6 on the first reading of story 5, a highly
significant effect ( p < :01).
Rashotte and Torgesen (1985) were interested in the impact of word
repetition on transfer. The children were 8.5 to 12 years old. No target
goal was set. Instead, improvement was measured by how much faster the
child read after a fixed number of rereadings. Children’s initial reading
speed ranged from 31 to 62 wpm, with an average of 50 wpm. Because all
the stories were at a second-grade difficulty level, reading accuracy was
already good, and there was scarcely any room for improvement.
The children read about 15 minutes each day, for a total of 7 days
(105 minutes). They were divided into three groups, and each group read
slightly different materials. Two groups read various passages from the
same story, rereading the same set of passages four times each session. For
group 1, 60 common words repeated many times across all the passages.
For group 2, the same story and the same passages were read, but the
common words were replaced by synonyms and there was little word
overlap. Both groups read passages from the same story on 28 occasions.
Group 3 read 28 different stories, an important control for the impact of
‘‘reading a lot.’’
The two rereading groups increased their speed by 34 wpm to 84
wpm from the first to the last session. The reading speed for the children
who read 28 different stories improved by only 5 wpm, proof that reread-
ing increases speed but that ‘‘reading a lot’’ does not. When they looked at
transfer in terms of speed, accuracy, and comprehension, only speed was
found to transfer from one story to the next, and this effect was stronger
for group 1, where passages contained the repeating words.
There were differences between the two studies that make them hard
to compare. Superficially it looks like the gains in speed (93 and 84 wpm)
were similar. But text difficulty varied. Herman’s training was spread over
3 months; Rashotte and Torgesen’s was concentrated into 7 days. In both
 200

studies, documenting improvements in comprehension was a problem.

Chapter 7

Rashotte reported no gains, but this may be because the stories were too
easy and comprehension was nearly perfect to start with. Herman did find
gains and transfer, but her comprehension measure was indirect and
vague: ‘‘errors in context sensitive word substitutions.’’ This measure
sheds no light on whether story content was understood or remembered.
Nevertheless, there is agreement on certain facts. Rereading is effec-
tive almost immediately and children can reach a target speed when re-
quired to do so. Reading speed shows a transfer effect, particularly when
there is some overlap in the words. Rereading is considerably more effec-
tive than reading the same number of different stories. Reading for speed
does not occur at the expense of accuracy but actually enhances it. Im-
provement in accuracy is much easier to demonstrate when story content
is close to reading level or grade level (not too easy). Like speed, im-
provements in accuracy will transfer from one story to the next. Measures
of comprehension were confounded in both studies, and better methods
were called for.
In 1987, Dowhower published the most comprehensive study to date.
She investigated transfer effects for speed, accuracy, and comprehension,
along with measures of prosody. She also investigated two types of re-
reading experience. One group of children read stories out loud without
assistance, and the other group read along with an audiotaped version of
the same story. The target goal in both cases was 100 wpm.
Dowhower was interested in beginning readers who were just making
a transition from word-by-word decoding to more fluent reading. She
screened 89 beginning second-grade students and selected 17 who fit the
profile of ‘‘accurate but slow’’ readers. Children read the stories aloud,
whether working on their own or with an audiotape. The group working
with the audiotape listened to the story first, then rehearsed it out loud,
with the goal of being able to read in synchrony with the tape.
The study design was complex. There were five stories (numbered
1–5), all at a second-grade level and 400 words long. There were two
additional stories (200 words each) at the same level of difficulty: story A
was read only once at the very first session (baseline), and story B was read
only once at the very last session (the final transfer test). The rereading
part of the study began at session 2. The children read the first 200 words
of story 1 over and over until they reached the target of 100 wpm. This
 201

initially took many sessions. Dowhower timed each rereading to check

Reading Fluency
how close the children were to achieving the target speed. Reliability
checks were made throughout by using a second observer. When the
children reached the target speed, they were asked to read the second half
of the story—the last 200 words (the first transfer test). This transfer test
preserved story context. Word overlap between the two halves of the story
was not controlled.
When the children finished reading the transfer passage (once), they
were asked to choose another story, and the process began again. This
continued until all five stories were read at the target speed of 100 wpm.
At the very last session, the children read the unrelated story (story B—
final transfer) one time, and the experiment ended. Each child met with
Dowhower for 15 minutes most days of the week, and this continued for
as long as necessary to achieve the target goal on each story. The study
lasted a total of about 7 weeks, approximately 7.5 hours per child. This is
twice as long as the longest study above (Herman).
Gains and transfer effects were measured for speed, accuracy, and
comprehension. Comprehension was measured by asking different ques-
tions about each story on the first and last trials. Statistical comparisons
were made between the initial and final reading of the same story, the
initial reading and transfer portion of the same story, all five initial read-
ings, all five final readings, and the readings of stories A and B. Also,
measures of prosody were coded from the child’s tapes, and included
things like inappropriate pauses, reading phrase length, and intonation.
The results of the study are shown in table 7.1.
All contrasts in the table are significant for both groups of children,
and there were no differences between reading alone or with an audiotape
on any measures of speed, accuracy, or comprehension. Speed improved
on the transfer passage (the last 200 words) from story 1 to story 5 by about
10 wpm. Accuracy was already good at the outset (the children were
chosen for being accurate) and only got better. The results for compre-
hension were particularly impressive, improving from 57 percent (story 1)
to 72 percent (story 5).
Transfer effects were also high when the children switched to a new
story, as shown in the comparison of the first reading of story 1 to the first
reading of story 5. Both groups began at 41 wpm and improved to 58–65
wpm, which is close to normal for beginning second graders. Again,
 202

Table 7.1
Chapter 7

The impact of rereading on transfer of speed, accuracy, and comprehension. (Data

from Dowhower, 1987)
Story 1 Story 5

First reading of the transfer passage

Assisted Reading Practice
Speed 53 wpm 64 wpm
Accuracy 188 words 189 words
Comprehension 58% 66%
Unassisted Reading Practice
Speed 57 wpm 66 wpm
Accuracy 181 words 185 words
Comprehension 58% 66%
First reading of Story 1 and Story 5
Assisted Reading Practice
Speed 41 wpm 58 wpm
Accuracy 184 words 189 words
Comprehension 56% 79%
Unassisted Reading Practice
Speed 41 wpm 65 wpm
Accuracy 180 words 186 words
Comprehension 58% 66%
Story A Story B

Assisted Reading Practice

Speed 35 wpm 62 wpm
Accuracy 178 words 191 words
Comprehension 64% 80%
Unassisted Reading Practice
Speed 38 wpm 67 wpm
Accuracy 179 words 188 words
Comprehension 68% 82%
Note: All passages were 200 words long.
 203

accuracy was good at the outset and there was not much room for im-

Reading Fluency
provement. Comprehension improved markedly even though the story
contexts differed, from 57 percent correct (story 1) to 66–79 percent cor-
rect (story 5).
The most impressive result was the savings across the five stories in
how many trials (rereadings) it took to reach the criterion of 100 wpm.
Both groups had great difficulty achieving this target goal on the first
story, taking an average of 15 attempts. By story 5, the number of at-
tempts was down to 4.5. What was once difficult had now become easy.
There was a large improvement between story A and story B, the
stories read only once at the beginning and end of sessions. Speed in-
creased from about 36 wpm to 65 wpm for both groups. Accuracy in-
creased from 178 words correct to 190 correct out of 200 words.
Comprehension on story B was 81 percent correct.
The systematic effect of transfer across all sessions is illustrated in
figure 7.1, which shows the progression from the very first story (story
A), through each initial reading of stories 1 through 5, to the final story
(story B).
Nearly all measures of prosody showed significant improvement as
well for both groups. However, the children reading with an audiotape
improved more noticeably (and significantly) on several measures, show-
ing that ‘‘reading with expression’’ improves more easily with a model. It
is interesting that having a model did not have a differential effect on
measures of speed, accuracy, or comprehension. One would imagine that
hearing a story read with expression would assist comprehension by en-
hancing meaning.
Dowhower recommended that teachers use the rereading technique
(either method), because it obviously works and children like it. She sug-
gested using the assisted method first (audiotape), especially for children
who read extremely slowly, and shifting to unassisted reading when chil-
dren reach 60 wpm, when they seem to do better on their own. She ob-
served that very slow readers were less frustrated working with the tape
than working alone. She also pointed out that this is not a quick fix. The
children did not make significant gains at the transition from story 1 to
story 2, except marginally for speed. She had several suggestions for fur-
ther research, such as looking at the relationship between prosody and
comprehension, examining the different populations of children who
 204
Chapter 7

Figure 7.1
Mean scores for rate (WPM), accuracy, and comprehension for the first reading (trial 1) at the initial pretest
(Story B). From Dowhower, 1987, p. 400.
 205

benefit best from this technique, and considering whether rereading

Reading Fluency
should be used as a ‘‘crash course’’ in getting children up to speed or as
part of normal classroom instruction, interspersed with other lessons.
I want to add another suggestion. All these extremely slow readers
dutifully achieved 100 wpm, and this got easier and easier with each story.
When target goals are set, the children can never go beyond them. A
moving target might produce longer-lasting results. For example, by the
third story, the children needed only seven trials to get from a starting
speed of 55 wpm to 100 wpm. Achieving 100 wpm, however, did not have
much impact on the starting speeds of the next two stories (60 wpm and
61 wpm), despite the fact that the children could read at 100 wpm with
good comprehension and no loss of accuracy. If the bar is raised a little
each time, children might start coming closer to 100 wpm on the first
reading.

Reading Faster While Noticing More

There is a speed/accuracy trade-off for any timed task, and people can
consciously trade accuracy for speed, as well as vice versa. As a rule, one
will suffer to the extent that the other is favored. Levy and her colleagues
challenged this notion by asking whether people could read faster while
monitoring the accuracy of the text at the same time. This tests the intu-
itive belief that when people read quickly, they will inevitably miss details.
In a series of studies on college students, Levy and her colleagues
(Levy, DiPersio, and Hollingshead 1992) developed a proofreading task to
measure the amount of detail picked up by expert readers as they im-
proved in speed with each rereading. This study was designed to prove,
one way or the other, a theory originally put forward by Levy and Kirsner
(1989) that expert readers process both perceptual and linguistic informa-
tion simultaneously as they read. The strongest test of this theory would
be to find a secondary task that required the students to focus attention on
details of the text while they tried to increase their reading speed. In the
proofreading task, students had to cross out errors consisting of misspelled
words, as well as syntactic or semantic errors that made the sentence
meaningless. To measure the level of perceptual awareness, the inves-
tigators looked at whether novel (unexpected) changes in typescript or
word spacing would disrupt reading and hamper reading speed.
 206

Levy and colleagues got the usual rereading benefits in reading speed
Chapter 7

and also found that students could do the secondary task equally well as
they read faster and faster. Misspelled words were easier to detect (at
around 85 percent accuracy) than words that breached rules of syntax or
semantics (around 70 percent). An error in word spacing or an abrupt
change of font produced some interesting effects. They found that reading
speed would slow even though subjects did not consciously notice the
change. Apparently when a detail is inessential (does not affect proof-
reading or comprehension), it impinges at an unconscious level but does
not emerge into consciousness.
Levy and associates demonstrated that readers process perceptual and
linguistic/cognitive information in parallel, and they do this so efficiently
that much of the lower-level processing takes place below conscious
awareness. Good readers are mainly aware of content and meaning, but
they can easily become aware of perceptual detail with no loss of under-
standing if their attention is directed to it.
Levy wondered whether this effect could be demonstrated with chil-
dren, including those with poor reading skills. A similar study was carried
out on 144 children in the third through the fifth grades (Levy, Nicholls,
and Kohen 1993). Good and poor readers read one story four times and
had to locate spelling and word errors. The instructions to the children
stressed speed, and they were aware they were being timed with a stop-
watch. After the fourth rereading, they transferred to a new story, and
transfer effects were measured. The stories varied in difficulty level (easy,
medium, hard).
Poor readers read much more slowly, but both reader groups in-
creased their reading speed by about the same amount, around 50 seconds
faster from the first to the fourth reading. Transfer effects for reading
speed were proportionally the same, and were quite large. The children’s
success rate in spotting errors was similar to the college students’, and
detection patterns were the same as well. It was easier to spot spelling
mistakes than notice errors that violated meaning. Error-detection rates
increased modestly for both good and poor readers across rereadings, but
improvement was inconsistent. Comprehension transfer did not occur
until the fourth and fifth grades, and this varied as a function of story
difficulty.
 207

The authors concluded that ‘‘the more fluent rereading observed at

Reading Fluency
each grade level, and for both good and poor readers, was achieved with-
out ‘guessing’ or ‘sampling’ the printed pages. Word recognition became
more efficient, not attenuated, as fluency was acquired’’ (Levy, Nicholls,
and Kohen 1993, 321).
The studies by Levy and her colleagues added a new dimension to the
rereading literature by showing that not only do adults and children im-
prove in accuracy as they read faster, but they can also carry out a sec-
ondary task (proofreading) with no loss of accuracy. Furthermore, even
poor readers can do this, and though they were worse overall, they had
slightly higher gains in both speed and accuracy than normal readers did.
Levy and associates attempted to manipulate text difficulty in this
study and did not succeed, perhaps because they relied on the publisher’s
criterion for difficulty (‘‘grade level’’) instead of objective criteria, such as
word frequency, word length, and so forth. In two subsequent studies with
Faulkner (Faulkner and Levy 1994, 1999), this problem was remedied.
The 1994 study is the most important study on rereading since Dow-
hower’s investigation, because it nails down the final elements that deter-
mine what makes rereading successful.
The 1994 study contained four different experiments that manipu-
lated the type of transfer task and text difficulty. The subjects were third,
fourth, and sixth graders and college students, divided into good and poor
readers. Poor readers were selected on the basis of accuracy (word recog-
nition) and not reading speed. The first two studies involved an in-depth
analysis of the impact on transfer of story context and word overlap (rep-
etition of the same words). Each task consisted of two readings: the initial
story and a transfer story. There were four kinds of transfer stories: the
same story (rereading), a story with a high overlap of the same words (word
overlap), a story with a high overlap in content/context but little overlap in
words ( paraphrase), and a story unrelated in content or words (unrelated ).
The analysis focused on the reading rates and accuracy for the trans-
fer story only. The results were the same regardless of grade level. Re-
reading the same story produced the fastest times and the fewest errors for
everyone compared to the other types of transfer stories. When the re-
maining transfer stories were compared to each other, good readers im-
proved most on the paraphrased version, and got no boost from word
 208

overlap. Poor readers benefited from both, but most from the overlapping
Chapter 7

words.
The results showed that when the text is difficult, word repetition is
helpful and content overlap is less so. When the text is easy, similar con-
tent enhances reading speed and accuracy, and specific words are not as
important. In other words, people are more likely to read difficult text at
the level of the word (more focus on decoding), and process meaning less
well. When people read at an optimum difficulty level, they read for
meaning alone, and particular words do not matter as much.
If this is true, then story difficulty ‘‘causes’’ reading speed and deter-
mines the amount of meaning extracted from the text. This corresponds
to the research on eye-movement control. Rayner (1986) found that text
difficulty was the most likely cause of erratic eye-movement patterns. But
does this always hold? The children in these studies were reading stories
at their grade level. As a result, the stories were easy for the good readers
but quite difficult for the poor readers. For this reason, Faulkner and
Levy, in a second part of the study, varied text difficulty by tying it to
reading skill, comparing good and poor readers across the age span. They
used the same four types of transfer stories as before: rereading, word
overlap, paraphrase, and unrelated.
They found that text difficulty level alone could make good readers
look exactly like poor readers, reading at the same slow rate and level of
accuracy. However, while poor readers did indeed read much faster when
stories were easy for them, they never read at rates remotely like those of
good readers. In one comparison, second-grade good readers and sixth-
grade poor readers were both given a fourth-grade story to read. The two
groups of children read at the same rate (120 and 117 wpm). But when the
fourth-grade good readers read a very easy story 2 years below their grade
level, reading speed soared to 151 wpm. In another study by Faulkner and
Levy (1999), fourth-grade good readers read material at or above grade
level at the rate of 159 wpm on the first reading and 182 wpm on the
second. Yet poor readers at the college level read easy material at the rate
of only 142 wpm.
These were not training studies. Children read stories far fewer times
(only twice in the last set of experiments), and results will not be compa-
rable to the true rereading studies. It is clear that a couple of rereadings
will not be sufficient to bring poor readers up to speed even on simple
 209

stories. And while text difficulty is very important in determining reading

Reading Fluency
speed, it only goes part of the way in accounting for poor readers’ ex-
tremely slow reading speed. Something else is going on.
Poor readers may get caught in a text-difficulty trap, in which reading
slowly becomes a way of life. This would happen if they got off to a bad
start and began lagging behind their peers. The reasons would include
such things as slow oculomotor development, language delays, lack of
early experience with letter shapes and sounds, missing school through
illness, poor instruction, and so on. Once behind, children who stay at
grade level will always be reading material at a difficulty level beyond their
comfort zone. Their reading speed will be slow and will remain slow even
if decoding accuracy improves to average or better. This pattern would
explain Wimmer’s ‘‘slow readers,’’ or at least a portion of them.

Summary
The good news is that children do not have to stay slow, and we now have
the formula for success. The formula is the same for slow readers (accu-
rate or inaccurate) and beginning readers alike.
Set target reading speeds well above the child’s level. So far, no one
has reported failing to achieve target levels 50 to 60 wpm higher than the
child’s baseline speed. I believe targets should be reset each time the
child’s reading speed improves by a certain amount. The final goal is to
have the first reading of the story at a normal or superior rate for the child’s
age. At this point, rereading exercises can cease. Children need to have
multiple rereading experiences (many stories), not just a few. Dowhower’s
time frame of about 7 hours’ work over 7 weeks was effective and seems
optimal. Time pressure was not excessive, and sessions were not spread so
far apart that there were no carryover effects. The ultimate goal is the
desired target speed on the first reading, and this goal determines how
long the rereading sessions last.
The difficulty level of the text is critically important, because speed is
tied to difficulty level. Very slow readers should start off reading passages
at or just above their reading level, not their grade level. Once reading
speed improves, stories should increase in difficulty. Passages with over-
lapping words are best for struggling readers and very young readers, and
are most likely to produce carryover (transfer) effects from one story to
the next. Overlapping context helps as well. This creates a situation in
 210

which transfer is successful and boosts confidence and motivation to

Chapter 7

continue.
This is far from the last word on this topic, because more research is
needed to sort out the best way to apply these new findings in the class-
room and in remedial settings. But the technique of rereading has finally
come of age. Nevertheless, fundamental research questions remain. What
causes children with good reading instruction to be slow, as found in the
Salzburg studies? So far, the evidence is pointing to a low verbal IQ and a
weak verbal memory. But other critical variables have not been studied,
such as the rate of oculomotor development and the nature of the skills
the child acquired prior to going to school.
 8
VOCABULARY AND COMPREHENSION
INSTRUCTION

Printed words are oral language written down. Print is a filter through
which people can exchange oral messages across space and time. To make
these distant conversations possible, two things are necessary: you must
understand the spoken language, and you must know how the code works
(decoding accuracy and fluency).
Decoding and fluency are the gateway to reading comprehension,
but they do not work in isolation from a child’s vocabulary and oral-
comprehension skills. This can be observed in a number of ways. A word
might be accurately decoded (vampire) but have ‘‘nowhere to go,’’ because
the child does not know what it means. A word that is not in a child’s
vocabulary (sympathy) might lead to a distorted, though ‘‘legal,’’ decoding
(sim-pa-thigh?). A word that is known may fail to be decoded correctly due
to anomalies in the spelling code, glacier read as ‘‘glassier.’’
Comprehension means more than a good vocabulary. It involves a
number of core language skills, such as the ability to use syntax to antici-
pate words in a sentence and assign unknown words to the appropriate
part of speech. It includes an aptitude for monitoring context, making
inferences on the basis of background knowledge, as well as familiarity
with oral or literary forms (genres). Children with good oral comprehen-
sion who read the phrase ‘‘the bunnies huddled in the dense green grass’’
may not know the meaning of the words huddle and dense, but they will
know that huddle is something the bunnies are doing (‘‘verb’’), and dense is
a property of the grass (‘‘adjective’’). They know this implicitly because of
where these words occur in the phrase.
Other comprehension problems can arise from the nature of the com-
munication or the text itself. Young children have particular trouble with
a ‘‘story grammar’’—the special sequential structure of a story and its
 212

fictional nature. This is surprising in view of the fact that children hear so
Chapter 8

many stories. Yet when children are asked to ‘‘tell a story,’’ most cannot
initiate or order the structural elements until around the age of 8 (Hudson
and Shapiro 1991). They routinely omit the ‘‘flags’’ that signal a story
beginning and ending (‘‘once upon a time’’; ‘‘they lived happily ever
after’’). They fail to provide a setting or any fictional characters. (Young
children’s stories are invariably autobiographical, with the children them-
selves in the title role.) They fail to create a problem or obstacle to carry
the story forward (story line), which means there is no resolution (story
apex). And despite what parents think, young children cannot retell a story
they have heard scores of times. They invariably get the story sequence
scrambled (Nelson 1998).
Reading comprehension is the end game of learning to read and nec-
essarily involves everything that comes before it: a good vocabulary and
good oral-comprehension skills, plus accurate and fluent decoding skills.
If children do badly on a reading-comprehension test, any of these four
things, alone or in combination, could be the culprit. Children who get
low scores on a reading-comprehension test solely because they cannot
decode are a very different from children who have low scores due to weak
oral-comprehension skills.
In English-speaking countries, reading researchers have focused much
more attention on decoding than on reading comprehension. In large part,
this is a consequence of the enormous number of children who fail at this
level. Nevertheless, the ultimate purpose of being able to read is to un-
derstand the message conveyed by the print. This is certainly the primary
goal of most teachers, even in the earliest grades, as has been seen many
times in this book. So far, we have learned that time spent on verbal lan-
guage skills is time taken away from learning how to decode and spell. But
we also know that decoding and basic spelling skills can be learned quickly
if they are taught appropriately. And there are excellent techniques to
improve reading fluency.
The tests used in the National Assessment of Educational Progress
(NAEP) to estimate reading competency in the United States, measure
reading comprehension, not decoding accuracy (Mullis, Campbell, and
Farstrup 1993; Campbell et al. 1996). When NAEP reported in 1993 and
1996 that 43 percent of fourth graders in America were ‘‘functionally
illiterate,’’ this did not mean these children could not decode (though that
 213

may have been true as well). It meant they could not locate information or

Vocabulary and Comprehension Instruction

use the information conveyed by the text to interpret meaning.
The question here, of course, is whether the failure of these children
was due to problems with decoding, to comprehension, or to a combina-
tion of the two. This point needs to be stressed, because it is frequently
overlooked in the research on reading comprehension, despite the fact that
the relative contribution of oral comprehension and decoding to scores on
a reading comprehension test has been highlighted in a number of studies.
For beginning readers, decoding accuracy is more strongly correlated
with reading comprehension than oral comprehension is ( Juel, Griffith,
and Gough 1986), but this quickly reverses. Hoover and Gough (1990)
studied 254 children who were tracked from grade 1 to grade 4. Decoding
was the best predictor of reading comprehension at first grade (r ¼ :84),
and oral comprehension was not a factor. At second grade a different pat-
tern emerged. Now decoding and oral comprehension were equally im-
portant, with oral comprehension gaining ground from second to fourth
grade (.71 to .87). However, the product of the two scores (decoding times
oral comprehension) was more powerful than either alone (r ¼ :85 to .91),
accounting for 72 to 83 percent of the variance.
Aaron (1991) tested 180 children in grades 3 through 8 on the Wood-
cock Reading Mastery tests of word recognition and passage comprehen-
sion (Form G was used for reading and Form H for listening). The
correlation between oral and reading comprehension grew larger over
time, from r ¼ :65 at third grade to r ¼ :75 at eighth grade. The impact
of decoding skill was not nearly as great; it decreased over time, rang-
ing from r ¼ :36 to r ¼ :21. As in the previous study, the combined
score (oral comprehension plus decoding) was a much stronger pre-
dictor than either alone, with fairly consistent values across the age range
(mean ¼ :85).
Aaron also discovered a large increase in reading speed over this age
range. When words were read in context, children averaged 7.5 letters per
second at third grade and 12.5 letters at eighth grade. Reading speed for
isolated words was slower and less variable (6.0 to 7.6 letters per second).
This shows that older children have learned to make good use of context
in anticipating words.
Joshi, Williams, and Wood (1998) confirmed the listening-reading
relationship on 273 bilingual children in grades 3 through 6. They used
 214

the same Woodcock tests, but modified the listening version of the test to
Chapter 8

a ‘‘cloze’’ format (‘‘supply the missing word’’). The correlation r between

oral and reading comprehension ranged from .68 (grade 3) to .75 (grade
6), with a mean of .67, values almost identical to Aaron’s. They also found
that correlations fluctuated depending on the tests. They gave third and
fifth graders three different standardized oral-comprehension tests and
correlated these scores with a composite measure of reading comprehen-
sion. Correlations for the Wechsler were modest (around r ¼ :45), but
were stronger for the Woodcock (r ¼ :60 for third grade and .65 for fifth
grade), and stronger still for the Peabody comprehension test (r ¼ :69
and .90).1
Nation and Snowling (1997) confirmed these findings in a factor anal-
ysis. Children were given a variety of decoding and oral-comprehension
tests. A factor analysis is a mathematical tool for determining which test
scores cluster or group together (known as ‘‘loading’’), and is essentially a
measure of the redundancy of the tests. The four decoding tests loaded on
one factor at very high values (all above .86). Oral comprehension loaded
on a second (different) factor at .95. Reading comprehension loaded on
both factors equally (.62 and .67). Once more, we see that skill in read-
ing comprehension is due to a combination of oral comprehension and
decoding.
This is unassailable evidence that good oral-comprehension and
decoding skills are not only critical to being able to understand what you
read, but they are the only two skills that matter, accounting for nearly all
the variance on tests of reading comprehension. There is no support here
for the notion that reading comprehension is a special or isolated skill. We
will come back to this issue later in the chapter.
These are critical findings with important ramifications for research.
Studies of instructional methods for improving reading comprehension

1. As a general rule, if correlations are real, the better the test, the higher the
values will be. This is a function of the items on the test, their difficulty level,
and the number of items at each level. A good test increases the likelihood of
a normal distribution, which will increase correlational values. The Peabody is
considered one of the best tests available for oral and reading comprehension.
 215

must control decoding skill. In view of the enormous overlap between

Vocabulary and Comprehension Instruction

oral and reading comprehension, methods of enhancing ‘‘reading com-
prehension’’ need to specify what is unique to reading comprehension—if
anything.
In the remainder of the chapter I will look at the research on two
types of instruction: vocabulary training and methods for improving read-
ing comprehension. I will be departing somewhat from the format of the
previous chapters, because the National Reading Panel’s analysis of the
vocabulary and comprehension studies was not handled in the same way as
the previous topics. No meta-analysis was carried out in either case, and
few effect sizes were reported. For these reasons, I have singled out certain
studies for special attention and have computed effect sizes myself in all
cases where this was appropriate.

Vocabulary Instruction
There is a popular theory that listening to stories and ‘‘reading a lot’’
causes vocabulary and comprehension skills to improve. This is assumed
to be true because written text contains more complex and rare words than
appear in everyday conversations. Haynes and Ahrens (1988) found that
children’s literature contains 50 percent more rare words than prime-time
television or college students’ conversations. When adult readers encoun-
ter unknown words, they try to work out meaning from syntax, context,
and word derivation. This is a lifelong process that is never completed.
Thus there are two propositions one could hold about the rela-
tionship between vocabulary and reading. One is that ‘‘vocabulary causes
reading,’’ because the more words are stored in memory, the easier it is to
decode them. The second is that ‘‘reading causes vocabulary,’’ because if
you ‘‘read a lot,’’ you learn more new words. Both lines of reasoning could
be correct (and probably are), and this has important consequences for
reading instruction in the classroom. (We have already seen that ‘‘reading
a lot’’ does not cause fluency or decoding accuracy.) But there are more
fundamental issues.

Vocabulary Development
Research on vocabulary instruction has to build on basic knowledge of
how vocabulary skills develop. So far, there has been a disconnect between
 216

the scientific knowledge of language development and the methods pro-

Chapter 8

moted by the educators and researchers. A prime example of this problem

is Chall’s descriptions of the 1960s basal-reader lessons, which showed
that teachers took up most of the reading lesson explaining the meaning of
words in the lesson that children already understood. We see it today in
whole-language classrooms, where children are supposed to acquire new
vocabulary (among other things) from listening to stories, and to guess
word meaning from context cues as they read. We saw it in Sumbler’s
observational study on 20 kindergarten classrooms, where teachers spent
the bulk of the language-arts period reading stories and doing vocabulary
work, time that was negatively correlated with every reading test score,
including reading comprehension. Evans and Carr’s data showed that this
negative correlation to reading comprehension increases across grades 1
through 3 (r ¼ :70 to .80), the opposite of what teachers believe.
This is strong evidence that not only do language activities take time away
from learning the code, they do not even contribute to comprehension
skills.
Given the fact that educators hold such strong views about the merits
of vocabulary training, we need to establish a baseline, or starting point,
for what children bring to the table. The average 5-year-old has a vo-
cabulary of approximately 10,000 words and a 6-year-old, about 13,000
words. Throughout childhood, the average child learns about 3,000 words
each year, though this number varies enormously from one child to an-
other (Nagy and Herman 1987; White, Graves, and Slater 1990). This is
about 10 new words per day. Children would have to learn at this rate in
order to acquire even a garden-variety adult vocabulary of 55,000 words,
the number of words that John and Mary Smith need to participate in
everyday conversations. This pattern of rapid, untutored vocabulary ac-
quisition begins in infancy at around 18 months and has been christened
fast mapping by developmental psychologists. Children’s books and other
texts (printed school English) go well beyond the 55,000-word level. They
contain about 88,500 words, according to Nagy and Herman’s estimate.
Children are not taught these words. They pick them up from
parents, friends, and other sources like television. If children stay true to
the patterns observed in early childhood, they remember words that in-
terest them, words that make it possible to discuss important events from
their point of view (Nelson 1998). The rapid expansion of vocabulary is
 217

a central problem for vocabulary instruction and for empirical research.

Vocabulary and Comprehension Instruction

The amount of time it takes to teach new vocabulary directly, compared
to what children are able to learn indirectly, must be taken into account.
If it takes 6 months of daily drill to teach 40 vocabulary words so that all
children know them, is this a productive way to spend class time? This
point was also raised by Nagy and Herman.

The Importance of Parents’ Input

But there is much more going on than ‘‘developmental rate.’’ A remark-
able study was carried out by Hart and Risley (1995), who monitored the
enormous individual differences in vocabulary growth and the possible
causes of these differences. Having studied children in Head Start pro-
grams, they became frustrated by the fact that it was hard to make a
noticeable difference in the children’s language skills. They believed the
problem started much earlier and decided to study parent-child interac-
tions in the home. These interactions were videotaped for 1 hour every
month from the time the children were 9 months old until they were 3.
Ultimately, 42 children and their families made it to the end of the study.
Thirteen families were from the highest socioeconomic group and des-
ignated ‘‘professional.’’ Twenty-three families were middle class, and six
families were on welfare.
The focus was on the child’s spoken vocabulary, and on how parents
(usually the mother) spoke to the child. The mother’s communications
were scored for the number of words per hour, and for simple number
counts of nouns, adjectives, past-tense verbs, and wh-questions used in
each session, as well as her style of verbal interaction with her infant. The
investigators tallied the number of imperatives, statements of approval and
disapproval, examples of positive and negative feedback, and so forth. In
the final analysis, these measures were grouped into five categories: lan-
guage diversity (number of different nouns and modifiers), positive feedback
tone (repetitions, extensions, expansions, confirmations, praise, approval),
negative feedback tone (imperatives, prohibitions, disconfirmations, criti-
cisms, disparagements), symbolic emphasis (the degree to which parents
made connections between things and events, as indicated by richness
of nouns, modifiers, and number of past-tense verbs), guidance style
(the number of invitations—‘‘Shall we?’’—divided by the number of
imperatives—‘‘Stop it!’’), and responsiveness (the number of responses to
 218

the child—‘‘Oh, you want Mommy to take the ball’’—divided by the

Chapter 8

number of initiations to the child—‘‘Let’s play with the blocks’’).

There were enormous differences between the high-, middle-, and
low-socioeconomic groups on a simple measure of the mothers’ verbal
output to their children. The average number of words per hour addressed
to the child between the ages of 13 and 36 months was over 2,000 for the
professional mothers, 1,250 for the middle-class group, and 616 for the wel-
fare mothers. This was despite the fact that welfare mothers spent, overall,
more time in the same room with the child than the other groups did.
There were differences, as well, as a function of the child’s age. Pro-
fessional mothers not only talked much more to their babies (1,500 words
per hour at 9–12 months) than the other mothers did, but this verbal
barrage increased systematically with the child’s age, leveling off by 30
months at around 2,500 words per hour. The middle-class parents spoke
less overall, and their initial rate was lower and increased more modestly
(1,000 to 1,500 words). The range for the welfare mothers was virtually
nonexistent (600–750 words). Based on these numbers, it was estimated
that by age 3, a child from a professional family would have heard 33
million words, a middle-class child, 20 million, and a child of a welfare
mother, 9 million.
But this did not tell the whole story. The three groups were no-
ticeably different in the way mothers interacted with their children. Pro-
fessional mothers used a much richer vocabulary. Interactions with their
children were consistently positive, at twice the rate of middle-class
mothers and five times the rate of welfare mothers. Professional women
rarely used negative feedback of any type. They were highly responsive
and far less inclined to be directive. Middle-class parents were ‘‘similar but
less so’’ in terms of the positive measures. Welfare mothers had a very
different style of verbal interaction with their children. Almost 80 percent
of the feedback to the child was negative and prohibitive (‘‘Stop it.’’
‘‘What did I tell you?’’ ‘‘Put that back or else.’’ ‘‘I said no!’’). They fre-
quently discouraged or disparaged their youngsters, calling them ‘‘stupid’’
or ‘‘dumb.’’ Encouragement was rare, and sometimes absent altogether.
This did not mean that mothers on welfare were cruel and did not love
their children. They were affectionate and met their baby’s needs.
The children’s vocabulary development was strongly related to the
sheer quantity of the words they heard. At age 3, children in professional
 219

families had an average spoken vocabulary of 1,115 words (true count),

Vocabulary and Comprehension Instruction

middle-class children, 750 words, and children of welfare mothers, 525.
IQ differences between the groups were large as well (117, 107, and 79
respectively), which raises the question of how much of this effect was
genetic. Verbal intelligence is highly heritable, genes accounting for over
50 percent of the variance, with shared environment accounting for the
remainder (Plomin et al. 1999). Professional parents with high verbal IQs
would be more verbal (talk a lot), use a more complex vocabulary, and
handle the interaction with their child more sensitively (more ‘‘intelli-
gently’’). There is certainly evidence for this interpretation. Mothers’
vocabulary scores were strongly correlated with their children’s actual
(recorded) vocabulary and with their children’s IQ—possible support for
the effect of heredity.
But heredity is not destiny. When they looked at the quality and
character of the mothers’ interactions with their children, these inter-
actions turned out to be more important than the parents’ vocabulary. All
five categories of parenting style were highly predictive of the child’s
vocabulary, IQ score, and performance on general language tests at age
3. Not only this, but the parents’ communicative style predicted their
children’s language skills almost as well when the children were followed
up at age 9.
Although the mother’s communicative style cannot be completely
disentangled from her verbal IQ, we can see the impact of style on the
child’s language development. And there is certainly no reason why any
parent or any teacher cannot copy what works best. Here are the key
implications:

The sheer quantity of parents’ verbal input (total number of words per
hour) predicts a child’s spoken vocabulary later in time. (Other research
has shown that speech must be child directed and not adult directed.)

The quality of the communicative style (its richness, as well as the type

of feedback the child receives) was a stronger predictor of the child’s ver-
bal development than socioeconomic status was. This was seen in the data
for individual children, where socioeconomic status mattered much less
than how the mother interacted with her infant.

Five key communicative styles were identified. These are, in order of

importance:
 220

Guidance style. Provide gentle invitations to play and engage in positive

Chapter 8

interactions. Avoid prohibitions.

Symbolic emphasis. Make connections between words and things and other
words.
Feedback tone. Positive feedback is good. Negative feedback is bad.
Language diversity. Use different nouns and adjectives as much as possible.
Responsiveness. Tune in. Follow the child’s lead. Avoid telling the child
what to do.

These guidelines are just as important in the classroom as in the home

if the goal is to teach vocabulary and oral comprehension. Vocabulary
training requires lots of repetition. It should include attempts to link new
words to the child’s vocabulary and knowledge via symbolic emphasis and
use of analogy. New vocabulary needs to be embedded in a context-rich
mélange of nouns and adjectives. Other items on the list are simply good
pedagogy: the importance of offering positive and not negative feedback, as
well as of finding ways to engage interest without demanding or ordering it.
Another basic issue is the question of the orderly progression of
learning to read as it relates to vocabulary lessons. This question has
become obscured by whole-word teaching methods, especially whole lan-
guage. In the early twentieth century, children’s readers were composed
of words with ‘‘regular’’ (decodable) spellings that children understood.
Story line, topic, content, and syntax were also geared to the age of the
child. As children acquired decoding skills, books gradually increased in
complexity and difficulty until children were encountering words they did
not understand. Whole language has muddied the waters by substituting
children’s literature for simple readers and making this the only resource
for teaching reading. Not only do the children in whole-language class-
rooms lack decoding skills (because they are not taught any), but they are
faced with comprehension difficulties as well.
If reading is taught properly, there is a means-end reversal in the re-
lationship between reading skill and vocabulary acquisition. As seen above,
children’s oral vocabulary is in advance of what they are able to read. After
this, print vocabulary begins to outpace oral vocabulary. When this hap-
pens, books become an additional resource for children’s vocabulary de-
velopment. Do children pick up new vocabulary from reading books
in the same way they pick up new words from listening to people? The
 221

short answer is that we do not know, because this is hard to demonstrate

Vocabulary and Comprehension Instruction

empirically. However, there are clues from studies investigating whether
children learn new vocabulary from being read to by an adult.
At this point we encounter the Achilles heel of vocabulary research:
how to measure vocabulary acquisition. The typical study involves expos-
ing children to a set of new words, with or without special training, and
measuring whether they learned them. Standardized tests are not good
measures of vocabulary acquisition, because the training is specific to par-
ticular words. For this reason, researchers design in-house tests using a
multiple-choice format. Children are asked to point to one of four pic-
tures, or to circle the correct word among four words after hearing a word
or definition spoken by the teacher.
Multiple-choice tests are forced-choice tests and susceptible to guessing.
To control for this possibility, the binomial test must be carried out, as
noted in chapter 6. Recall that this test computes the minimum score that
is significantly greater than chance at a specific probability (i.e., p ¼ :05).
The formula takes into account the number of choices for each item and
the number of items on the test. As a basic principle, tests with few items
and few choices are very bad tests. To score significantly above chance
at p < :05 on a two-choice test (true-false) containing ten items, a person
must get eight or more correct (not five, as many people believe).
Despite the fact that multiple-choice tests are the most common
measure used in vocabulary research, I have not come across any studies
where the researchers knew how to address this problem, and most were
unaware it was a problem. We can see the consequences of this oversight
in the following group of studies.

Do Children Learn New Vocabulary from Listening to Stories?

Do children, in fact, do what teachers and parents assume and learn new
words from listening to stories? Robbins and Ehri (1994) found (inadver-
tently) that much more is required to imprint new words on memory than
listening to stories. They read a story to 38 kindergarten children (indi-
vidually) in which 11 target words (new words) appeared either one or two
times. A day or so later, the same story was read again. The story was read
with no expansion or explanation about the target words.
After the second reading, the children were asked to listen to a series
of sentences that contained either the target word or a novel word that
 222

was not in the story (‘‘foils’’). They had to point to one of four pictures
Chapter 8

that best fit the meaning of the sentence (25 percent correct by chance).
The success rate for the foil words (words not in the story) was just over 3
words. The success rate for the target words was 4.4 words, not much
better. Neither score was significantly above chance (5.6 correct is signif-
icant at p ¼ :05). Only the high-vocabulary group scored this well, and
just barely.
There is no evidence from this study that the majority of children
learned any words, despite the authors’ conclusions to the contrary. Even
the high-vocabulary group learned only two or three words.
Senechal and Cornell (1993) reasoned that if parents or teachers
used a more interactive style, this would enhance memory for new words.
There were 80 four-year-olds and 80 five-year-olds in the study. All were
middle or upper middle class. The design of the study was similar to the
one above, with target words embedded in a story. The story-reading
session (only one) took about 30 minutes. The target words were odd and
unfamiliar (as they were in Robbins and Ehri’s study): angling, corridor,
elderly, gazing, infant, lineman, reposing, sash, satchel, snapshot. All had a
familiar synonym: fishing, hall, old, looking, baby, repairman, resting, window,
purse/bag, picture.
The children were tested initially on their knowledge of the target
words and their synonyms. All children knew the familiar synonyms, but
none knew the target words. The next day, they met with an experimenter
(individually) and listened to the story. The story was read in one of four
ways, which varied in how much repetition or questioning was involved.
Immediately after this, the child took recall and recognition tests, and one
week later, the recognition test was given again.
We can dispense with the four treatments, because they made no dif-
ference whatsoever to either age group. Children remembered just as well
(or badly) when they heard the story read with no help or with expla-
nations of the target words. We can also dispense with the data for the
recall-memory test, because children did not do what they were supposed
to do. When they were shown a picture representing a target word, in-
stead of saying the word from the story, they gave the common word (the
synonym) almost exclusively. Only 18 of the 160 children provided any
target words throughout testing.
 223

The recognition-memory test suffered from the same problem as in

Vocabulary and Comprehension Instruction

the Robbins and Ehri study: the fact that children would get 25 percent
correct by chance. To their credit, Senechal and Cornell did address this
problem, but their solution (subtracting pretest scores) does not take into
account the number of items on the test. A child needed to score 5.0 or
higher to do statistically better than chance in this study. As seen in the
accompanying table, none of the scores met this criterion. (Scores are
averaged across treatments.)

Pretest First posttest Second posttest

4 years 1.6 2.9 3.5
5 years 1.8 3.5 4.3

These results show that a handful of children learned one to two words
from initial testing to follow-up, but most children were just guessing.
Even viewed in the most generous way, it takes 30 minutes per individual
child to add one word to their vocabulary, and even then children will not
use the word spontaneously, making it hardly worth the effort.
The target words in this and the previous study were decidedly odd.
Children add words to their vocabulary so they can say something they
could not say otherwise. There is no reason they would prefer the archaic/
technical/literary target words in these stories as replacements for com-
mon synonyms they already know. This may be one explanation why
these experiments did not work.
If experimenters reading to a child have no impact on vocabulary ac-
quisition, would preschool teachers do any better? And if so, what type of
teacher-child interaction works best? These were the questions addressed
by Dickinson and Smith 1994. This study only serves to illustrate just how
difficult an assignment this was. This was a naturalistic study in which the
teacher’s style of interaction during story time was uncontrolled. It was
coded after the fact from videotapes in 25 different Head Start classrooms.
At the end of the year, 25 children (5 years old) were randomly selected
from these classrooms and tested on vocabulary and story comprehension.
 224

The videos were coded for the amount of talk during story time for
Chapter 8

teachers and the children in each class, plus styles or types of interaction.
The investigators coded 21 different measures, which were combined
to represent three types of classroom interactions. The first was labeled
co-constructive (5 classrooms), in which a high number of analytic con-
versations took place between the teacher and the children, prior to,
during, and after reading the story. The second type was called didactic-
interactional (10 classrooms). Talk was limited generally, and what talk
did occur consisted of repetition (saying a sentence again) or answering a
question. The final type was performance-oriented (10 classrooms). Here,
talk largely occurred prior to and following reading the story, and the
story was read with little interruption. The preamble to reading the story
was often extended, analytic, and evaluative. When the story was finished,
questions were asked about story recall and understanding. Sometimes this
involved reconstructing the entire story piece by piece.
This was Dickinson and Smith’s interpretation of what the data
showed. But a table of probability values indicated that only five measures
discriminated among the classrooms at a conservative p < :01. Most of
these measures had to do with how much talk was going in the classroom
by both teachers and children prior to, during, and after story reading.
The only other discriminating measure was the proportion of teacher and
child clarifications about the story. Using these more conservative mea-
sures, classroom ‘‘style’’ boiled down to two things: the total amount of
talk by teachers and children, and how this talk was distributed between
the prestory and poststory phases and during the story.
Vocabulary and story-comprehension measures were compared for
the three types of classrooms. Children in the performance-oriented
classrooms had significantly higher vocabulary scores than children in the
didactic-interactional classrooms ( p < :01). (One assumes no other com-
parisons were significant because no other values were provided.) No dif-
ferences were found for story-comprehension scores.
There was a fatal design flaw in this study. Vocabulary was not measured
before the children entered the study. Without this baseline, there is no way to
know whether the children’s vocabulary levels ‘‘caused’’ (animated and
extended) the type of interaction that went on in the classroom, which
seems highly likely, or whether what went on in the classroom ‘‘caused’’
 225

the vocabulary. The authors opted for the second interpretation and did

Vocabulary and Comprehension Instruction

not consider the first.2
These are examples of some of the studies carried out on younger
children, looking at the impact of simple story telling and story time on
vocabulary acquisition. Basically, nothing was found. Children perform at
chance on tests of story vocabulary, and while the classification of classroom
‘‘style’’ may have some validity, its impact on vocabulary acquisition cannot
be determined without knowledge of the baseline measures of vocabulary.
Some insight about why children do not learn more from listening to
children’s books was provided by a study by Brett, Rothlein, and Hurley
(1996). Earlier, Elley (1989) had reported that when difficult words were
defined and explained, receptive vocabulary improved noticeably. Brett
and colleagues set out to explore this systematically. Teachers were pro-
vided with two storybooks along with scripted definitions for 10 target
words from each story, plus synonyms to help clarify meaning. The
teacher would read ‘‘As the boy went back to the village, he felt despon-
dent,’’ followed by ‘‘As the boy went back to the village, he felt helpless.’’
Target words were abstract, and included words like indulgent, reverie,
imminent, disdain, sauntered, obscure, emanated, immerse, exemplary. There
were 175 fourth graders in the study, divided into three groups. The first
group heard the two books read over a 10-day period (5 days each), and
the 10 target words were defined as described above. The second group
heard the same books but had no explanation about the meaning of these
words. The third group did not hear either book.
A multiple-choice vocabulary test was constructed for each set of tar-
get words. Children took this test prior to hearing the stories, immediately

2. A series of regression analyses were carried out in which the children’s

language scores were ‘‘correlated’’ with the various measures of classroom
behaviors. I am unaware of any statistical tests in which individual scores from
one set of tests can be correlated with group scores on a different set of tests.
The investigators’ claim that the ‘‘joint proportion of prompted or respon-
sive analysis, prediction and vocabulary utterances by teachers and children’’
accounted for 50 percent of the variance in the children’s vocabulary is
unsupportable.
 226

after the 5-day book-reading period, and also 6 weeks later. Only the
Chapter 8

children who were taught the meanings of the target words scored sig-
nificantly above chance. They did well on both books immediately after
training and also 6 weeks later, scoring around 50 percent correct. The
control group that heard the books read for the same amount of time with
no explanation, did no better than the control group that had never heard
the books, both groups scoring at chance.
These results showed that simply listening to a story does not impact
vocabulary acquisition. If new words are explained and synonyms provided
in context, there is some success. Nevertheless, even a 5-day training
period was not long enough for the children to remember more than half
of 10 new words. Meanwhile, they would have acquired 50 words on their
own without any training.

Direct Vocabulary Instruction in the Classroom

The volume of studies on vocabulary instruction peaks at around fourth
grade, according to a tally provided by the National Reading Panel. This
is the age where direct instruction in vocabulary becomes an issue, no
doubt as a consequence of the format of basal-reader programs. Very little
is understood about what methods work best and why.

Stahl and Fairbanks The first attempt at a meta-analysis of the classroom

studies on vocabulary instruction was carried out by Stahl and Fairbanks
(1986), who deserve a good deal of credit for their efforts. However,
it must be noted that neither Stahl and Fairbanks nor the NRP vocab-
ulary group discuss the ubiquitous problem of controlling for guessing on
multiple-choice tests. This means that effect sizes will be based on bogus
values and will be unreliable. This is a far greater problem than the varia-
tions in training methods, which was a major complaint in the NRP report.
Stahl and Fairbanks’s review targeted the extreme variability in
instructional methods as a key issue in coming to grips with this research.
They established five categories for sorting the studies based on the
degree to which the instruction was largely definitional, largely context
based, or mixture of the two. They also identified another feature of the
instruction in terms of the richness or depth at which new vocabulary
words were processed. They isolated three characteristics of methods:
 227

purely associational, comprehension oriented (using words in sentences,

Vocabulary and Comprehension Instruction

finding synonyms), and oriented toward generation (using words in novel
utterances). The third category was the number of exposures of the words:
one, two, or many. The fourth category dealt with the ‘‘setting,’’ which
included factors like whether the target audience was a group (classroom)
or an individual, and, finally, how much time was spent teaching the
words. Total time varied enormously from one study to the next, ranging
from several weeks to less than an hour.
A study had to meet two criteria to be included in the database: it
had to include a control group, and there had to be sufficient statistics to
compute effect sizes. The researchers located 52 studies that met these
criteria, including a total of 94 comparisons or ‘‘cases.’’ The studies were
further divided into group versus individual studies, by the number of
minutes allocated to each new word, and by the method of instruction.
The studies split about evenly among two types of control groups, one
where the control children had no exposure to the words, and one where
students had exposure (saw them in print or on worksheets) but got no
special training. Test scores for both vocabulary and passage comprehen-
sion were reported.
The results of the meta-analysis were described as ‘‘trends’’ rather
than anything definitive due to the high variability and the lax screening
criteria. The total effect size for vocabulary training compared to a no-
exposure control group was ES ¼ :97, a substantial effect. Mixed methods
of instruction worked better in all conditions than merely providing defi-
nitions or having students deduce meaning from context. However, as
expected, effect sizes were substantially reduced when standardized tests
were used to measure vocabulary and passage comprehension (ES ¼ :30
and .29). Methods using multiple repetitions of the words had double the
effect sizes, and this was true for all measures including passage compre-
hension. Effect sizes did not vary as a function of individual versus group
instruction.
Stahl and Fairbanks discussed a variety of limitations with the studies,
not the least of which was the vagueness with which the training program
was described. However, this is not nearly as serious as computing effect
sizes from unreliable test scores, and we have no knowledge of which test
scores were reliable.
 228

The NRP Report The NRP subgroup’s analysis of the training studies, 14
Chapter 8

years on, was disappointing. It would have been helpful to have an update,
perhaps adopting Stahl and Fairbanks’s classification scheme and compar-
ing this to more recent research. Instead, the NRP declared all research
null and void for purposes of a meta-analysis. The initial screening criteria
were publication in a scientific journal, inclusion of a control group that
was either matched or randomly assigned, and a proper statistical analysis.
The initial search turned up 197 papers on ‘‘vocabulary’’ plus ‘‘instruc-
tion,’’ and after screening, 50 studies remained in the pool. On further
analysis of these 50 studies, the panel decided that no research met the
NRP criteria that explicitly addressed measurement issues. In the execu-
tive summary of this report, the reason for ruling out a meta-analysis was a
‘‘heterogeneous set of methodologies, implementations, and conceptions
of vocabulary instruction’’ (p. 4-3).
Instead, the panel provided brief descriptions of 40 studies, set out
in various categories. Ten studies overlapped categories and appeared in
more than one place. Included among them were the studies I presented
above on storybook reading, where the data were invalid due to the failure
to control for guessing. The panel was unaware of this problem and their
report on these studies is inaccurate (see p. 4-21).
The NRP stated that Senechal and Cornell showed that ‘‘a single
book reading significantly improved children’s expressive vocabulary.’’
However, there was no significant effect for receptive vocabulary in this
study, and even Senechal and Cornell reported that there was no impact
on expressive vocabulary, because they found that children used few target
words to label the illustrations . . . and there was not enough variability in
the data to conduct statistical tests.
The NRP stated that Robbins and Ehri’s method ‘‘helped teach chil-
dren meanings of unfamiliar words.’’ Yet the data were invalid in this
study as well. They claimed that Dickinson and Smith showed that ‘‘the
amount of child-initiated analytic talk was important for vocabulary gains’’
when it did nothing of the sort. Gains were never measured in this study
(no baseline). The panel’s final summary of these 40 studies was even
more troubling, because these and other inaccurate conclusions were gen-
eralized further.
 229

Direct Vocabulary Instruction Does Work

Vocabulary and Comprehension Instruction

At this point, I want to provide more detailed information on a few of the
better studies in the NRP report, studies that do demonstrate that vocab-
ulary training works. I will be focusing on studies that were directed to an
analysis of teaching methods and on the amount of exposure to the words
(repetition and duration).
The most in-depth series of studies was carried out by Beck and her
colleagues. Beck, Perfetti, and McKeown (1982) studied 56 fourth graders,
mainly African-American students (70 percent) from lower-income fami-
lies. They were taught 104 new words taken from the Ginn 720 series for
fourth grade. These words were quite difficult and covered 12 semantic
categories. Included were words like philanthropist, filch, famished, scruti-
nize, diligent, meander, banter, gregarious, audible, introvert, compromise, and
exotic.
One classroom received the training and a control classroom did not.
The training group was taught 104 words at the rate of 8 to 10 words in
daily 30-minute sessions, over a period of 12 weeks (30 hours total). This
averages to 17 minutes per word. After this, they continued learning 43 of
these words for an additional 18 days, spread out over 6 weeks (45 hours
total). This averages to 27 minutes per word. At testing, knowledge of the
target words was contrasted with similar words that were not taught
(control words) and also contrasted with the control group, which partici-
pated in their usual language-arts program. Training involved a mix of
approaches, including learning definitions, word associations (synonyms),
and creating sentences and contexts for the target words.
Posttest scores on a multiple-choice (in-house) receptive vocabulary
test showed that the 43 words (longer exposure) were remembered with
86 percent accuracy, and the 61 words seen less often, with 78 percent
accuracy (values significantly above chance). There was a modest transfer
effect to the words in the stories not specifically taught, because the
trained group scored 10 percent higher than the control group (41 versus
30 percent) and 10 percent higher than their own pretest score. The
training also transferred to a standardized test (Iowa Test of Basic Skills).
The experimental group gained 10 percentile points on the vocabulary
and reading subtests from pretest to posttest, a gain not shown by the
control group. These transfer effects were significant.
 230

These are interesting results for several reasons. First, it took 30

Chapter 8

hours (over 18 weeks) to bring the class up to an 85 percent accuracy rate

on a measure of receptive vocabulary for the 43 overlearned words (27
minutes per word). Children also did well on the less trained words (close
to 80 percent correct), but even this took 17 minutes per word. Mean-
while, with no instruction, these children will have picked up over 1,000
words during this period. As for whether the 104 words would enter
productive vocabulary (be used), we have no idea, because this was not
monitored.
Second, the transfer effect is important, because it suggests that the
methods and strategies used in training to help children comprehend
new words transferred to nontarget words in the stories, and to words
on a standardized test the children had not seen before. This is valuable
information.
The children also participated in a semantic-decision task (‘‘Which
word is a person?’’) and in a sentence-verification task (true-false) in which
children could score 50 percent correct by chance. Unfortunately, failure
to correct for guessing invalidates these data, and none of the statistical
comparisons (experimental versus control groups) will be valid.
The impact of training on story recall was also measured. Three
stories were created. One used the most exposed target words, one used
the less exposed words, and one used control words (not taught). Nobody
did well. The group that got the training performed exactly like the con-
trols on the stories with the less exposed or untaught words, and did only
slightly better on the story with the most practiced words (33 percent
versus 21 percent correct). Peppering stories with 61 or 43 overlearned
words did not make them any more memorable than the stories with
novel, unfamiliar words. This suggests that the new vocabulary had not
been integrated into productive vocabulary (vocabulary use), and/or that
the stories themselves were hard to recall for the reasons discussed earlier
in the chapter.
Beck’s group (McKeown et al. 1985) continued to refine their teach-
ing approach by examining timing and forms of instruction, again study-
ing fourth graders. They contrasted two types of training. One type was
described as ‘‘rich,’’ involving elaborative and varied techniques. The sec-
ond was described as ‘‘extended’’ because it involved activities and home-
 231

work outside the classroom. One classroom was designated ‘‘rich only,’’

Vocabulary and Comprehension Instruction

one classroom ‘‘rich þ extended,’’ a third classroom used a traditional
approach consisting of dictionary definitions and generating synonyms,
and a fourth classroom had no training.
There were 24 target words taken from the previous list. They used
two exposure levels: 12 repetitions for one set of words and 4 for another.
One group learned the 24 words in twelve 30-minute lessons (6 hours),
about 15 minutes per word. The other group learned the 24 words in four
30-minute lessons (2 hours), about 5 minutes per word. The training took
3 weeks, in contrast to the previous study, which lasted from October to
March.
The measures were those used previously. Test scores for high-
exposure words for all training groups (including the traditional classroom)
were close to ceiling (95 percent correct) on the vocabulary-knowledge
test. The absence of any effect of the different types of training was also
found for the less exposed words, though scores were lower (84 percent
correct). Obviously, scores are well above chance here, and they were also
above chance on a test of semantics.
Group differences were mainly found on tests of productive vocabu-
lary where children had to interpret and describe contexts related to the
target words. The two ‘‘rich’’ groups did not differ, both scoring 75 per-
cent correct on questions related to the most exposed words, and 50
percent correct on the less exposed words. The ‘‘traditional’’ group had
significantly lower scores (60 percent and 40 percent).
Finally, the children were asked to recall stories written with the
target words. Again, performance was poor. The two ‘‘rich’’ groups had
higher scores on stories with the high-exposure words, but not much
higher—averaging only 26 percent correct. Again, this result is curious. It
appears that story memory poses a distinct problem in and of itself, and
this may be relatively independent of vocabulary. It should also be noted
that there was no advantage for the ‘‘extended’’ learning experience on
any test.
There are some interesting comparisons between the two studies.
In the second study, children scored higher on the vocabulary test after
fewer learning sessions (fewer minutes per word). There are two possible
explanations for this effect. One was a better teaching method. The other
 232

reason relates to Lloyd’s (1992) discovery that fast and intense training is
Chapter 8

more effective for young children. Because McKeown et al. changed three
variables that relate to timing—the number of words taught (24 versus
104), the exposure time for each word (shorter), and the length of the
training (shorter)—there is more than one explanation for why all three
methods (even the traditional method) produced almost perfect receptive-
vocabulary scores (95 percent correct) for high-exposure words, and nearly
as high (85 percent correct) for the low-exposure words. Was this due to
learning fewer words, the compressed learning time, or both? The impact
of the ‘‘rich’’ teaching approach versus the traditional approach appeared
only on measures of productive vocabulary, which suggests that the teach-
ing method matters most in enhancing recall memory.
These studies suggest that any method that calls attention to meaning
and engages the student cognitively, produces gains in receptive vocabu-
lary. Exposure duration and intensity of learning impact how well and
how much is remembered. This general hypothesis may or may not be
accurate, but it is certainly worth further study.
Jenkins, Matlock, and Slocum (1989) also found a ‘‘frequency’’ and
‘‘method’’ effect in a short-term study on vocabulary instruction. There
were 135 fifth-grade children in the study, primarily middle class, with ex-
ceptionally high vocabulary and reading-comprehension scores on stan-
dardized tests. The goal was to teach 45 target words. The children were
taught with either of two methods. In one, word meanings were taught
directly, and in the other, meaning was derived from context. In addition,
words were seen once, three times, or six times. Training took place over
9, 11, and 20 days depending on the amount of exposure. Children spent
15 minutes per day learning these words, about 5 hours for the long-
exposure group (6.5 minutes per word).
‘‘Training for meaning’’ included memorizing definitions, using the
target words in a sentence, and substituting synonyms for the target words.
‘‘Context training’’ involved applying a sequence of strategies: (1) substi-
tute a word or expression for an unknown word, (2) check other context
clues to support this choice, (3) determine whether the substitution fits all
context clues and if not, (4) revise the word and start again.
Knowledge of the target words was tested for both productive and
receptive vocabulary. The context-learning group did not do well on
any of the tests at any exposure condition (low, medium, or high). They
 233

scored about 1 correct for finding synonyms for isolated words, and were

Vocabulary and Comprehension Instruction

only marginally better for finding synonyms for words in context (3.25
words across all conditions). The context group did not score significantly
above chance on receptive-vocabulary tests, so the authors’ statistical anal-
yses of these data will be invalid.
The group taught the meaning-based method did much better, and
their performance was strongly affected by the number of exposures to the
word. In the synonym test, they scored three times higher after six expo-
sures than after one, whether words appeared in isolation or in context.
And they scored twice as high after six exposures on easy and difficult
receptive-vocabulary tests (85 percent and 75 percent correct). The high-
exposure group scored well above chance throughout.

Summing Up
The results from these well-executed studies are very consistent. Fre-
quency of exposure to new words makes a big difference only if students
have some guidance (instruction) and gain a deeper understanding of what
these words mean. Deriving meaning from context analysis is not effective.
It appears to be too abstract, even for bright fourth graders. This finding
contradicts the major tenet of whole language, that children can easily
derive meaning from the pictures and context clues and do this while they
teach themselves to read.
Vocabulary can be taught, and there is solid evidence on which
teaching methods have value. We also see that repetition is critical, and
that short-term (intense) teaching works better than lessons spread out
over a long period of time. As to whether teaching new vocabulary is
‘‘worth it,’’ the fact that these abstract words would be unlikely to be
acquired spontaneously suggests that this is a good idea. Using the right
method and approach, the cost is only around 5 minutes per word. Stahl
and Fairbanks pointed out that learning just 300 words a year will increase
vocabulary size about 10 percent. This is around two new words per day
of classroom days (175 days). However, there is considerable debate about
the effectiveness of teaching isolated words rather than learning new vo-
cabulary in the context of general comprehension training. As we will see
in the next section, a good comprehension program dramatically enhances
vocabulary, even when this is not a specific feature of the lessons.
 234

Teaching Reading Comprehension

Chapter 8

There is an implicit assumption behind much of the research on reading

comprehension that it is different in kind from oral comprehension. This
can be seen in the absence of any consideration of the connection between
them in almost every study on this topic. It is evident in the NRP com-
prehension panel’s opening remarks. They cite Markman (1978) as a pri-
mary catalyst for interest in doing research on reading comprehension.
The tenor of Markman’s remarks is that certain comprehension skills are
specific to reading, noting, for example, that readers often fail to detect
logical or semantic inconsistencies in text. The NRP panel paraphrased
some of her ideas:

Reading comprehension was seen not as a passive receptive process, but as an

active one that engaged the reader. Reading came to be seen as intentional
thinking during which meaning is constructed through interactions between
text and reader. . . .
Reading comprehension was seen as the construction of the meaning of a
written text through a reciprocal interchange of ideas between the reader and
the message in a particular text. (p. 4-39)

It is not clear whether this is intended to mean that before Markman

had these insights, people believed that reading comprehension was ‘‘pas-
sive’’ and that readers ‘‘failed to contruct meaning,’’ or that Markman
believes that reading comprehension is unique. But it is obviously the case
that listeners are equally likely to fail to ‘‘detect logical or semantic incon-
sistencies’’ in speakers’ utterances, as well as the fact that listeners always
engage in active interpretation and reciprocal exchanges of ideas during a
conversation.
What then is special about reading comprehension? Why does it need to
be taught? We never get an answer to this question.
The numerous attempts to discover ways to teach reading compre-
hension have spawned a variety of methods, but there is little substantive
work in this field. The NRP classified 16 different approaches, and even
this was difficult. Various strategies have been proposed that involve a
wide range of different techniques for targeting certain modes of thought,
such as predicting what will happen, noticing contradictions, making
inferences, asking specific types of questions, and so forth.
 235

Because these methods are for the purpose of impacting reading com-

Vocabulary and Comprehension Instruction

prehension, another key variable comes into play that gets short shrift
in this research. Poor comprehenders are the subject population in many
of these studies. As noted earlier, children are poor comprehenders for
different reasons. Poor decoding skills influence reading comprehension.
Slow but accurate readers also have reading-comprehension problems.
There is a striking relationship between reading fluency and measures of
oral syntax (Willows and Ryan 1986). Rayner (1998) cites 40 studies on
the connection between eye-movement patterns, reading speed, and syn-
tax. Children with low verbal skills are also poor comprehenders. Without
proper control over these variables, a research study has no validity. After
reviewing much of the literature, I found that this is often the case.
Furthermore, when you teach ‘‘reading-comprehension’’ skills to chil-
dren with good or excellent listening comprehension but who score badly
on reading-comprehension tests, what exactly are you teaching? How is
the teacher supposed to know where the child is in the reading process if
listening comprehension, decoding accuracy, and reading speed are never
measured?

A Meta-analysis
Apart from these concerns, there are problems with how reading com-
prehension is measured in these studies. Unlike the research on vocabu-
lary instruction where specific words are taught and tested, training in
reading comprehension must generalize to other text to prove the validity
of the method. The results of a large review show that it does not. Rosen-
shine, Meister, and Chapman (1996) did a meta-analysis on 26 studies that
dealt with ‘‘questioning’’ types of instruction. As they put it, ‘‘Teaching
students to ask questions may help them become sensitive to important
points in a text and thus monitor the state of their reading comprehen-
sion’’ (p. 183).
The method had to include a large proportion of time spent generat-
ing questions to help students understand a passage for the method to be
included in their analysis. Also included were studies using ‘‘reciprocal
teaching,’’ in which both teacher and students collaborate to interpret a
passage. Rosenshine and colleagues excluded all studies where the children
were tested on the same passage they had been trained on. The overall
result was an effect size of .86 for experimenter-generated tests—test
 236

passages that had a similar structure to the passage on which the children
Chapter 8

were trained. However, this training did not generalize, and there was
little transfer to standardized measures of comprehension (ES ¼ .36).
The studies were broken down further into five different types. One
was ‘‘signal words’’ in which the student is prompted by words like who,
what, where, when, why, and how. The second type involved ‘‘generic
questions.’’ Here the student is trained to ask a variety of questions, such
as how two things are alike or different, what the main idea is, or how
events or actions are related to one another. The third type was ‘‘main
idea only,’’ in which children find the main idea and then ask questions
about it. ‘‘Question types’’ comprised the fourth category. Here students
are first directed to find specific information, then to relate two or more
pieces of information, and finally to answer questions where information
must be inferred or deduced by logic and background knowledge. (The
NAEP tests are based on this approach.) The fifth type encompassed
questions about story grammar like those reviewed earlier: ‘‘Who is the
main character in this story?’’
On the experimenter-designed tests, the ‘‘generic-questions’’ approach
generated the largest effect sizes, followed by the ‘‘signal-words’’ tech-
nique. Finding the ‘‘main idea’’ did not fare well. Other types included too
few studies to make effect sizes meaningful. As for performance on stan-
dardized tests, the ‘‘signal-words’’ group produced an effect size of .36,
and the ‘‘question-type’’ group an effect size of zero. Even more interest-
ing, Rosenshine, Meister, and Chapman found no evidence for the impact
of length of training. Paradoxically, studies with positive results had fewer
sessions overall (4 to 25) than those with nonsignificant results (8 to 50).
This fact alone calls most of this research into question. With a good
method, learning time ought to translate into better learning, not worse.
The authors commented on the contrast between experimenter-
designed tests and standardized tests. Experimenter-designed tests were
more highly structured, with a clear ‘‘main idea’’ and obvious sup-
porting detail. Standardized test passages were more typical of nor-
mal text, without such a clear and obvious structure. It should also be
noted that standardized tests are normed and control for age, whereas
experimenter-designed tests are not. Experimenter-designed tests usually
employ multiple-choice questions and are subject to guessing. As we have
 237

seen, researchers in this field generally do not correct for guessing. In

Vocabulary and Comprehension Instruction

view of these facts, much of this research is likely to be meaningless.

The NRP Report

The NRP group located 203 studies that survived their final screening
protocol. However, because these 203 studies included over 16 different
methods, with few, if any, studies carried out in the same way (same
training, same duration, same tests), the panel felt that a meta-analysis
would be out of the question. Instead, they categorized the methods and
made some observations. This was not a productive exercise, and their
analysis was contradictory and uncritical to the point of being biased.
Of the 16 types, the panel reported that only 7 had a ‘‘firm scientific
basis for concluding that they improve comprehension in normal readers’’
(National Reading Panel, 2000, 4-42). (The report did not reveal what
this basis was.) Six of the types included comprehension monitoring, coopera-
tive learning, graphic organizers, question answering, question generation, and
story structure. Several of these approaches can also combine to form a
seventh, a multiple strategy approach.
It should be noted that the NRP screening protocol did not include
important methodological considerations that would bias the outcome
of a study, such as situations where (1) the subject pool consisted of
poor readers whose decoding skills were ignored while they were taught
to ‘‘comprehend’’ what they could not read, (2) the tests measuring
gains used the same passages on which the children were trained, (3)
experimenter-designed multiple-choice tests were utilized and the data
were not corrected for guessing, and (4) no measure existed to ensure the
method was implemented as described. Comprehension training programs
are very complex, often taking pages to describe. Documentation is needed
(in the form of classroom observations or video) to show that the teacher
actually taught the method as intended.
Aside from these concerns, the NRP authors provided no evidence on
the generality of the studies they believed were valid. There were 96
studies in the 7 categories, yet only 20 percent of these studies were cited
as using standardized tests. The NRP did not provide the data from these
tests. Comprehension instruction must impact reading comprehension for
a variety of texts, not just for one unique text.
 238

Instead, the panel accepted the evidence from the experimenter-

Chapter 8

designed tests as valid. In their summary of the methods under review,

they came out strongly in favor of the multiple strategy approach, particu-
larly one that included a reciprocal teaching method. The main support for
this conclusion came from a meta-analysis carried out by Rosenshine and
Meister (1994) on 16 reciprocal-teaching studies. They reported an effect
size of .88 on experimenter-designed tests, but only .32 for when stan-
dardized tests were used (9 of the 16 studies). In other words, this method
did not generalize to other texts.
It is disconcerting for the NRP authors to claim that the 203 studies
remaining in their data pool could not be submitted to a meta-analysis,
and then to cite a meta-analysis carried out by someone else on a fraction
of these studies as the sole support for their evaluations of the methods.
This is especially problematic in view of the fact that a subsequent meta-
analysis by the same group (Rosenshine, Meister, and Chapman 1996)—
which the NRP did not cite—compared reciprocal teaching to other
instructional methods, and found it had no advantage, either on standard-
ized tests (ES ¼ .23 versus ES ¼ .27) or on experimenter-designed tests
(ES ¼ .88 versus ES ¼ .70). Effect sizes were not significantly different in
either case.3
It is of some interest to provide evidence on one of the better (meth-
odologically sound) studies using reciprocal-teaching techniques. Johnson-
Glenberg (2000) compared two types of reading-comprehension training
with each other and with an untutored control group. Children in the
third through fifth grades were trained in small groups for 10 weeks with
either a reciprocal-teaching (RT) approach or a visualizing/verbalizing
(V/V ) program developed by Bell (1986). They received an average of 28
half-hour sessions (14 hours total). The children had good to high decod-
ing skills but poor comprehension skills.
The reciprocal-teaching approach was adapted from Palincsar and
Brown (1984) and consisted of four strategies: summarization, clarifica-
tion, prediction, and question generation. The lessons began with exer-

3. I did not include cases where there was one study only per method in these
effect sizes.
 239

cises in summarizing very short passages, and then gradually expanded

Vocabulary and Comprehension Instruction

so all strategies were engaged using longer text passages. In reciprocal
teaching, the teacher and the children have equal input in the process.
The visualizing method trains imagery by having students describe ver-
bally in great detail how they ‘‘see’’ certain characters, settings, objects,
and events, in their mind’s eye.
The children were tested on 12 different measures prior to and after
training, including four standardized tests. Interrater reliability was com-
puted for all in-house tests and was consistently high. The investigators
also computed effect sizes for pretraining and posttraining scores. Chil-
dren improved over time on most tests, with the exception of the control
group on the WRAT word recognition (standard scores), a measure of
reading speed, explicit questions on story recall, and open-story recall.
WISC vocabulary scores were controlled (covaried) in the group
comparisons, and this significantly dampened the differences between
groups. Nevertheless, the RT groups outscored controls on six measures:
WRAT word recognition, question generation, explicit-questions story
recall, implicit-questions story recall, listening story recall, and visual
questions. I computed effect sizes for these specific comparisons, and these
were, respectively, ES ¼ :34; 1:0; :78; :65; :35, and 1.55. The V/V groups
outscored the controls on three measures: question generation, implicit-
questions story recall, and visual questions (ES ¼ :54; :76, and .37). Inter-
estingly, the V/V groups did not score higher than the other two groups
on the remaining visual measures: the paper-folding test and a test of
visual imagery. The RT groups also outscored the V/V groups on ques-
tion generation, explicit story recall, and listening story recall (ES ¼
:51; :66; :53). However, despite the strong effects of training on the mainly
in-house tests reported above, no group differences were found on the
only standardized measure of reading comprehension (Gates-MacGinitie)
or on the Detroit Test of Learning Aptitude.
This was a fairly rigorous test of the RT approach, and while it had
an advantage over the V/V training, differences were not that great. It is
important to note that Johnson-Glenberg reported highly significant
gain scores for the trained groups on all 12 measures before vocabulary
was statistically controlled. These differences were seriously curtailed after
these controls were introduced. Here is another wild card that is certain
 240

to influence outcomes in these studies. Vocabulary is rarely controlled in

Chapter 8

most studies on reading comprehension.

There is some evidence here to support the NRP comprehension
panel’s recommendation in favor of RT, but the evidence does not sup-
port their general conclusions, which they furnished in a summary to this
section. They provided a hearty endorsement of teaching comprehension
strategies in general: ‘‘The past 2 decades of research appear to support
the enthusiastic advocacy of instruction of reading strategies’’ (p. 4-46).
This positive outcome was attributed to methods in which teachers

demonstrate, explain, model, and implement interaction with students in

teaching them how to comprehend a text. In studies involving even a few
hours of preparation, instructors taught students who were poor readers but
adequate decoders to apply various strategies to expository texts in reading
groups, with a teacher demonstrating, guiding or modeling the strategies, and
with teacher scaffolding. (p. 4-47)

The effectiveness of this reciprocal interaction between teacher and

students was alleged to be consistent with ‘‘socially mediated learning
theory.’’ No specific study or evidence was referenced in support of these
conclusions. The only source cited was Rosenshine, Meister, and Chap-
man 1996, which found nothing of the sort.
Following this endorsement, the next and final section of the panel’s
report was highly critical of the research on reading comprehension, and
largely contradicted everything that came before. Lysynchuk et al. 1989
was the main source for this critique. This publication screened 37 studies
on comprehension instruction and coded 24 variables of methodological
relevance. There was a substantial fail rate on 9 of them. These were, in
order from the very worst to the least: used the correct unit of analysis for
statistics (83 percent fail); monitored the strategies the students actually
used, as opposed to the ones they were supposed to use (73 percent
fail); provided information on training time (70 percent fail); and checked
whether students were doing what they were supposed to do (63 percent
fail). There were several variables with a 35 percent fail rate: random as-
signment of children to groups, exposing the controls to the same mate-
rials as the experimental subjects, ceiling and floor effects on the tests, and
lack of interrater reliability checks on in-house tests.
 241

Overall, this is a very gloomy analysis from all points of view. I do not

Vocabulary and Comprehension Instruction

regard this research, taken as a whole, as providing any proof whatsoever
of the validity of comprehension instruction. Instead, I want to examine
the only two studies I could locate that are both methodologically rigorous
and that illustrate the true power of good comprehension training. Both
methods involved extensive efforts over long periods of time and produced
solid gains on standardized tests.

Reading-Comprehension Programs That Work

Brown et al. (1996) employed a method known as transactional strategies
instruction (TSI), a form of reciprocal teaching in which everyone coop-
erates to ask and answer questions. This process is teacher directed. As
they put it, ‘‘The short-term goal of TSI is the joint construction of
reasonable interpretations by group members as they apply strategies to
texts’’ (p. 19).
Students are trained to ‘‘predict upcoming events, alter expectations
as text unfolds, generate questions and interpretations while reading, vi-
sualize ideas, summarize periodically, and attend selectively to the most
important information’’ (p. 20). Students think aloud while they are learn-
ing to do all this.
The authors report that it takes several years of training and experi-
ence before teachers can implement this method effectively.
Five reading groups using TSI were compared to five groups taught
with an eclectic approach (specific to teacher) over the course of one
school year. Students were second graders, reading below grade level.
They were matched on Stanford Achievement Test scores and randomly
assigned to ten groups of six children each (N ¼ 60). I was unable to
locate information on the number and duration of lessons per week.
A number of in-house tests were designed to measure outcomes.
However, I will focus only on the standardized tests. After training, the
TSI children were significantly advanced on the SAT comprehension
subtest and the ‘‘word-skills’’ test (word recognition). I computed effect
sizes and found the ES value was .67 for word-study skills and .89 for
comprehension, compared to controls. These are substantial effects and
show that this program generalizes beyond the texts used in training.
However, it cannot be known how much is due to method and how much
to exposure, because Brown and colleagues did not report how much time
 242

was devoted to teaching this program. Also, in the case of children with
Chapter 8

poor reading skills, there is a strong possibility that a fair proportion of

these children had lower-than-average verbal IQ scores, and verbal IQ
was not controlled.
The last study (Block 1993) was the only study I was able to locate
in which the method of instruction had great generality for improving
critical-thinking skills beyond the immediate task. It was also the only
study that solved nearly all the methodological problems, and the only
study where there was a clear understanding that reading comprehension
is a consequence of oral comprehension. The paramount goal of transfer
and generalization was set out in the introduction. I paraphrase:

1. If instruction helps students use strategic processes consistently and

reflectively as they read, they should comprehend and use more informa-
tion in their daily lives.
2. Students who have trouble comprehending may need a new type of
instruction, involving new strategies and thinking/reading competencies.
3. Students who are confused about important concepts, inferences, and
relationships in a text may never have encountered the inductive or de-
ductive reasoning patterns the author used to create the text. If these
thinking strategies are explained prior to reading, thinking repertoires
should expand.
4. Students may be more able to generalize this new knowledge to solve
decoding and comprehension problems on their own.

Block uncovered eight strategies that are important and need to be taught.
These reflect such things as basic cognitive operations, analytic thinking,
decision strategies, problem solving, metacognitive analysis (awareness of
one’s current state of knowledge), creative thinking, plus skills for working
in groups, and skills for working effectively alone. Sixteen lessons were
designed to teach these strategic skills, and these lessons were carefully
field tested across all grade levels on a large number of children prior to
this study.
The lessons were structured so that one critical-thinking technique
was introduced per lesson, plus strategies for improving comprehension.
This constituted part 1 of the lesson. In part 2, children selected reading
material to apply this new knowledge. They could choose from a large
 243

selection of books among a variety of genres: fiction, nonfiction, poetry,

Vocabulary and Comprehension Instruction

autobiographies, folk literature, periodicals, and science fiction. Children
were given simple and useful handouts for how to attack an unknown
word, and to increase their understanding of what they read. They were
encouraged to ask questions and to learn to recognize what a good (pro-
ductive) question is. There is no space here to review this program in
depth. The description of the method was detailed and precise.
There were 16 classrooms in the study ranging from grades 2 through
6 (352 children). Half the classrooms received the special lessons, and
half acted as controls. A trained research assistant was present in every
classroom during the language-arts period and worked together with the
teacher. In the experimental classrooms, these assistants helped with the
prescribed lessons, and in the control classrooms they helped the teacher
present lessons that did not include the strategy instructions. The lessons
continued throughout one school year, 3 hours per week for 32 weeks, for
a total of 96 hours.
Treatment fidelity was measured by videotaping the last lesson in
each classroom. Tapes were coded ‘‘blind’’ by raters who did not know
which classroom got the training. There were ten significant and impor-
tant differences in the children’s verbal and physical behavior between the
two types of classrooms. The trained children had more highly skilled
interactions (verbal exchanges) and far less nonproductive behavior (near
absence of interruptions, random remarks, noise, use of jargon or slang,
and off-task behaviors/boredom), which is remarkable in and of itself!
On the reading-comprehension subtest of the Iowa Test of Basic
Skills, the trained group scored significantly higher than the control group
( p < :001), and they improved in vocabulary as well ( p < :001), although
this was not a focus of the training. There was little impact on grammar.
This was expected, because grammar was not taught. I computed effect
sizes for these results. The effect sizes were 2.24 for comprehension, 2.62
for vocabulary, .32 for grammar, and 1.34 for Total Reading. These fig-
ures are very impressive, one could even say extraordinary, in view of the
typical results in this field.
Block used a number of additional tests to measure generalization
effects. Children from the trained group were much more likely than
control children to represent analytic thought in their writing ( p < :0001),
and to report that what they had learned at school was useful outside
 244

school (92 percent reporting ‘‘useful,’’ versus zero for the controls). The
Chapter 8

trained children had significantly higher self-esteem, though the groups

had not differed at the outset of the study. On tests of problem solving,
the trained groups used a greater number of critical-thinking strategies,
were more precise in their statements about these strategies, and offered a
larger number of alternative solutions to the various problems. The results
in these comparisons were highly significant ( p < :0001 in all cases).
Block’s outstanding work was published in 1993 and, so far, seems
to have had little impact on the research community. No other method
comes close to achieving the goals of what a good comprehension pro-
gram is supposed to do. Not only does it generalize to standardized tests
(with huge effect sizes), but to all forms of analytic thought, academic
performance, and classroom behavior, and it works across a broad age
range, requiring little teacher training. Furthermore, it produces a huge
impact on vocabulary, even when measured by standardized tests. This is
quite beyond anything achieved in the vocabulary-training studies cited in
the previous section.
Overall, the research is this field is extremely problematic. There
are certainly good instructional methods, but for the most part, they
have been ignored by the research community. Instead, educators and
reseachers seem more interested in designing their own unique approach
than in building on one another’s work. As a result, instructional methods
are highly idiosyncratic and difficult to classify.

General Conclusions
The past few chapters have shown that there are some remarkable instruc-
tional methods for almost every type of reading skill, methods that pro-
duce close to 100 percent success for every child. This is the good news.
Wouldn’t it be exciting if everyone knew about these methods, especially
teachers? Unfortunately, due to the enormous volume of published and
unpublished research, these excellent methods are very hard to locate.
It is difficult to be neutral about the fact that there is such a vast
quantity of poor research (‘‘junk science’’) clogging the reading-research
databases. In the real world of science, the most rigorous and most im-
portant studies tend to find their way to the top journals. Here, quite the
opposite is true. The flagship journals, of which there are only two, are
just as likely to publish research that is methodologically flawed as not.
 245

(Among a very large pile of papers on comprehension I had to set aside

Vocabulary and Comprehension Instruction

because of fundamental problems with methodology, 80 percent were
published in the top two journals.) This creates the impression that either
the editors and reviewers know next to nothing about research methodol-
ogy, or that they are overly influenced by who is ‘‘important’’ in the field
and who is not. As a result, talented unknowns must seek publication in
journals near the bottom of the food chain. This makes the problem dou-
bly difficult, because readers are just as likely to find high-quality work
hidden amidst the dross almost anywhere they look. Readers should not
need a Ph.D. in statistics and research methodology because they are
forced to analyze the research design, redo the statistics, and critically
assess authors’ conclusions from the data. That is the reviewers’ job and
the responsibility of journal editors.
The combination of quantity (too many articles without any peer
review on the databases, and too many journals) plus the lack of any
mechanism for determining quality (no way to rely on journals; no way to
rely on databases), means that the scientific study of reading is basically
going nowhere and has been going nowhere for over 30 years. One might
even argue that it is going backward, in view of the fact that the majority
of researchers in the field still seem to believe that children need an
extensive phoneme-awareness training program in addition to an effective
phonics program. That belief is so powerful that the NRP ignored the
evidence in its own database and in its analysis of what children need to be
taught.
When we come to spelling, a topic not even considered by the NRP,
the situation is far worse, as we will see in the following chapter.
 9
H O W D O E S A N YO N E L E A R N T O S P E L L ?

In a deeply opaque writing system like the English alphabet code, most
phonemes have multiple spellings. Only eight are reliable, and another
six, relatively so. But even here, half of the ‘‘predictable’’ phonemes have
single- and double-letter spellings: /b/, /d/, /l/, /p/, /t/, /g/, /m/, and
/n/, as in cab/ebb, lad/ladder, curl/hill, tap/tapped, bat/batter, fog/egg, ham/
hammer, win/winning). Some phonemes can be spelled nine or ten differ-
ent ways. Because spelling requires recall memory and reading requires
only recognition memory, spelling is much more difficult than reading. It
is easy to read a word like hill, but quite another matter to remember
whether to double the l when you spell it (boil, ball, deal, will, pal, pull, bail,
doll ).
As we saw in chapter 3, there have been four attempts, historically, to
systematize the English spelling code, and I was only able to locate three
studies where these systems were tested empirically. Typically, spelling
instruction consists of lists of random words that go up on the board on
Monday for the spelling test on Friday. The yardstick of spelling difficulty
is syllable length, as well as ‘‘regular’’ versus ‘‘rare’’ spellings, but seldom
emphasizes the structural elements of the code—such as the spelling pat-
terns linked to phoneme position within words.
Research on classroom spelling programs is so rare that there was no
section on spelling in the NRP report. Graham (2000) managed to locate
only 60 studies on spelling by scouring the journals back to the 1920s.
These studies compared the two dominant approaches: ‘‘natural’’ learning
(self-taught spelling) versus ‘‘traditional’’ instruction, consisting mainly of
random word lists. Most of these studies were methodologically flawed,
but the general message was that children cannot teach themselves to spell
simply by reading or through creative writing, and that ‘‘traditional’’
 248

programs were superior. As for what these programs contained, little was
Chapter 9

said. Graham noted that, apart from rote memorization, spelling lessons
were made more enjoyable by ‘‘including student choice in the selection
of spelling words and methods of study, guided discovery in learning the
patterns underlying the spelling of words, opportunities to work with
peers, and use of games’’ (p. 245).
How ‘‘student choice’’ and ‘‘guided discovery’’ are supposed to work
in the absence of any knowledge on this subject was not explained.
The fact that many children do learn to spell is, therefore, a bit of a
mystery, and how children succeed at this task has been the central ques-
tion in research on spelling. This question occupies us for the next two
chapters, and there are some surprising answers.
The ‘‘how’’ question is approached differently depending on the
researcher’s background knowledge. For the most part, researchers who
study spelling have little or no understanding of how writing systems work
and no knowledge of the structure of the English spelling code. Because
of this, spelling research is based on a set of implicit assumptions. It is
assumed that it is ‘‘natural’’ for children to teach themselves to spell, and
that spelling skill proceeds in stages. It is assumed that children learn to
spell by reinventing the spelling code (invented spelling). Using this logic,
a poor speller is someone with a developmental delay or a deficit. Yet if
children are not taught something as complex as the English spelling
code in a structured and meaningful fashion, how can anyone learn it? To
someone with greater knowledge, the good speller seems unnatural and the
poor speller seems normal.
There a deeper issue here, which I call the many-word problem. Even if
the world’s best spelling program could be devised, it would never be
possible to teach the spelling of every word. Because the English spelling
system is so opaque, and only a handful of phoneme-to-grapheme corre-
spondences are consistently reliable, the only way it can be mastered is
through its probability structure, the reoccurring regularities in spelling
patterns as outlined in chapter 3. For the brain to set up this structure, it
needs exposure to thousands of examples of correctly spelled words. A
good spelling program can jump-start this process by grouping words with
these redundant patterns, but it will never succeed in teaching every word.
There will always be words that a fluent English reader is unable to spell,
words that have to be looked up in a dictionary.
 249

Nevertheless, some people can master most of this structure unaided,

How Does Anyone Learn to Spell?

while others cannot. Do good spellers have special visual or linguistic skills
that make this possible? These kinds of questions have given rise to three
major lines of inquiry: the study of ‘‘spelling stages’’; an exploration of
how linguistic knowledge might affect spelling skill; and the study of
structural redundancies in the spelling code, which is intimately tied to
computer models of reading. But before we turn to these issues, we need
to look at more fundamental skills that might underpin an aptitude for
spelling.

Spelling Predictors
There are three important factors that will impact spelling scores regard-
less of method.

IQ IQ is a consistent contributor to spelling skill. IQ has an impact on

spelling ability from the outset, and this connection increases with age.
Kindergartners’ aptitude for producing modestly accurate ‘‘phonetic spell-
ings’’ is correlated with the WISC Full Scale IQ at r ¼ :51 (Mann 1993).
Bond and Dykstra (1967) found the identical value in their monumental
study on over 9,000 first-graders, Full Scale IQ correlating with spelling at
r ¼ :51. Muter and Snowling (1997) found this connection was a long-
lasting one. Full Scale IQ measured at age 4 and 5 predicted scores on a
standardized spelling test at age 9 (r ¼ :42). In a large-scale study by Sha-
nahan (1984), scores on the Stanford IQ Vocabulary subtest (receptive
vocabulary) correlated with spelling at second grade (.45), and at fifth
grade (.62). These are remarkably consistent findings, and show that IQ
predicts spelling skill across the age range, accounting for about 25 per-
cent of the variance.

Sex Girls outnumber boys in high-reader groups and are relatively

scarce in low-reader groups (see my book Language Development and
Learning to Read ). Girls have an even greater advantage in spelling. First-
grade girls were superior spellers in Bond and Dykstra’s study in every
methods comparison. Allred (1990) tested over 3,000 children in grades 1
through 6 from various regions across the United States. Girls did better
on both spelling-dictation and spelling-recognition tests, and this was
highly significant ( p < :001) at every grade. And while spelling skill varied
 250

significantly by region, the sex difference did not. (No region  sex inter-
Chapter 9

actions were found.)

The Spelling-Reading Connection Even with a transparent writing system,

the correlation between reading and spelling will not be perfect, because
skill in reading and spelling depends on different types of memory. This
connection will be weaker with an opaque writing system, since spelling
is much more difficult than reading. Bond and Dykstra reported a strong
correlation between spelling and word recognition (r ¼ :63) and reading
comprehension (r ¼ :66). Shanahan (1984) tested over 250 second graders
and 250 fifth graders on the Stanford Diagnostic Reading Tests and the
Gates-MacGinitie reading-comprehension test. Spelling was measured by
a well-designed in-house spelling test (reliabilities around .90). Spelling
was correlated with basic decoding skills at r ¼ :66 (second grade) and
r ¼ :60 (fifth grade), and with reading comprehension at r ¼ :58 (second
grade) and r ¼ :56 (fifth grade). Again, the correlational values in these
studies are in remarkable agreement. However, had IQ and sex been con-
trolled, these values would be lower.
Taking these uncontrolled factors into account, reading and spelling
are unlikely to share more than about 25 percent common variance. This
is too low to support the notion that reading and spelling skill are tightly
linked, but high enough to show a consistent connection. Nevertheless,
this leaves 75 percent of the variance in spelling scores unaccounted for.
There are a number of options for how and why reading and spelling part
company. People may have different aptitudes in recognition and recall
memory. Or they may exhibit different facilities for reading and spelling as
a consequence of instruction (or noninstruction), causing them to focus on
the wrong perceptual units of the writing system. This would lead to two
separate strategies, one for reading and one for spelling.
Bearing these important facts in mind, we turn next to the research on
spelling.

Stage Models under the Microscope

The dominant model in spelling research is the ‘‘stage model’’—the
theory that spelling skills progress in developmental stages. The fact that
the majority of children learn to spell is taken as evidence that people have
a special aptitude for spelling. (There are stage models for reading as well,
 251

but they do not dominate the field of reading research to nearly the same

How Does Anyone Learn to Spell?

degree.) Unfortunately, much of this work is as weak methodologically as
it is logically. Researchers consistently fail to control factors known to af-
fect spelling skill, such as age, IQ and sex. And they tend to ignore reading
and spelling instruction altogether, a point also emphasized by Ehri (1992,
330): ‘‘One important issue that has been neglected in studies of spelling
stages is the relationship between stages and instruction. The concept
of developmental stages implies that the growth patterns are relatively
impervious to specific instructional experience. However, in the case of
spelling development, the nature and extent of instruction very likely
makes a big difference to development.’’ This oversight is puzzling, be-
cause inappropriate or absent classroom instruction causes the spelling
errors that researchers use as data to construct their stages. The basic
instructional failings are these:

Teaching letter names rather than correspondences between letters and

sounds

Teaching the alphabet code backward: letter to sound only (visual ori-

entation) rather than sound to letter

Excessive reliance on sight-word memory (random-word lists)

Teaching practice that actively divorces reading and spelling instruction

from one another (taught on different days, different words, different

modes of learning)

Teaching a hodgepodge of spelling ‘‘rules’’ that get broken most of the

time

Failure to let the structure of the spelling code dictate the sequence

of instruction, because nobody knows what the structure of the spelling

code is.

To put this in context, let’s begin in the real world. Molly’s mother began
teaching her the alphabet at age 3. Molly learned to chant letter names in
alphabetical order. She learned to match letter names to letter shapes. She
was cycled through a ‘‘letter-of-the-week’’ regime in which a new mag-
netic letter appeared on the refrigerator door each Monday. By the time
Molly entered school at age 5, she and most of the children in her middle-
class neighborhood had a fair-to-good knowledge of letter shapes and
letter names. The children could name the letters they were shown, point
 252

to letters that were named, and some children (mostly girls) could write
Chapter 9

their name and a fair number of letters from memory.

Molly believed that letter names were extremely important in learning
to read, otherwise why had she spent so much time learning them? When
she arrived in kindergarten, this belief was reinforced. Her teacher spent
more time reviewing letter names and letter shapes and taught some sight
words. In first grade, Molly was ‘‘immersed in print,’’ told to use all her
‘‘cueing systems,’’ taught some ‘‘sight words,’’ and encouraged to guess
words by context clues, as well as to ‘‘sound out’’ words using ‘‘phonics
clues.’’ For Molly, as well as for most of the other children, the only
sound-out clues at their disposal were letter names. In the time set aside
for creative writing, she was encouraged to write stories using her own
‘‘invented spelling’’ system. Molly had two available strategies for spelling
words in her stories: visual memory (sight words) and letter names. By an
amazing coincidence, she used these two strategies to spell words.
In 1971, these strategies were documented for the first time when
Read published an analysis of preschool children’s invented-spelling
errors. He later discovered that the most common spelling mistakes con-
sisted of using letter names instead of phoneme-grapheme correspon-
dences (Read 1986). This type of error was more likely to be observed
with vowels (KAM for came). Read viewed this as a ‘‘psycholinguistic pro-
cess’’ in which children bring their awareness of the language to bear on
the task. Read’s article is the most frequently cited work in the research
literature on spelling stages, where his findings are taken to mean that
children spell with letter names because they are in the letter-name spelling
stage.
Note that for a period of four years, Molly received training in activ-
ities supposedly related to learning the English writing system. Yet in all
that time, she had never been taught anything she needed to know, nor
was she exposed to any useful exercises that could instill this knowledge:
(1) A writing system is a code. Reading means decoding letters into speech
sounds (sounds in your speech). Spelling means encoding speech sounds
into letters. (2) Reading and spelling are reversible processes. (3) There is
a one-to-one correspondence between many sounds and their spellings,
and a one-to-many correspondence for others. (4) There are patterns (re-
curring regularities) in the spelling code to help us remember how words
are spelled.
 253

Molly was never taught these things because neither her mother nor

How Does Anyone Learn to Spell?

her teacher knew them.
This is the backdrop against which researchers frame their theories
and carry out their studies. They seem to believe that Molly’s situation is
reasonable, because if it was not reasonable, there would be little purpose in
studying it. I can show that it is anything but reasonable by shifting the
focus a little. Molly’s mother also taught her ten number symbols and ten
number names. She taught Molly that numbers ‘‘stand for’’ quantities. She
bought workbooks with pictures of pies, pieces of string, and rows of hip-
popotamuses, and as well as floating balloons for Molly to count in order
to connect quantity with a number code. Molly learned there was a direct
correspondence between the quantity 5 (five balloons, five hippopota-
muses, five pieces of pie or string) and the number 5. This is all very good,
because a number system divorced from an understanding of quantity is
not much use.
Then Molly went to school. It was during the era of ‘‘invented arith-
metic.’’ Children were supposed to discover the principles of addition,
subtraction, multiplication, and division by being immersed in numbers.
They were encouraged to invent formulas or systems for solving numeri-
cal word problems. Researchers descended on classrooms to annotate and
classify the errors that children made while they invented arithmetic, or
rather, tried to rediscover the invention of arithmetic unaided. ‘‘Arithme-
tic stages’’ were carefully documented. Researchers were intrigued by the
discovery of a universal initial stage of ‘‘protoarithmetic,’’ in which num-
bers were ordered in rows or columns in the same increasing order as the
quantities they represented. One person called it the ‘‘protoarithmetic’’
stage, another the ‘‘serial-order’’ stage, and another the ‘‘initial combina-
torial equity’’ stage.
I may joke about reading and spelling stages (and I do not intend to
spend much time on them), but there are genuine issues here that have
been muddled together with the issue of stages. I will address these issues
as I review this work.

Stage Models of Spelling Acquisition

A Little History of Stages
Stage models in psychology came into prominence at the beginning of the
twentieth century largely through the work of Freud and Piaget. Both
 254

were trained in biology, Freud as a neurologist and Piaget as a zoologist.

Chapter 9

During their training, they learned about biological stages of develop-

ment, such as the larval stage of an insect. Later, in searching for a frame-
work for their discoveries, they borrowed the concept of stages from
biology, just as Hans Selye borrowed the concept of psychological and
physiological stress from physics. In his psychoanalytic model, Freud
claimed to document stages of psychoemotional/sexual development that,
if not negotiated appropriately, could lead to subsequent neuroses.
Piaget studied the development of logical reasoning. He discovered
that when children acquired a certain level of intellectual capacity, they
were able to think about more than one spatial dimension at a time
(length, breadth, height), or to realize that quantity is ‘‘conserved’’
through transformations in form, and that this process is reversible. The
classic experiment is on the conservation of volume—water being poured
from a short, fat container into a tall, thin one, and back again. Piaget
viewed these new capacities as cognitive developmental stages and as spe-
cific to logical reasoning about objects and object properties. He believed
that language development was more continuous than stage-like (Piaget
[1964] 1993).
Stage models have special properties (Flavell 1963). Stages follow in a
fixed sequence. They tend to come into being rather abruptly, then stabi-
lize, resembling a staircase when plotted over time. Skills acquired at ear-
lier stages are integrated with later ones, but children in a higher stage
abandon naive solutions used at earlier stages (stages do not go backward).
According to Piaget, if a child solves the problem of conservation of
quantity and reversibility, this will generalize to similar problems (a block
of clay smashed into a long thin tube), if not immediately, then certainly
with brief instruction. Nor does the child revert to being unable to solve
these problems merely through lack of opportunity.
Both Freud and Piaget viewed stages in a biological sense as intrinsic
to the organism, and brought about through interactions with the world.
These models have to do with problem solving, psychosocial in the first
case, and logical/mathematical in the second. This is a far cry from the
ability to master human inventions, especially because one of the hall-
marks of a true stage of development is that it cannot be taught!
Stage models contrast with two other models in developmental psy-
chology. One is a continuous developmental model in which a biological
 255

program unfolds in tandem with exposure to a minimally suitable envi-

How Does Anyone Learn to Spell?

ronment. A cumulative sequence of subskills emerges that combine and
continue to work together. Infant language development is a prime exam-
ple, where children acquire more and more aptitudes and sensitivities as
time goes by (see Language Development and Learning to Read ). The dif-
ference between this model and a stage model is that skills gradually
improve over time as each skill is added to the one before. Multiple sub-
skills work simultaneously in a reciprocal fashion to increase the accuracy
and fluency of speech perception and production.
The second model is learning, which is not specific to childhood
(reading and spelling can be taught at any age). New skills and insights
are acquired through discovery, demonstration, training, and repetition.
Learning is cumulative in the same way as continuous development; one
thing builds on another. But learning is not a biological program. It is
prone to setbacks and failure that arise for various reasons, such as being
exposed to misleading instruction, adopting inappropriate strategies, and,
in some cases, the lack of a critical level of certain subskills. Children or
adults learn to read and spell. They do not go through spelling stages or
developmental spelling levels.
To gain a fuller understanding of how a stage model has been abused
by researchers studying ‘‘spelling development,’’ we turn next to some of
these studies.

Research on Spelling Stages

Stage models for spelling (and reading) began to pop up in the early
1980s, spurred on by Read’s report on children’s invented spelling. These
models were based on an analysis of children’s spelling errors in their
creative writing, or errors produced by specially designed spelling tests.
The most influential models were those developed by Henderson and his
colleagues (Henderson and Beers 1980; Henderson 1982, 1992), by Frith
(1980, 1985) in the United Kingdom, and by Ehri (1986, 1989b, 1995).
Although the basic assumption behind these models is similar (chil-
dren teach themselves to spell), the way the stages are derived is quite
different. Henderson’s model is based on the surface structure of the error
patterns, what Ehri (1992) described as orthographic features. In Hender-
son’s model there are five stages: Prephonetic, Letter Name, Within
Word Pattern, Syllable Juncture, and Derivational Constancy. They
 256

reflect the child’s growing awareness of spelling detail, as well as the

Chapter 9

characteristics of the English language. Common words are short, less

common words have more than one syllable, and the least common words
have lots of syllables. Multisyllable words tend to be ‘‘morphological’’ in
structure (combining units of meaning: un-happi-ness), and this structure is
reflected in the spelling code.
Frith’s stages are more akin to global strategies than stages, and are
less specific. The child begins as a ‘‘logographic’’ (whole-word) reader and
speller, graduates to an ‘‘alphabetic’’ stage (phonetic), then to an ‘‘ortho-
graphic’’ stage, when spelling ‘‘rules’’ and conventions are mastered. In
this model, reading and spelling stages do not necessarily develop in
parallel. Dissociations appear, leading to three types of children: good
readers/good spellers, good readers/poor spellers, and poor readers/poor
spellers. The existence of the middle group led Frith to propose that some
children read by eye (visually) and spell by ear (phonetically).
Ehri (1992, 326) contrasts her model with the others as being more
oriented toward ‘‘the nature of the correspondence between written (or-
thographic) units and spoken (phonological and morphemic) units that
spellers exhibit in their invented spellings.’’ In other words, she views
‘‘spelling development’’ as a search for predictable mappings between a
growing awareness of sounds in words and the symbols that map to these
sounds. As such, Ehri’s model ties directly to the phoneme-awareness re-
search. The model also links spelling to reading, because reading experi-
ence is believed to have a direct impact on spelling. There are four stages
in her model: Precommunicative (preliterate), Semiphonetic (only por-
tions of sounds are represented, and letter names are used), Phonetic
(represents all sounds, some spellings accurate, some letter names), and
Morphemic (uses functional multiphoneme patterns).
A stage-model approach has a number of pitfalls. Framing a research
problem in terms of a stage model leads to a dangerous (unscientific)
mode of investigation in which the data are constrained at the outset by
the classification system. This is especially problematic when there is so
much latitude in the classification process. Not only this, but because the
spelling errors must fit the categories and the categories must be stages,
the description and interpretation of these error patterns are adapted to
comply with the stage model, rather than simply reflecting what the
data show. To illustrate this, I provide evidence from research by C. S.
 257

Beers and J. W. Beers (Beers and Beers 1992), former colleagues of

How Does Anyone Learn to Spell?

Henderson.
The first thing one notices in their report is that their stages are dif-
ferent from Henderson’s, and are based more on phonological and lin-
guistic knowledge: Prephonetic, Early Phonetic, Phonetic, Structural, and
Correct. Yet, in summarizing the earlier work in the field (i.e., Hender-
son’s), Beers and Beers wrote: ‘‘There are indeed identifiable stages of
orthographic awareness through which children pass as they become more
proficient in their writing, [and] children proceed through these stages
at varying rates’’ (p. 231). Apparently, these stages are not sufficiently
‘‘identifiable’’ to be constant from one decade to the next, or from one
group of collaborative researchers to another.
Beers and Beers tested first through sixth graders on a nonsense-word
spelling test in which the words had to be altered to represent the plural,
regular past tense, or progressive verbs. Test items took the form: ‘‘John
saw a blinch in the woods. Then he saw two .’’ Errors were tallied
and assigned to ‘‘spelling stages’’ for each age group.
For purposes of this illustration, I will focus on the top stage (Cor-
rect), which would reflect a high degree of accuracy for most children who
were at the Structural stage (Beers and Beers), or the Derivational Con-
stancy stage (Henderson)—the stage of morphologically based spelling
‘‘rules.’’ Specific ages are never claimed for any stage model (including
Piaget’s). Nevertheless some age constraints must apply, otherwise stages
cannot be ‘‘developmental.’’ In view of the 6-year age spread, ‘‘mastery’’
scores (percent correct) should appear for most children at a particular
age, and for the remainder shortly thereafter. Furthermore, because all the
test items reflect knowledge of ‘‘orthographic rules’’ for adding common
suffixes, they belong in the same stage.
Neither of these things happened. There were huge age differences
for ‘‘stages’’ depending on the particular transformation required. Mastery
appeared at first grade for adding -s for the plural, at second grade for
adding -es, but the transformation sky ! skies did not materialize until
sixth grade. Which one is the correct measure of the ‘‘spelling stage’’ for
plurals, or do we need a stage for each kind of plural?
Similar anomalies appeared for past-tense spellings. First graders did
poorly. There was a big jump at second grade, no change whatsoever for 3
years, then sudden mastery in sixth grade. Is this a stage with a 4-year
 258

plateau, or did the teacher teach past-tense spellings in sixth grade? There
Chapter 9

seem to be two stages for mastering the ‘‘add -ing rules’’ (drop e, double
the consonant). For some reason it is easier to master the ‘‘drop e, add -ing
rule,’’ though certainly not for everyone. Only 70 percent succeeded by
second grade. Following this, there was no change for the next 4 years. The
letter-doubling trick appears to be much later ‘‘developmentally.’’
These peculiar results are just as likely to reflect what was going on
the classroom. Because no information was provided about this, these
results are uninterpretable. In any case, they do not support a stage model
that places ‘‘structural’’ or ‘‘derivational’’ spelling errors at a single stage.
In fact, no stage model can explain these data, not even one that pro-
posed a different stage for every type of plural and every type of verb
transformation.
A classification process must follow standard scientific principles;
otherwise this work does not count as science. Evidence for a stage model
requires that at least these criteria be met:

An objective (not subjective) classification scheme must exist.

Interrater agreement (reliabilities) should be provided.

All errors must be classified, not just some of them.

A numerical tally (frequency count) of each type of spelling error should

be provided in a published report.

Shifts in patterns of spelling errors across age must be documented in

either cross-sectional or longitudinal studies.

And for a classification scheme to count as ‘‘science,’’

It must be sufficiently robust that it can be replicated by others in the

field.

It should be adopted unanimously by everyone in the field (this does not

mean it cannot be modified).

Imagine what the field of chemistry would be like if every investigator

developed his or her own unique table of chemical elements.
As I reviewed the numerous studies on spelling stages, other concerns
arose about the validity of stages as a proper interpretation of the data.
Here are some of the problems:
 259

Error scores are the basis for the coding scheme, but errors provide no

How Does Anyone Learn to Spell?

information on how children spell words correctly.

Because stages are defined by the error-classification process itself, the

logic is circular: larger to fewer spelling errors ¼ lower to higher ‘‘spelling

stages.’’

The data are often represented inappropriately, as tallies (head counts)

of children ‘‘at each stage.’’ Stages can only be demonstrated by shifts for
individual children to prove that spelling stages are not mixed. That is,
children could not be in a ‘‘visual’’ stage once they were in a ‘‘phonetic’’
stage.

I provide examples of four well-known studies to illustrate these prob-

lems. Two claim support for stages and two found no support.
Morris and Perney (1984) were interested in the relationship between
spelling skill (stages of development) at the beginning of first grade, and
reading scores at the end of the year. They designed a spelling test that
had words containing consonant and vowel digraphs, consonant blends,
and a variety of e-controlled and vowel þ r spellings. All spellings were
‘‘regular.’’ First graders took this spelling test (the same test) in Septem-
ber, January, and May. Reading tests were also given in May.
A scoring system was devised for spelling errors that fit their four-
stage (six levels) model: The Prephonetic stage was a simple letter tally,
with three levels corresponding to zero, one, or two letters in the correct
sequence. The Phonetic stage consisted of getting a single vowel letter in
the right location plus any minor error (BAC for back). The Transitional
stage consisted of ‘‘legal’’ orthographic patterns (CK ), but with a minor
error. The Transitional stage was combined with the Correct stage in all
analyses, reasons unstated, probably due to too few correct spellings. Each
word was scored as being in one of these stages and awarded points. The
circular logic is obvious. Spelling errors are coded in accordance with
the number of correctly spelled phonemes in the correct sequence. The
spelling-error coding scheme ¼ the ‘‘stages.’’1

1. Morris and Perney’s scheme is different from Henderson’s and from that of
Beers and Beers, yet Morris formerly worked with Henderson.
 260

A tally was made of how many children were in a particular stage in

Chapter 9

January as a function of where they were in September. These numbers

reveal that children can get stuck at a stage or skip one or two stages
entirely. Given the fact that these children had the same exposure, and
everyone started school knowing all the letters to spell these words, this
weakens the argument for stages. Stages mean a similar developmental se-
quence given the same environment. Nothing in Morris and Perney’s data
suggests this.
Nunes, Bryant, and Bindman (1997) in the United Kingdom studied
‘‘developmental stages’’ by focusing on one spelling issue: adding -ed to
indicate past tense. They set out guidelines for what a stage model meant
to them:

Any stage model should pass at least three tests. One is that all, or very nearly
all, the children should clearly belong to one of the stages in each session. . . .
Second, the developmental stages should be related to external criteria: The
children at more advanced stages should be the older or educationally more
successful children in the sample. The third test is the most stringent and
unfortunately the least often applied. . . . Children should move in one direc-
tion but not in the other. (p. 642)

Despite this clear statement, the descriptive language in the report is quite
at odds with the notion of stages. The language reflects children’s slowly
emerging awareness of spelling conventions. The researchers wrote that at
first children ‘‘ignore’’ spelling conventions, then ‘‘they begin to realize’’
there is an -ed spelling convention but cannot apply it, later they ‘‘grasp
[its] grammatical significance’’ but misapply it to irregular verbs, and
finally they ‘‘learn about exceptions.’’
This gradual process of becoming aware of spelling patterns is not a
description of anything ‘‘stagelike.’’ It is a description of learning.
This was a longitudinal study with 363 children tracked from second
through fourth grade. They were given a spelling-dictation test in which
ten words were regular past-tense verbs ending in /d/ or /t/ that take the
-ed spelling (load, loaded; wilt, wilted ), ten were irregular past-tense verbs
ending in /d/ or /t/ ( found, felt), and ten were common nouns ending in
/d/ or /t/ (bird, belt).
 261

The spelling test was given three times: at the start of the study, 7

How Does Anyone Learn to Spell?

months later, and 13 months after that. Spelling errors were classified as
unsystematic errors (no connection to the word stem), phonetic errors in
which the letter d or t was simply added to the root word, and general-
izations of the -ed spelling to the foil words (irregular past-tense verbs, and
nouns).
The results were surprising. Generalization errors of the -ed spelling
to irregular past-tense verbs were common ( felted ), and the number of
children making these errors actually increased over time, from 34 percent
at the first testing to 38 percent at the second and 42 percent at the third.
More remarkable, this effect also appeared for nouns (birded ). Roughly 30
percent of the children made noun-generalization errors at all three test-
ings. Overall, 71 percent of the children made at least one generalization
error for irregular verbs, and 56 percent did so for nouns. A stable or
increasing error rate over a 20-month period is strong evidence against
stages. (It is also strong evidence against learning!) It seems that the
nature of the task, and the experience of taking the identical test three
times, led children to believe that the correct response was to add -ed
wherever possible.
Faced with these unexpected results, the authors ignored this simple
explanation and created new ‘‘stages’’ on the basis of the overgeneraliza-
tion errors. A five-stage model was superimposed on the pattern of error
scores, very much in the style of Morris and Perney. Children were
classified according to their predominant type of spellings error (a.k.a.
‘‘stages’’) and assigned to stages (a.k.a ‘‘spelling errors’’). These categories
were even less convincing than those used by Morris and Perney. As
an example, the stage 2 classification was ‘‘Spell half or more of the end-
ings of irregular verbs and nonverbs correctly. . . . Make at least five pho-
netic transcriptions of regular past verbs but should produce very few
(less than three) ed spellings for any of the words’’ (pp. 641–642; my em-
phasis). Some ‘‘stages’’ even permitted a 50-50 split between one stage and
another.
Nunes, Bryant, and Bindman tracked the fate of children from the
first to the second testing 7 months later, which provides some sense of
the movement of individual children through stages. Of the children who
started out in stage 1, 28 were still there 7 months later, 11 moved up one
 262

stage, 8 moved up two, and 2 were in stages 4 and 5 respectively. Children

Chapter 9

who were initially at stage 2 behaved even less stagelike, 43 got stuck
there, 5 went backward to stage 1, none went up one stage, 5 went up two,
and 6 went up three. Children’s movement from one stage to the next is
scarcely progressive or orderly. The follow-up testing produced similar
results. Of the 58 children initially assigned to stage 1, 15 children (25
percent) were still there 20 months later.
A second hypothesis in this study was that knowledge of grammar
(past tense) would directly impact children’s ability to spell past-tense
verbs. However, their in-house grammar test was far too advanced for
7- to 9-year-olds. Test scores were near zero and changed little over 20
months. This did not prevent the authors from doing statistics on the
data. The authors interpreted their findings as follows:

These generalizations are at the heart of our new model of the development of
spelling. This proposes that a child’s first step in spelling is to adopt a phonetic
spelling strategy; the next step is to notice and to try to incorporate exceptions
to these rules, but without a complete understanding of their grammatical
basis; the next step is to understand fully this grammatical basis for some of
the spelling patterns that do not fit well with the letter-sound rules; and the
final step is to learn about the exceptions to the grammatically based rules.
(p. 647)

There is no evidence that children are progressing in ‘‘steps.’’ There is no

evidence that these children ‘‘understand fully [the] grammatical basis’’ of
the past-tense -ed rule, because ‘‘awareness’’ of this knowledge was never
measured. The data actually run counter to this argument, since error
scores increased over time. It is equally likely that children learn to spell
these words through exposure and practice, not because they are in an
‘‘orthographic spelling stage.’’
Morris and Perney’s work as well as that of Nunes and colleagues is
typical of the research in this area. The logic is circular, and ‘‘stages’’ are
often rerigged to suit the experimental hypothesis. There is a strong bias
toward stage models and few studies in the literature where this model is
challenged. We turn next to two such studies.
Canadian psychologists (Varnhagen, McCallum, and Burstown 1997)
decided to take a long, hard look at the validity of ‘‘spelling stages.’’ They
 263

adopted Gentry’s (1981) classification system, and used Morris and Per-

How Does Anyone Learn to Spell?

ney’s (1984) error-scoring method. This was a cross-sectional study on
272 children in first through sixth grade. The children were asked to write
a composition about ‘‘A Special Day.’’ They were told not to worry about
spelling errors and not to correct them.
The analysis of spelling errors was limited to words with e-controlled
vowel spellings (lake), and with the three phonetic forms of regular past-
tense verbs: helped (help’t), grabbed (grab’d), listed (listed). The stages were
as follows: Precommunicative (random), Semiphonetic (minimally pho-
netic), Phonetic (consonants correct but letter name for vowel, or all
phonemes correct but ends with -d for past tense), Transitional (correct
phonetically but incorrect orthographically), and Correct spelling. Three
raters independently scored a portion of the errors, and interrater relia-
bility was high, above .90 in all cases. (This critical step is missing in most
of the research on stages.)
The first finding was that the classification system did not hold up.
There were no Prephonetic errors. Semiphonetic errors were in evidence
at first grade but not after that. Stage 4 (Transitional errors) were virtually
nonexistent, and nearly all the children jumped from stage 3 to 5. The
only two categories that reliably classified the spelling errors and that
varied systematically as a function of age were the Phonetic and Correct
categories.
When spelling scores were plotted by age, there was zero evidence for
stages. Instead, there was a simple trade-off between Phonetic and Correct
spellings. Phonetic errors gradually declined (sloped gently down) and
Correct spellings gradually increased (sloped gently up) as mirror images
of each other. The same pattern appeared for every type of spelling error,
e-controlled vowel spellings and past-tense suffixes alike. These patterns
are typical of learning curves and are not remotely stagelike. Furthermore,
these are the only age-related functions I have seen in the research on
spelling stages.
When the investigators looked at the individual data for vowel þ e
spelling errors, they classified 27 children as being in three stages at the
same time. They arrived at similar results with past-tense spelling errors.
They concluded that there was no evidence for stages, only evidence
for a transition from a purely ‘‘surface’’ phonetic approach to an increas-
ing knowledge of spelling patterns. Children are learning to spell, and this
 264

is influenced by classroom instruction. As they noted: ‘‘Different rates

Chapter 9

in this progression appeared to be related to the spelling curriculum’’

(p. 160).
In the second study from this group (Steffler et al. 1998), the notion
of stages was abandoned and replaced by ‘‘strategies,’’ which were solicited
from the children. Children took a spelling-dictation test and responded
to the question ‘‘How did you spell that word?’’ after every word. This
was a cross-sectional study on children in grades 2 through 5.
Four strategies appeared with sufficient frequency to tabulate them.
The most common responses were ‘‘I just knew it’’ (Retrieval ) and ‘‘I
sounded it out’’ (Phonetic). Some children reported using analogies to
similarly spelled words—‘‘Clip is lip with a C in front of it’’ (Analogy)—as
well as orthographic rules: ‘‘e makes the A say its name’’ (Rule). Idiosyn-
cratic responses were scored as Other. Analogy, Rule, and Other strategies
were rare at all ages and diminished with age. By grade 5, the proportion
of analogy strategies had dropped to 2 percent (opposite the prediction of
stage models).
Rule-based strategies were reported only for CVCe words (18 percent
at grade 2, 13 percent at grade 3, zero at grade 5), and were very helpful
indeed. Again, this is opposite to the prediction of many stage models,
which are supposed to end up at the ‘‘rule-based’’ stage. Second graders
who reported they knew the e worked with the preceding vowel scored 95
percent correct on CVCe words (home, late), as opposed to children who
used Analogy (37 percent correct), a Phonetic approach (48 percent cor-
rect), or Retrieval (73 percent correct). The two most common strategies,
Phonetic and Retrieval, traded off with age. A Phonetic strategy domi-
nated at second grade, and a Retrieval strategy dominated at fifth grade.
The term retrieval is difficult to interpret. The process is too instan-
taneous to be a ‘‘strategy.’’ The authors’ treatment of this was somewhat
vague and contradictory. They used definitional terms like automatic as
well as choose and adaptive approach, suggesting that a retrieval strategy is
under voluntary control. This led them to infer that ‘‘retrieval’’ replaces a
phonetic strategy, marking the end of ‘‘phonetic spelling’’ for a particular
group of words. But it is just as likely that rapid phonetic analysis continues
and combines with visual memory to equal ‘‘retrieval,’’ a sense of instant
awareness. As noted often in this book, what seems instantaneous in sub-
 265

jective time is not instantaneous in neural time. People are not aware of

How Does Anyone Learn to Spell?

their eye movements when they read, yet their eyes move nonetheless.
Taken as a whole, these studies provide no support for a stage
model, even assuming there were no methodological problems. I will leave
the final word to Ehri (1992), taken from her commentary on the ‘‘fea-
ture’’-oriented stage models. I believe these comments apply to all stage
models:

We might also consider whether a focus upon spelling features necessitates a

conception of stages to organize the features, or whether it might be sufficient
to describe the emergence of the features themselves and leave it at that. This
would resolve the controversy over which features are hallmarks of particular
stages. Also, it would eliminate expectations that certain features emerge at the
same time, which data show are false. (p. 328)

Ehri called for longitudinal studies to look at

individual children’s spellings, how consistent their appearance is at first and

later on, what effect one feature has on other features when it appears (i.e.
overgeneralization errors), and how the emergence of features corresponds to
or is influenced by explicit instructional experiences. . . . Presently our basis for
believing that there are stages of spelling development is primarily conceptual
and quite limited empirically. (pp. 328–329)

There are more longitudinal studies in the literature today than there
were in 1992, but the data continue to show that stage models are false.
Stage models of spelling will continue to be limited empirically because
they are circular and logically untenable.

What Do Good Spellers Know That

Poor Spellers Do Not?
Another popular approach to the study of spelling is based on the theory
that good spellers have certain knowledge that poor spellers lack, such as
knowledge of the linguistic structure of English words and of how this
structure relates to the spelling code. Knowing that the past tense is usu-
ally represented by -ed is one example.
 266

This type of study attempts to pinpoint specific skills peculiar to good

Chapter 9

spellers that are absent in poor spellers, using word lists that feature par-
ticular types of spellings. Poor spellers will, by definition, be worse on
every type of test (they are selected to be worse). The goal is to show dif-
ferential performance on the tests, a worse performance on some tests than
on others. For instance, one might anticipate that differences would be
small to nonexistent on one-syllable ‘‘regularly’’ spelled words, but greater
on words where spellings obey a convention or ‘‘rule’’ (past-tense end-
ings) or are determined in some way by morphology (language structure).
The validity of this research is entirely dependent on whether the
word lists actually measure what they purport to measure. Knowledge of
the structure of the spelling code is imperative for in-house spelling tests
to be valid. Otherwise, there is no way to know whether performance is
due to a visual or phonetic strategy, to ignorance of the orthographic or
morphological spelling ‘‘rules,’’ or to insufficient exposure to print. Un-
fortunately, few researchers have any knowledge of this structure.
As an example of this problem, it is assumed that morphological
spelling clues like those bequeathed to us by Samuel Johnson in 1755 are
of great benefit to readers and spellers alike ( Johnson [1755] 1773). Lin-
guists, echoed by reading researchers, often point to ‘‘linguistic’’ con-
nections between word forms and the spelling code. But they are highly
selective. The word sign is said to contain the gn spelling because it is
morphologically related to signal and signature. This may be true, but it
does not matter a fig unless this morphological clue is consistent and pre-
dicts these transformations: deign, deignal, deignature; reign, reignal, reign-
ature; design, designal, designature; impugn, impugnal, impugnature; benign,
benignal, benignature. As you can see, it does not. In order to use this
‘‘morphological clue,’’ a person would have to remember that the gn in
sign occurs because of signal or signature, ‘‘BUT NOT’’ (a blocking rule—
or exception to a rule you have to remember) for other words with the gn
spelling. It is far simpler to remember that gn is a spelling alternative for
the sound /n/ (except for signal and signature), and be done with it. There
are 15 common words with this spelling.

Research on College Students

Fischer, Shankweiler, and Liberman (1985) were the first to try this ap-
proach using good and poor spellers at the college level. They designed
 267

a spelling test consisting of various types of words like those described

How Does Anyone Learn to Spell?

above. They predicted that poor spellers would be differentially worse on
words where spellings were most influenced by linguistic features, and
differ least on words with the most transparent (‘‘regular’’) and most
opaque spellings (rare sight words). They designed a spelling test with
three levels of ‘‘transparency of orthographic representation.’’ They con-
sulted Hanna et al. 1966 (see chapter 3), a book on spelling rules by
Witherspoon (1973), and two dictionaries. These sources provided spelling
‘‘rules’’ of this dubious variety: ‘‘Words of more than one syllable, ending
in a single consonant preceded by a single vowel, if accented on the last
syllable usually double the final consonant before a suffix beginning with a
vowel’’ (Fischer, Shankweiler, and Liberman 1985, 439). Clymer (1983),
who showed that spelling ‘‘rules’’ or ‘‘generalizations’’ are meaningless,
would be amused by the word usually.
Fischer and colleagues set up the spelling tests correctly using spelling
alternatives (possible spellings for 40þ phonemes), not decoding alterna-
tives (possible readings of a letter). The various levels were described as
follows:

Level 1. Words where the phonetic realization is close to the orthographic

representation and has a high frequency in written English.
Level 2A. Words that contain an ambiguous segment departing from
straightforward phonetic mapping and that require an application of or-
thographic conventions or a sensitivity to regularities at the surface pho-
netic level. (This refers to such things as knowing when to double final
consonants when adding a suffix.)
Level 2B. Also has an ambiguous segment, reflecting ‘‘abstract morpho-
phonemic knowledge.’’ For the most part, these transformations were
assumed to call on an awareness of how root words and affixes combine.
That is, a good speller was supposed to know that the final r is doubled in
confer/conferring but not in confer/conference as a consequence of syllable
stress.
Level 3. Words were ‘‘demon’’ spelling words that contained segments not
normally found in English spelling and often of foreign derivation.

Words on the lists were mainly multisyllable words and were balanced
across levels for syllable length and frequency in print.
 268

The distinction between Levels 1 and 3 is clear. There is no question

Chapter 9

that words like blunder, alternate, and unemployment are more transparent
and easier to spell than words like diphtheria, sergeant, annihilate, and
pygmy. And these words were noticeably different from Level 2 words.
The problem arose in the contrast between Level 2A and 2B words. For
example, letter doubling was supposed to work by an ‘‘orthographic rule’’
at Level 2A in words like sobbing, clannish, and thinned. Yet these words
were not qualitatively different from words of ‘‘morphophonemic’’ origin,
such as omitted, regrettable, and equipped (Level 2B). Thus the distinction
between Levels 2A and 2B was based on bogus phonics-type rules.
The results showed that Level 1 was easiest, Level 2 next easiest, and
Level 3 hardest, for good and poor spellers alike. Both groups had the
same error patterns, and in the same proportions, phonetic substitutions
being the most common errors (around 88 percent), confirming the find-
ings of Varnhagen and colleagues. The two groups parted company on
Levels 2A and 2B. Poor spellers were equally bad on both word lists, but
good spellers spelled the morphophonemic (2B) words more accurately
than the orthographic (2A) words (contrary to expectation).
Fischer and associates also tested good and poor spellers on a
nonsense-word spelling-recognition test. The students had to choose
which of two words was most likely to be spelled correctly. Poor spellers
had the most difficulty when an added prefix or suffix required a modifi-
cation to the root word.
When they looked at other possible contributors to these results, the
poor spellers were found to score well below good spellers on the WRAT
reading test and on the Stanford reading-comprehension test. The groups
did not differ in vocabulary (WAIS vocabulary), showing that these
students’ spelling problems were not related to verbal IQ. And equally
important, poor spellers did not do worse on a visual-memory test for
abstract visual patterns, evidence that visual memory is not the source of
the poor spellers’ difficulties.
Fischer and colleagues surmised that because the greatest discrepancy
between good and poor spellers appeared on the ‘‘morphological’’ spelling
test (Level 2B), ‘‘linguistic sensitivity’’ was at the root of their reading and
spelling difficulties. However, it is just as likely that these students are poor
spellers because they are poor readers and do not read (or write) fre-
quently enough to observe the more difficult spelling patterns very often.
 269

Australian psychologists, Holmes and Ng (1993) were unable to rep-

How Does Anyone Learn to Spell?

licate Fischer, Shankweiler, and Liberman’s findings for Level 2A and 2B
type words, and attributed this to an artifact in the word lists used by
Fischer and colleagues. Holmes and Ng found the least difference on reg-
ular words, and the greatest difference between the spelling groups on the
idiosyncratic words (contrary to Fischer, Shankweiler, and Liberman’s
prediction). Othographic and morphophonemic word lists were interme-
diate in difficulty. Overall, Holmes and Ng’s results showed that good and
poor spellers were most alike when words are easy to spell, and least alike
when words are hard to spell, suggesting that poor speller’s problems are
due to insufficient exposure to words with unusual spellings.
They reasoned that poor spellers’ relatively weaker performance on
idiosyncratic words might be because there are more ways to spell them
incorrectly. They did a second experiment to control for this possibility.
Groups of poor and good spellers were given a reading task in which the
students had to decide (yes/no) whether a word was a real or nonsense
word (blurge), using ‘‘regular’’ and ‘‘idiosyncratic’’ spellings. The time to
initiate a response (voice-onset latency) and errors were recorded. Poor
spellers needed more time to decode longer, more complex words than
good spellers did, and another experiment was carried out to pinpoint the
problem.
Nonsense words were created from real words by systematically
changing one letter in the initial, middle, or final position. Good and poor
spellers differed only when the change appeared in the middle of the word.
It took poor spellers 100 milliseconds longer to read these words. It
appears that poor spellers focus mainly on the outer segments of words
as they read, rather than decoding phonetically from left to right. If this
is the case, it means that exposure is confounded with decoding strategy
—the failure to pay close attention to internal spelling patterns in multi-
syllable words.
Visual-processing skill was also measured. The students had to decide
whether pairs of words, nonwords, random-consonant sequences, and
random-symbol sequences were the ‘‘same’’ or ‘‘different.’’ Poor spellers
were significantly slower on judgments involving words and nonwords,
though both groups were very accurate on this part of the experiment.
Matching pairs of random letters or symbol sequences was incredibly
slow, twice as slow (4,000 ms versus 2,200 ms) in some cases. But there
 270

were no differences between good and poor spellers on either visual task
Chapter 9

for speed or for errors. Once more, the evidence shows that poor spellers
do not have visual-processing problems.
These are important results. First, they show that it takes only half the
time to judge whether a pair of words is the same as it does to decide
whether a random string of the same letters is the same. This means that
processing speed improves as a function of exposure, and the redundant
syllable patterns in speech, and how these are represented by spelling
code, make this possible.
Holmes and Ng also found that poor spellers were far more likely to
have low vocabulary scores, but did not differ in nonverbal IQ. The stu-
dents took the Author Recognition Test (Stanovich and West 1989) to
estimate exposure to print. The task is to check off famous authors’ names,
which are mixed with names of unknown persons. The differences were
enormous. Good spellers correctly identified an average of 18 authors out
of 40, poor spellers only 7.5 (pure guessing), proof that poor spellers do
not read nearly as much as good spellers.
These studies provide a excellent profile of the poor speller at the
college level. They score well below good spellers on standardized reading
tests; they also read far less and have a weaker vocabulary. They have no
visual-perception problems, but they do have an unusual visual-scanning
patterns for reading multisyllable words, focusing most attention on the
outer segments of the word. None of this, of course, sheds much light on
cause. For example, a weak vocabulary or poor decoding skills might
depress interest in ‘‘reading a lot,’’ so that print exposure is a result of
reading skill, and any ‘‘causal’’ link to spelling is indirect.

Ressearch on Children
Waters, Bruck, and Malus-Abramovitz (1988) adapted this research design
to study children. They set up five types of spelling lists suitable for chil-
dren age 8 to 12 years (see pp. 172–173). Precise descriptions of each type
were provided. Here they are in slightly reduced form:

Regular words. ‘‘Must contain spellings that directly reflect the surface
phonology of the word, and which can be derived through the application
of sound-spelling correspondences. Sounds have few spelling alternatives’’
(original emphasis). Later in the article, they referred to regular words as
 271

words where ‘‘spellings can be derived on the basis of invariant sound-

How Does Anyone Learn to Spell?

spelling correspondences.’’
Regular words*. These are similar to regular words but contain a phoneme
represented by several legal spelling alternatives. (Example: street, streat,
strete.)
Orthographic words. Contain a phoneme or segment that is spelled accord-
ing to orthographic conventions. (Example: when to use ch or tch for final
/ch/.)
Morphologic words. Knowledge of word structure and conventions for
attaching affixes to words, and of spellings based on relationships to other
words. (Examples: sign, grammar.)
Strange words. Contain at least one orthographic segment that rarely
appears in English spelling. (Examples: yacht, ocean.)

This is a well-designed scheme, so it was surprising how little the words

on the spelling lists reflected these descriptions. Regular words were not
‘‘regular.’’ ‘‘Surface phonology’’ means that connections between individ-
ual phonemes and their individual spellings are unaffected by surrounding
phonemes (context free). Five words ( program, unit, finding, push, album)
are not spelled with the most probable (regular) spelling and they can be
spelled phonetically a number of ways ( proagram, ewnit, fineding, and so
on). Nor are they context free, as these hyphenated examples show: prog-
ram, un-it, fin-ding.
Regular words* were defined as ‘‘regular’’ (most probable spellings)
but containing a phoneme with more than one spelling. But six of these
words did not have high-probability spellings: money (munny), true (trew),
duty (dooty), absolute (absoloot), event (eavent), and fever ( feever). Two words
lacked any viable spelling alternatives: foolish (? fulish/foulish) and woodpile
(? wouldpile/wudpile). I defy the reader to spell them differently.
Similar anomalies appeared on the Orthographic and Morphologic
word lists. Morphologic words were more like ‘‘spelling demons.’’ They
contained uncommon spelling patterns (bomb, sign, obey), schwa vowels
(German, destructible), and words spelled with double letters (really, hope-
fully, dissolve). It is highly unlikely that children (or adults) spell these
words by reflecting on morphological ‘‘rules’’ having to do with analogies
to bombadier or signature, or because they are aware of a final -l doubling
rule for adverbs.
 272

Waters and colleagues remarked that there were flaws in Fischer,

Chapter 9

Shankweiler, and Liberman’s word lists, but there were also flaws in the
word lists used by Waters and associates, making it difficult to interpret
the data. The most straightforward description of the word lists is that
they increase in difficulty with each level as they depart from a surface
phoneme-grapheme relationship.
This study involved 158 children at four grade levels (3 through 6),
consisting of the top and bottom thirds in spelling performance. Poor
spellers were also very poor readers. Results for all groups combined
(main effects) showed that high-frequency words were easier to spell than
low-frequency words (frequency in print), Regular words were spelled
more accurately, Regular* and Orthographic levels did not differ from
each other, and the remaining levels differed significantly from each other
in the expected direction. Grade level was also significant.
However, while poor spellers did worse overall, there was no differ-
ential impact as a function of the types of spelling words. This result is
particularly interesting, not only for purposes of this study, but because a
stage model would predict that these spelling ‘‘levels’’ would be acquired
by good and poor spellers at different times or rates. This was one of three
requirements of a stage model as specified by Nunes, Bryant, and Bind-
man 1997.
The patterns across age were very revealing. First, good spellers made
systematic gains across all spelling levels between grades 3 and 4, while
gains for poor spellers during this time period were virtually nil. Second,
the size of the gains was comparable for all categories of spelling words
with the exception of ‘‘Strange words.’’ Third, not only did poor spellers
score far below good spellers on every test, but they made proportionately
fewer gains at every grade level. They are late out of the starting gate and
run more slowly as the race progresses.
This is more evidence against developmental stages (not that Waters
and colleagues were interested in stages). A stage model would predict
variations in acceleration as a function of age and the category of words
the child was asked to spell, assuming these categories were meaningful.
Instead, there were similar rates of improvement across all levels.
Waters and associates also compared the performance of the same
children on a spelling-recognition task. The children had to choose the
 273

correct spelling from three foils (trane, train, trayn). Poor spellers had a

How Does Anyone Learn to Spell?

disproportionate boost in accuracy compared to good spellers. They
improved at five times the rate of good spellers on Regular* words, double
the rate on the Morphologic words, and by 25–50 percent for Strange
words. Of course poor spellers had more room for improvement, but
results were the same after the top spellers were removed from the
data analysis. Poor spellers clearly have more accurate spellings in mind
(available in recognition memory) than they are able to retrieve (recall
memory). There is also a hint in these results that poor spellers may be
learning the code in only one direction, from print to sound, because they
were so much more accurate in judging spellings (visual-recognition
memory) than in producing them (recall memory).
As a rule, children tend to score similarly on reading and spelling
tests. It is uncommon for a child to be very good at one and very poor at
the other. Bruck and Waters (1990) managed to locate 13 children among
175 sixth graders who scored normally on a reading-comprehension test,
and extremely poorly on a spelling test. They were matched to two con-
trol groups that were either good at both or poor at both. We will call
them the good, poor, and mixed groups. The mixed group was charac-
terized by very weak reading and spelling scores but normal reading-
comprehension scores. They also had above-average verbal ability vocab-
ulary standard scores: good, 107; mixed, 105; poor, 86.
In a series of carefully designed tests, Bruck and Waters showed that
both the mixed and poor groups read difficult or ‘‘incongruous’’ words
equally slowly and inaccurately. Both groups relied heavily on context
to read words, but the mixed group was far more successful, no doubt
because of having superior vocabulary skills. However, one should not
get carried away with theorizing about this unusual population of chil-
dren. The mixed group in this study represented only 3 percent of the
175 children originally tested, and the children in this group were
not good but poor decoders/poor spellers (their only advantage being in
comprehension).

Summary
Poor spellers are more likely to have a limited vocabulary and weak
decoding skills, and do not read as much as good spellers do. However, as
 274

a group they are not differentially worse than good spellers on tests
Chapter 9

designed to tap knowledge of specific orthographic ‘‘rules’’ or morpho-

logical patterns. Their spelling ability falls short across the board, no
matter how the tests are constructed. There is indirect evidence that
reading and spelling skill are linked (supporting the correlational re-
search), because it appears difficult to find children who perform normally
on a reading test and abnormally on a spelling test. However, this research
suffers from the problem of not knowing how these children and young
adults came to be poor readers and spellers in the first place. In other
words, we have no idea how they were taught.

Some Instructional Tips

Finally, we do know something about certain aspects of spelling instruc-
tion that do or do not work. Some of this research was described briefly in
chapter 4.

Writing Words Helps You Learn How to Spell Them

Cunningham and Stanovich (1990b) taught first graders to spell a list of
words, learning a third on the computer, a third with letter tiles, and a
third by writing them out. On half the trials the children said the letter
names as they typed, arranged letter tiles, or wrote the letters, and half the
time they said nothing. The children practiced these three tasks for four
days, 30 minutes each day. Writing the words was superior to the other
two conditions, though spelling was hardly exemplary. Out of 10 words in
each condition, the average score was 1.5 (tiles), 1.6 (computer), and 3.1
(writing). Saying letter names had no effect. These results support Hulme
and Bradley (1984), who also found that children learn to spell faster and
more accurately by writing words than by arranging letter tiles.
Research has shown that it is easier to remember something in the
same environment in which you learned it (a phenomenon known as
context-dependent learning). To test for this effect, Cunningham and Sta-
novich taught children a set of words in three modes: tiles, computer, and
writing. The children were tested with tiles or by computer, stacking the
deck against the writing group. Despite this manipulation, the writing
group still excelled. It is the learning phase that matters most, not how
learning is tested. Visually guided movement, like copying, boosts mem-
ory in a way that purely visual learning does not.
 275

The authors commended Montessori for her wisdom in insisting that

How Does Anyone Learn to Spell?

children write as a major method for learning to spell. Montessori also
believed that spelling should precede reading for this and other reasons.2
Cunningham and Stanovich did not point out Montessori’s other in-
sight (and that of Dale 1898)—that children should say the sound each
letter stands for as it is being written (not the letter name). So far, there
are few empirical studies on Montessori’s important ideas.

Learning Letter Names Can be Hazardous to Your Spelling

Much of the research on spelling is based on spelling errors, errors that
are often a consequence of ineffective instruction. After Read documented
the numerous letter-name spelling errors in children’s creative writing, a
‘‘letter-name stage’’ appeared in stage models, as if these errors were part
of the inevitable scheme of things (Gentry 1981; Henderson 1982).
The stage-model research showed that letter-name substitutions pro-
duced numerous spelling errors. A thorough documentation of letter-
name substitution errors has been carried out on a large sample of
children’s writing by Treiman (1994). She replicated Read’s findings en-
tirely: the ubiquitous use of vowel letter names in CVCe words (CAK),
the use of the letter name ar (R) as a vowel: car ¼ CR, and the use of
consonant letter names for final open syllables in words like baby ¼ BAB.
In their study on spelling ‘‘stages,’’ Varnhagen, McCallum, and Burs-
town (1997) commented that their ‘‘Phonetic’’ category was swamped
with letter-name errors, particularly in CVCe words: lake spelled LAK,
hide spelled HID, and home spelled HOM. The proportion of first graders
making errors on these particular words was 79 percent, 44 percent, and
50 percent, respectively, and this persisted into second grade: 22 percent,
15 percent, and 0 percent. Recall that only two spelling categories held up:
Phonetic errors and Correct spelling. Because the Phonetic category con-
tains a large number of letter-name spelling errors, if these errors could be
eliminated by not teaching letter names in the first place, the Phonetic

2. As noted earlier, Montessori believed that if spelling is taught first, children

can automatically decode (read) what they have just written and will ‘‘dis-
cover’’ how to read. This is an excellent way to ensure the code is reversible.
 276

category ought to disappear. And if it did, the only category left would be
Chapter 9

Correct spelling!
Treiman discovered that letter-name knowledge can have a negative
impact on children’s spelling accuracy. A study was designed to find out
whether children knew the letter names in their own name, and whether
this would generalize to knowing letter-sound correspondences (Treiman
and Broderick 1998). Kindergarten and first-grade children were asked to
identify letters in the alphabet by name and by sound. Children knew far
more letter names than sounds. Only the letter name for the initial letter
in the child’s first name was consistently (and significantly) likely to be
known, and this was true for both age groups. However, knowing this
letter name did not ensure that these children knew the first sound in their
names.
The second experiment was a replication of the first, with a writing
component added. Preschoolers (4:3 to 5:9 years old) were tested for
letter-printing accuracy, plus letter-name recognition. Scores were sig-
nificantly higher for the letter that began the child’s first name, but, again,
this did not lead to knowledge of the sound the letter stood for. Despite
the fact that young children are very familiar with the shape and ‘‘name’’
of the first letter of their name, knowing this provides no clue to the
sound it represents even by first grade. The supposed ‘‘generalizability’’
effect has been the only argument for the importance of teaching letter
names.
Treiman and Tincoff (1997) designed a special spelling test to pin
down letter-name spelling errors in multisyllable words. Kindergartners
and first graders were asked to spell nonsense words like tuzzy, tuggy, and
tuzzigh. The first word ends in a letter name—zee—and the others do not
( ghee and zigh). They found children commonly spelled open syllables
(CV-CV ) with single consonant letters (b, z, d, g). These ‘‘letter-name’’
errors were three times more likely to appear in the kindergartners’ spell-
ings than in first graders’ spellings. This shows that letter names are
something a child has to unlearn to be able to spell. Treiman and Tincoff
observed that the error patterns showed that children were processing
words at the level of the syllable, not the phoneme, and they were match-
ing letters and sounds at the level of the syllable as well. They emphasized
the significance of this fact: ‘‘These letter-name spellings reveal that the
alphabetic principle is fragile for beginning spellers. Children find it dif-
 277

ficult to represent speech as a sequence of individual phonemes. They

How Does Anyone Learn to Spell?

sometimes take an easier path—the path of symbolizing groups of pho-
nemes as units’’ (p. 447). In other words, letter names mislead children in
two ways, by directing attention to the wrong phonological unit for the
writing system, and by causing them to misrepresent the sound-to-print
correspondences.
Yet despite the obvious negative effect of letter-name knowledge on
spelling, Treiman and Tincoff did not believe this was cause for alarm or a
reason to change what we are doing: ‘‘Knowledge of letter names thus has
both positive and negative effects on early spelling. It can help children to
spell alphabetically by helping them learn the links between phonemes
and letters and by setting up expectations about how words are spelled.
However, knowledge of letter names sometimes leads children to deviate
from the alphabetic principle’’ (p. 449). This is a surprising statement in
view of the finding by Treiman and Broderick (1998) that letter-name
knowledge does not spontaneously generalize to letter-sound knowledge!
There is a vast literature on letter-name knowledge and letter-naming
speed that proves conclusively that knowledge of letter names per se has
nothing to do with reading or spelling skill (see Language Development and
Learning to Read ). Digit-naming speed and accuracy are just as highly
correlated with reading skill (decoding accuracy and speed) as letter-
naming speed. Because naming digits has nothing to do with reading, this
means the apparent connection between letter-name knowledge and
reading is artifactual. Something else is causing the relationship. What
this something might be is unknown. It could be an aptitude for paired-
associate learning, or it could be the fact that when mothers and teachers
teach letter names, they also teach digit names.
We have more direct and practical evidence as well. Wimmer (1993)
pointed out that Austrian teachers actively discourage the use of letter
names when teaching children to read and spell, and focus exclusively on
the phoneme-grapheme relationship instead. Every child learns to read
and spell with a high degree of accuracy. Of course, the fact that the
German spelling code is transparent matters a lot. But we know from
these studies that letter names do not matter.
Jeffrey and Samuels (1967) and Samuels (1972) showed long ago that
learning letter-sound relationships cut the learning time to decode words
spelled with those letters by approximately 50 percent as compared to
 278

learning letter names. Furthermore, letter-name knowledge never gen-

Chapter 9

eralized to the sounds inside the names. Learning letter names was no
more beneficial than memorizing names of geometric shapes or cartoon
characters! In fact, it was no more beneficial than being taught nothing
at all.
The message is clear: Discourage and eliminate the use of letter names and
encourage the teaching of phoneme-grapheme correspondences.
 10
THE MANY-WORD PROBLEM:
MORE TO SPELLING THAN MEETS THE I

The studies reviewed in the previous chapter failed to show either how
children learn to spell, or why some children fail and others do not. The
concept of stages of spelling development is untenable on both logical and
scientific grounds. The notion that poor spellers lack some linguistic
knowledge or insights that would allow them to access ‘‘orthographic
rules’’ or ‘‘morphological levels’’ of the spelling codes has not been sub-
stantiated. Poor spellers do worse across the board and have the most
difficulty with rare, irregularly spelled words that appear infrequently in
print—words whose spelling patterns do not reflect ‘‘orthographic rules’’
or ‘‘morphological structure.’’ The most parsimonious explanation of the
poor spellers’ problem is lack of exposure to print. I address the reasons
this might be the case later.
One of the major problems with these research efforts is that they
reflect two main sources of confusion. The first relates to these funda-
mental questions: What processing skills are involved in mastering our
spelling system? Is it phoneme awareness, knowledge of phoneme-
grapheme correspondences, rote visual memory, memory for redundant
visual patterns, or all the above? The second source of confusion is igno-
rance of the structure of the English spelling code. Unless we can get
beyond these basic holes in knowledge, we are unlikely to make much
headway toward the ultimate goal, which ought to be figuring out how we
teach children to spell. The question ‘‘What’s wrong with people who
can’t spell?’’ is, after all, a rather foolish question when researchers do not
know the structure of the spelling code or how to teach it.
An analysis of the English spelling code was provided in chapter 3.
This showed that the spelling code contains multiple components or
 280

aspects. It is a phonemically based system, but a strict one-to-one rela-

Chapter 10

tionship holds over a very limited portion of phonemes. It is strongly

linked to the phonotactic structure of the syllable, and scores of common
spelling patterns reflect this structure. There are transformations for add-
ing suffixes that come very close to ‘‘rules,’’ and when they do, they are
well worth teaching, not as rules, but through suitable examples and rep-
etition. However, the bulk of the code is made up of hundreds and hun-
dreds of patterns, systematic (redundant) grapheme sequences that are
highly context dependent, context meaning the order of sounds and letters
and the relative position of those letters: batch and botch, but watch, not
wotch; teach, not teech; church, not churtch or cherch; fault and fall, not falt
and faul.
Apart from the more obvious structural elements in the spelling code
as outlined in chapter 3, the central concern for researchers is the many-
word problem. How do people learn these patterns, and what sort of pro-
cessing is going on in the brain to allow them to do this? The many-word
problem encompasses the fact that it would be impossible to teach chil-
dren to spell every word in the English language that deviated from a
common set of phoneme-grapheme correspondences. This is a problem
that does not exist in countries with a transparent alphabet. If the brain
did not spontaneously record and update the thousands of redundant pat-
terns in our spelling code, no one could learn to read or spell unless they
had a photographic memory. For a brain to solve the many-word prob-
lem, it must see a lot of print. The problem certainly cannot be solved by
taking scores of spelling tests consisting of random word lists, or by writ-
ing compositions riddled with misspelled words.

Measuring Print Exposure

If a large part of spelling skill is dependent on the brain’s ability to code
the structural redundancies of our complex spelling code, then how can
we know if this is true? One way is to measure print exposure and study its
impact on spelling. Stanovich and his colleagues explored the interactive
or ‘‘knock-on’’ effects of reading a lot on spelling skill. In particular, they
were concerned that the one-sided view that phoneme awareness explains
everything misses a critical aspect of reading and spelling skill—the fact
that unless phoneme analysis is connected to the spelling patterns of the
written code, it is of little use.
 281

The initial research (Stanovich and West 1989) was designed to look

The Many-Word Problem

at the contribution of both phonological decoding and orthographic
knowledge to spelling skill. The general view had been that phonological
analysis was mainly responsible for the integration of phoneme sequences
and spelling patterns. However, Stanovich and West felt that while this
may be true, the ability to store orthographic patterns in memory might
be an independent skill. If this is the case, then a measure of exposure to
orthographic patterns would be most likely to tease the two apart.
They hit on a novel solution to measure exposure to print by design-
ing a test to estimate how much people read. Questionnaires on reading
behavior are fatally prone to ‘‘halo effects,’’ people reporting they read far
more than they actually do. To avoid this problem, Stanovich and West
designed the Author Recognition Test (ART) and the Magazine Recog-
nition Test (MRT). We met the ART in the previous chapter.
The tests are straightforward, simple to administer, and easy to take.
The Author test consists of a list of 50 authors (well known by virtue of
book sales and critical reviews), mixed with 50 foils of names of unknown
persons. The Magazine test lists 50 names of real magazines and 50 of
made-up magazines. The task involves a simple yes-no choice.
In one study, college students took the two tests described above plus
a Reading Habits questionnaire, asking for a list of favorite authors and
magazine subscriptions. Students took two spelling tests, a standardized
test (WRAT spelling) and Fischer’s test described in the previous chapter.
Of the self-report measures, the ART was far and away the best pre-
dictor of spelling skill, with the following correlations: composite spelling,
r ¼ .46; segment errors on the WRAT, r ¼ .38; and Fisher’s test, r ¼ .43.
Correlations between the spelling measures and self-reported reading
habits and scores on the MRT were essentially zero.
In a second study, involving a much larger number of students, Sta-
novich and West used a battery of tests to tease apart phonological
decoding from ‘‘orthographic’’ analysis. Reading was measured by the
Word ID (word recognition) and Passage Comprehension subtests of the
Woodcock. Other tests were Phonological Choice (select the word from
a pair of misspelled words that ‘‘sounds like’’ a real word: trane, clane),
Nonsense Word decoding test, Orthographic Choice (choose the cor-
rectly spelled word: trane, train), and Homophone Choice (which of these
two words is a fruit? pair, pear).
 282

The students were divided into skilled and less skilled readers on the
Chapter 10

basis of their reading test scores and compared statistically on 25 tests.

Three tests failed to distinguish the two reader groups: the MRT, Ortho-
graphic Choice, and Homophone Choice. Clearly, the MRT is not a dis-
criminating test, nor were the tests designed to measure ‘‘orthographic’’
visual memory. Because good readers read a lot and poor readers read
very little, the absence of any difference on these tests is evidence against
their theory that exposure to print enhances memory for orthographic
patterns.
The most discriminating spelling measure was a count of word-
segment errors on the WRAT spelling test, followed by ‘‘exception
words.’’ Poor readers differed most from good readers on words with un-
usual spellings that appear infrequently in print, supporting the results of
Holmes and Ng, as well as of Waters, Bruck, and Malus-Abramovitz (see
the previous chapter). Spelling test scores were strongly correlated with
Word ID (r ¼ .50 to .63) and somewhat less with Passage Comprehension
(r ¼ .26 to .44). These values are similar to those reported on children
(Bond and Dykstra 1967; Shanahan 1984). Taken together, these results
confirm the consistent connection between reading and spelling skills,
with around 25–30 percent of shared variance.
The main purpose of the study was to look at the separate contribu-
tion of print exposure to spelling, with phonological decoding skill con-
trolled. Phonological Choice and Nonword decoding were entered first in
a regression analysis, followed by the ART (the measure of print expo-
sure). The two decoding measures accounted for 25 percent of the vari-
ance on WRAT spelling, and the ART contributed a small (3 percent) but
significant amount of variance beyond this. This cannot constitute cause.
Verbal IQ is consistently correlated with spelling and was not controlled.
The ART scores could be partially attributable to ‘‘reading a lot’’ because
you have good verbal skills. If verbal ability and decoding skill ‘‘cause’’
print exposure, the connection between print exposure and spelling will be
spurious. The authors were well aware of this problem and of the fact that
correlational research cannot sort this out.
The failure to control for verbal IQ, and the problem of determining
the direction of causality in a correlational study did not get solved in
subsequent research from this group. In later studies on print exposure,
the focus shifted to children. Cunningham and Stanovich (1990a) devel-
 283

oped a new test called the Title Recognition Tests (TRT). The children

The Many-Word Problem

had to identify real titles of children’s books from a list of foils. But this
only made the causality issue murkier, because, as the authors acknowl-
edged, children’s awareness of book titles could reflect a literary environ-
ment, or print exposure, or both—something heard as well as something
seen (e.g., bedtime stories).
Because the TRT is not a clean measure of print exposure, and be-
cause of the methodological problems in these studies (small sample size,
unstable results from one study to the next), the research on children
(Cunninghan and Stanovich 1990a; Cunningham, Stanovich, and West
1994) sheds no light on whether ‘‘reading a lot’’ improves one’s knowl-
edge of the spelling code.
The idea that print exposure plays an important role in learning the
redundant patterns in the spelling code seems so intuitively plausible that
most people want to believe it. Yet, so far, the evidence is too weak to
show that print exposure plays a causal role in enhancing this skill.

Spelling without Phonology: A Study of Deaf Children

The goal of Stanovich and West (1989) was to tease apart phonological
decoding from knowledge of orthographic structure to study the impact of
print exposure on orthographic knowledge alone. They did not succeed.
But there is another way.
Aaron et al. (1998) devised an ingenious way of tackling this problem
by comparing the spelling of profoundly deaf children to that of normally
hearing children on a variety of spelling tests. Profoundly deaf children
have little or no phonological awareness so it is not clear how they learn to
spell, which many do with some success. Because they are obliged to rely
solely on visual memory, what kind of visual memory is it?
Aaron et al. pointed to a number of anomalies in the research lit-
erature on this topic, largely due to the lack of control over the degree
and type of hearing impairments (some children had spoken language).
There were disagreements in the literature about how to account for the
fact that some deaf children and adults spell rather well. One theory was
that they relied on pure rote visual memory, making them logographic
spellers. Others believed that they acquire ‘‘phonological sensitivity’’
through speech training or lip reading, while still others suggested that
they learn by mastering morphological spelling clues or ‘‘rules.’’ Aaron
 284

et al. proposed a fourth theory—that deaf readers rely on the statistical

Chapter 10

probabilities of spelling sequences in words. They referred to this as a

‘‘frequency’’ model, meaning frequency in print.
They located 35 profoundly and congenitally deaf children in grades
5 through 12 who passed a rigorous screening. The children had no
speech, or completely unintelligible speech, and could not communicate
vocally. They were matched in reading age on the SAT and Woodcock
reading tests with normally hearing children. The hearing children were
evenly distributed by reading age across grades 2 through 5. This estab-
lished a baseline. If spelling varied systematically between the deaf and
hearing children on different types of spelling tests, this would be inde-
pendent of reading and specific to spelling.
A variety of spelling tests was used to test the various theories de-
scribed above. The first was the critical test of phonological encoding.
Aaron and colleagues used a homophone-production task, in which the
children were asked to write in the missing word in a sentence (correct
spelling) and then add a ‘‘homophone’’ (same sounds, different spelling):
‘‘Clouds are white, but the sky is b .’’ (Answer: blue, blew.) In another
homophone test, children had to write all the pairs of words they could
think of that ‘‘sounded’’ alike (or were ‘‘spoken’’ alike) but were spelled
differently. Deaf children had great difficulty understanding this concept
and needed extra instruction. (Instruction was done in sign and was pro-
vided by a highly qualified teacher.)
The deaf children scored 90 percent correct on spelling words in
context, equivalent to the grade 5 children. However, their success rate
dropped to 10 percent when they were asked to supply a homophone, far
below the grade 2 readers, who scored 47 percent correct. In the second
homophone task, deaf children scored 20 percent correct and grade 2
readers 60 percent correct. Inspection of the individual data indicated that
two deaf children did rather well on the last task. Their data were ex-
cluded from the remaining analyses, leaving a group of 33 children whose
phonological sensitivity was essentially zero.
Next, the visual theories were put to the test. The children read sen-
tences with a missing word that was spelled with a digraph with an ines-
sential letter. Thus sno still sounds ‘‘snow’’ without the w. Deaf children
spelled these words about as well as grade 5 readers (average 12/13 cor-
rect). When they looked at the proportion of spelling errors due to
 285

missing letters, hearing children made 80 percent of their errors in this

The Many-Word Problem

category and deaf children only 38 percent. This shows that hearing chil-
dren rely more on phoneme-grapheme patterns, and deaf children rely on
visual memory, which is not supported by phonological processing.
But what kind of visual memory is it? Visual memory was tested by
showing a series of word sets flashed on a screen for 30 seconds. Follow-
ing each set, the children wrote down as many of the words as they could
remember and tried to spell them correctly. The deaf children succeeding
in writing down 90 percent of the words, the hearing children around 80
percent. The deaf students were much older, so this was expected. How-
ever, despite their superior memory for the words, the deaf students made
far more spelling errors, 48 percent correct compared to 70 percent, 76
percent, and 89 percent for the hearing children at reading grades 3, 4,
and 5. Deaf children made more transposition errors, visual errors, and
unclassified errors, and virtually no phonetic errors (2 percent). This
shows that visual memory alone is inadequate to spell common English
words, even though deaf students remembered more words.
The final and critical task manipulated the statistical properties of
spelling patterns in the words. The investigators used short nonwords
spelled legally or illegally based on bigram and trigram frequencies (the
number of times two- and three-letter sequences appear in words). The
question was whether deaf students learned to spell by remembering re-
dundant (probable) spelling patterns rather than by pure rote memory
(random-letter sequences). If so, they would not do any better than hear-
ing children on spelling words that are phonotactically illegal (unpro-
nounceable), like prta, hrda, and tneh. The task took the same form (word
sets) as the previous experiment. Children had to remember the words in
each set, write them down, and try to spell them correctly.
Again, deaf students succeeding in recording (recalling) more words
than even the fifth-grade children. The totals across all word sets combined
were 391 (deaf ) versus 259 (normal) for pronounceable words, and 309
versus 162 for the unpronounceable words. Apart from this, the deaf and
normal children were remarkably alike. Both remembered significantly
more legally spelled nonwords and spelled them more accurately. Deaf
children performed much like the grade 3 and 4 reading groups, scoring
67 percent correct on legal nonwords. However, both deaf and hearing
children did much worse (48 percent correct) in spelling illegal nonwords.
 286

Deaf and hearing children find legal spelling patterns much easier to
Chapter 10

remember and spell. Not only this, but deaf children do not have superior
visual rote-memory skills to hearing children. It appears that deaf and
hearing children alike rely on both types of visual memory—pure rote
memory and memory for redundant orthographic patterns—and they do
so to the same extent.
A qualitative analysis of spelling errors on this task revealed that
hearing children made mainly phonological errors, and deaf children
hardly any. Their errors were visual and included letter reversals, and
missing or substituted vowels and consonants. Deaf children made vowel
and consonant errors in the same proportion, whereas the hearing chil-
dren’s errors mostly involved vowel spellings.
This is strong evidence that profoundly deaf children learn to spell by
relying on the statistical properties of visual spelling patterns. These chil-
dren had no phonological skills and no special advantage in rote visual
memory. Yet they were able to spell nearly as well as grade 5 readers on
some tasks. The other important result was that the hearing children were
using a combination of three different skills: phonological processing for
phoneme-grapheme relationships, pure rote visual memory, and memory
for the statistical probabilities of ‘‘orthographic’’ patterns in words. It
is this combination of skills that allows us to remember words and spell
them accurately, and that accounts for why spelling improves with age. As
Aaron pointed out, deaf children are at a distinct disadvantage: ‘‘A lack
of acoustic phonological skills sets an upper limit at about the fourth
grade level, beyond which spelling skills may not progress. This may be so
because beyond a certain level, children encounter many cognate verbs,
adverbs, and multi-syllable words whose spelling is influenced by mor-
phophonemic conventions’’ (p. 18).
A related finding on a very different population of children was re-
ported by Siegel, Share and Geva (1995) in Canada. They tested 257 poor
readers who scored below the 25th percentile on the WRAT reading test,
and 342 normal readers who scored above the 35th percentile.
The task was a spelling-recognition exercise: look at pairs of nonwords
and select the one that looked ‘‘most like a word.’’ The children were also
tested on the Word Attack subtest of the Woodcock. The normal and poor
readers differed on the Word Attack test in the expected direction. But the
poor readers were superior (statistically) on the spelling-recognition task,
 287

supporting the findings of Waters, Bruck, and Malus-Abramovitz (1988).

The Many-Word Problem

The poor readers behaved more like the deaf children, overly relying on
visual memory for the common patterns of letter sequences, and were less
aware of phonemes and phoneme-grapheme correspondences.

How Does the Brain Solve the Many-Word Problem?

Now we are at the heart of the matter. Aaron et al. proved that deaf and
hearing children alike learn the statistical probabilities of orthographic
patterns. But what exactly are those statistical patterns? Their study con-
trasted illegal letter strings with pronounceable letter strings based on
bigram and trigram frequencies. But our spelling code cannot be mapped
accurately by bigrams and trigrams. Readers have to be on the lookout
for single letters (t), digraphs (th), and multiletter phonograms (ough in
though), and constantly shift among them as they scan each word. If people
rely on the statistical probabilities of orthographic sequences in addition to
phoneme-grapheme correspondences, what are those sequences, and how
does this work? If we had precise answers to these questions, if we knew
what the brain did automatically, perhaps we could optimize instruction to
work with the brain and not against it, as we are doing at the moment.
Twenty years ago, cognitive psychologists began a quest to discover
how reading skill (speed and accuracy) is impacted by redundant ortho-
graphic patterns. This was the beginning of the discovery of the ‘‘third
way’’ to decode and spell, as revealed by Aaron’s elegant study on deaf and
hearing children. In this research, however, the expression ‘‘orthographic
patterns’’ refers to the reading code and not the spelling code (keeping in
mind that writing systems are spelling codes, not reading codes). The re-
search was intended to discover how spelling patterns map to pronuncia-
tions, and not how sounds in the language map to the spelling patterns.
This is a nontrivial distinction, because the spelling code has a logic and
a true probability structure, and the reading code does not. This will be-
come more apparent in the analysis of the orthographic patterns that
researchers assume impact decoding speed and accuracy.
Much of this research has been intimately connected with the devel-
opment of computer models of reading. An early and influential model
was the dual-route model. According to this theory, there are two inde-
pendent pathways in the brain that process a printed word to its final
destination (Coltheart 1978; Coltheart et al. 1993). One path (lexical entry)
 288

is ‘‘logographic’’ (purely visual) and maps directly to semantic memory

Chapter 10

and the stored phonological representation of the word. The other path
involves translating from letters to individual phonemes and assembling
them into a word. The translation is said to proceed via grapheme-phoneme
correspondence rules (GPC rules). This route bypasses semantic memory
and links to the output phase just prior to ‘‘saying’’ the word. GPC rules
are the decoding counterpart of phonics-type spelling rules.
The concept of GPC rules had its roots in the work of Chomsky and
Halle (1968), who proposed that phonology and morphology were linked
by a set of correspondence rules. If a rule failed in a particular instance,
this violation would be marked (presumably by the brain) with a ‘‘blocking
rule’’ (‘‘BUT NOT’’). Venezky’s (1970, 1999) attempt to discover ‘‘rules’’
of the spelling-to-sound code (chapter 3) fits this line of thinking. The
dual-route model holds that GPC rules can be revealed by a complete
analysis of printed words of the type Venezky embarked on. These rules
are presumed to be deduced by the learner (or taught) and will explicitly
guide the learning process. The dual-route model also includes the dictum
that once a GPC analysis of a word is sufficiently fast, the word is trans-
ferred to the lexical path and becomes a sight word, recognized instantly
(which, as chapter 2 showed, is impossible).
A similar idea was developed to explain the behavioral data on adult
readers and children learning to read. It was believed that people have two
(equally efficient) reading styles, reading whole words by sight (logo-
graphically), or reading phonetically. Baron and Strawson (1976) chris-
tened them ‘‘Chinese’’ and ‘‘Phoenician’’ readers. This reflects a mistaken
belief that the Chinese (or anyone else) have a logographic writing system,
and that people can learn to read by memorizing whole words as if they
were telephone numbers.
The dual-route, Chinese/Phoenician models had their detractors, but
none as important as Robert Glushko. His pioneering work is directly
linked to the latest computer models of reading based on ‘‘parallel dis-
tributed processing’’ (Seidenberg and McClelland 1989; Plaut et al. 1996).
These models mimic what is thought to occur in the brain when someone
learns to read. These are statistical models. They ‘‘learn’’ by virtue of
processing the structural redundancies in the input, along with the feed-
back from the environment about success rates.
 289

On the Trail of Structural Redundancies in the Spelling Code

The Many-Word Problem

Glushko became disenchanted with the dual-route model as a graduate
student and set off in a different direction. The result was a new way to
think about how reading (decoding) occurs (Glushko 1979). It was
Glushko’s work that led to the realization that there was a third process at
work during decoding, distinct from rote visual memory and grapheme-
to-phoneme correspondences.

A New Logic Glushko’s insights stemmed from a logical analysis of the

problem. A dual-route model is illogical whether routes are considered
separately or in combination. Glushko proposed that all the information
about a word—visual, phonological, orthographic, semantic—is processed
at the same time in parallel. This means that no matter how many ele-
ments contribute to successful decoding, processing is not carried out in
separate (disconnected) pathways.
Furthermore, the proposal that ‘‘orthographic rules’’ could be distin-
guished separately from an analysis of the whole word was patently false.
For one thing, how does one identify the particular size of the unit that
qualifies it as an ‘‘orthographic rule’’? Glushko did not give an example,
but there are many to choose from. Consider the ‘‘orthographic rule’’ that
a consonant should be doubled when adding -ing. Where does one draw
the line between this rule and the remaining letters in the word? Should
this rule contain a drop -e clause, or an antidoubling clause, or are these to
be separate rules? And because this rule must also specify where a reader
has to look to find out whether to double, it needs two additional sub-
clauses and a blocking rule to make it work: (1) if the word ends in e, drop
the e before adding -ing (dine, dining; bridge, bridging); (2) double the con-
sonant when the preceding vowel is a ‘‘simple’’ checked vowel (bat, bat-
ting) BUT NOT when it isn’t (beat, beating).
And these were not the only problems. Dual-route theorists use the
term orthographic rules to mean three different things. One meaning has to
do with linguistic descriptions, the second, knowledge of language struc-
ture, and the third, procedures or mechanisms of pronunciation. A lin-
guistic description of the spelling code was supposed to be a reflection of a
reader’s implicit or explicit knowledge of it. But as Glushko pointed out,
‘‘There is no necessary relationship between linguistic descriptions of the
 290

orthographic and phonological regularity in the language and a reader’s

Chapter 10

knowledge of such language structure’’ (p. 675).

This is like my argument against the notion that people are aware of
the ‘‘morphological rules’’ that govern spelling patterns. There is no rea-
son to expect that spelling patterns are perceived by the speller (or reader)
in a linguistic framework rather than simply as probable spellings of those
words. This is not to say that some morphological rules or patterns cannot
be taught or might be useful to know, but, rather, that they are not auto-
matically deduced from spoken language or from text. Fischer, Shank-
weiler, and Liberman (1985) found that when they interviewed the college
students about their strategies on the spelling test, they were completely
unaware of any orthographic or morphological rules.
Glushko also pointed to difficulties with Venezky’s 1970 analysis, and
with his notion that readers generate GPC rules: ‘‘Since each word in
the language is weighted equally in deciding whether a correspondence is
regular or not, descriptive rules like these would be extremely difficult for
readers to induce’’ (Glushko 1979, 675).
There are two issues here. First, Venezky’s analysis was not proba-
bilistic. He did not specify which print-to-sound ‘‘decodings’’ are most to
least likely and to what degree. Second, the fact that only linguists with
extensive training are able to generate these ‘‘orthographic rules’’ makes it
highly unlikely that the average reader would do so spontaneously, some-
thing even Venezky (1967; 1999) acknowledged.
Glushko made several new proposals concerning how decoding could
occur. First, the decoding process must take into account the contextual
dependencies of the entire word. No purely surface decoding via GPC
rules will work, just as no attempt at whole-word decoding, independent
of the sequential redundancies in the orthography, will work.
Second, decoding proceeds by a variety of means, operating in parallel
and integrating orthographic and phonological information. ‘‘Ortho-
graphic information’’ consists of all relevant featural components and their
sequential patterns. Glushko used the analogy of natural language, in
which the speaker and listener use all linguistic elements at the same time
(semantic, syntactic, phonological, prosodic) as well as the context in
which the conversation is taking place.
Third, he adopted the term activation to refer to a hypothetical ‘‘state’’
in which the brain automatically processes this information:
 291

I suggest that it is more appropriate to focus on the functional properties of

The Many-Word Problem

the lexical and orthographic knowledge bases, and to replace the unnecessarily
specified mechanisms with a more neutral term such as ‘‘activation.’’
In this simpler framework, I propose that words and pseudowords are
pronounced through the integration of orthographic and phonological in-
formation from a number of sources that are activated in parallel, much as
readers comprehend sentences by integrating lexical, syntactic, and contextual
information. As letter strings are identified, there is parallel activation of or-
thographic and phonological knowledge from a number of sources in memory.
This knowledge may include the stored pronunciation of the letter string,
pronunciations of words that share features with the letter string, and infor-
mation about the spelling-to-sound correspondence of various subparts of the
letter strings. (p. 678)

The central problem, then, became one of determining which of these

sources or features of the word’s spelling were more or less relevant.
The goal in this research has been to find an objective or quasi-
objective way to quantify the ‘‘orthographic patterns’’ that lead to success
in decoding English words. Many approaches have been attempted, but all
have failed. One reason is a lack of understanding of the structure of
English words (phonotactics) and the way the spelling code maps to this
structure. A second reason follows from the first. Because the analysis
is exclusively in the direction of decoding and never in the direction of
encoding, the ‘‘structural patterns’’ do not reflect the way the code was
written and how it developed. The fact is that spellings are assigned to
words. Words are not assigned to spellings. A third reason is one Glushko
pointed out. Orthographic structure consists of ‘‘the contextual depen-
dencies of the entire word’’ and ‘‘all relevant featural components and their
sequential patterns’’ (p. 678). It has proven extremely difficult to meet the
goal of defining ‘‘all relevant features.’’
There have been two main approaches to this problem, one quantita-
tive and the other subjective. The quantitative approach makes no prior
assumptions about ‘‘patterns,’’ but relies on frequency counts of letter
sequences (bigrams and trigrams). The subjective approach is based on the
observation that many final VC or VCC units in one-syllable words are
spelled consistently, providing orthographic clues for how to read the
word. In cognitive models, this component is known as the word body.
 292

Teachers have known it for at least two centuries as a word family, and
Chapter 10

more recently it was rechristened the rime. I will refer to the visual repre-
sentation of the VC or VCC letter sequences in one-syllable words as the
orthographic rime and the phonological ‘‘decoding’’ of that rime as the
phonological rhyme. It is very important to keep this distinction straight, or
this research will seem even more complex than it already is.

Orthographic Rimes Have Consistent and Inconsistent Neighbors Glushko’s

interest in orthographic structure led him to the orthographic rime. He
set up lists of words based on whether the rime was consistent or incon-
sistent in terms of how it was decoded—its pronunciation. This was the
logic: the orthographic rime (ile in mile) will automatically activate words
stored in memory that have the same rimes both visually and phonologi-
cally. In Glushko’s theory the orthographic rime maps directly onto the
phonological rhyme. The theory does not take into consideration the fact
that a phonological rhyme would also map to an orthographic rime (/s/ /t/
/ie/ /l/—style), if, indeed, rimes are even relevant in the first place.
The word lists consisted of regular words (those with many consistent
neighbors) and exception words (those with inconsistent neighbors).
Words were matched for word frequency. For example, dean is regular
(there is only one way to read words ending ean); deaf is an exception
word because it inconsistent due to the word leaf. Two sets of nonsense
words were also created by changing the initial consonant of word pairs:
hean and heaf. The hypothesis was that regular words/nonwords would be
decoded faster and more accurately than exception words/nonwords.
It was as at this point that Glushko’s otherwise impeccable logic be-
gan to fail. Already, one can see the problem. Because the initial conso-
nant was not part of this scheme, other readings besides the ones he
expected were not acknowledged. What if heaf was read as hef by analogy
to head, producing the same outcome as deaf, but for a different reason?
What if hean was read as hen as an analogy to head instead of to dean?
More problematic was Glushko’s use of the term error. The ‘‘errors’’
were not really errors in any objective sense. Instead, errors occurred
when the subjects did not read a nonword as Glushko expected them to
(with a consistent pronunciation), even though the rendering was perfectly
legal. For instance, the mathematical likelihood of reading heaf as hef or
heef is identical, because there are only three real words spelled with this
 293

rime, deaf, leaf, and sheaf. This selective scoring biases the error rate in the

The Many-Word Problem

direction of the hypothesis, which appears to be what happened.
And there were problems with the word lists, the data, and the infer-
ences from the data. I will run through the studies briefly to illustrate
these effects. College students in the first experiment saw 172 words and
nonwords plus 100 filler words presented randomly (one at a time) by
computer. Their task was to read each word as accurately and as quickly as
possible into a microphone, and the time to start speaking was recorded
(voice-onset latency).
Glushko provided an item analysis for the response-time data, one of
the few people in this field to do so. These values are extremely illumi-
nating. For one thing, contrary to Glushko’s prediction, the students read
14 exception words faster than the regular words, and there were 4 ties.
This is 42 percent of the words on the lists. And even though the results
were ‘‘significant’’ in the predicted direction, regular words were read, on
average, only 29 ms faster than exception words ( p < :05). Yet the differ-
ence between the word pairs that went opposite to prediction was much
larger, 51.4 ms.
What could explain this? Certainly not regularity and exceptionality
or consistent and inconsistent ‘‘rimes.’’ Perhaps this was a word-frequency
effect. High-frequency words like both, done, most, none, some, and were
might be read faster because they are ‘‘activated’’ more quickly, to use
Glushko’s terminology. Some of these words fit little ‘‘families’’ phono-
logically: done, none, some (one, come), most, post ( ghost, host). But there are
no easy answers for why foot was read faster than feet, hood was read faster
than heed, or comb was read faster than cold. The fact that spook—the only
inconsistent ‘‘rime’’ in the ook family—was read 90 ms faster than spool is
a mystery, and this goes directly against the idea of consistency as being
the determining factor in decoding these words.
Experiment 2 was designed to counter the objection that there were
too many overlapping words on the lists (words with the same rimes),
which may have biased the way people responded. This problem was
remedied by using nonwords only. Some pairs were taken directly from
the old list, some had the same rime but different initial consonant, and
some were new. I prepared a table with the following information: (1) the
pairs used in experiment 2; (2) whether the same or a similar pair was used
in experiment 1; (3) response-time differences between experiments 1 and
 294

2; and (4) whether or not the response times followed prediction. The
Chapter 10

table illustrated some disconcerting facts about these kinds of word lists.
In the first place, it is tempting for experimenters to drop out words that
go against the prediction and retain those that do. I am not saying that this
is a conscious act or that Glushko did this; otherwise all word pairs would
have gone in the prediction direction. But he did have a better hit rate for
the reused word pairs than for the new word pairs he created: 16 out of
26 old pairs going with the prediction (62 percent), versus only 3 out of
7 new pairs (43 percent). The response-time data represents an additional
concern, because the values for the same word pairs fluctuate wildly from
experiment 1 to experiment 2.
Glushko himself appeared to have considerable doubts about his clas-
sification scheme. In the third experiment, he reclassified words into three
categories. He used the word have to explain how this worked. Have is both
an exception word and inconsistent, in that the spelling ave is only decoded
/av/ in this word, whereas it is usually decoded /aev/ (cave, gave). Thus, have
is the only word with this pronunciation. The word gave, on the other
hand, is no longer regular by the old scheme, but inconsistent by virtue of
the existence of have. Another group of words were both regular and con-
sistent, having lots of neighbors and no competitors. Two sets of consistent
words were derived from the exception and inconsistent words by changing
a single consonant (haze and wade), and these became the control words.
In general, Glushko’s predictions for this complex set of contrasts was
borne out statistically, but it is doubtful that the categorization process fits
his activation theory, or that the findings are even supported by the data.
First of all, the pairs of regular/consistent control words (haze, wade) were
supposed to be read at more or less the same speed, having been derived
from the same word ( gave), both having lots of neighbors. But only 66
percent of these word pairs were processed at comparable speeds, when all
should have been.
Furthermore, because the activation theory predicts that neighbors
will automatically be activated together and boost the likelihood of a par-
ticular reading of a word, regular/inconsistent words like wave (lots of
consistent neighbors) ought to be read more quickly than an exception
word like have (all neighbors inconsistent). This, after all, is the basis for
the consistency effect. Yet reaction time was 492 ms for have and 528 ms
 295

for wave. Similarly, the exception word love took 472 ms to read, and the

The Many-Word Problem

regular word grove took 642 ms. Altogether, 12 regular/inconsistent words
took longer to read than exception words, and 4 were tied—39 percent
going against prediction.

Some Conclusions Comparing the reaction times of some ‘‘exception’’

words that were used in both experiments 1 and 3 revealed some star-
tlingly erratic values. Often the differences between the reaction times
for the same word were larger than the comparisons between different
words, which seriously compromises this work. Perhaps voice-onset la-
tency relayed through a microphone is an unreliable technology, and/or
people are not consistent in naming speed.
There are a host of factors involved in how quickly people can read
words, and they take us far beyond the realm of consistent or inconsistent
‘‘word bodies.’’ These are the number of times you have seen the word
(familiarity), the emotional saliency of the word (love has high emotional
salience), and idiosyncrasies due to word type.
More important issues have to do with the spelling code, which is very
different from the reading code. For example, have is a completely ‘‘regu-
lar’’ spelling; there is no other way to spell it, given that final /v/ is always
spelled ve. The e in the ve spelling does double duty in some words, also
marking the vowel. The fact that the e plays two roles in words like gave
makes the ve spelling unusual. A number of ‘‘exception’’ words on Glush-
ko’s lists are not exceptional in the least, because this is the only way they
can be spelled. These are words like give, foot, hoof, pint, plow, and soot.
(I challenge the reader to spell them differently.) Then there are words
where vowel spellings are linked to the initial consonant: b and p take the
u spelling for /oo/ (bush, bullet, put, push—book is the exception). The
vowel spelling is controlled by the w in words like wand and work (rather
than wond and werk).
Glushko confessed in a footnote that things were a lot more complex
than simple rime patterns would allow, and perhaps his remarks are the
best conclusion to this work: ‘‘A more general activation and synthesis
model, with a broader experimental base, would allow for the contribution
of neighbors in all positions, and would differentially weight them in dif-
ferent tasks’’ (p. 684).
 296

At this point, I must address the problem created by the notion that
Chapter 10

orthographic rimes are major sources of brain activation. In Glushko’s

initial work and in the subsequent research on computer models of read-
ing, the word body or rime means a letter string mapping to a pronuncia-
tion. Whether activation generates a brainstorm of related words with
identical rimes is unknown. This is complicated by the fact that we read so
fast that we are not aware of how we do it. Nevertheless, it takes the brain
450–700 ms to read a simple word—quite a long time for brains. Simple
reaction time (see a light, press key) ranges from 150 to 200 ms, so a brain
can do quite a lot in that much spare time.
Something undoubtedly gets activated, but what? If both ortho-
graphic rimes and phonological rhymes light up, chaos would ensue. A
number of Glushko’s words illustrate this quite nicely. A person reads
great and the visual system activates words with the same visual rime: eat,
beat, cheat, defeat, heat, meat, neat, peat, pleat, seat, treat, wheat. A few
milliseconds later, the phonological system gets a match (/eet/). Mean-
while, memory traces have linked great to the phonological rhyme (/aet/),
activating phonological rhymes for the words: bait, crate, date, inflate,
freight, gate, hate, mate, plate, slate, state, spate, straight, trait, wait. Because
the systems work in parallel, I assume the spellings above (visual rimes) are
supplied rather quickly as well. What is the brain to do with this? A sys-
tems collapse seems inevitable! Yet Glushko found that the reaction time
for great (an exception word) was 41 ms faster than for greet, a highly
regular spelling with no ‘‘enemies.’’
It is unlikely that the system works this way. Decoding may occur
largely through generating the most likely connection between a visual
image and its phonological image, and where both images are much more
highly specified than by word fragments like rimes or rhymes, which on
the face of it is an extremely quaint notion. How does anyone learn to
read or spell words that do not rhyme?
I have taken time to survey the problems in constructing these word
lists to illustrate the danger of making inferences from these results about
how the brain processes spelling patterns. I did this, as well, because
Glushko’s word lists march inexorably on and are still being used today to
confirm or disconfirm the validity of influential computer models of
reading.
 297

More Attempts to Pin Down Orthographic Structure

The Many-Word Problem

Glushko’s work led to an explosion of research on this topic, and there is
no space to cover this here. Here, I want to deal briefly with the fate of the
orthographic rime and then review more recent discoveries that show
what is really going on.

Juel and Solso Glusho’s article was followed by a more quantitative at-
tempt using bigram (letter-pair) frequencies in a series of studies by Juel
and Solso (Solso and Juel 1980; Juel and Solso 1981; Juel 1983). They
provided an informative and thoughtful analysis of the problem, though
this was largely a failed attempt because bigram frequencies are a very bad
fit to the spelling code. In 60 percent of the words in their lists, at least
one bigram unit was discordant with the word’s phonology.
Juel and Solso’s analysis did go some way toward identifying impor-
tant structural elements of the spelling code. They identified two types of
information contained in the spelling patterns. Their language has been
changed to some extent to be consistent with the use of terms in this book.

1. The probability (or number of occurrences) with which a specific letter

pair (bigram) appears in specific word positions in a syllable, estimated on a
large corpus of words. They called this orthographic redundancy.
2. The probability with which a letter pair appears in any position in a
large number of different words. They called this orthographic versatility.

These categories tie letter sequences to positions within words, but do not
specify what the reader does with them. This led to a third orthographic
category:

3. The probability with which a particular letter pair represents a partic-

ular phoneme or phonemes. Using their terminology, this would reflect
‘‘orthographic consistency,’’ but they called this coding or decoding.

As can be seen, this is essentially a visual logic, and does not include the
probability structure of the spelling code from the phoneme out. Thus,
there are two missing variables on the list. One is the frequency of occur-
rence in print of a particular spelling for a particular phoneme. The other
 298

is the probability or likelihood that spelling alternatives for the same pho-
Chapter 10

neme occur in a large corpus of words. This problem was never solved in
future research, and remains unsolved as this book goes to press.
There were several interesting findings in this work. Juel (1983)
reported that the probability that a bigram consistently represents a pho-
neme was much more important for beginning readers (second graders).
Orthographic versatility was more important to older children (fifth
grade) and adults, and they read ‘‘versatile’’ letter pairs (bigrams that ap-
pear in lots of positions in lots of words) much faster. This result seems
counterintuitive, because one would expect that positional knowledge of
spelling patterns (chat, catch) would be extremely important and should
improve with age, and not that positional stability is less beneficial. How-
ever, this may be an artifact of the bigram technique and of their word
lists.
After investigating the instructional implications of these findings,
she reported that first graders taught by a letter-to-sound method learned
the letter-sound relationships faster for ‘‘versatile’’ letter pairs (those that
appear in many slots in words) than for nonversatile letter pairs. Juel
(1983, 325) had this to say:

There is a possible instructional implication that emerges from the current

study. A real advantage that adult readers have over children is their sensitivity
to versatile letter combinations. Versatile letter combinations appear to be
more quickly processed both visually and phonologically. Presenting children
with large numbers of words that contain similar letter combinations would
seem facilitative to developing a set for versatility.

Versatility represents a simple frequency measure, the fact, for example,

that the letters ad appear in a lot of words in lots of different positions.
However, children should never be taught letter pairs unless they are
legitimate digraphs. Tailoring reading instruction to words representing
certain classes of bigrams would be extremely counterproductive because
they so strongly misrepresent the phonetic nature and structure of our
spelling code.

Zinna, Liberman, and Shankweiler Zinna, Liberman, and Shankweiler

(1986) were the first to study children using Glushko’s approach. The
 299

children were asked to read real and nonsense words, and were scored for

The Many-Word Problem

accuracy, not speed. Zinna and colleagues provided a thorough analysis of
how the children decoded the vowel digraphs on each word on their test,
which tells us more than average scores and significance levels do. As in
the Glushko study, words were assigned to ‘‘consistent’’ and ‘‘inconsis-
tent’’ neighborhoods on the basis of the VC rime: beach was consistent
because the spelling each is likely to be decoded ‘‘eech’’; head was incon-
sistent because ead can be decoded ‘‘heed’’ or ‘‘head.’’
The results showed that children are just as idiosyncratic as adults in
how they respond to ‘‘rime consistency’’ and ‘‘neighborhoods,’’ indicating
that orthographic rimes have little connection to what the brain is doing. I
report here on the nonsense words, because there were ceiling effects on
real words.
Some neighborhoods were more consistent than others. The consis-
tent oo group mooth, looth, troom, poom, shoon, smoon, woon behaved in an
orderly way, with nearly everyone opting for the /oo/ vowel sound. The
nonwords ending each, ean, eam were consistently read as /ee/. But none
of the remaining rimes in the ‘‘consistent’’ group were read consistently.
The children decoded the rimes oup, oung, oud with five different vowel
sounds. Children read the three oup words (the soup neighborhood) as
‘‘oop’’ 30 percent, 50 percent, and 73 percent of the time, depending on
the word—not remotely consistent. Decoding was unstable for words with
the rimes ield, ief, and iece as well.
On the other hand, many rimes in the ‘‘inconsistent’’ group produced
highly consistent responses, most children opting for /ool/ in bool and
^ as in wool ). Preferences were strong (around 90 percent)
smool (not /ool/
for reading eak words as /ee/ and not as /ae/ (steak). Children were split, as
expected, on the ow words, reading them as /ou/ (cow) or /oe/ (low) equally
often.
In reviewing these words, I can confirm that the orthographic pat-
terns fit the authors’ intentions. The variability in responding was due
to the children and not to the word lists. Some children seemed to be
decoding by analogy to a word family, others by analogy to CV initial
letters (chead was read as an analogy to cheap and not head 77 percent of
the time). Some nonwords contained whole real words, which may have
encouraged the children to seek out little words inside bigger words: zoo in
zook and moo in mook won the battle with the powerful ook, family whose
 300

only enemy is spook. The cow in cown and how in hown may be responsible
Chapter 10

for the children’s preference for /ou/ versus /oe/ readings, which garnered
nearly 100 percent of responses.
In analyzing these responses, Zinna and colleagues concluded that
‘‘the influence of the initial segment appears to account for most of the
variability’’ (p. 474). They suggested that because people read from left
to right, the first sequence activates a word: prou ¼ proud instead of
proup ¼ soup. But this explanation does not hold up consistently, not even
for proup, which was read proop 50 percent of the time. In short, children’s
decoding did not obey either the CV onset or the VC ‘‘rime,’’ and seemed
to involve the whole word.

Treiman and Colleagues The first attempt to provide a quantitative ac-

count of the consistency with which all orthographic units represent
language did not appear until 1995. This was the work of Treiman and
her colleagues (Treiman et al. 1995). The authors stated that while the
English writing system is alphabetic, there are structural redundancies in
its orthography. It was their belief that the most important structural ele-
ment was the VC unit or ‘‘rime’’:

We argue here that a consideration of orthographic and phonological units

that are larger than single graphemes and single phonemes can shed new light
on the nature, use, and acquisition of the English writing system. Specifically,
we claim that letter groups that correspond to the rimes of spoken syllables, or
units that include the vowel and any following consonant, play an important
role in adults’ and children’s pronunciation of printed words. (p. 107; original
emphasis)

And they were quite clear about this direction of the relationship: ‘‘The
consonant that follows the vowel helps to specify its pronunciation’’
(p. 108). They referred to this as a ‘‘special dependency.’’
These statements are troubling, because the goal is so circumscribed,
suggesting a lack of objectivity in the quest to classify orthographic
structure.
The article was in three parts. The first was a tabulation of the fre-
quency count of the ‘‘neighbors’’ of the orthographic units: C, V, C, CV,
VC in 1,329 CVC words. These units were analyzed correctly, in that
 301

each phoneme could be represented by one to four letters: sit, soap, sight,

The Many-Word Problem

sought.
The tabulation proceeded as follows. Each letter, digraph, or pho-
nogram in each position in the 1,329 CVC words was given a value ac-
cording to whether that letter—in the same position in all comparison
words—was pronounced alike or differently in a much larger corpus of
multisyllable words. A ‘‘neighborhood’’ (the same letter or letter se-
quence) was quantified according to its consistency, the frequency with
which this occurred.
Pronunciations that matched were ‘‘friends,’’ and those that mis-
matched were ‘‘enemies.’’ Each of the orthographic units was scored for
the number of friends relative to the number of friends plus enemies by
the formula F/F þ E. This provides a single value for ‘‘neighborhood consis-
tency’’ for all letter units in each set: C, V, C, CV, VC.1
The tabulated data showed that the most consistent reading of any
letter in a CVC word is the initial consonant (95 percent consistent), fol-
lowed by the final consonant (92 percent consistent). The VC unit was
more consistent than the CV unit (80 versus 55 percent), and more con-
sistent than the vowel alone (62 percent). These values are for the proba-
bility of the spelling in all words in the corpus. The values for frequency
of occurrence in print were somewhat lower for all units involving vowels,
but single consonants maintained the same high values. When the number
of different pronunciations and their probabilities across all monosyllabic
words in the corpus were calculated, the most regular (predictable) were
the final consonant and the VC unit. Next came the initial consonant and
the CV unit. The vowel pronunciation was least predictable.
Treiman and colleagues drew the following conclusions from this
effort:

1. This was a formidable undertaking, because orthographic units had to be

‘‘decoded’’ to be compared. The pronunciation determined how each item was
scored, and this cannot be done by computer. However, there was no ex-
planation about how this critical process was carried out, nor any reliability
checks on the coding process. On the one hand, this effort appears highly ob-
jective (frequency counts), and on the other hand, there is an unknown degree
of subjectivity that was not controlled.
 302

The results for individual graphemes paint a rather bleak picture of the En-
Chapter 10

glish writing system. . . . English is not very regular. For vowels especially,
a single grapheme often maps onto several phonemes. . . . If we incorporate
large orthographic and phonological units into our description of the English
writing system, however, the picture becomes more encouraging. The pro-
nunciation of orthographic units that contain a vowel grapheme and a final con-
sonant grapheme are more consistent than the pronunciations of single vowel
graphemes. (p. 112; emphasis added)

There are two problems with this and the statement above. First, it is no
more likely that the VC consistency is due to the consonant controlling
the vowel pronunciation, as the authors claim, than for the vowel to con-
trol the consonant spelling (co-occurrence is not cause). Second, these
1,329 common CVC words, constitute only .07 percent of all words in a
college dictionary of 200,000 words, scarcely a sufficient number of words
to advocate incorporating larger orthographic and phonological units in
descriptions of our writing system. Finally, there is no evidence for, and
considerable evidence against, the assumption that people’s behavior will
follow this particular statistical pattern in these particular words, as earlier
results have shown. The obvious next step was to find out if it did. Stu-
dents from two universities (27 from Wayne State and 30 from McGill)
were asked to read the 1,329 CVC words as quickly and accurately as
possible.2
I will present the results that were consistent between the two student
groups. Unless results can be generalized across similar populations, they
have no validity. Of the 42 variables entered into a regression analysis,
only those shown in the accompanying table accounted for significant

2. Reaction time (voice-onset time) and error scores were averaged across
subjects for each word, for the two groups of students separately. Thus, there
were a total of 1,329 reaction-time scores, one for each word. These were used
in a multiple-regression analysis to look at the relationship between the speed
involved in saying a word and the various properties of that word. In effect,
‘‘words’’ were substituted for subjects in the statistical analysis, and all the
‘‘variance’’ (variability) came from the words and not from the students. This
is a very unusual procedure.
 303

amounts of variance for both the Wayne State and McGill students. These

The Many-Word Problem

values represent unique variance for each variable, with shared variance
controlled.

Wayne State McGill

Word familiarity 3.8% 1.3%
Word frequency 2.2 .3
No. of letters in C, V, C (summed values) 2.7 1.4
Consistency: initial consonants 2.2 .9
Consistency: VC .8 1.3
Variance due to producing the initial 5.2 22.8
phoneme

The table illustrates two things. First, the figures in the bottom row reflect
a serious problem with the technology used to measure voice-onset la-
tency. Variation in the physical act of producing the initial phoneme in a
word accounted for the largest amount of variance in voice-onset time and
was quite different for the two groups. In essence, this is a huge source of
noise in the data. As such it should have been subtracted or set aside, yet
the authors added it in their analysis.
Second, whatever it is that accounts for the speed it takes to read a
word, this study has not found it. Word frequency or familiarity will not
play a big role in college students’ ability to read high-frequency, simple
CVC words, and we see that it does not here. But even so, it accounted for
more variance than anything else that was measured. The number of let-
ters per phoneme and the consistency rating for the initial consonant both
accounted for more variance than the VC rime, which accounted for less
than 1 percent in the Wayne State data and 1.3 percent for the McGill
students. Although these tiny values were ‘‘significant’’ due to the fact that
the 1,329 words acted as ‘‘subjects’’ in the study, they have no practical
significance.
Treiman and colleagues reached the following conclusion:

Readers appear to have picked up the statistical regularity documented in Part

1, that VC units are fairly reliable guides to pronunciation. Given a word
 304

whose VC had a single pronunciation, people pronounced the word relatively

Chapter 10

quickly and accurately. When the VC had multiple pronunciations, perfor-

mance was comparatively poor. This was a robust finding, emerging in all of
the regression analyses for both the Wayne State and McGill participants. . . .
Together with the effect of VC consistency, this finding suggest that, at
least in the speeded naming task, printed words are processed to a large extent
in terms of orthographic units that correspond to onsets (C units) and ortho-
graphic units that correspond to rimes (VC units). (pp. 121, 122)

The data do not support this extreme position.

The final section included a study of children in first, second, third,
and fifth grades. Word lists were set up on the basis of high or low CV
and VC consistency scores (proportions of friends and enemies). There
were four kinds of lists: (1) words with high-CV/high-VC consistency
scores; (2) words low on both; (3) high-CV/low-VC; (4) low-CV/high-
VC.
Treiman and associates predicted that children would make fewer
errors on words with highly consistent VC rimes, because, according to
the theory, the final consonant is the primary cue to how the vowel is
pronounced in these words. The data appeared to confirm this, because
words in the ‘‘high-VC’’ lists were read significantly more accurately by
first, second, and third graders. Fifth graders were equally accurate on all
four lists.
My analysis of the word lists showed that something quite different
was going on. In many words, the vowel controlled how the consonant was
spelled in VC units (watch, beach), or the initial consonant controlled how
the vowel was spelled ( /b/ and /p/ take the u spelling for /oo/, as in bush,
push). Words like these were scattered throughout the lists and did not fit
the designations used by Treiman and colleagues. In a large number of
words, there was no relationship between any of the phonemes or gra-
phemes in the word, which is typical of the ‘‘simple-vowel’’ CVC words
on their lists: bob, mob, thin, gum, gun, wet, met. In perfectly ‘‘regular’’
CVC words like these, any letter can be swapped out, and this does not
affect how the other letters are decoded.
I carried out a thorough analysis of all the words on these lists (see
appendix 2) and found only 2 words out of 60 where the final consonant in
a VC rime definitely influenced how the vowel is decoded: ball and tall.
 305

In all other cases there was little or no connection. This means that the

The Many-Word Problem

‘‘significant’’ advantage in reading accuracy for ‘‘consistent’’ VC rimes is
an artifact of the words on the word lists. What might this factor be?
It turns out that the main reason children made more mistakes read-
ing low-consistency VC words was because these words were considerably
harder to read. They were longer (had more letters), and they were
spelled with less common spelling alternatives: bear, shall, none, ton, won,
cough, and rough. Four words on one list used the Old French spelling se
for the phonemes /s/ and /z/, and the vowel could be decoded in two
ways: chose, dose, pose, lose. It is not surprising that children had more trou-
ble reading these kinds of words. If these anomalous spellings cause a
decrease in reading accuracy, is this really because the ‘‘rime’’ is inconsis-
tent? Can the absence of something be a cause?
So far, there are no satisfactory measures for studying which ‘‘ortho-
graphic’’ patterns are processed by the brain to help retrieve information
in order to decode words. These measures are not only unsatisfactory,
but misleading, especially when the interpretation of the data tends to be
biased.

Computer Models of Reading

For reasons unknown, Glushko left off his study of the statistical prop-
erties of orthographic patterns and went on to other things. As things
have progressed, perhaps this was a wise decision. Nevertheless, Glushko’s
fundamental questions are still important since computer models of read-
ing have started to come of age. The current problem is not so much
whether a computer program can learn to read, but which of the various
models fits the way humans read. Unless the model behaves like people
do, one cannot infer that the computer is doing anything remotely ‘‘brain-
like.’’ After all, a computer can play chess by running through every pos-
sible move, ten moves in advance, in the time it takes to say nanosecond, but
this is not what chess masters do.
For historical reasons (one assumes), the major players in parallel-
distributed or neural-network models of reading (Seidenburg and Mc-
Clelland 1989; Plaut et al. 1996) have relied on Glushko’s original logic
and have attempted to validate their models on tasks identical to or similar
to Glushko’s, even to the point of using Glushko’s original lists of words.
We have already seen that those lists are problematic. After a decade or
 306

so, one would imagine that everything had been nailed down, and that all
Chapter 10

the head scratching that Glushko suffered through was over and done
with. Not so.

A Can of Worms Gets Some More Worms

In 1990, Jared, McRae, and Seidenberg published what was intended to be
the definitive study to get this sorted out, especially in view of the fact that
so many people were finding it difficult to replicate Glushko’s results. At
the same time, they were interested in finding reliable behavioral support
that a model based on strict grapheme-phoneme correspondence rules was
untenable. Orthographic patterns were important, but what were they?
One of the central problems was that the computer models of ‘‘be-
havior’’ (referring to Seidenberg and McClelland 1989) did not mimic
human behavior and the tests designed to measure it, such as Glushko’s.
The computer is given 3,000 words to learn to read, along with some ini-
tial ‘‘training’’ on basic letter-to-sound correspondences. The computer
gradually organizes ‘‘patterns of activation’’ encoded by a set of weights
(interneuron substitutes). These eventually link a spelling to its most
probable decoding based on common patterns (statistical redundancies)
that appear in the entire corpus of 3,000 words. As the computer learns,
it gets feedback on each trial for whether it did or did not print out the
correct phonetic rendering of the word. The outcome is fed back into the
system, and the weightings are modified. Various mathematical models
have been tested, the most recent outlined by Plaut et al. (1996).
The difference between these models and behavioral measures is that
all relevant words contribute to an outcome, plus the data are continuous,
precisely weighted values and constantly updated. By contrast, behavioral
research relies on discrete categories of words in which there is competi-
tion between inconsistent and consistent neighbors, or on other variables.
In the computer models, ‘‘neighbors’’ (friendly or unfriendly) cooperate to
a supply pool of information. Thus, ‘‘The weights come to encode facts
about the consistency of spelling-sound correspondences in the training
corpus. The models’ performance is evaluated in terms of a phonological
error score that indicates the discrepancy between the computed pho-
nological code and the correct code’’ ( Jared, McRae, and Seidenberg
1990, 689). The two most important points about this statement include
the computation of consistency (how often a match occurs or does not
 307

occur), and the fact that phonology is wrapped into the process—a process

The Many-Word Problem

that serves the function of integrating letter patterns and speech sounds.
There is no need nor reason, therefore, for ‘‘dual routes’’ or different
‘‘pathways.’’
The work of Brown (1987) had seemed to show that the largest
impact on decoding speed was not consistency of the ‘‘rime’’ but the fre-
quency with which orthographic rimes appeared in print. Jared and asso-
ciates suggested that perhaps the notion of consistency is correct, but that
this will only appear in certain kinds of words, having something to do
with the ‘‘neighborhood size’’ of enemies relative to the neighborhood size
of friends. For instance, if an orthographic rime has few enemies that are
low frequency in print, there will be little interference in decoding the
word. On the other hand, if the rime has a lot of enemies all of high fre-
quency in print, decoding will take longer. It must be stressed that by
neighborhood ‘‘consistency,’’ Jared and colleagues meant the orthographic
rime only.
In a series of four experiments, Jared, McRae, and Seidenberg more
or less ‘‘proved’’ the theory outlined above, or at least thought they did.
But this theory held for a limited corpus of words, with far too many
constraints.3
The results showed that the ‘‘consistency effect’’ was isolated to words
with orthographic rimes that had lots of enemies, but it failed to appear
when enemies were few. The investigators believed they had settled the
dilemma that had plagued this research for so long: the fact that the

3. In the definitive experiment, the following constraints had to apply for the
‘‘orthographic rime theory’’ to work. Words were selected that fit two main
categories: words with high-frequency friends (rime segments that appear
often in a lot of words) and words with low-frequency friends. They were
divided further into words with consistent and inconsistent rimes (one possible
decoding versus more than one). The inconsistent words were split further
into words with high-frequency enemies (rimes with different pronunciations
that appeared often in print), and those with low-frequency enemies (infre-
quent in print). Each word was then matched to a control word on everything
except consistency. Also controlled were a variety of other measures such as
word frequency and word length.
 308

frequency in print of the orthographic rimes had not been controlled.

Chapter 10

They argued that these results provided a strong refutation of a dual-

route model and GPC rules, because a rule-based system cannot explain
the impact of the ‘‘rime’’ consistency and frequency effects on word-
recognition speed.
Further experiments added some nuances. Jared and colleagues at-
tempted to tease apart the impact of the visual orthographic rime and the
phonological rhyme on the ‘‘consistency effect.’’ Results showed that the
consistency effect was found only when the students responded verbally
(not manually), leading the researchers to conclude that the consistency
effect was due to phonological factors. It is not clear what this means.
Structural redundancies are certainly visual as well; otherwise deaf chil-
dren could not use them to read and spell.

More Worms
We need to put Jared, McRae, and Seidenberg’s results in perspective,
especially because they underpin the rationale for the latest computer
models on reading from this group (Plaut et al. 1996). It is insufficient to
argue merely that these studies did or did not prove a consistency theory
about word-recognition speed and accuracy, because all the constraints
that made this statement possible should be part of the equation. An ac-
curate statement of what happened in this study follows.
Given words of equal frequency in print, all one syllable long and
containing the same number of letters, the same bigram-frequency count,
and the same orthographic-error score, consistent words (rimes pro-
nounced only one way) are read faster and more accurately than inconsis-
tent words (rimes pronounced more than one way) only if the inconsistent
words have enemies seen frequently in print. This effect is most marked if
the frequency of friends is low as well. This result is not affected by the
simple tally of friends or enemies.
There are two ways to look at this outcome. Here is the authors’
version, presented in light of the fact that they controlled all the relevant
variables listed above: ‘‘It is very likely that the observed effects are due to
the one stimulus property that systematically differed between groups:
consistency of spelling-sound correspondences’’ (p. 707).
I was more struck by the fact that as the controls kept mounting, the
corpus of words became so small that the investigators had to repeat word
 309

lists twice to have enough words to do the study. Thus, even if these ef-

The Many-Word Problem

fects are robust and replicable, are they relevant when they pertain to a
very special group of words constituting a minute fraction of all words?
Jared and colleagues attempted to link these results to computer
models of word recognition, especially that of Seidenberg and McClelland
1989. This model has no representations for individual words. Instead,
properties of the model give rise to consistency effects:

Words that are similarly spelled and similarly pronounced (e.g., rhyming
neighbors such as FEAT and TREAT) have similar effects on the weights:
therefore exposure to one word improves performance on the other. . . .
Words that are orthographically similar but phonologically dissimilar have
mutually inhibitory effects: training on TREAT has a negative impact on the
weights relative to GREAT and vice versa. The net effect of the entire en-
semble of learning experiences is poorer performance on inconsistent items
compared to entirely consistent ones. ( Jared, McRae, and Seidenberg 1990,
709)

Is reading really about patterns of reciprocal activation between ortho-

graphic rimes and phonological rhymes? There is no convincing evidence
so far that it is. And Jared and associates introduced this little conundrum
near the end of their article:

Throughout this paper we have assumed, as others have before us, that pro-
nunciation is largely determined by properties of neighborhoods defined in
terms of word bodies. . . . However, there are some words in English that have
word bodies found in no other words (e.g. SOAP is the only OAP word). . . .
Performance on SOAP, for example, may be affected by exposure to words
such as SOAK and SOAR. Therefore, our studies should not be taken as
indicating that only neighborhoods defined in terms of word-bodies are rele-
vant to naming. (p. 711)

But what if there is a simpler explanation? What if the word soap is read

s
/ / / oe/ p
/ /

For some reason the alphabet code has gone astray in this research!
 310

There is a far simpler explanation of Jared, McRae, and Seidenberg’s

Chapter 10

results than the one they provided. This has to do with the fact that the
words in their consistent and inconsistent lists contained different vowels
and unequal ratios of the same vowels. When words on these two lists are
balanced for counts of the same vowel sounds, and their spellings are
examined, rime-based explanations fall apart, a problem similar to the one
with Treiman et al.’s word lists.
My analysis was extensive, and I provide one small example here,
using the words from experiment 3 ( Jared, McRae, and Seidenberg 1990,
698). The words on the consistent and inconsistent lists had only 6 vowels
in common out of a total of 19 vowels. The vowels on the consistent lists
were more likely to be ‘‘simple’’ (short) vowels: /a/, /e/, and /i/. That is,
they tended to be the vowels that had the fewest spelling alternatives and
that were the easiest to decode. As to what this has to do with rimes or
rhymes, the answer is nothing. But we can probe deeper. The alleged
function of the orthographic rime is that the final consonant provides a
clue to the pronunciation of the vowel sound. Did the final consonant
help in these word lists?
We can take as an example the /oe/ sound, which was used the most
on any list, for a total of nine words. Here is a list of the /oe/ words from
the ‘‘consistent’’ list on the left, with the total number of words that are
read consistently ( friends) in brackets. They are paired with words on the
‘‘inconsistent’’ list—that is, words with different pronunciations of the
same orthographic rime (enemies), with the number of enemies in brackets.
Counts are based on a corpus of 3,000 words.

Consistent words Inconsistent words

dome (4) some (2)
foes (11) shoes (1)
prone (9) gone, one (5)
wove (8) move, shove (7)
comb (1) tomb, bomb (3)
dough (2) tough, cough, through (5)
dose (3) nose (6)
 311

If the orthographic rime (VC, VCC) is relevant to decoding, the final

The Many-Word Problem

consonant(s) must affect the pronunciation of the vowel, and must do so
by a statistically meaningful value. Only one rime on the list has odds in its
favor sufficient to cause it to be read faster. Reading foes as foze has an
11:1 probability over fooz. But remember Glushko? He found that the
word shoes was read nearly 100 ms faster than the high-frequency word
goes. If goes cannot conquer shoes, certainly foes cannot either. (There is no
way to know what happened in this study, because response times were
not reported.) When I carried out this type of analysis on all words in the
lists where vowel sounds overlapped, final consonants had no effect on how a
vowel was read. Jared, McRae, and Seidenberg’s results are therefore an
artifact of the particular vowel sounds in the words and their particular
spellings.
There is not the slightest evidence from any of these studies that the
orthographic rime is an important structural pattern, so it comes as no
surprise that it vanishes without a trace in a new computer model.

The Orthographic Rime Gets Lost in Attractor Space

In 1996, Plaut et al. (Seidenberg and McClelland are coauthors) provided
a major update on their original model (Seidenberg and McClelland
1989). In particular, I want to discuss the results for the third model in a
series of four, a model described as an ‘‘attractor network.’’ (The first and
second models are most like their original model, and the fourth model
included a semantic component.) An attractor model is an advance on
the other models because it allows for feedback within the network,
which Plaut et al. called ‘‘recurrent connections,’’ a feature of real neural
systems.
They described it this way:

Processing in a network is interactive when units can mutually constrain each

other in settling on the most consistent interpretation of the input. . . . A
common way in which interactivity has been used in networks is in making
particular patterns of activity into stable attractors. In an attractor network,
units interact and update their states repeatedly in such a way that the initial
pattern of activity generated by an input gradually settles to the nearest at-
tractor pattern. (p. 82)
 312

An ‘‘attractor’’ essentially represents a point in a ‘‘multidimensional state

Chapter 10

space’’ in which, at any given time, the pattern of activity across all units
engaged in processing corresponds to this single point. This is similar to
a ‘‘node’’ in a network that contains all the information about particular
input-output relationship. Furthermore, ‘‘A set of initial patterns that
settle to this same final pattern corresponds to a region around the at-
tractor, called its basin of attraction. To solve a task, the network must
learn connection weights that cause units to interact in such a way that the
appropriate interpretation of each input is an attractor whose basin con-
tains the initial pattern of activity for that input’’ (p. 82). Whether neural
networks behave like attractors or work by superposition and Fourier
analysis/synthesis (Pribram 1991), or both, is a problem for the future.
What is interesting now is how well the attractor model mimics human
performance. For one thing, it can ‘‘decode’’ nonwords never seen before
just as well as humans do. This had been a failing of the earlier models.
Another bonus of this model is that it can be directly compared to the
reaction-time data. Previous models could only be compared to error
scores.
Two properties of the model are of greatest interest here. The first is
that the mutual redundancies of orthographic ‘‘rimes’’ and phonological
‘‘rhymes’’ are mapped together in an attractor basin. The authors explain
it this way: ‘‘The reason is that, in learning to map orthography to pho-
nology, the network develops attractors that are componential—they have
substructure that reflects common sublexical correspondence between or-
thography and phonology. This substructure applies not only to most
words but also to nonwords, enabling them to be pronounced correctly’’
(p. 83; original emphasis).
The second property of interest is that an attractor model can be sys-
tematically and continuously degraded until its ‘‘boundary state’’ for
mapping a particular feature(s) of the input begins to collapse. This allows
for a direct measure of each feature’s robustness, as well as of which fea-
tures work together to produce a correct outcome (the correct pronunci-
ation of a word).
The model (a computer program) was set the task of learning how to
read 3,000 words. These words varied in things like neighborhood con-
sistency (friends/friends þ enemies), word length, number of phonemes in
the word, and a measure of word frequency. The results for the time it
 313

took to learn to read a word showed that all four factors accounted for

The Many-Word Problem

unique variance: 10 percent for word consistency, 5.6 percent for log fre-
quency, .8 percent for orthographic length, and .1 percent for phonologi-
cal length. But this analysis did not reveal what orthographic features
created the consistency effect.
To look at the orthographic features, the investigators reduced the
activity in the ‘‘network’’ of each feature separately and in combination
until the system started to misread words. This was precisely quantified as
a ‘‘boundary value,’’ which ranges from 0 to 1. Zero means that a letter or
letter combination (digraph or consonant cluster) is relevant to the pro-
nunciation of the word, but not relevant to the pronunciation of adjacent
letters; 1 means it is absolutely essential to decoding other letters. This
provides a pure measure of orthographic probability. The system was
tested on the four types of words (see figure 10.1).
For regular consistent words (words that are consistently decoded one
way), the units corresponding to initial consonant(s), vowel, and final con-
sonant(s) were largely independent, which, as Plaut et al. (1996, 88) put it,
‘‘indicates that the attractor basins for regular words consist of three sep-
arate, orthogonal sub-basins (one for each cluster).’’ As already noted,
letters in CVC words with ‘‘simple’’ vowels are ‘‘context free.’’ Any letter
can be swapped without affecting how the other letters are pronounced
( pot, top, dot, pod, pit, tip, and so on). Thus, the model confirms what is
known about the spelling code at this level.
As words became less ‘‘regular’’ (regular/inconsistent words), the units
still remain independent, except the vowel needs some minimal support
from both initial and final consonants. In the two remaining types of
words, the vowel must have mutual support from both consonants to be
recognized. This shows that the initial and final consonants have two values
in ‘‘attractor space,’’ one for their independent contributions and another
for their contributions to a correct decoding of the vowel. For ambiguous
words, the vowel had a higher boundary value than the supporting con-
sonants, showing the vowel was a stronger cue to how it was decoded. But
for the exception words, the supporting consonants had higher boundary
values than the vowel, showing they are critical to decoding it.
There are two important points here. First, this analysis supports my
observation that Jared, McRae, and Seidenberg’s results were largely due
to the different kinds of vowels on the lists, which would grossly alter
 314
Chapter 10

Figure 10.1
The amount of activity in the attractor network (boundary state) for initial consonants (onset),
vowel, and final consonants (coda) necessary for the computer to read words correctly. In words with
regular spellings (trip) all phonemes are context free (decoding is not influenced by surrounding spellings).
In words with irregular spellings (dead/read, break/beach) the vowel cannot be decoded independently of
the surrounding consonants. From Plaut et al. 1996. Copyright APA. Reprinted with permission.
 315

these sequential dependencies. Second, even though these words were

The Many-Word Problem

chosen for the presence or absence of consistent VC rimes, the VC unit
played no more important role than the CV unit. Both the initial and final
consonant(s) were necessary to the same degree in helping decode the
vowel in words where it was ambiguous. There is no basis in fact to the
notion that the VC rime unit makes some special contribution to decod-
ing. It contributed far less than the independent contributions of isolated
initial and final consonants.
These results confirm my analysis of the word lists presented in this
chapter, which led to the same conclusion. This analysis was based on
the orthographic structure of the spelling code, not on the orthographic
structure of the ‘‘reading code.’’ This investigation clarifies key theoretical
problems and raises fundamental questions for future research:

1. Glushko’s original insight that all the letters in a word contribute to a

statistical template of orthographic redundancy was mainly correct except
for a certain class of CVC-type words.
2. If the attractor network was taught the structure of the spelling code
directly (sound to print), would it learn to read more quickly? Would it
learn to read at all? (Humans would.)
3. Can a network trained entirely on the ‘‘reading code’’ also spell? This
would be an interesting test of the model, because humans are inconsistent
here.

These studies provide strong support for a third way to decode print in
addition to knowledge of phoneme-grapheme correspondences and rote
visual memory. This third way is a gift of the brain’s capacity to automat-
ically encode the statistical redundancies of any kind of pattern. The third
way was first proposed by Glushko, and has been operationalized mathe-
matically in research by Plaut et al. Aaron and colleagues provided further
proof in their discovery that deaf children rely on visual statistical redun-
dancies to read and spell.
We have come full circle. Chapters 1 and 2 showed that knowledge
of the English spelling code is the key to understanding how English-
speaking peoples learn to spell and how they decode (read) orthographic
patterns. This knowledge is as important for designing spelling and
 316

reading programs for the classroom and clinic as it is for behavioral

Chapter 10

research and for designing computer models of the decoding process. In

view of the fact that the classroom research shows we now know how to
design effective reading and spelling programs, we need a new road map
for where to go next. The research questions for the twenty-first century
are entirely different questions from those addressed so far, and it is time
to move on.
 11
NEW DIRECTIONS FOR THE
TWENTY-FIRST CENTURY

It is our good fortune that the National Reading Panel’s fishing expedition
led to such a harvest of treasures. We now have precise knowledge about
how to teach every child to decode, read fluently, and comprehend what
they read. No child needs to be left behind. This is exciting news indeed.
It has been long in coming. We have been held up by two major road-
blocks: the difficulty of unearthing properly controlled research on suc-
cessful programs amidst the vast wasteland of publications on this topic,
and the difficulty of identifying the basic elements of reading instruction
that consistently lead to success. The NRP has gone some distance toward
solving the first problem. And I hope this book will help in solving the
second.
The NRP’s analysis showed that research on reading instruction is
still being cast in the phonics versus whole-word (‘‘reading wars’’) frame-
work. In virtually all 66 ‘‘cases’’ in the NRP database, the control group
was taught with a whole-word method. These methods were identified
in the NRP tables as ‘‘whole word,’’ ‘‘basal,’’ ‘‘whole language,’’ or
‘‘unspecified’’—meaning ‘‘regular classroom program’’ (i.e., whole word).
Few studies contrasted two types of phonics program.
It is time to move on. The goal of this chapter is help us do that. First,
I want to summarize the important discoveries presented in this book, and
second, I would like to make some suggestions for how to move reading
research into the twenty-first century. I will begin with the prototype.
This is the first set of objective guidelines for reading instruction based
on the historical and scientific evidence. They establish the ground rules,
though they are still incomplete. Much more research is needed to pin
down specific details, like the order (if any) in which phonemes should be
taught, how fast they should be taught, what type of lessons work best,
 318

and which materials, special exercises, and activities matter, and which do
Chapter 11

not.

The Prototype: Putting It All Together

Several chapters have addressed the historical and empirical evidence for
how a writing system should be taught, and how an alphabetic writing sys-
tem, in particular, should be taught. The evidence is overwhelming that
whole-word writing systems cannot work, never did work, and never will
work. The evidence is equally clear that all writing systems are based on
one (and only one) phonological unit below the level of the word.
Two principles determine which phonological unit is chosen for a
particular writing system. First, the unit must fit the phonotactic structure
of the language. Second, the choice follows the principle of least effort,
and represents a trade-off between perception and memory. The unit
chosen should be the easiest to hear (segment from the speech stream) but
not impose an excessive memory load. This obeys the following simple
rule: too many units ¼ too many symbols ¼ unmemorable. English has
over 55,000 phonotactically legal syllables. Chinese has 1,270. This is the
reason the Chinese use a syllabary writing system, and we use an alphabet.
These facts rule out all reading programs based on the whole word,
and all reading programs that teach phonetic units not represented by the
code. For an alphabetic writing system, this means syllables, consonant
blends taught as single units, and word families (rhyming endings).
The lessons from the 5,000 years of the history of writing systems are
fully supported by the evidence from the National Reading Panel report.
Whole-word programs are consistently inferior to phonics programs, and
multisound phonics has no advantage over the whole-word programs. Not
only this, but phoneme-awareness training that also includes multiple
phonological units is consistently inferior as a predictor of reading skill to
teaching phoneme awareness alone.
Thus we have two lines of support for the first set of guidelines that
constitute the general prototype. These are reiterated here:

1. Make sure the complete structure of the writing system has been
worked out (or thoroughly understood) before a method of instruction is
developed.
 319

2. Teach the specific sound units that are the basis for the code. (Do not

New Directions for the Twenty-First Century

teach other sound units that have nothing to do with the code.)
3. Teach the arbitrary, abstract symbols that represent these sounds.
These symbols constitute the code.
4. Teach the elements of the system in order from simple to complex.
5. Ensure that the student learns that a writing system is a code and that
codes are reversible.
6. Make sure that encoding (spelling) and decoding (reading) are con-
nected at every level of instruction via looking (visual memory), listening
(auditory memory), and writing (kinesthetic memory).

On the basis of this knowledge, phonics versus whole-word research is

a nonissue and has been a nonissue for 5,000 years. The real and impor-
tant research question is which phonics methods work best? Research in the
twenty-first century should be directed toward the study of the effective-
ness of the various types of phonics programs outlined in chapter 5. First,
we need to compare general types of phonics where it is reasonable to do
so. Second, we need to compare methods within types, such as the differ-
ent forms of linguistic-phonics programs.

Types of Phonics Programs: Which Work Best?

The NRP came up with one list of phonics types under the general rubric
of ‘‘explicit, sequential’’ phonics. I came up with another. Here is my list
once more:

Junk Phonics The practice of teaching aspects or elements of the al-

phabet code in a chaotic, nonsensical, and unstructured fashion. Most
basal phonics falls in this category.

Visual Phonics The short version (a) teaches the 26 ‘‘sounds’’ of the
26 letters of the alphabet. The long version (b) teaches the 27 to 200þ
‘‘sounds’’ of the letters, digraphs, and phonograms.

Whole-to-Part Phonics (also known as embedded, analytic, intrinsic).

The practice of easing children into phonemes by starting with whole
words, then syllables and word parts (word families, blends), then individ-
ual phonemes taught overtly or covertly (embedded).
 320

Multisound Phonics Same as above, but the different-size sound units

Chapter 11

are mixed up and taught together.

Linguistic Phonics Incomplete (a) (called synthetic) teaches from the

sound to the letter. Teaches the 40þ phonemes of English and their main
spellings, plus some spelling alternatives. Complete (b) includes (a) above
plus 136 spelling alternatives.

We can rule out ‘‘junk phonics’’ at the outset based on the evidence
from the NRP report. Children taught with basal-reader phonics gener-
ally did poorly compared to children taught with other types of phonics
programs. We can eliminate whole-to-part and multisound phonics pro-
grams as well, based on the historical evidence and on their dismal show-
ing in the NRP report.
This leaves visual phonics and linguistic phonics. There is consider-
able confusion about the nature of these two types of programs, especially
the distinction between teaching from the letter versus teaching from the
phoneme, and the issue of which one is ‘‘synthetic’’ phonics. The NRP
reading committee, along with most U.S. phonics advocates, either do not
recognize this distinction or do not know it exists.
Let’s clarify the distinction once more. Linguistic phonics anchors the
code in the sounds of the language, and bases initial reading instruction
on a basic code. This avoids several pitfalls created by our highly opaque
spelling system. The code is temporarily set up with a manageable number
of units (40þ phonemes), and this number does not change as the lessons
progress. Children see that an alphabet code works from a finite number
of speech sounds to their individual spellings. They observe that the code
can reverse (all codes are reversible mapping systems by definition). Later,
spelling alternatives can easily be pegged onto the system without chang-
ing the logic: ‘‘There’s another way to spell this sound. I’ll show you some
patterns to help you remember when to use this spelling.’’
Visual phonics does the opposite. Because letters and letter sequences
(digraphs, phonograms) drive the code and phonemes do not, this deprives
the code of any anchor or endpoint. As the number of ‘‘sounds’’ begins to
expand (there are over 250 ‘‘sounds’’ that letters ‘‘can make’’), there is no
way to get back to the 40 phonemes that are the basis for the code. The
child is soon awash in letter patterns and there is no discernible structure,
no limit, no logic, and no pivot point around which the code can reverse.
 321

Thus, the code loses one of its essential properties—its reversibility—and

New Directions for the Twenty-First Century

cannot function as a code. There is no way to teach spelling alternatives,
except as letter-sound alternatives, flipping the logic of the code on its head:
‘‘There’s another way to say this letter.’’ ‘‘This letter makes another sound
here.’’ As these letter sounds proliferate, there is a high risk of confusion
and failure.
We have seen many examples of the confusion created by visual phon-
ics in this book. When Griffith, Klesius, and Kromrey (1992) stated that
children were taught ‘‘90 different individual sounds’’ in the basal-reading
program used in their study, there is no doubt that they were referring to
90 different, individual phonemes in the English language.
Berninger et al. (1998) attempted to teach children to spell using
Venezky’s recommended sequence (see chapter 3), which relies solely
on visual logic. This sequence has 47 spellings representing an unknown
number of ‘‘sounds,’’ and no spellings to represent 7 missing phonemes.
Using this sequence, children were taught with several different methods.
Children learning Venezky’s ‘‘visual phonics’’ did no better on a spelling
test than children taught purely by rote (sight words). And none of the
training groups (regardless of method) did any better on a standardized
spelling test than a control group that was taught nothing!
For these reasons, visual phonics leads to the unanswerable questions
posed by the NRP, such as: How many letter sounds should be taught? If
this was a legitimate question, the proper answer should be all of them.
We know from the analysis of the spelling code in chapter 3 that there are
40 basic-code spellings, 136 remaining common spellings, and 80 high-
frequency sight words with unusual spellings. But in visual-phonics logic,
this would mean there are 256 common sounds in the English language
represented by 256 spellings. And if one was a purist and wanted to in-
clude Venezky’s tallies on the outer reaches of the code, we could add
100þ more rare letter sounds.
Framed in visual-phonics logic, the NRPs question would be a genu-
inely formidable one: Which of the 356 decodings should be taught?
Framed in linguistic-phonics logic (the logic of an alphabet code), this
question is meaningless.
The NRP attempted to answer its question as follows: ‘‘It is clear that
the major letter-sound correspondences, including short and long vowels
and digraphs, need to be taught’’ (National Reading Panel, 2000, 2-135).
 322

But what is ‘‘clear’’—and clear to whom? And what is ‘‘major’’? In the first
Chapter 11

place, English vowels do not vary systematically by length (or by height).

In the second place, which digraphs are they referring to? More to the
point, what sounds of these digraphs are they referring to, given that (in
visual phonics) digraphs have so many sounds? (The digraph ou can stand
^ /u/, /o/, and /er/, as in soup, proud, soul, could,
for /oo/, /ow/, /oe/, /oo/,
young, cough, and journey.)
The visual versus linguistic-phonics issue has another face. There is
an influential and vocal minority that does recognize the ‘‘letter-first’’
versus ‘‘phoneme-first’’ distinction. Nevertheless, this group believes that
children should be taught from letters to phonemes. They argue that pho-
nemes are abstract and transient, whereas letters are concrete (visible and
tangible). From their perspective, it is okay to tell children that letters
‘‘have sounds’’ during early reading instruction, because this will not
matter in the long run. It is not clear why letter shapes would be more
tangible and less abstract to children than the sounds of their own speech.
The danger of vision-first phonics can be seen in any remedial-
reading setting. Many children actually believe that letters ‘‘make sounds’’
and do not have the slightest idea where these ‘‘sounds’’ come from. I
have met many functionally illiterate adults (including some of my college
students) who are amazed to learn that a writing system is a code, and that
this code represents sounds in their speech. There is no need or reason to
take this risk.
In practice, a phoneme-first versus letter-first approach requires a
simple shift in orientation to achieve maximum effect as learning pro-
gresses. The instructional implications are so minor initially that this
is probably why people think they do not matter. In linguistic phonics,
the teacher begins the lesson by introducing a phoneme. The teacher can
produce this or can ask the children to listen for a phoneme featured in a
story or poem. Once the children are familiar with the phoneme and can
produce it, they are shown the letter that represents its most common
spelling. In visual phonics, the teacher points to a letter shape and/or
draws it on the board, and the children are told the sound it ‘‘makes.’’
This is referred to thereafter as a ‘‘letter sound.’’ This shift in logic blocks
understanding of where this sound comes from (that it is part of natural
speech and is one of forty sounds that are the basis for the code), leading
children to believe that the sound comes from the letter.
 323

The differences between visual phonics and linguistic phonics need to

New Directions for the Twenty-First Century

get sorted out, semantically, logically, and empirically. An important first
step would be to compare a visual-phonics method like those advanced by
Venezky or Beck to a linguistic-phonics method like Lippincott or Jolly
Phonics.

Linguistic Phonics: What Works Best?

More fundamental research questions have to do with the linguistic-
phonics programs themselves. We need to pin down the elements that
make a difference. Given their complexity and variety, this is not a
straightforward task. We can begin by establishing what we know for sure
and move forward from there. To do this, here is the final version of the
prototype, which takes into account the classroom observations, spelling
research, and studies in the NRP database. The first two activities in the
list are italicized, because they produced negative or zero correlations
between time on task and reading/spelling test scores in several large-scale
studies.

The Prototype for Teaching the English Alphabet Code

1. No sight words (except high-frequency words with rare spellings).
2. No letter names.
3. Sound-to-print orientation. Phonemes, not letters, are the basis for the
code.
4. Teach phonemes only and no other sound units.
5. Begin with an artificial transparent alphabet or basic code: a one-to-one
correspondence between 40 phonemes and their most common spelling.
6. Teach children to identify and sequence sounds in real words by seg-
menting and blending, using letters.
7. Teach children how to write each letter. Integrate writing into every
lesson.
8. Link writing, spelling, and reading to ensure that children learn that
the alphabet is a code, and that the code works in both directions:
encoding/decoding.
9. Spelling should be accurate or, at a minimum, phonetically accurate (all
things within reason).
10. Lessons should move on to include the advanced spelling code (the
136 remaining common spellings and 80 sight words).
 324

Given that at least five programs reviewed by the NRP follow all or
Chapter 11

nearly all of these guidelines (minus the complete spelling code), the dif-
ferences in the effectiveness of these programs have to do with specific
features such as pacing and depth, special activities, and curriculum mate-
rials. Research varying these details, element by element, would be ex-
tremely time consuming and tedious, but not all elements are in question.
Let’s begin with the definitive proof about what children can learn, at
what age, and in what time period, to provide a baseline or set of criteria
for what can be accomplished.
One important ‘‘rule’’ needs to be stressed. If your goal is to teach the
alphabet code, all components of the prototype must be taught at the same
time, because each component bolsters the remainder. It is counterpro-
ductive to ease children into reading by teaching these elements slowly in
disconnected and unrelated bits.
The youngest children in these studies were in the Jolly Phonics/Fast
Phonics First classrooms in the United Kingdom. They were between 4:8
and 5:0 years old at the start of training. The training period lasted from
10 to 16 weeks depending on the study, and the total number of hours
in whole-class lessons varied from 26 to 60 h. In these training sessions,
children learned the 40þ phonemes, their basic-code spellings, and how
to write every letter and digraph for the 40 phonemes. Children learned
to identify phonemes in all positions in a word, segment and blend pho-
nemes in words, and read, write, and spell common words. Immediately
after training, these children were 8 months above U.K. age norms on
standardized reading and spelling tests. The control groups scored exactly
at age norms. These gains increased at the second and third testings to 1
year above norms, and by the fourth and fifth grades children were 2 years
above norms on tests of decoding. Effect sizes were consistently around
1.0 or higher compared to the children learning via analytic phonics.
Now we have a useful baseline. If children this young can learn to
read and spell 40 phonemes and their common spellings in 10 to 16
weeks, certainly any older child can do the same. We also know from
Lloyd’s account of the development of Jolly Phonics that children are
eager learners and willing to work hard. Because these programs involve
only the elements in the prototype and the learning period is so com-
pressed, children see gains almost immediately. They begin reading words
 325

after the first six sounds are introduced. This is highly motivating, espe-

New Directions for the Twenty-First Century

cially because children start school with the strong expectation that the
‘‘main event’’ is learning to read.1

Jolly Phonics Actions The interesting difference among the U.K. studies
is the shorter training time in the Scottish study (26 hours versus 50 to
60) while achieving similar gains that held up well over time. A large
part of the savings came from omitting the Jolly Phonics ( JP) actions.
On the other hand, Sumbler in Canada reported that of all the phonics
activities they assessed, the time spent learning and using JP actions pro-
duced the highest correlation to reading test scores. The problem here is
that JP actions are confounded with lessons on phoneme-identification as
well as segmenting and blending activities, as they are taught and used
simultaneously. Nevertheless, Sumbler’s data suggest that JP actions
matter, while Johnston and Watson’s data suggest they do not.
Nellie Dale, who developed the first successful linguistic-phonics pro-
gram, warned teachers to ‘‘never teach anything you have to discard
later.’’ Does this warning apply to the JP actions, or might learning them
be helpful even though they will be discarded? This is an interesting re-
search question from a number of perspectives. Lloyd believes that the
action patterns supply a ‘‘motor’’ component in a multisensory type of
learning, and this helps automate learning sound-symbol correspondences.
One could argue, as well, that the action patterns help focus attention on
the differences between the phonemes, increasing auditory discrimination.
(Children must listen carefully to know which action pattern to use.) I
pointed out in chapter 5 that the actions may help the teacher during
whole-class instruction by providing visual signals for who is participating
and who is not, or who is ‘‘getting it’’ and who is not.

Lessons on Speech-Motor Features A similar question applies to the speech-

motor patterns used extensively in the Lindamood program, an idea that

1. At the end of his first week in kindergarten, my son had this tearful com-
plaint: ‘‘I’ve been in school a whole week. It seemed like a hundred years, and I
haven’t even learned to read!’’
 326

originated with Dale. Dale had the onerous task of engaging 70 children
Chapter 11

from the front of the room. She reported that focusing attention on the
parts of the mouth, tongue, and vocal cords that produce each phoneme
makes the phoneme easier to hear and identify, and therefore faster to
learn. It also makes it easier to tell phonemes apart (auditory discrimina-
tion improves).
Unlike the JP action patterns, speech-motor patterns are executed
spontaneously. Because there is no need to learn them, only to be made
aware of them, this will take up little extra time. Dale used these patterns
strictly for auditory support as each phoneme was taught. In the Linda-
mood program, however, the motor patterns form a major part of training
and are taught in depth. Each motor pattern for 44 phonemes is given a
special name, which children have to memorize. For example, a bilabial
plosive (/b/, /p/) is called a lip popper. The voiced plosive (/b/) is called a
noisy lip popper.2
We now have the ingredients for an interesting and important study
on whether these ‘‘add-ons’’ assist learning. A study could be designed in
which there are three groups of beginning readers all taught with the same
linguistic-phonics program. One group learns the basic program only, a
second group learns this plus Jolly Phonics actions, and the third group,
the basic program plus speech-motor patterns (à la Dale). Any benefit
from these additional components would be expected to impact learning
rate—the time it takes to master 40þ phoneme-grapheme associations)
and automaticity—the speed to match a letter to a phoneme and vice versa.
Other important measures, of course, would include reading, spelling, and
phoneme awareness (segmenting, blending).
The automaticity issue is extremely important. Lloyd (1992) observed
that the most persistent differences between children was the time it took
to ‘‘automate’’ the phoneme-grapheme associations. Without automaticity,
children have difficulty blending sounds into words (auditorally or from

2. Memorizing these names and which phonemes they apply to is not easy, as
anyone who has used the Lindamood program can attest. Using them during
lessons is even harder: ‘‘In this word the ‘quiet lip popper’ comes after the
‘little skinny,’ not after a ‘scraper.’ ’’
 327

print), because they lose track of phoneme sequence and which phoneme

New Directions for the Twenty-First Century

came first. She found it was helpful for children to accent the first pho-
neme in the word, causing it to stick in memory and act as an anchor for
the sequence.
Lloyd’s observations on the extensive timeline to achieve automaticity
are borne out in a study on letter-name knowledge and letter-naming
speed. Mann and Ditunno (1990) found it took to the end of first grade
for all children to achieve a error-free performance on naming letters.
However, naming speed (automaticity) still showed high individual vari-
ability and continued to improve. These are surprising results, because
children begin learning letter names from a very young age. This is usu-
ally the first ‘‘reading activity’’ taught both at home and at school. This
study showed that children can take up to 2 years or more to memorize
the 26 ‘‘names’’ of the letters of the alphabet, and it takes much longer to
automate this knowledge. (Since knowing letter names serves no purpose
in learning to read, this is another powerful argument for not teaching
them.)
In view of the individual differences in learning rate, and the long
timeline to achieve automaticity, the children in our hypothetical study
need to be followed for several years. This is the only true test of whether
these extra components matter in the long run. For example, Johnston and
Watson (1997) reported that at 7.5 years, 9 percent of the Jolly Phonics
children scored more than one standard deviation below national norms.
Although this compared favorably with the control group (30 percent of
them below 1 s.d.), one wonders why any child scored this low. Unfortu-
nately, we know nothing about these children. Did they miss lessons due
to illness, or did they transfer in from another school? Did they fall be-
hind because they did not automate sound-letter correspondences at the
same rate as the other children? Did they become confused because this
program does not teach the complete spelling code?

Do Beginning Readers Need Special Phoneme-Awareness Training?

The evidence showed that a separate phoneme-awareness training pro-
gram for beginning reading instruction is largely a waste of time. No
phoneme-awareness program reviewed in chapter 6 had any advantage
over a linguistic-phonics program, either for reading and spelling, or even
for improving phoneme awareness.
 328

There is, however, a good argument for special training in phoneme

Chapter 11

awareness in the clinic. Poor readers have extremely maladaptive decoding

strategies, guessing whole words from the first letter only, assembling
little word parts into something like a word, or refusing to read altogether
(McGuinness 1997b). An ineffective decoding strategy leads to habits that
can be hard to break. It is almost a given that these children (or adults)
have few or no phoneme-analysis skills. Because print can be aversive,
causing anxiety and even panic, initial phoneme-awareness training is
more effective in an auditory mode using blank tiles. A three-step process
is necessary: developing phoneme awareness with blank markers, learning
phoneme-grapheme correspondences, and reading simple (easily decod-
able) text.

Reading Fluency
The research on reading fluency is a curious mix of nonsense and bril-
liance. The vast majority of publications uncovered by the NRP were
devoted to the premise that ‘‘reading a lot’’ increases accuracy, reading
speed, and comprehension. Research (or quasi-research) on this topic
turned out to be so methodologically flawed that only 14 studies out of the
1,000 or so screened by the panel had any merit. These studies showed
essentially nothing. There is a central problem with this premise, because,
without a highly sophisticated research design, there is no way to account
for reading habits, and no way to tell whether children read more because
they are fluent readers, or whether they turn into fluent readers because
they read a lot. The best research on this topic has been provided by
Stanovich and West (1989), and even this study could not solve all the
research-design problems. This study and similar studies are reviewed in
Language Development and Learning to Read.
Fortunately, the ‘‘reading a lot’’ issue is a red herring, because there is
a much better way to ensure that children read fluently, accurately, and
with understanding. This is through the technique of rereading. I want to
provide a brief summary of this work along with suggestions for future
research.
Due to the large individual variation in the rate at which the
phoneme-grapheme associations become automated, a small portion of
children have difficulty decoding at sufficient speed to comprehend what
 329

they read. Advances in research on rereading techniques as a means of

New Directions for the Twenty-First Century

alleviating this difficulty have been enormous, and we know exactly how
to cure this problem. The basic format for this training has largely been
worked out by Dowhower (1987), with the final missing pieces added by
Faulkner and Levy (1994, 1999).
The research just cited is notable for its methodological rigor, some-
thing sadly lacking in the field as a whole. Among other things, good
methodology means asking the right questions. In view of the fact that
good research on reading instruction has barely begun, it is informative to
look closely at the questions all the researchers in the field had to identify
to prove that rereading has a real and long-lasting effect.

Research Questions on the Technique of Rereading

1. Does reading get faster with or without a target goal—that is, a crite-
rion of words per minute?
2. Do children make greater gains reading alone or with an audiotape?
3. Is rereading more effective than reading the same number of different
stories?
4. What is the optimum practice time necessary to produce lasting
changes?
5. How should this time be allocated: concentrated into a few days, or
spread out over weeks or months?
6. How many different stories should be reread to ensure a lasting effect?
7. Does the difficulty level of the text make a difference?
8. Does overlap in the words from one story to the next make a
difference?
9. Does overlap in story content/context make a difference?
10. Does prosody (reading expression) improve along with speed?
11. What kinds of transfer effects are there and how should they be
measured?

Five main goals or outcomes had to be measured for rereading to be

deemed a success. Reading speed should increase with no loss of accuracy
(or an increase in accuracy). Oral rereading should cause an improve-
ment in prosody (melody, inflection). Comprehension should improve.
Most importantly, there should be transfer effects. Reading speed should
 330

increase from one story to the next. If a criterion reading speed is set, this
Chapter 11

should be achieved more quickly with each new story. Accuracy and com-
prehension should improve with each new story as well.
I have provided a list of suggestions, based on the data, to help ensure
that rereading is a success. Here they are again. Target reading speeds
should be set well above the child’s initial reading speed. So far, no one
failed to achieve targets 50 to 60 wpm higher than the child’s baseline. It is
my personal view that targets should be reset each time the child’s reading
speed improves by some critical amount. The final goal should be to have
the first reading of a story at a normal or superior rate for the child’s age. At
this point, rereading exercises can cease.
Children need multiple rereading experiences (across many stories),
not just a few. Dowhower’s time frame of about 7 hours of work over 7
weeks was effective and seems optimal. Time pressure was not excessive,
and sessions were not spread so far apart that there were no carryover
effects. The ultimate goal is the desired target speed on the first reading,
and this goal determines how long the rereading sessions last.
The difficulty level of the text is critically important, because speed
is tied to difficulty level. Very slow readers should start off reading passages
at or just above their reading level, not their grade level. Once reading
speed improves, stories should increase in difficulty. Passages with over-
lapping words are best for struggling readers and very young readers, and
are most likely to produce carryover (transfer) effects from one story to
the next. Overlapping context helps as well. This creates a situation in
which transfer is successful and boosts confidence and motivation to
continue.

Vocabulary and Comprehension Instruction

The last categories covered in the NRP report are vocabulary and com-
prehension. The contrast between the studies on basic reading instruction
and studies on comprehension, in particular, is interesting. New phonics
programs are created all the time by making minor changes in curriculum
detail and emphasis (and given colorful new names by their authors), yet
the basic platforms change very little. As we saw above, phonics programs
reduce to very few common types. Comprehension programs, on the other
hand, are designed from the ground up by individuals who seem to have
never heard of one another’s work. The NRP was so overwhelmed by the
 331

variety of these programs that it was unable to categorize them. Instead it

New Directions for the Twenty-First Century

set them up loosely in 16 types, largely characterized by semantics and not
by anything objective.
Fortunately, none of this matters, because one program is so extraor-
dinary that all other programs (for vocabulary or comprehension) pale in
comparison. I will be summarizing the elements of this program below.
First, I would like to say something about the efforts to improve children’s
vocabularies, especially because teachers spend so much time working on
this, as the classroom observational studies showed.
We know that children rapidly and automatically acquire new vocab-
ulary on the order of around 8 to 10 new words per day. We also know
that children from impoverished environments do not have the same vo-
cabulary skills as children from more affluent homes. It makes sense for
teachers to want to help the children who need special assistance. This was
one of the major goals of Hart and Risley’s study, presented earlier. The
impetus for that study was the failure of Head Start to have any impact on
vocabulary and verbal skills generally. The main question is what is the
best way to approach this problem. Let’s look once more at the bottom
line for vocabulary training.
It is possible to teach vocabulary in the short term, though it took a
lot of research to prove it. McKeown et al. (1985) found that by keeping
the word list short (24 or so words at a time), compressing the learning
time, and providing a rich or in-depth explanation of the words, children
needed about 5 minutes per word to ensure nearly perfect performance on
a vocabulary test for those words. Using a similar format, Jenkins, Mat-
lock, and Slocum (1989) confirmed these results, based on 6.5 minutes per
word. We do not know, however, whether these words ever become part
of productive vocabulary—the vocabulary children actually use. Nor does
this represent a major shift in vocabulary size, because children learn
about 8 words a day just by minding their own business.
It turns out that these vocabulary gains are minor compared to what
children can achieve with proper training in oral and reading compre-
hension, as seen by the brilliant work of Block (1993). Block’s program
had an enormous spin-off for vocabulary skills, even though this was not
a focus of the training. The effect size was unlike anything seen in the
vocabulary-instruction research: 2.62 on a standardized test. In the NRP
studies overall, the effect size for vocabulary gains (all types of tests) was
 332

ES ¼ .97. But this fell substantially when standardized tests only were
Chapter 11

used (ES ¼ .30).

Block’s program is the gold standard for the field as a whole. Most
important, it shows that short-term interventions are not the way to
improve vocabulary and comprehension skills. To make a difference, the
training needs to fit into the scheme of ordinary school lessons, and con-
tinue for months and possibly years. Block’s program took up 3 hours
a week over an entire school year (about 96 hours of classroom time).
Ninety-six hours sounds like a lot of hours, until you consider that this is a
mere fraction of the total time children spend at school. On average, les-
sons take up about 5 hours per day, 5 days a week (25 hours), for a school
year of 32 weeks (800 hours). Block’s program took up about 12 percent
of total classroom time, yet it literally transformed these children. One
wonders what schoolchildren could achieve if the remaining 22 hours of
the week (700 hours of the school year) were spent in such rewarding and
enlightening activities. This is a stunning indictment of our educational
system.
The other notable feature of Block’s study was the outstanding meth-
odology. The program is also unique for setting up a purpose or set of
goals. These were not parochial or limited academic goals, but life-
changing goals, and I list them once more:

1. If instruction helps students use strategic processes consistently and

reflectively as they read, they should comprehend and use more informa-
tion in their daily lives.
2. Students who have trouble comprehending may need a new type of
instruction, involving new strategies and thinking/reading competencies.
3. Students who are confused about important concepts, inferences, and
relationships in a text may never have encountered the inductive or de-
ductive reasoning patterns the author used to create the text. If these
thinking strategies are explained prior to reading, thinking repertoires
should expand.
4. Students may be more able to generalize this new knowledge to solve
decoding and comprehension problems on their own.

The eight strategies that Block believed were crucial for improving chil-
dren’s comprehension skills and increasing progress toward these goals
 333

included strengthening basic cognitive operations like analytic thinking,

New Directions for the Twenty-First Century

decision strategies, problem solving, metacognitive analysis (awareness of
one’s current state of knowledge), and creative thinking. Children also
needed social skills for working in groups and skills for independent
learning.
What made the lessons effective was their structure, a structure that
evolved over months and years of pilot work. Lessons worked best if one
strategy was taught at a time, and then followed up by group or indepen-
dent activity. Materials were carefully chosen that were relevant to the
lesson and added insight for problem solving.
Not only did children have dramatic increases in vocabulary, but
reading-comprehension skills soared as well. The training generalized to
oral comprehension, so that students became more articulate, asked better
and more penetrating questions, wrote more coherent papers and stories,
and improved dramatically in behavior and confidence.
Scores on the reading-comprehension subtest of the Iowa Test of
Basic Skills showed astonishing gains over a control group. The effect size
for reading comprehension was 2.24, and it was 1.34 for the Total Read-
ing score (though reading was not taught directly). These are extraordi-
nary results in view of the typical findings in this field.
There were other improvements as well. Children in the program
were much more likely to represent analytic thought in their writing to
an astonishing degree ( p < .0001). They reported that what they had
learned at school was useful outside school (92 percent reporting ‘‘use-
ful’’), whereas none of the control group found that what they learned was
useful (which makes quite a statement about the school system). The
trained children had significantly higher self-esteem, though the groups
did not differ initially. The trained group used a greater number of
critical-thinking strategies, were more precise in their statements about
these strategies, and offered a larger number of alternative solutions to the
various problems. These comparisons were highly significant ( p < .0001
in all cases).
I would urge readers interested in this program to read the books by
Block and her colleagues, because these books outline the program, the
ideas behind it, and the effort that went into designing it (Block, Gam-
brell, and Pressley 1997; Block and Mangieri 1997; Block and Pressley
2001; Block 2002).
 334

New Frontiers
Chapter 11

Linking Linguistic Phonics to the Advanced Spelling Code

Even the most successful reading programs are missing a major com-
ponent of reading and spelling instruction. This is the advanced spelling
code, comprising the remaining spelling alternatives that need to be taught
once children have mastered the basic code. The advanced code consists
of an additional 136 common spellings alternatives.
There are few programs that teach the advanced spelling code ( Jolly
Phonics teaches 22 additional spellings for vowels only, and Lindamood,
16). I am aware of three programs, and only two of these programs are
published at this time. The study by Smith (1999), reported in chapter 3,
used a spelling program I designed (McGuinness 1997a, 1998a). Adoles-
cents, who were extremely poor readers and spellers, had 12 hours of in-
dividual instruction on the advanced code, up to and including all Latin
suffixes. These children not only made large gains in spelling (9 standard-
score points), but even larger gains in reading and reading comprehension
(10 to 14 standard-score points). Total Reading scores were at the 50th
percentile (normal) on standardized tests by the end of the sessions.
Another published program—by C. McGuinness and G. McGuinness
(1998)—was originally designed for the clinic for one-on-one sessions
with poor readers. It has been adapted for parents to teach poor and be-
ginning readers, and for beginning readers in the classroom. In a study on
87 consecutive clients to the clinic (C. McGuinness, D. McGuinness, and
G. McGuinness 1996), gains were comparable to those reported above,
except spelling and comprehension were not measured. Nearly all poor
readers read at grade level in about 12 hours or less based on standardized
tests of word recognition and word attack.
Because this program was designed initially for poor readers (age 6
or older), it was possible to speed up the introduction of the advanced
spelling code by overlapping it with the basic code about halfway through.
This works well for older children, but it is not clear whether a sudden
shift in logic (one sound–to–one spelling versus one sound–to–many
spellings) is as successful with younger children. Young children may be-
come confused by this.
So far, there are no methodologically valid studies on the classroom
version of this program. Research carried out by local school districts
shows strong gains, but, so far, these studies do not meet NRP standards.
 335

Finally, there is a comprehensive new reading/spelling program for

New Directions for the Twenty-First Century

parents and teachers designed for children age 4–6 years (McGuinness,
forthcoming). This program integrates a linguistic-phonics program with
the advanced spelling code. Lessons are sequenced by spelling complexity.
For example, children begin learning the phonemes that can appear in any
position in a word without changing their spelling. This makes it possible
to teach ‘‘transitivity’’ from the outset, the fact that the same sound can
appear in all positions in a word—something young children have trouble
with. The first three phonemes in the lessons (/p/, /o/, and /t/) make four
CVC words, so children can begin reading almost immediately. After six
phonemes (and one sight word), children can read little phrases, such as ‘‘a
tan man on a mat.’’
Lessons follow the same format and are self-explanatory (foolproof ),
so no training is necessary. Each lesson begins with a story read by the
parent (or teacher) that introduces the phoneme for the lesson. An excerpt
from ‘‘Poppy Pig’’ was provided earlier. This is followed by a structured
listening exercise to teach phoneme identity and transitivity (all positions
in a word). Next, the child sees the letter for that phoneme and learns to
trace, copy, and write it from memory. As phonemes accumulate, words
appear. These are visually segmented and read in a segmented fashion,
then blended into the word. Spelling dictation follows. As more words are
added, phrases and stories appear as ‘‘little readers.’’ The recommended
speed for parents is one phoneme every other day, or a minimum of three
phonemes per week. Teachers can move faster. Lessons are cumulative,
and include only the sounds and the spellings taught to that point (no
surprises). At the end of the lessons (after approximately 5 to 7 months),
the child can read all the parent stories, and should have reading and
spelling skills at the second- to third-grade level.
As far as I am aware, this is only program that weaves the structure
of the spelling code (from easy to hard) into the lessons from day one. It
includes the advanced code up to the Latin level, in both single-syllable
and multisyllable words.
These three programs, all of which feature the advanced spelling code
in one format or another, need extensive research to demonstrate their
merit. In terms of the gains reported so far, it is predicted that at least in
the long term they will outdistance the basic linguistic-phonics programs,
which provide minimal or no instruction in the advanced spelling code.
 336

Computer Models of Reading and the Real Spelling Code

Chapter 11

What is the connection, if any, between the structure of the advanced

spelling code set out in chapter 3, and the computer models discussed in
the previous chapter? The stated goal of computer simulations of read-
ing is that if the computer mimics human behavior, one can infer that
it reflects something about how the brain processes information and
‘‘learns.’’ The major problem with this assumption is that while computer
models may seem objective and scientific, the computer can only do what
it is programmed to do, and it is rated on its performance by what humans
deem to be correct. There is no guarantee that the researchers will have
an accurate knowledge of the structure of our spelling code, any more
than there is any guarantee that they understand how the brain works.
This is an important issue because the ‘‘parallel-distributed model,’’
which is part and parcel of this work, is purported to work like the human
brain. If any mathematical model succeeds in mimicking human behavior,
even though the model is wrong, or the scoring of the model’s performance is
wrong, this may have dangerous consequences. The attempt of these
researchers to link the computer’s performance to ‘‘types of dyslexia’’ is a
prime example of this danger. (See Coltheart 1978; Coltheart et al. 1993;
Plaut et al. 1996; Seidenberg 1992.)
In the last chapter, I noted two central problems with these computer
simulations. One has to do with the types of words the computer is given
to read, and the other is that the models are based on visual phonics,
in which the probabilities of spelling patterns are derived from letters.
The true structure of the English spelling code is based on phonemes.
Spellings are assigned to phonemes in words; words are not assigned to
spellings.
If you map the structure of the spelling code the way it was written
(see chapter 3), this brings a huge array of patterns (statistical redun-
dancies) into focus that would otherwise not be noticed by human ob-
servers, even though they are certainly ‘‘noticed’’ by the brain. In the
Plaut et al. 1996 studies, several patterns not discernible to the research
team appear to have been discerned by the computer program that was
optimized to record probabilities.
Nothing illustrates this better than the computer’s nonword decoding
‘‘errors.’’ The computer was trained on 3,000 real words (CVC, CCVC,
 337

CVCC only) for 2,000 repetitions. The number of times each word ap-

New Directions for the Twenty-First Century

peared was controlled by its frequency in print. The ‘‘attractor model’’
was then tested for transfer of training on 86 nonwords from Glushko’s
(1979) lists.
According to the team, the computer ‘‘failed’’ on 19 of these non-
words. But these ‘‘failures’’ were more in the eye of the beholder than
being anything objective. If the computer substituted a different legal
decoding for the one the researchers wanted, this was scored as an error.
Based on the real spelling code, the computer misread only 3 of the 19
words. Here are some examples. The responses are phonetic renderings of
the actual output.

Computer Errors?
Computer sees Computer says Researchers say
BLEAD, WEAD bled, wed bleed, weed
BOST, SOST boast bossed
Here are the actual probabilities (options) for making analogies to real
words:

BLEAD, WEAD:
Analogy to onset/vowel (CV )
as /e/ (none)
as /ee/ bleach, bleak

Analogy to rime (VC)

as /e/ bread, dead, dread, head, lead, read, spread, thread, tread
as /ee/ bead, lead, plead, read

BOST, SOST:
Analogy to onset/vowel
as /oe/ bold, both
as /o/ Bob, bog, bond, box

Analogy to rime
as /oe/ ghost, host, most, post
as /o/ cost, frost, lost
 338

In these examples, the computer selected an equally probable or more

Chapter 11

probable decoding than the researchers, one based on actual probabilities.

In another corpus of words, the vowel spelling was the issue. The
phoneme /u/ can be spelled o-e in a number of high-frequency words
(come). The o-e spelling is also the most probable spelling for the sound
/oe/ in a large number of common words (home). The former represents
‘‘frequency in print,’’ the latter, ‘‘probability’’ across all words in a corpus
(total vocabulary). For whatever reason, the computer preferred the fre-
quency solution. It decoded LOME, MONE, PLOVE, WONE as lum,
mun, pluv, wun, a match with common words like above, come, done, dove,
glove, love, none, one, shove, some. Nevertheless, these renderings were
scored as errors. If the computer was a child, she would say this ‘‘wasn’t
fair!’’3
The point here is profound. It goes back to the first premise in the
general prototype, which states that you must have a complete under-
standing of the structure of a writing system before you teach it. Ergo, you
must have a complete understanding of the structure of a writing system
before you model it. At the very least one needs to be aware that our
spelling system is a spelling system, not a decoding system. Until this occurs,
computer models of reading will not be reliable. And if they are not reli-
able on these very simple words, how can they cope with words like arti-
ficial, parallel, and intelligence, or with the hundreds of thousands of words
that do not have ‘‘word bodies’’ and do not share ‘‘rimes.’’
Before we leave this topic, I want to try to clear up a serious miscon-
ception about neuroanatomy that permeates the cognitive and computer
models of reading. This is the notion that ‘‘processing’’ goes on in ‘‘path-
ways.’’ The use of the term pathway to refer to types of neural processing
during reading originated in the ‘‘dual-route’’ model of Coltheart (1978).
Since then, this term has become ubiquitous, even in the computer models
that disprove Coltheart’s theory (Plaut et al. 1996). Ehri speaks of ‘‘estab-

3. The authors did state that if they scored the errors on the basis of whether
any word in the original corpus of 3,000 had the ‘‘same body’’ (rime) as the
nonword, the computer’s errors reduced to 9. But the network model is based
on statistical redundancy (probability), not isolated cases, so this does not
solve the problem.
 339

lishing access routes’’ as if learning is about co-opting or carving out

New Directions for the Twenty-First Century

‘‘channels’’ in the brain, rather like digging a trench to lay a pipeline or
stringing a telephone cable.
Here are the facts. Neurons come in two main flavors. There are
cells with dendrites and axons, and cells with only dendrites. Dendrites
are short, branching fibers. Axons are long (nonbranching) fibers. Neural
processing goes on in networks of dendrites. This processing is recursive,
with feedback from every neuron to every other neuron in the network via
the dendrites. The final output of this computational effort arrives at the
layer of neurons with axons. Axons relay this output to other regions of
the central nervous system. Axons are the neural pathways. Axons are like
telephone lines, and like telephone lines, they do not process anything. They
merely transmit. Learning does not occur in ‘‘pathways,’’ but via anatom-
ical and neurochemical changes in dendrites and adjacent cell membranes.
(This information is available in any good undergraduate textbook on
neurobiology; e.g., see Pinel 1997.)

How Does the Brain Read?

Early in the book, I described a new type of sight-word recognition that
some researchers believe occurs late in reading acquisition. This is essen-
tially another armchair theory about how the brain ‘‘reads.’’ It is con-
tradicted not only by brain science, but also by studies that reveal the
complex processing that goes on at an unconscious level before we produce
a conscious, behavioral response. This is yet another misleading theory
that can lead to dangerous consequences, as we will see.
Trying to infer how the brain works from behavior involves reasoning
backward from the behavior to what might be occurring in the brain. This
can lead to a fallacy in which brain processing is thought to be analogous
to how we experience or characterize the behavior. The mistake is that
brains do not work like humans think, nor do they work like products of
human thought (machines, telephones, digital computers). The complex-
ity of brain processing is far beyond anything our conscious mind can
fathom.

Beware Old Messages in New Bottles The comparative analysis of writing

systems proves that a whole-word (logographic) writing system is impos-
sible. Research on children learning to read continues to show that whole-
 340

word (sight-word) methods of instruction are ineffective. Time spent

Chapter 11

memorizing sight words was found to be negatively correlated with read-

ing skill (Evans and Carr 1985; Sumbler 1999). Meanwhile, some reading
researchers, including those who know that whole-word methods do not
work, have adopted a new kind of sight-word model. I will call this late-
stage sight-word reading, to distinguish it from basic (logographic) sight-
word reading.
These two types of sight-word reading are exemplified by Frith’s
(1980, 1985) stage model. At stage 1, the logographic stage, children
memorize words by sight as random letter sequences. As reading vocabu-
lary grows, the alphabetic stage begins, and children figure out how to
match phonemes to graphemes. As this becomes efficient, a final sight-
word stage emerges, the orthographic stage. Words that are decoded by
‘‘orthographic rules’’ are ultimately processed instantly (by sight) via a
visual pathway. This model is directly tied to Coltheart’s dual-route com-
puter model of reading.
Adams and Huggins (1985) proposed a similar idea. They studied the
effect of word frequency and context on decoding accuracy in elementary
school children, and measured what they called functional sight vocabularies.
They found that reading accuracy was strongly affected by word frequency
and improves more when words are read in context than in isolation.
Based on these results, they proposed a stage theory of sight-word acquisi-
tion. A word becomes a sight word in three stages. At the third or highest
stage, ‘‘The word [is] securely represented in the reader’s visual lexicon. It
is this third stage of mastery that permits the level of word recognition
automaticity that is so characteristic of skilled readers and so central to
theories of their behavior’’ (p. 275). Adams and Huggins proposed this
third stage of ‘‘automatic direct access’’ despite the fact that they measured
accuracy and not speed.
Ehri (1991) added a new twist to the ‘‘late-stage’’ level. Sight-word
reading is the final product of internalizing the ‘‘reading code’’ (grapheme-
to-phoneme correspondences): ‘‘Sight-word reading involves establishing
visual-phonological access routes rather than strictly visual logographic
routes into lexical memory’’ (p. 402).
Ehri’s (1998) version of sight-word reading apparently created con-
siderable confusion, which she felt obliged to resolve. Instead, the hole
only got deeper:
 341

Misconceptions about sight word reading persist. One is that only irregularly

New Directions for the Twenty-First Century

spelled words are read by sight. This is not true; all words, once they have been
read a few times, become sight words. . . . The term ‘‘sight’’ indicates that the
sight of the word triggers that word in memory, including information about
its spelling, pronunciation, and meaning.’’ (p. 91; emphasis added).

Ehri provided no operational definitions for what constitutes a ‘‘sight

word,’’ retreating to semantics with words like quickly and automatically. A
sight word is defined by what it is not. If a reader must consciously decode
a word, this means it has not yet achieved sight-word status: ‘‘Words not
known by sight may be read in other ways, although these ways require
attention and consume more processing time. Readers might apply their
knowledge of grapheme-phoneme relations to convert the spelling into a
pronunciation’’ (p. 90).
The notion that all words are ultimately read by sight because it seems
that way is rapidly gaining acceptance in the field despite the absence of
any proof. Aaron et al.’s (1999) remarks illustrate this quite clearly: ‘‘Even
though the importance of sight-word reading is recognized almost univer-
sally, how it is accomplished remains an unsolved issue’’ (p. 91; emphasis
added). They are referring here to ‘‘late-stage’’ sight-word reading, not to
‘‘logographic’’ reading.
The danger that this idea will find its way into the classroom is real.
Aaron et al. (1999) had several suggestions for how teachers could enhance
this kind of sight-word reading through vocabulary training.

A Century-Old Fallacy of Inferring Brain Processing Time from Response

Time The idea that reading speed mirrors brain-processing speed has
a long history, and has already had a profound effect on reading instruc-
tion. The story begins in 1886 when Cattell published a little paper in the
journal Mind on comparing the time it took to name letters and to read
simple, common words. He found that adults can read common words as
fast as they can name letters, and they can read words in context twice as
fast as they can read them in isolation.
This paper was cited as the ultimate ‘‘scientific proof’’ for the impor-
tance of teaching sight words as the major method of reading instruction
by the advocates of the look-say movement in the early twentieth century.
Later, Cattell’s paper came to the attention of reading researchers and
 342

his basic findings were replicated and extended. Biemiller (1977/1978) re-
Chapter 11

ported that second and third graders took longer to name words than
to name letters, but that the speeds were identical by fourth grade and
remained so through college, while overall naming speed increased at each
grade. Reading words in context was much faster than reading words in
isolation, and this effect appears by second grade.
Samuels and colleagues speculated that if a word is read as a single
unit (holistically), then word length would not matter (see Samuels,
Laberge, and Bremer 1978; Samuels, Miller, and Eisenberg 1979). They
found that word length (three to six letters) did have a strong effect on
naming speed at second grade, and continued to do so at higher grades
when words were over five or six letters long, but there was no word-
length effect for college students. They concluded that younger readers
process words letter by letter, whereas college students read ‘‘holistically.’’
This was interpreted as some kind of ‘‘activation’’ process: ‘‘What is ap-
parently being changed as one progresses through the grade levels is a
reduction of the contribution of activation of each letter code to the word
code’’ (Samuels, Laberge, and Bremer 1978, 719).
However, as Samuels, Laberge, and Bremer noted, ‘‘activation units’’
cannot explain how college students process all words at the same speed
regardless of length. Instead, they proposed that ‘‘there must be a tradeoff
in the activation threshold with word length, such that the letter codes of
three-letter words produce more activation than the letter codes of the six-
letter words’’ (p. 719; emphasis added).
To such complexities do our theories lead us!
I have surveyed a mere sprinkling of the many studies over the past
120 years in which researchers concluded that if you can read a word as
fast as you can name a letter, this is proof that the word has become a
‘‘sight word,’’ processed holistically, instantly, and automatically.
No one stopped to consider that the same results could reflect the
limits of motor processing (speech-output rate) across the developmental
span, not visual-processing speed. In other words, the similarity in speeds
to name letters and words may be caused by limits on the time it takes to
program an output and produce it, not how long it takes to perceive or
recognize it.
The idea that you can infer something about perception, cognition,
and brain processing from a single measure of simple response time is
 343

extremely naive. This problem was recognized early in the twentieth cen-

New Directions for the Twenty-First Century

tury by psychologists studying reaction time, and scientists who measure
eye-movement patterns during reading have made this point many times.
No single measure can reliably predict a reader’s efficiency in decoding
text. Rayner (1998, p. 377) summed up this work as follows: ‘‘Any single
measure of processing time per word is a pale reflection of the reality of
cognitive processing.’’
Eye-movement research shows that extremely precise measures of
fixation time per word (225 ms is average), recorded by the most sophis-
ticated eye-movement camera, are too imprecise to make inferences about
how processing occurs. Imagine how much this imprecision multiples when
it includes the entire reading/speaking act: focus, scan, transform visual
image to word, program output of word, execute output program. Nor
would this include the fact that individuals have such highly inconsistent
response times for reading the same words, as shown by Glushko’s data.
There can never be a metric to pin this process down. There is no
way to estimate how the brain reads from response times as sluggish as
450–800 ms, the time it takes to read a single word aloud. Nor is there
any evidence (or reason) that phonological decoding stops because motor
(vocal) responses speed up. The speed and complexity of the brain’s neural
systems are at a rate and quantity far outside the limits of what a simple
measure of behavior can show.
The late-stage sight-word theory is based on the assumption that
because a word seems to be recognized instantly, it is processed instantly as
well. The sense of ‘‘seems to be’’ is beautifully captured by the children’s
spontaneous remarks in Varnagen et al.’s study on spelling strategies: ‘‘I
just knew it.’’
Well, the mind plays tricks on us, as Freud observed long ago in
Project for a Scientific Psychology (Freud 1966; Pribram and Gill 1976).
Freud, who was trained as a neurologist, was apparently the first to realize
that the brain cortex makes conscious reflection possible, and then only
portions of it depending on what you are conscious of. Span of con-
sciousness, or ‘‘attention span,’’ is restricted to one domain or event at any
point in time. We cannot be aware of everything at once (McGuinness
and Pribram 1980). Most of what goes on in the brain cortex, and all of
what goes on in subcortical regions, operate outside our conscious aware-
ness or our conscious control.
 344

A century later, we know that the brain processes all incoming sen-
Chapter 11

sory signals but only alerts us to what is relevant after processing it. We can
ask our brain to hold something ‘‘in mind’’ if we choose to focus on it,
but usually at a metabolic cost (Pribram and McGuinness 1975). Further-
more, conscious awareness of external events can be incredibly slow,
especially when they are novel and unexpected. When President Reagan
was shot, a film recorded nearly a minute of silence before any reaction
was seen or heard (brains were churning while bodies were frozen in
inaction).
There is another brain principle at work. With experience and learn-
ing, neural processing increases in efficiency (in terms of precision, speed,
and neural organization) to the point where multiple or ‘‘parallel’’ pro-
cessing goes on automatic pilot, outside our awareness. There are in-built
properties of neural systems that allow for such rapid processing that we
can never become aware of it, no matter how hard we try. People do not
hear or feel phoneme coarticulation in speech. Yet we know it is there all
the same. We do not hear the microsecond differences between the arrival
time of sounds to each ear that allows us to localize objects in space.
Our sense that we read whole words instantly by sight via some
direct pipeline from the eye to meaning is an illusion. No matter how
much something ‘‘seems like’’ it happens instantly, it does not. Con-
scious awareness and brain processing run on different clocks. The brain
can map grapheme-phoneme correspondences, analyze patterns of ortho-
graphic redundancies, register degrees of word familiarity, perceive con-
text cues, and work out possible decodings of odd or unpredictable
spellings, in parallel. Everything else the brain does works this way. In
ordinary conversations, we map phoneme sequences into words, process
syntax and semantics, perceive quality and tone of voice, note vocal in-
flection, and watch mouth movements, facial expressions, body posture,
and mannerisms all at the same time. The brain has an amazing capacity
for ‘‘multiplexing’’ or ‘‘multitasking’’ hundreds or thousands of operations
simultaneously. Brains are redundant parallel processors even when they
do not need to be.

Your Brain Can Read Before You Know What It Has Read We have barely
begun to scratch the surface of how the brain processes the printed page.
Recent breakthroughs have shown that readers’ expectations based on
 345

syntax and semantics govern how the text is read, and what they look for

New Directions for the Twenty-First Century

or ignore. This is contrary to the view presented above, in which reading
is essentially passive and linear: see word . . . automatically decode word . . .
move to next word, and so on. The crucial tests for this new insight are
based on function words, those ubiquitous little words that one would
expect to be read most automatically of all: a, the, on, in, for, and.
Healy (1976) developed a useful technique for measuring what people
are conscious of as they process text, called the missing-letter effect. People
were asked to locate a target letter as they read continuous text at a normal
rate. Letters in function words were more likely to be missed. The error
rates for detecting t in words like the were much higher than for detecting
t in think.
At that time, it was discovered that readers focus less often and for
shorter periods on function words than on other words (Rayner 1977).
Various explanations were proposed for this effect. One was that func-
tion words are redundant and can be skipped. Another was that because
readers focus on meaning, function words are missed because they are
semantically empty.
Greenberg and Koriat (1991) thought differently. They discovered
that Hebrew readers detect function words largely on the basis of their
syntactic and semantic role in the sentence. They followed up this re-
search with U.S. college students and adapted it to the English language
by varying the linguistic roles of prepositions.
In one experiment, the word for appeared in sentences either in its
simple prepositional role, or in one of two common phrases. In ‘‘for
better or worse’’ it acts as a preposition, while in ‘‘for or against’’ it acts
as a modifier (adjective/adverb). The expectation was that the error rate
(detecting the target letter f in for) would be the same when for was in its
usual prepositional role, but that error rates would drop when for acted as a
modifier. This is exactly what was found. The error rate for for as a prep-
osition was 23 percent in a sentence and 25 percent in a common phrase,
but only 6 percent when it acted as a modifier.
Greenberg and Koriat replicated this effect using the preposition on
in sentences where on functioned like a preposition or as an adjective in
compound terms such as on switch, on call, onlooker, on side. The error rate
for finding the target letter n was 42 percent when on was a preposition,
but only 17 percent when it was an adjective.
 346

The investigators looked for the source of this effect by experiment-

Chapter 11

ing with the location of information in a sentence. Sentences were con-

structed where the context of the initial phrase was the same, but the word
for was used differently. In one case (‘‘for or against’’) it was a modifier;
in the other (‘‘for abortion’’) it was a preposition. They predicted that
detecting function words would vary as a function of context, even when
its function was in the future (in the second phrase), evidence of a ‘‘late-
stage’’ analysis of sentence structure. In other words, people either antici-
pate the second phrase before they read it, or they actually read it before
they are conscious of reading it.
The students missed 35 percent of the targets when for was a simple
preposition and 27 percent of the targets in the phrase ‘‘for abortion,’’ but
only 3 percent of the targets in ‘‘for or against.’’ Because prior context
could not have influenced these results, and the word for was detected
quite easily in one phrase but not in the other, this ruled out the theory
that function words are part of some ‘‘unitized word frame’’ (sequential
dependencies).
To pin this down more precisely, Greenberg and Koriat constructed
sentences with controls both on content and on functional phrases. Mean-
ing, for example, was held constant across all critical phrases, while con-
tent varied. They created sentences in which the word for could be linked
to something that either preceded or followed it, as in the following ques-
tions: ‘‘Are you for abortion or against it?’’ and ‘‘With regard to this issue
of abortion, are you for or against it?’’ In the first sentence, for is the
preposition of the object word abortion, and the statement is explicit. In the
second sentence, the word abortion comes early and for comes late, serving
as a modifier (‘‘for or against’’). In this case the meaning is implicit (it
refers back to something earlier).
Greenberg and Koriat predicted that the explicit phrase, where for
plays its usual prepositional role, would produce more errors than the
implicit phrase. This would show that the missing-letter effect is tied to a
function word’s specific syntactic role, and also clarifies how sentences are
processed. The error rate was 22 percent for detecting the letter f in for in
the explicit phrase, but only 14 percent in the implicit phrase.
Taken together, these results support an active, structural approach to
reading sentences, which psychologists call top-down processing or analysis by
synthesis:
 347

We propose that although the coding of structure and the coding of meaning

New Directions for the Twenty-First Century

go hand in hand during reading, the coding of structure generally leads the
way. Readers try to establish a tentative structural frame for the phrase and
then use it to guide the interpretation and integration of the constituent units
into a meaningful representation. The establishment of these frames appears
to occur at a relatively early stage in text processing on the basis of a cursory
and shallow visual analysis. . . . Letter detection is more difficult in function
words which normally serve to anchor and support phrase structure than in
the semantically informative content words. However, when function words
are forced into a content role, their constituent letters remain available.
(p. 1058)

The author viewed the ‘‘structural frame’’ as local in the sense of the im-
mediate context. The location of a particular word that disambiguates
meaning changes the depth of the perceptual analysis in surrounding
words, even when it ‘‘appears in mind’’ before it is read and/or is expected
in the future.
This research shows that reading is far more complex and more amaz-
ing than anyone could possibly imagine. It highlights the fact that this
complexity, which, in these studies, required visual search, decoding, the
processing of meaning, and syntactic analysis, is also accompanied by an
attempt to carry out a global structural analysis of every sentence and
anticipate which words need to be briefly scanned or receive full attention.
And this goes on completely outside conscious awareness. A good reader is only
conscious of meaning. The print on the page is all but invisible.
We are not the conductors of this symphony. Our brain is. This is
why it is foolish to imagine that anyone can decree how a particular word
is read by the brain, simply because reading seems automatic, and then
presume to know what this process entails.
 Appendix 1: How Nations Cheat on

International Literacy Studies

In 2001, the media in the United Kingdom reported that English students
outscored Sweden in a recent international reading survey, citing PISA
2000 (Programme for International Student Assessment, OECD 2001).
Yet a mere four years earlier, in another OECD study on twelve coun-
tries, Sweden came first in the 16-to-25 age group while the United
Kingdom, Australia, New Zealand, Ireland, and the United States barely
outdid Poland. Contradictory results also appeared in a recent study on
nine-year-olds from thirty-two countries—PIRLS 2001 (Progress in In-
ternational Reading, April, 2003). England was third in this study after
Sweden and the Netherlands. The United States was ninth. The apparent
spurt in literacy rates for England and the United States is not mirrored
by well-conducted domestic studies, such as the National Assessment of
Education Progress (NAEP).
In the most recent NAEP report on fourth-grade children (NAEP
2002, U.S. Department Education 2003. See USDE Web site), 38 per-
cent of the children were ‘‘below basic’’ (functionally illiterate), 32 percent
were at ‘‘basic level,’’ 23 percent were ‘‘proficient,’’ and 6 percent were
‘‘advanced.’’ These values are similar to earlier results of 43 percent, 33
percent, 20 percent, and 4 percent, respectively. Certainly, a 38 percent
functional illiteracy rate is unlikely to catapult the United States into ninth
place internationally. NAEP studies are extremely rigorous. In NAEP
1992 (Mullis et al. 1993), 140,000 children were tested on an individual
basis. Test items were secured, and outside testers monitored protocols
and compliance rates.
Earlier international reading studies sponsored by the OECD were
also well conducted. Yet a senior official on this project told me that
 350

some countries put pressure on them to obscure or disguise the data.

Appendix 1

One country withdrew from the study because they did not like their
results.
Today, the situation has deteriorated to the point where results are
meaningless. PISA 2000 is the most recent of the OECD studies, involv-
ing thousands of 15-year-olds from thirty-two countries. Pages 232 to
236 of the report provide information on population sampling along with
tables on exclusion and compliance rates (tables A3.1, p. 232, and A3.2,
p. 235). Compliance was monitored in the following ways: (1) Target
schools representing a demographic sample were provided to local school
districts. (2) If a school would not or could not participate, a replacement
school from a second list had to be chosen. To avoid selection bias, the
compliance rate at step 1 was set at 85 percent minimum. (3) A fixed
number of students per class had to participate based on classroom size
(usually about 35 children), with a minimum compliance of 80 percent
student participation.
Exclusion rates refer to children in the population who cannot be
tested because of mental retardation, being blind or deaf, and so forth.
The guidelines were: ‘‘The percentage of 15-year-olds excluded within
schools had to be less than 2.5 percent of the nationally desired target
population.’’ The last part was in boldface type. This section of the re-
port also expressly stated that ‘‘special education’’ was not a criterion for
exclusion.
The PISA consortium, a body of various government officials and
statistics experts, were apparently unable to enforce even these lenient
guidelines. Failure to comply with exclusion rates was ignored completely.
Poor compliance rates were explained away with the statement (p. 236)
that countries with poor compliance rates supplied documentation that
adequately explained why rates were low! The Netherlands, with a 27
percent compliance rate, was the only country dropped from the study.
The compliance rates for the remaining worst offenders are provided
below. The score in column 1 must be 85 percent to meet the require-
ments of the study. If not, reading scores will be unreliable, especially
when the percentage of missing schools is added to the percentage of
missing students. Even with replacement schools, the United States tested
only 55 percent of students they should have tested.
 351

Countries with poor compliance rates:

How Nations Cheat on International Literacy Studies

With
Target replace-
schools ment Students
United States 56% 70% 85%
United Kingdom 61% 82% 81%
Belgium 69% 86% 93%
New Zealand 78% 86% 88%
Poland 79% 83% 88%
Australia 81% 94% 84%
Japan 82% 90% 96%

However, as shown by the following countries, there is no excuse for low

compliance rates.

Countries with excellent compliance rates:

Korea 100% 100% 99%
Sweden 100% 100% 88%
Austria 99% 100% 92%
Italy 98% 100% 93%
Finland 97% 100% 93%
Spain 95% 100% 92%
Mexico 93% 100% 94%

I compiled lists representing the true facts about exclusion and compliance
rates for all countries. Composite reading test scores are included, taken
from table 2.3a, p. 253 in the report.

Countries failing to meet requirements for both exclusion and compliance:

Exclusion Reading
rate score
Ireland 4.6% 527
Luxembourg 9.1% 441
Netherlands 4.4% NA
New Zealand 5.1% 529
Poland 9.7% 470
United Kingdom 4.9% 523
United States 4.1% 504
 352

Countries failing to meet the exclusion cut-off only, scoring above 3

Appendix 1

percent:
Canada 4.9% 534
Denmark 3.1% 497
France 3.5% 505
Sweden 4.7% 516

Countries failing to meet the 85 percent compliance rate only:

Australia 528
Belgium 507
Japan 522

The following countries played by the rules.

Countries meeting all requirements ranked by reading test scores:

Finland 546 Germany 484
Korea 525 Liechtenstein 483
Austria 507 Hungary 480
Iceland 507 Greece 474
Norway 505 Portugal 470
Switzerland 494 Russia 462
Spain 493 Mexico 422
Czech Republic 492 Brazil 396
Italy 487
In this last list, what is most significant is the absence of English-
speaking countries. From the tables overall one notices the highly inflated
scores for the ‘‘cheaters’’ as compared to the countries that did not cheat.
It is evident that the top-scoring countries in the final list are there for
a reason. Finland and Korea have the most transparent writing systems
in the world. Nearly as good are the Scandinavian, German, Italian, and
Spanish writing systems. It’s a pity Sweden violated the exclusion crite-
rion, because there was no need to. They would have scored near the top
in any case.
Think about how misleading this report is. If one merely looked at
the tables of reading test scores, being unaware of, or skipping over, the
complex tables on exclusion and compliance rates, it would appear that
the United States had scored in the mid range, and the United Kingdom
ranked with Korea.
 353

Control over sampling and compliance rates was also a problem in the

How Nations Cheat on International Literacy Studies

PIRLS 2001 study on 150,000 students from thirty-five countries. This
study was sponsored by Boston College, with support from the Interna-
tional Association for the Evaluation of Educational Achievement. One
example of the problem is the fact that Singapore tested 7,000 children
while the United States tested only 3,763. Although Sweden came out on
top in this study, not much else fit expectations. The Netherlands and
England (major culprits in the PISA 2000 study) came in second and third,
followed by Bulgaria and Latvia. Iceland and Norway scored a dismal
twenty-first and twenty-fifth, the penalty for playing by the rules.
As project directors obviously have lost control over these studies,
international testing has become an exercise in futility. It is an enormous
waste of time and resources, and proves nothing.
 Appendix 2: Misuse of Statistics

Statistical Problems in Bond and

Dykstra 1967

Use of Means Instead of Individual Data in ANOVA Statistics

The use of simple means to represent the data from a small population
(20 to 30 children per classroom) is liable to serious distortion. This would
occur if one or two children in a class had extreme scores, a common
situation in reading research. The mathematics that underpins analysis-
of-variance (ANOVA) statistics, and the probability tables formulated to
work with these statistics, are based on variances and not on means.
The ANOVA computes a ratio of the between-groups variance to
within-groups variance. In this study, each classroom should contribute a
range of scores. These scores are used to compute the ratio of the variance
due to treatments (between-groups variance) to the variance due to the
children across all treatments and all classrooms (within-subjects variance).
Between-groups variance is the numerator, and within-subjects variance
is the denominator or error term. These terms should reflect a normal
distribution (normal variance of the all the children’s reading ability).
Replacing individual scores with classroom mean scores means that scores
are unlikely to be normally distributed.
The probability of obtaining a significant result (large F ratio) is
determined by the size of the difference between the groups, and by the
power in the study. ‘‘Power’’ translates to the number of subjects. Thou-
sands of subjects is very high power, and tens of subjects is low power.
By using the means instead of individual scores, all the variance (the vari-
ability) from the children was eliminated and replaced by the variability
between classroom means. This not only changed the focus of the Bond
and Dykstra 1967 study from one comparing children to one comparing
classrooms, but dramatically reduced statistical power so that the proba-
bility of obtaining a significant result was far too low.
 356

About Units of Measurement It is currently fashionable to use classroom

Appendix 2

means as the unit of measurement in cases where the ‘‘treatment’’ (the

teaching method) is delivered to the whole class rather than to individuals
(or groups). From a logical standpoint it is unclear why this is necessary,
and from a mathematical standpoint it makes no sense. ANOVA statistics
were invented for field trials in agricultural research. If plants are sub-
stituted for children, the analogy is nearly perfect, except that the depen-
dent variables are height and weight. In field trials, plants are set out
in plots and each plot gets the same treatment (specific proportions of
nutrients). Plots are counterbalanced or randomized across the terrain to
control for such things as sunlight, drainage, wind, mineral content of the
soil, and so forth. After treatment, each plant contributes one score to the
analysis. This is essential to compute the variance within plots as a func-
tion of treatment unless the number of plots is vast. Using mean height
per plot would eliminate this variance, and statistical analysis would be
invalid.

Analysis of Covariance
The analysis of covariance (ANCOVA) is justified when subjects differ
widely in a particular skill at the beginning of a study where the same skill
is being trained, or when subjects differ widely on something likely to con-
tribute to this skill, such as IQ or vocabulary. In other words, covari-
ance would be used where it would be logical to pair or match subjects
between groups, or when a difference between groups occurs that does
not seem due to chance. Even if one assumes that Bond and Dyskstra’s
study met these requirements, one must be able to answer the following
question: Did the groups consistently differ on the baseline measures?
Unfortunately, this question cannot be answered, because the incorrect
data (means again) were used for the statistical analyses. Furthermore,
when comparing the statistics for all projects combined, versus individual
project analyses, the results from the baseline tests suffered from a ‘‘now
you see it now you don’t’’ effect. In the basal-reader versus i.t.a. compar-
isons, letter-name knowledge was significantly better for i.t.a. Yet when
the five projects were analyzed separately, the letter-name effect disap-
peared for every group. However, a group difference in phoneme discrimi-
nation appeared that had not been there before. These results are typical
 357

of what happens when means instead of individual scores (variances) are

Misuse of Statistics
used in a statistical analysis.
ANCOVA is based on correlational statistics with the twist that it
allows you to make all groups comparable on the baseline measures, and
then adjust the outcome measure accordingly. The mathematics requires
the data to be linear and normally distributed, with similar variances on
each test. By reducing the data to means, none of these requirements can be
met. In addition, the computations in an ANCOVA analysis must include
values for sampling error and residual variance (the square of the standard-
deviation estimate), neither of which can be computed with means.

Degrees of Freedom
Degrees of freedom translates to the number of subjects in a study, minus
the levels or conditions under each treatment or factor, minus 1. In com-
puting ANOVA statistics, the total variance for treatments is divided by
the number of levels for that treatment (for i.t.a. versus basals, df ¼ 1), and
the total variance for subjects across treatments (within-subjects error) is
N  1  1. If the degrees of freedom for the error term is very large (lots
of subjects), then the final value for within-subjects error (the denomina-
tor in the F ratio) is very small. This, in turn, would make the final F value
too large and more likely to be significant. In this study, individual proj-
ects were analyzed one at a time, and the degrees of freedom should have
represented only the classrooms in that project. Instead, the value for the
degrees of freedom represented all the projects in a method group. In the
i.t.a. comparison, the degrees of freedom for the error term was five times
larger than it should have been for every comparison. And the same prob-
lem continued throughout.
As an illustration, by analyzing each project separately, this reduces
to a two-factor random-group design (gender  treatment). Project 1 had
32 classrooms (64 means), so degrees of freedom would be 1 (treatment)
1 (sex) and 61 subjects. The table values given, however, were 1 and 292,
which is clearly incorrect. This was not a typographical error, because the
same mistake appeared on all other tables where individual projects were
analyzed separately (see tables 27, 38, 49, 60, and 71). The same mistake
appeared in the analyses for the individual tests as well. This means one
cannot rely on the results for any of the outcome measures.
 Appendix 3: An Analysis of Word

Lists from Treiman et al. 1995

CVC Spelling Word Lists are Sorted by High and Low

Consistency of Initial (CV ) and Final ( VC) Segments
Words are listed by spelling type, starting with vowel þ r.

Low CV/High VC
bar, jar, beer, dear, rear, turn, ball, tall, lock, cook, loss, bob, mob, heap, thin

High CV/High VC
deck, pick, suck, Dutch, mess, gum, gun, met, wet, file, game, hope, rode, role,
sung

Low CV/Low VC
bear, pull, push, wash, watch, none, ton, won, cough, rough, lose, dog, fog, death,
bade

High CV/Low VC
word, worm, chose, dose, lease, pose, gas, yes, doll, shall, hood, mood, wood, limb,
mead

Problems with High-Consistency VC Units

1. One list contains six vowel þ r words: bar, jar, beer, dear, rear, and turn.
The /r/ is a vowel in its own right (/r/ in her). When it is combined with
another vowel (VV ) as in for, this creates a diphthong: /oe/-/er/. A diph-
thong is taught as one unit (a digraph or phonogram) and not as two
separate phonemes (/oe/ and /r/), which would make no sense. The words
listed above are actually CV words, not CVC words, and so have no VC
 360

unit. The letter r cannot ‘‘control’’ the pronunciation of the vowel because
Appendix 3

it is part of the vowel.

2. That this is the case is revealed by the word turn, which is treated as a
CVC word in which the /r/ (spelled ur) counts as one vowel.
3. The word bear is on one of the low-VC lists, yet has the identical rime
to dear and rear on the high-VC list. It is not clear how ear can be con-
sistent and inconsistent at the same time. And, of course, these are CV
words in the first place and have no VC unit.
4. There were eight words on the high-VC lists where the vowel controls
the final-consonant spelling, exactly the opposite of the claim. These
words are lock, cook, deck, Dutch, pick, suck, loss, and mess. Vowel phonology
determines the spelling of final /k/, /ch/, /f/, /j/, /l/, and /s/. ‘‘Simple
vowels’’ usually take the spellings ck, tch, dge, ff, ll, and ss. All other vowels
or consonants take k, ch, ge, and l as well as ce, se, or s. The final consonant
does not control how the vowel is pronounced in these words. The pro-
nunciation would be unchanged if the other legal spellings were used: lok,
dek, duch, pik, suk, los, and mes. It is the spellings that are controlled, not the
pronunciation.
5. There were nine CVC words of the ‘‘simple’’ (‘‘short’’) vowel type: bob,
bog, thin, gun, gum, met, wet, sun, and sung. In these words, all phoneme
spellings are ‘‘context free.’’ No phoneme (or its spelling) has the least
effect on how the others are decoded or spelled. They are pronounced
consistently, not because the VC unit makes them consistent, but because
nothing is affected by what it sits next to.
6. The words with V þ e vowels are also stable: file, game, hope, rode, and
role. The vowel is unaffected by changing the final consonant: file, fine,
five / gage, gale, game, gape, gate, gave / hope, hole, hone, home, hose / rode,
role, robe, rose, rote, rove.
7. There are only two words where a final consonant controls vowel pro-
nunciation. These are the ‘‘l-controlled’’ vowels. When the letter l follows
a, this causes it to be read as /o/, as shown: ball, tall (though not always:
shall, fallow, gallows, tallow, pal, pallid ). This would meet their criteria for
‘‘high consistency,’’ but they are the only two words on any of the lists
that do.
8. One of the high-VC words is heap. The p is supposed to signal that ea is
read /ee/ and not /e/ (head ). The orthographic eap family consists of four
words—heap, cheap, reap, and leap—but this does not hold up in multisyl-
 361

lable words (weapon). The VC unit is not productive here in view of the

Analysis of Word Lists

fact that the phonological eep family has two spellings, those above plus beep,
cheep, creep, deep, keep, peep, seep, sheep, sleep, steep, sweep, weep.

Problems with Low-Consistency VC Units

When we turn to the words with low-VC units—orthographic rimes
where the consonant is unlikely to help decode the vowel—we encounter
entirely different kinds of words. In the first place they are considerably
harder to read. The words are longer and have more irregular spellings. It
is not surprising that children had more trouble reading these words. But
was this a function of the low-VC consistency? Can the absence of some-
thing be a cause?

1. There are far more irregularly spelled words on the low-VC lists. I
have mentioned bear, which is ‘‘irregular’’ compared to dear and rear.
There is shall (one of the renegades in the all family), as well as none, ton,
and won, with uncommon spellings of the vowel /u/. Cough and rough both
appear on the same list, causing confusion because ou stands for different
vowel sounds, and gh is a low-probability spelling for /f/. Did children
have trouble reading these words because of their irregular spellings, be-
cause of the fickleness of ough (bough, bought, though, through), or because
the VC ‘‘rime was too inconsistent’’ to provide a clue to how to pro-
nounce the vowel, as they claim?
2. These lists contained many words in which the initial consonant con-
trols the vowel spelling: pull, push, wash, watch, word, worm. As noted
above, /p/ (and /b/) take the u spelling for the vowel /oo/, ^ as in pull, push,
pudding, pulley, pullet, put (and in bull, bullet, bully, bush, butcher). The sound
/w/ takes the a spelling for the vowel sound /o/: wad, waffle, wan, wand,
wander, want, wash, wasp, watch, water, also true for /sw/ blends (swab,
swamp, swap, swat) and even for /skw/ blends (squab, squabble, squall,
squander, squash, squat). The sound /w/ takes the or spelling for the vowel
sound /er/, as in word, work, world, worm, worry, worse, worst, worth, worthy.
It is noteworthy that four of these strongly CV-controlled spellings were
not on the high-CV consistency list, showing that these lists reflect little
understanding of the spelling code.
3. Four words use the Old French spelling se for /s/ and /z/: chose, dose,
pose, and lose (choze, doass, poze, and looze), and are unstable for both
 362

consonant and vowel, thus being highly likely to cause decoding errors.
Appendix 3

To add to the confusion, uncommon final /s/ words were also on the list:
gas, yes. They are part of a tiny group of words ending in /s/ (not plural)
that use the s and not ss spelling (bus, gas, us, this, yes).
4. The words hood, mood, and wood were on the same list. This creates
confusion because this is the main spelling (basic-code spelling) for two
vowels (nothing to do with ‘‘control’’ by a final consonant), as in food,
mood, noodle, poodle / good, hood, stood, wood.
5. Two other words seemed designed to cause reading errors—limb and
mead—because few children will have heard of these words.
6. Finally, the words dog and fog are on these lists. These are part of the
‘‘context-free’’ CVC group, and it is not clear why the og spelling is ‘‘in-
consistent’’ with any other pronunciation.
 Glossary

alphabet A writing system based on phonemes (individual consonants

and vowels).

analogy decoding An instructional approach in which children are en-

couraged to decode an unknown word by searching in memory for a word
with the same ending sounds, and then swapping the initial sound(s).
(‘‘Hand looks like land, except it starts with an aitch.’’)

artificial transparent alphabet Used during initial reading instruction.

Each phoneme in the language is represented by a letter, digraph, or arti-
ficial symbol.

basal reader A reading program that includes all elements of a curricu-

lum. Designed by education publishing houses for use in the classroom.

basic code Used during initial reading instruction. Each phoneme in the
language is represented by its most common or least ambiguous spelling.

binomial test A statistical test that computes a numerical score that will
exceed chance (guessing) at a specified level of probability. The computa-
tion takes into account the number of items on a test and the number of
alternatives for each response.

categorical perception The inability to hear the acoustic transitions

between two similar consonant contrasts (ba/pa) and the tendency to hear
only one or the other.

ceiling effects A situation in which a test or task is so easy that most

people get perfect scores.
 364

character In Chinese writing, a symbol standing for a syllable (or word).

Glossary

classifier In Chinese writing, a symbol acting as a marker for a semantic

category.

coarticulation Phonemes overlap one another in speech. The phonemes

coming later in the word modify the production of phonemes coming
earlier in the word, creating a complex acoustic envelope of sounds.

code Any system in which arbitrary symbols are assigned to units within
a category. The number symbols 1–10 represent units of quantity. Letters
represent units of speech sounds (phonemes).

consonant A phoneme that involves contact and movement between

one or more speech articulators. Voiced consonant engages the vocal
folds. Unvoiced consonant does not engage the vocal folds.

consonantal alphabet A writing system in which symbols are assigned

only to consonants. Suitable for languages where consonant sequences
stand for meaning, and vowels signal grammar and are inferred from
context.

consonant blend or consonant cluster Two or more consonants in

sequence in a word: str in street.

construct validity The aspect of test design in which items faithfully

reflect the construct being measured and not some other construct.

covariance analysis A statistical tool to adjust posttest scores to reflect

individual differences in these scores at initial testing (e.g., intelligence is
covaried in order to look at the correlation between phoneme awareness
and reading, independent of intelligence).

decoding The act of translating symbols that represent units of some-

thing back into those units (translating from letters to phonemes).

degrees of freedom In statistics, the number of scores in a study that

are free to vary given one known value.

determiner A symbol standing for a semantic category (plant, man,

water).
 365

diacritic A special mark or extra letter written above, below, or beside a

Glossary
letter to indicate pronunciation.

digraph Two letters standing for one phoneme: ch in church.

diphone system A system of writing in which one symbol represents

each consonant-vowel (CV ) unit in a language.

diphthong A vowel sound that elides two vowels in rapid succession and
counts as one vowel (/e/ þ /ee/ ¼ /ae/ in late).

dyslexia Greek for ‘‘poor reading.’’ Taken to mean a genetic predispo-

sition to difficulties in learning to read in some countries (U.S.). Means
poor reading, whatever the cause, in other countries.

eclectic reading method A method of teaching that includes a variety

of approaches not necessarily relevant to what needs to be taught.

effect size A conversion of the difference in scores between two groups

to standard-deviation units. This provides a standard metric for compar-
ing two or more studies with similar research designs and measures.

encoding The act of transcribing units within a category into arbitrary

symbols assigned to each unit (phonemes to their spellings).

expressive vocabulary A spoken vocabulary. The number of words

someone is able to say and have understood.

factor analysis A complex type of correlational statistics in which the

relations between scores on a variety of tests are explored in geometric
space. Tests are sorted according to their similarity to one another (prox-
imity in this space), and factor scores or ‘‘factor loadings’’ are computed
for each test in relation to every other test. Tests that ‘‘load’’ together (are
highly correlated) constitute a factor. A factor is an abstraction, and the
researcher must determine what the tests have in common. In general, a
factor loading is not meaningful unless it is .80 or higher.

factor loading A final value assigned to a test after a factor analysis has
been carried out. This represents the power of that test to represent a
factor in correlational values.
 366

floor effects When a test or task is so difficult that most of the scores
Glossary

are at zero.

frequency Used here as a measure of the frequency of occurrence of

a particular spelling in printed text. Frequency is the number of times it
would occur in a specified number of words in text.

functional illiteracy A designation of reading difficulty used in national

testing in which the reader is unable to find, use, and interpret the mean-
ing of printed text.

hieroglyphic A writing system used on religious or public monuments

for sacred or political purposes.

hiragana One of two systems for marking diphones used in the Japa-
nese writing system.

homophones Words that sound exactly alike but have a different

meaning.

invented spelling An instructional method in which children are en-

couraged to recreate the spelling system based on whatever knowledge
they might possess.

isomorphic Same form.

item analysis In test construction, an analytic technique to determine

the power and reliability of individual items to measure a construct, such
as which items produce consistent or inconsistent responses, and which
are too easy or too difficult.

kanji Symbols standing for whole words used in the Japanese writing
system.

katakana One of two systems for marking diphones used in the Japanese
writing system.

logograph An abstract symbol standing for a whole word.

look-say A reading method in the early twentieth century in which chil-

dren memorized isolated printed words as random letter sequences by
visual memory alone.
 367

mapping A process by which units of one type are assigned to units of

Glossary
another type.

meta-analysis The weighted mean of series of effect-sizes. This pro-

vides a global overview in standard deviation units of many studies using
similar research designs and measurements.

morpheme The smallest unit of a word that conveys meaning (boat

contains one morpheme; boats and boating contain two morphemes).

morphology A division of grammar in which morphemes determine

grammatical form.

normal distribution A mathematical distribution of a series of mea-

surements in which the form of the distribution can be entirely deter-
mined by the mean and the standard deviation. This distribution takes the
form of a bell.

onset A technical term that refers to the initial consonant or consonants

that precede a ‘‘rime.’’ str in street is an onset.

ontogeny The biological development of an individual.

opaque alphabet An alphabetic writing system in which there are mul-

tiple spellings for the same phoneme.

orthography ‘‘Standard spelling.’’ The patterns of permissible spellings

for sounds in words in a writing system.

paired-associate learning A type of learning that requires memoriza-

tion of arbitrary pairs of something (letter symbols for phonemes).

paleography The study of ancient writing systems.

percentile rank A conversion of a test score from a standardized test to

reflect the level at which this score exceeds a percentage of the population.
A percentile score of 90 means this person exceeded 90 percent of the
population.

perceptual span In eye-movement research, this is the distance in the

periphery from central visual focus that people can see and/or use infor-
 368

mation. In reading research this is measured by the number of letters to

Glossary

the left or right that influence speed of decoding.

phoneme The smallest unit of speech that people can hear; corresponds
to consonants and vowels.

phoneme awareness The ability to hear and remember the order of

phonemes in words.

phonetic Speech sounds in words.

phonics A generic term for any reading method that teaches a relation-
ship between letters and phonemes.

phonological awareness The ability to hear and remember a variety

of units of sound within words: syllable, syllable fragment (onsets/rimes),
phoneme.

phonology A system of speech sounds that make up a language.

phonotactics The permissible or legal phoneme sequences in words in a

particular language.

phylogeny The evolutionary patterns or lines of descent of an organism.

pictogram A stylized picture standing for a whole word. Used in ancient

times for business purposes (invoices, inventory lists) or as part of an early
writing system.

probability The likelihood of a particular event or occurrence as a func-

tion of a total range of possibilities (a is the most probable spelling for the
sound /a/ (cat)).

prosody The general term for variations in the acoustic properties of

speech for a given language that do not carry meaning directly, such as
fluency, melody, stress patterns, and inflection.

radical The part of a Chinese character that represents a semantic cate-

gory. Identical to a determiner.

recall memory Memory retrieved from a long-term memory store with-

out the benefit of prompts or clues. (An essay test involves recall memory.)
 369

receptive vocabulary The number of words that someone can under-

Glossary
stand when spoken by others.

recognition memory Memory retrieved from a long-term memory

store that is assisted by prompts and clues. (A multiple-choice test involves
recognition memory.)

reliability In test construction, a reliable test is one where people score

similarly on different occasions.

rime A technical term for the final portion of a word that ‘‘sounds like’’
other words (rhymes). and in band, bland, brand, hand.

schwa vowel A swallowed ‘‘uh’’ sound in a nonstressed syllable (hesitate,

abundance).

semantic or semantics The content or sequence of words in phrases

that conveys meaning. In linguistics—the study of how language repre-
sents meaning.

sight words Printed words that children are asked to memorize visually
as random string of letters. A true sight word contains rare spelling pat-
terns (‘‘yot’’ ¼ yacht).

spelling alternatives All possible spellings of a single phoneme.

standard deviation A measure of the variance (variability) in a set of

scores that represents the square root of the sum of the squared differ-
ences of each score from the mean.

standardized test A test administered to a very large number of people

(normed) over a wide age range, in which test scores are ‘‘standardized’’ to
fit a normal distribution. This is often converted to a standard metric, with
a mean of 100 and s.d. of 10. Individual test scores take into account age
(in months) and the distance from the mean in standard-deviation units.

statistical power The number of scores on a particular test or task. The

greater the number, the more likely there will be a normal distribution.
This translates to the number of subjects in a study: large number ¼ high
power ¼ more reliable result.
 370

syllabary A writing system based on the syllable.

Glossary

syllable A speech unit contained within a word, or constituting a whole

word, which consists of one vowel plus any consonants. I and straight are
one-syllable words; Basket and triumph are two-syllable words; and so on.

tonal language A language in which changes in pitch signify meaning.

transparent alphabet An alphabetic writing system in which there is

rarely more than one spelling for the same phoneme.

vowel A phoneme in which the articulators do not stop or curtail the

flow of breath. All vowels are voiced.

vowelBr A category of vowels in English in which the sound /r/ is

either a vowel on its own (her contains the sounds /h/ /r/), or it forms
a diphthong with another vowel ( for contains the diphthong /oe/ þ /r/).
The are nine such vowels in English.

whole language A philosophy which holds that learning to read is sim-

ilar to the acquisition of natural language. Children learn to by read by
exposure, reading along with the teacher, and by guessing words using the
context, pictures, and other cues.

word attack A type of reading test that consists entirely of nonsense

words.

word family A group of words that share the same ending sounds, that
are spelled the same, and that rhyme (bright, night, fight, sight).

word identification Same as word recognition.

word recognition A type of reading test in which a child decodes unre-

lated words one at a time.

writing system A systematic mapping of the elements of a unit of

speech into a set of arbitrary symbols, such that every word in the lan-
guage can be represented. No writing system marks whole words. Writ-
ing systems mark one of four sound units (and only one): syllable, CV
diphone, consonants only, consonants and vowels (phonemes).
 References

Aaron, P. G. 1991. Can reading disabilities be diagnosed without using intel-

ligence tests? Journal of Learning Disabilities, 24, 178–186, 191.
Aaron, P. G., Joshi, R. M., Ayotollah, M., Ellsberry, A., Henderson, J., and
Lindsey, K. 1999. Decoding and sight-word naming: Are they independent
components of word recognition skill? Reading and Writing: An Interdisciplin-
ary Journal, 11, 89–127.

Aaron, P. G., Keetay, V., Boyd, M., Palmatier, S., and Wacks, J. 1998. Spell-
ing without phonology: A study of deaf and hearing children. Reading and
Writing: An Interdisciplinary Journal, 10, 1–22.

Adams, G. L., and Engelmann, S. 1996. Research on Direct Instruction: 25 Years

Beyond DISTAR. Seattle: Educational Achievement Systems.

Adams, M. 1990. Beginning to Read. Cambridge, MA: MIT Press.

Adams, M. J., and Huggins, A. W. F. 1985. The growth of children’s sight

vocabulary: A quick test with educational and theoretical implications. Reading
Research Quarterly, 20, 262–281.

Allred, R. A. 1990. Gender differences in spelling achievement in grades 1

through 6. Journal of Educational Research, 83, 187–193.

Aslin, R. N., Saffran, J. R., and Newport, E. L. 1998. Computation of condi-

tional probability statistics by 8-month-old infants. Psychological Science, 9,
321–324.

Ball, E. W., and Blachman, B. A. 1988. Phoneme segmentation training: Ef-

fect on reading readiness. Annals of Dyslexia, 38, 208–225.

Ball, E. W., and Blachman, B. A. 1991. Does phoneme awareness training in

kindergarten make a difference in early word recognition and developmental
spelling? Reading Research Quarterly, 26, 49–66.
 372

Balmuth, M. 1992. The Roots of Phonics. Baltimore: York Press.

References

Baron, J., and Strawson, C. 1976. Use of orthographic and word-specific

knowledge in reading words aloud. Journal of Experimental Psychology: Human
Perception and Performance, 2, 386–393.

Barr, R. C. 1972. The influence of instructional conditions on word recogni-

tion errors. Reading Research Quarterly, 7, 509–529.

Barr, R. C. 1974/1975. The effect of instruction on pupil reading strategies.

Reading Research Quarterly, 4, 556–582.

Beck, I. L., Perfetti, C. A., and McKeown, M. G. 1982. Effects of long-term

vocabulary instruction on lexical access and reading comprehension. Journal of
Educational Psychology, 74, 506–521.

Becker, W. C., and Gersten, R. 1982. A follow-up of follow through: The

later effects of the Direct Instruction model on children in fifth and sixth
grades. American Educational Research Journal, 19, 75–92.
Beers, C. S., and Beers, J. W. 1992. Children’s spelling of English inflectional
morphology. In S. Templeton and D. R. Bear, eds., Development of Ortho-
graphic Knowledge and the Foundations of Literacy, 231–252. Hillsdale, NJ:
Erlbaum.

Bell, N. 1986. Visualizing and Verbalizing for Language Comprehension and

Thinking. Paso Robles, CA: Academy of Reading Publications.

Berninger, V. W., Vaughan, K., Abbott, R. D., Brooks, A., Abbott, S. P.,
Rogan, L., Reed, E., and Graham, S. 1998. Early intervention for spelling
problems: Teaching functional spelling units of varying size with a multiple-
connections framework. Journal of Educational Psychology, 90, 587–605.

Biemiller, A. [1977]1998. Relationship between oral reading rates for letters,

words, and simple text in the development of reading achievement. Reading
Research Quarterly, 13, 223–253.

Blachman, B. A., Tangel, D. M., Ball, E. W., Black, R., and McGraw, C. K.
1999. Developing phonological awareness and word recognition skills: A two-
year intervention with low-income, inner-city children. Reading and Writing:
An Interdisciplinary Journal, 11, 239–273.

Block, C. C. 1993. Strategy instruction in a literature-based reading program.

Elementary School Journal, 94, 139–151.
 373

Block, C. C. 2002. Improving Comprehension Instruction: Rethinking Research,

References
Theory, and Classroom Practice. Jossey-Bass Education Series, no. 20. San Fran-
cisco: Jossey-Bass.

Block, C. C., and Mangieri, J. N. 1997. Reason to Read: Thinking Strategies for
Life through learning. Pearson Learning report, no. 20.

Block, C. C., and Pressley, M., eds. 2001. Comprehension Instruction. New
York: Guilford Press.

Block, C. C., Gambrell, L., and Pressley, M. 1997. Training the Language
Arts: Expanding Thinking through Student-Centered Instruction. Boston: Allyn
and Bacon.

Bond, C. L., Ross, S. M., Smith, L. J., and Nunnery, J. A. 1995–1996. The
effects of the Sing, Spell, Reading and Write program on reading achievement
of beginning readers. Reading Research and Instruction, 35, 122–141.

Bond, G. L., and Dykstra, R. 1967. The cooperative research program in first-
grade reading instruction. Reading Research Quarterly, 2, 1–142.

Bond, G. L., and Dykstra, R. 1997. The cooperative research program in

first-grade reading instruction. (Reprint of the 1967 paper.) Reading Research
Quarterly, 32, 342–344.

Boronat, C. B., and Logan, G. D. 1997. The role of attention in automatiza-

tion: Does attention operate at encoding, or retrieval, or both: Memory and
Cognition, 25, 36–46.
Bradley, L., and Bryant, P. E. 1983. Categorizing sounds and learning to
read—a causal connection. Nature, 301, 419–421.

Bradley, L., and Bryant, P. E. 1985. Rhyme and Reason in Reading and Spelling.
Ann Arbor: University of Michigan Press.

Brady, S., Fowler, A., Stone, B., and Winbury, N. 1994. Training phonologi-
cal awareness: A study with inner-city kindergarten children. Annals of Dys-
lexia, 44, 26–59.

Brennan, F., and Ireson, J. 1997. Training phonological awareness: A study to

evaluate the effects of a program of metalinguistic games in kindergarten.
Reading and Writing: An Interdisciplinary Journal, 9, 241–263.

Brett, A., Rothlein, L., and Hurley, M. 1996. Vocabulary acquisition from lis-
tening to stories and explanations of target words. Elementary School Journal,
96, 415–422.
 374

Brown, A. S. 1988. Encountering misspellings and spelling performance: Why

References

wrong isn’t right. Journal of Educational Psychology, 80, 488–494.

Brown, G. D. A. 1987. Resolving inconsistency: A computational model of

word naming. Journal of Memory and Language, 26, 1–23.

Brown, I. S., and Felton, R. H. 1990. Effects of instruction on beginning

reading skills in children at risk for reading disability. Reading and Writing: An
Interdisciplinary Journal, 2, 223–241.

Brown, R., Pressley, M., Van Meter, P., and Schuder, T. 1996. A quasi-
experimental validation of transactional strategies instruction with low-
achieving second-grade readers. Journal of Educational Psychology, 88, 18–37.

Bruck, M., and Waters, G. S. 1990. An analysis of the component spelling and
reading skills of good readers–good spellers, good readers–poor spellers, and
poor readers–poor spellers. In T. H. Carr and B. A. Levy, Reading and Its De-
velopment: Component Skills Approaches, 161–206. New York: Academic Press.

Bus, A. G., and van IJzendoorn, M. H. 1999. Phonological awareness and

early reading: A meta-analysis of experimental training studies. Journal of Edu-
cational Psychology, 91, 403–414.

Byrne, B., and Fielding-Barnsley, R. 1989. Phonemic awareness and letter

knowledge in the child’s acquisition of the alphabetic principle. Journal of
Educational Psychology, 81, 313–321.

Byrne, B., and Fielding-Barnsley, R. 1990. Acquiring the alphabetic principle:

A case for teaching recognition of phoneme identity. Journal of Educational
Psychology, 82, 805–812.

Byrne, B., and Fielding-Barnsley, R. 1991. Sound Foundations. Sydney: Peter

Leyden Educational.

Byrne, B., and Fielding-Barnsley, R. 1993. Evaluation of a program to teach

phonemic awareness to young children: A 1 year follow-up. Journal of Educa-
tional Psychology, 85, 104–111.
Byrne, B., and Fielding-Barnsley, R. 1995. Evaluation of a program to teach
phonemic awareness to young children: A 2 and 3 year follow-up and a new
preschool trial. Journal of Educational Psychology, 87, 488–503.

Byrne, B., and Fielding-Barnsley, R. 2000. Effects of preschool phoneme

identity training after six years: Outcome level distinguished from rate of
response. Journal of Educational Psychology, 92, 659–667.
 375

Calfee, R. C., and Henry, M. K. 1985. Project READ: An inservice model for

References
training classroom teachers in effective reading instruction. In J. V. Hoffman,
ed., Effective Teaching of Reading: Research and Practice, 199–229. Newark, DE:
International Reading Association.

Campbell, J. R., Donahue, P. L., Reese, C. M., and Phillips, G. W. 1996.

National Assessment of Educational Progress 1994: Reading Report Card for the
Nation and States. Washington, DC: Office of Educational Research and Im-
provement, U.S. Department of Education.

Cattell, J. M. 1886. The time taken up by cerebral operations. Mind, 11, 220–
242, 377–392, 524–538.

Chall, J. 1967. Learning to Read: The Great Debate. New York: McGraw-Hill.

Chall, J., and Feldman, S. 1966. First grade reading: An analysis of the inter-
actions of professed methods, teacher implementation and child background.
The Reading Teacher, 19, 569–575.

Chaney, C. 1992. Language development, metalinguistic skills, and print

awareness in 3-year-old children. Applied Psycholinguistics, 13, 485–514.

Chang, Kwang-chih. 1963. The Archeology of Ancient China. New Haven, CT:
Yale University Press.

Chomsky, C. 1976. When you still can’t read in the third grade: After decod-
ing, what? Language Arts, 53, 288–296.

Chomsky, N., and Halle, M. 1968. The Sound Patterns of English. New York:
Harper and Row.

Civil, M. 1973. The Sumerian writing system: Some problems. Orientalis, 42,
21–34.

Clanchy, M. T. 1994. From Memory to Written Record: England 1066–1307.

Oxford: Blackwell.

Clay, M. 1985. The Early Detection of Reading Difficulties. Tadworth, Surrey:

Heinemann.

Clymer, T. 1983. The utility of phonic generalizations in the primary grades.

In L. M. Gentile, M. L. Kamil, and J. S. Blanchard, eds., Reading Research
Revisited, 113–119. Columbus, OH: Merrill.

Coltheart, M. 1978. Lexical access in simple reading tasks. In G. Underwood,

ed., Strategies of Information Processing, 151–216. San Diego, CA: Academic
Press.
 376

Coltheart, M., Curtis, B., Atkins, P., and Haller, M. 1993. Models of reading
References

aloud: Dual-route and parallel-distributed processing approaches. Psychological

Review, 100, 585–608.

Cooper, J. S. 1996. Sumerian and Akkadian. In P. T. Daniels and W. Bright,

The World’s Writing Systems. New York: Oxford University Press.

Cossu, G., Rossini, F., and Marshall, J. C. 1993. When reading is acquired but
phonemic awareness is not: A study of literacy in Down’s syndrome. Cognition,
46, 129–138.
Coulmas, F. 1989. The Writing Systems of the World. Oxford: Blackwell.

Cunningham, A. E. 1990. Explicit versus implicit instruction in phoneme

awareness. Journal of Experimental Child Psychology, 50, 429–444.

Cunningham, A. E., and Stanovich, K. E. 1990a. Assessing print exposure and

orthographic processing skill in children: A quick measure of reading experi-
ence. Journal of Educational Psychology, 82, 733–740.
Cunningham, A. E., and Stanovich, K. E. 1990b. Early spelling acquisi-
tion: Writing beats the computer. Journal of Educational Psychology, 82, 159–
162.

Cunningham, A. E., Stanovich, K. E., and West, R. F. 1994. Literacy envi-

ronment and the development of children’s cognitive skills. In E. M. H.
Assink, ed., Literacy Acquisition and Social Context, 70–90. New York: Harvester
Wheatsheaf.
Dahl, P. R. 1979. An experimental program for teaching high-speed word
recognition and comprehension skills. In J. E. Button, T. Lovitt, and T.
Rowland, eds., Communications Research in Learning Disabilities and Mental
Retardation, 33–65. Baltimore, MD: University Park Press.

Dale, N. 1898. On the Teaching of English Reading. London: Dent.

Dale, N. 1902. Further Notes on the Teaching of English Reading. London: G.

Philips and Son.

Daniels, P. T., and Bright, W. 1996. The World’s Writing Systems. New York:
Oxford University Press.

Dickinson, D. K., and Smith, M. W. 1994. Long-term effects of preschool

teachers’ book reading on low-income children’s vocabulary and story com-
prehension. Reading Research Quarterly, 29, 105–122.
 377

Dixon, M., and Kaminska, Z. 1997. Is it misspelled or is it mispelled? The

References
influence of fresh orthographic information on spelling. Reading and Writing:
An Interdisciplinary Journal, 9, 483–498.

Dowhower, S. L. 1987. Effects of repeated reading on second-grade tran-

sitional readers’ fluency and comprehension. Reading Research Quarterly, 22,
389–406.

Dykstra, R. 1967. Continuation of the Coordinated Center for First-Grade Reading

Instruction Programs. Report of Project No. 6-1651. Minneapolis: University of
Minnesota.

Dykstra, R. 1968a. The effectiveness of code and meaning-emphasis begin-

ning reading programs. The Reading Teacher, 22, 17–24.

Dykstra, R. 1968b. Summary of the second-grade phase of the Cooperative

Research Program in primary reading instruction. Reading Research Quarterly,
4, 49–70.

Ehri, L. C. 1980. The development of orthographic images. In U. Frith, ed.,

Cognitive Processes in Spelling, 311–338. London: Academic Press.

Ehri, L. C. 1986. Sources of difficulty in learning to spell and read. In M. L.

Wolraich and D. Routh, eds., Advances in Developmental and Behavioral Pediat-
rics, 121–195. Greenwich, CT: JAI Press.

Ehri, L. C. 1989a. Knowledge and its role in reading acquisition and reading
disability. Journal of Learning Disabilities, 22, 356–365.
Ehri, L. C. 1989b. Movement into word reading and spelling: How spelling
contributes to reading. In J. Mason, ed., Reading and Writing Connections,
65–81. Boston: Allyn and Bacon.

Ehri, L. C. 1991. Development of the ability to read words. In R. Barr, M. L.

Kamil, P. B. Mosenthal, and P. D. Pearson, eds., Handbook of Research in
Reading, vol. 2, 383–417. New York: Longman.

Ehri, L. C. 1992. Review and commentary: Stages of spelling development. In

S. Templeton and D. R. Bear, eds., Development of Orthographic Knowledge and
the Foundations of Literacy. Hillsdale, NJ: Erlbaum.

Ehri, L. C. 1995. Phases of development in learning to read words by sight.

Journal of Research in Reading, 18, 115–125.
 378

Ehri, L. C. 1998. Word reading by sight and by analogy in beginning readers.

References

In C. Hulme and R. M. Joshi, eds., Reading and Spelling: Development and Dis-
orders. Hillsdale, NJ: Erlbaum.

Ehri, L. C., and Wilce, L. S. 1987. Does learning to spell help beginners learn
to read words? Reading Research Quarterly, 22, 47–65.

Eldredge, L. 1991. An experiment with a modified whole language approach

in first-grade classrooms. Reading Research and Instruction, 30, 21–38.

Elley, W. B. 1989. Vocabulary acquisition from listening to stories. Reading

Research Quarterly, 24, 174–187.

Ellis, A. J. 1870. Transactions of the Philological Society, 89–118. London.

Engelmann, S., and Bruner, E. 1969. Distar reading program. Chicago: Science
Research Associates.

Evans, M. A., and Carr, T. H. 1985. Cognitive abilities, conditions of learn-

ing, and early development of reading skill. Reading Research Quarterly, 20,
327–350.

Falkenstein, A. 1964. Das Sumerische. Reprint from Handbuch der Orientalistik.

Leiden: Brill.

Faulkner, H. J., and Levy, B. A. 1994. How text difficulty and reader skill in-
teract to produce differential reliance on word and content overlap in reading
transfer. Journal of Experimental Child Psychology, 58, 1–24.
Faulkner, H. J., and Levy, B. A. 1999. Fluent and nonfluent forms of trans-
fer in reading: Words and their message. Psychonomic Bulletin and Review, 6,
111–116.

Fischer, F. W., Shankweiler, D., and Liberman, I. Y. 1985. Spelling profi-

ciency and sensitivity to word structure. Journal of Memory and Language, 24,
423–441.

Fisher, R., and Craik, F. I. M. 1977. The interaction between encoding and
retrieval operations in cued recall. Journal of Experimental Psychology: Human
Learning and Memory, 3, 701–711.

Flavell, J. H. 1963. The Developmental Psychology of Jean Piaget. Princeton, NJ:

Van Nostrand.

Flesch, R. [1955] 1985. Why Johnny Can’t Read. 3rd ed. New York: Harper
and Row.
 379

Foorman, B. R., Francis, D. J., Beeler, T., Winikates, D., and Fletcher, J. M.

References
1997. Early interventions for children with reading problems: Study designs
and preliminary findings. Learning Disabilities, 8, 63–71.

Foorman, B. R., Francis, D. J., Fletcher, J. M., Schatschneider, C., and Mehta,
P. 1998. The role of instruction in learning to read: Preventing reading failure
in at-risk children. Journal of Educational Psychology, 90, 37–55.

Foorman, B. R., and Liberman, D. 1989. Visual and phonological processing

of words: A comparison of good and poor readers. Journal of Learning Dis-
abilities, 22, 349–355.

Freud, S. 1966. Project for a Scientific Psychology. Ed. J. Strachey. Standard

Edition of the Complete Psychological Works of Sigmund Freud, vol. 1, 281–
397. New York: Norton.

Friederici, A. D., and Wessels, J. M. I. 1993. Phonotactic knowledge and its

use in infant speech perception. Perception and Psychophysics, 54, 287–295.

Frith, U. 1980. Unexpected spelling problems. In U. Frith, ed., Cognitive Pro-

cesses in Spelling. New York: Academic Press.

Frith, U. 1985. Beneath the surface of developmental dyslexia. In K. E.

Patterson, J. C. Marshall, and M. Coltheart, eds., Surface Dyslexia: Neuro-
psychological and Cognitive Analyses of Phonological Reading. London: Erlbaum.

Fulwiler, G., and Groff, P. 1980. The effectiveness of intensive phonics.

Reading Horizons, 21, 50–54.
Gelb, I. 1963. A Study of Writing. Chicago: University of Chicago Press.

Gentry, J. 1981. Learning to spell developmentally. Reading Teacher, 34,

378–381.

Gersten, R., Darch, C., and Gleason, M. 1988. Effectiveness of a direct in-
struction academic kindergarten for low-income students. Elementary School
Journal, 89, 227–240.
Geva, E., and Siegel, L. S. 2000. Orthographic and cognitive factors in the
concurrent development of basic reading skills in two languages. Reading and
Writing: An Interdisciplinary Journal, 12, 1–30.

Glushko, R. J. 1979. The organization and activation of orthographic knowl-

edge in reading aloud. Journal of Experimental Psychology: Human Perception and
Performance, 5, 674–691.
 380

Goodman, K. 1967, May. Reading: A psycholinguistic guessing game. Journal

References

of the Reading Specialist, 126–135.

Goswami, U. C. 1990. Phonological priming and orthographic analogies in

reading. Journal of Experimental Child Psychology, 49, 323–340.

Graham, S. 2000. Should the natural learning approach replace spelling in-
struction? Journal of Educational Psychology, 92, 235–247.

Graves, M. F., and Dykstra, R. 1997. Contextualizing the first-grade studies:

What is the best way to teach children to read? Reading Research Quarterly, 32,
342–344.

Greenberg, S. N., and Koriat, A. 1991. The missing-letter effect for common
function words depends on their linguistic function in the phrase. Journal of
Experimental Psychology: Learning, Memory and Cognition, 17, 1051–1061.

Griffith, P. L., Klesius, J. P., and Kromrey, J. D. 1992. The effect of phonemic
awareness on the literacy development of first grade children in a traditional or
a whole language classroom. Journal of Research in Childhood Education, 6, 85–92.

Haddock, M. 1976. Effects of an auditory and an auditory-visual method of

blending instruction on the ability of prereaders to decode synthetic words.
Journal of Educational Psychology, 68, 825–831.

Hanna, P. R., Hanna, J. S., Hodges, R. E., and Rudorf, E. H. 1966. Phoneme-
Grapheme Correspondences as Cues to Spelling Improvement. Washington, DC:
U.S. Department of Health, Education, and Welfare, Office of Education.
Hart, B., and Risley, T. R. 1995. Meaningful Differences. Baltimore: Paul H.
Brookes.

Hatcher, P. J., Hulme, C., and Ellis, A. W. 1994. Ameliorating early reading
failure by integrating the teaching of reading and phonological skills: The
phonological linkage hypothesis. Child Development, 65, 41–57.

Haynes, D. P., and Ahrens, M. 1988. Vocabulary simplification for children: A

special case of ‘‘motherese’’? Journal of Child Language, 15, 395–410.

Healy, A. F. 1976. Detection errors on the word the: Evidence for reading
units larger than letters. Journal of Experimental Psychology: Human Perception
and Performance, 2, 235–242.

Helfgott, J. A. 1976. Phonemic segmentation and blending skills of kinder-

garten children: Implications for beginning reading acquisition. Contemporary
Educational Psychology, 1, 157–169.
 381

Henderson, E. H. 1982. Orthography and Word Recognition in Reading. New

References
York: Academic Press.

Henderson, E. H. 1992. The interface of lexical competence and knowledge

of written words. In S. Templeton and D. R. Bear, eds., Development of
Orthographic Knowledge and the Foundations of Literacy, 1–30. Hillsdale, NJ:
Erlbaum.

Henderson, E. H., and Beers, J. W., eds. 1980. Developmental and Cognitive
Aspects of Learning to Spell: A Reflection of Word Knowledge. Newark, DE: Inter-
national Reading Association.

Henry, M. K. 1989. Children’s word structure knowledge: Implications for

decoding and spelling instruction. Reading and Writing: An Interdisciplinary
Journal, 2, 135–152.

Herman, P. A. 1985. The effect of repeated readings on reading rate, speech

pauses, and word recognition accuracy. Reading Research Quarterly, 20, 553–
564.
Ho, Ping-ti. 1976. The Cradle of the East. Chicago: University of Chicago
Press.

Hodges, R. E. 1981. Learning to Spell. Urbana, IL: National Council of

Teachers of English.

Hodges, R. E. 1982. Improving Spelling and Vocabulary in the Secondary School.

Urbana, IL: National Council of Teachers of English.
Hohn, W. E., and Ehri, L. C. 1983. Do alphabet letters help prereaders
acquire phonemic segmentation skill? Journal of Educational Psychology, 78,
243–255.

Holmes, V. M., and Ng, E. 1993. Word specific knowledge, word-recognition

strategies and spelling ability. Journal of Memory and Language, 32, 230–257.

Hoover, W. A., and Gough, P. B. 1990. The simple view of reading. Reading
and Writing: An Interdisciplinary Journal, 2, 127–160.

Howard, M. 1982. Utilizing oral-motor feedback in auditory conceptualiza-

tion. Journal of Educational Neuropsychology, 2, 24–35.

Howard, M. P. 1986. Effects of pre-reading training in auditory conceptual-

ization on subsequent reading achievement. Unpublished doctoral disserta-
tion, Brigham Young University.
 382

Hudson, J. A., and Shapiro, L. R. 1991. From knowing to telling: The devel-
References

opment of children’s scripts, stories and personal narratives. In A. McCabe

and C. Peterson, eds., Developing Narrative Structure, 89–137. Hillsdale, NJ:
Erlbaum.

Huey, E. B. 1908. The Psychology and Pedagogy of Reading. New York:

Macmillan.

Hulme, C. 1981. Reading Retardation and Multi-sensory Teaching. London:

Routledge and Kegan Paul.
Hulme, C., and Bradley, L. 1984. An experimental study of multi-sensory
teaching with normal and retarded readers. In R. Malatesha and H. Whitaker,
eds., Dyslexia: A Global Issue, 431–443. The Hague: Nijhoff.

Hulme, C., Monk, A., and Ives, S. 1987. Some experimental studies of multi-
sensory teaching: The effects of manual tracing on children’s paired associate
learning. British Journal of Developmental Psychology, 5, 299–307.

Jacoby, L. L., and Hollingshead, A. 1990. Reading student essays may be haz-
ardous to your spelling: Effects of reading incorrectly and correctly spelled
words. Canadian Journal of Psychology, 44, 345–358.

Jared, D., McRae, K., and Seidenberg, M. S. 1990. The basis of consistency
effects in word naming. Journal of Memory and Language, 29, 687–715.

Jeffrey, W. E., and Samuels, S. J. 1967. Effect of method of reading training

on initial learning and transfer. Journal of Verbal Learning and Verbal Behavior,
6, 354–358.

Jenkins, J. R., Matlock, B. A., and Slocum, T. A. 1989. Two approaches to

vocabulary instruction: The teaching of individual word meanings and prac-
tice in deriving word meaning from context. Reading Research Quarterly, 24,
215–235.

Jensen, H. 1969. Sign, Symbol, and Script. New York: Putnam.

Johnson, S. [1755] 1773. A Dictionary of the English Language. 4th ed. London:
Strahan.

Johnson-Glenberg, M. C. 2000. Training reading comprehension in adequate

decoders/poor comprehenders: Verbal versus visual strategies. Journal of Edu-
cational Psychology, 92, 772–782.

Johnston, R. S., and Watson, J. 1997, July. Developing reading, spelling and
phonemic awareness skills in primary school children. Reading, 37–40.
 383

Johnston, R. S., and Watson, J. 2003. Accelerating reading and spelling with

References
synthetic phonics: A five year follow up. Interchange 4, ISSN 1478-6788, 1–8.
Edinburgh: Scottish Executive Education Department.

Johnston, R. S., and Watson, J. Forthcoming. Accelerating the development of

reading, spelling, and phonemic awareness skills in initial readers. Reading and
Writing.

Jordan, T. R. 1986. Testing the BOSS hypothesis: Evidence of position-

insensitive orthographic priming in the lexical decision task. Memory and Cog-
nition, 14, 523–532.

Joshi, R. M., Williams, K. A., and Wood, J. R. 1998. Predicting reading com-
prehension from listening comprehension: Is this the answer to the IQ de-
bate? In C. Hulme and R. M. Joshi, eds., Reading and Spelling: Development and
Disorders, 319–327. Mahwah, NJ: Erlbaum.

Juel, C. 1983. The development and use of mediated word identification.

Reading Research Quarterly, 18, 306–327.
Juel, C., Griffith, P. L., and Gough, P. 1986. Acquisition of literacy: A longi-
tudinal study of children in first and second grade. Journal of Educational Psy-
chology, 78, 243–255.

Juel, C., and Solso, R. L. 1981. The role of orthographic and phonic structure
in word identification. In M. L. Kamil and A. J. More, eds., Perspectives in
Reading Research and Instruction: 30th Yearbook. Washington, DC: National
Reading Conference.
Jusczyk, P. W. 1998. The Discovery of Spoken Language. Cambridge, MA: MIT
Press.

Karlgren, B. 1923. Analytic Dictionary of Chinese and Sino-Japanese. Paris:

Geuthner.

Katz, L., and Frost, R. 1992. The reading process is different for different
orthographies: The orthographic depth hypothesis. In Orthography, Phonology
and Meaning, 67–83. Amsterdam: North-Holland.

Klesius, J. P., Griffith, P. L., and Zielonka, P. 1991. Whole language and tra-
ditional instruction comparison: Overall effectiveness and development of the
alphabetic principle. Reading Research and Instruction, 30, 47–61.

Kramer, S. N. [1956] 1981. History Begins at Sumer. 2nd ed. Philadelphia:

University of Pennsylvania Press.
 384

Kramer, S. N. 1963. The Sumerians. Chicago: University of Chicago Press.

References

Krevisky, J., and Linfield, J. L. 1990. The Awful Speller’s Dictionary. New York:
Random House.

Landerl, K., Wimmer, H., and Frith, U. 1997. The impact of ortho-
graphic consistency on dyslexia: A German-English comparison. Cognition, 63,
315–334.

Larson, S. C., and Hammill, D. D. 1994. Test of Written Spelling. Austin, TX:
Pro-Ed.

Leinhardt, G., and Engel, M. 1981. An iterative evaluation of NRS. Evaluation

Review, 5, 579–601.

Leslie, L., and Thimke, B. 1986. The use of orthographic knowledge in be-
ginning reading. Journal of Reading Behavior, 18, 229–241.

Levy, B. A., DiPersio, R., and Hollingshead, A. 1992. Fluent rereading:

Repetition, automaticity, and discrepancy. Journal of Experimental Psychology:
Learning Memory and Cognition, 18, 957–971.

Levy, B. A., and Kirsner, K. 1989. Reprocessing text: Indirect measure of

word and message level processes. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 15, 407–417.

Levy, B. A., Nicholls, A., and Kohen, D. 1993. Repeated readings: Process
benefits for good and poor readers. Journal of Experimental Child Psychology, 56,
303–327.

Liberman, A. M., Cooper, F. S., Shankweiler, D. P., and Studdert-Kennedy,

M. 1967. Perception of the speech code. Psychological Review, 74, 431–461.

Liberman, I. Y., Shankweiler, D., Liberman, A. M., Fowler, C., and Fisher,
F. W. 1974. Explicit syllable and phoneme segmentation in the young child.
Journal of Experimental Child Psychology, 18, 201–212.

Lie, F. 1991. Effects of a training program for stimulating skill in word analy-
sis in first-grade children. Reading Research Quarterly, 26, 234–250.

Lindamood, C. H., and Lindamood, P. C. 1969. Auditory Discrimination in

Depth. Allen, TX: DLM Teaching Resources.

Lindamood, P. 1991. Reports of Santa Maria-Bonita Project: 1985–1988. Fort

Osage follow on 1991. San Luis Obispo, CA: Lindamood–Bell Learning
Processes.
 385

Lloyd, S. 1992. The Phonics Handbook. Essex, England: Jolly Learning Ltd.

References
Lundberg, I., Frost, J., and Petersen, O. 1988. Effects of an extensive program
for stimulating phonological awareness in preschool children. Reading Research
Quarterly, 23, 263–284.

Lundberg, I., Olofsson, A., and Wall, S. 1980. Reading and spelling skills in
the first school years predicted from phonemic awareness skills in kindergar-
ten. Scandinavian Journal of Psychology, 21, 159–173.

Lysynchuk, L. M., Pressley, M., D’Ailly, H., Smith, M., and Cake, H. 1989.
A methodological analysis of experimental studies of comprehension strategy
instruction. Reading Research Quarterly, 24, 458–470.

Mair, V. 1996. Modern Chinese writing. In P. T. Daniels and W. Bright, eds.,

The World’s Writing Systems, 200–208. Oxford: Oxford University Press.

Mann, V. A. 1993. Phoneme awareness and future reading ability. Journal of

Learning Disabilities, 26, 259–269.
Mann, V. A., and Ditunno, P. 1990. Phonological deficiencies: Effective pre-
dictors of future reading problems. In G. T. Pavlidis, ed., Perspectives on Dys-
lexia, vol. 2 New York: Wiley.

Markman, E. M. 1978. Realizing that you don’t understand: A preliminary

investigation. Child Development, 48, 986–992.

Martinussen, R., and Kirby, J. 1998. Instruction in successive and phonologi-

cal processing to improve the reading acquisition of at-risk kindergarten chil-
dren. Developmental Disabilities Bulletin, 26, 19–39.

Mattingly, I. G. 1985. Did orthographies evolve? Remedial and Special Educa-

tion, 6, 18–23.

Mattys, S. L., Jusczyk, P. W., Luce, P. A., and Morgan, J. L. 1999. Phono-
tactic and prosodic effects on word segmentation in infants. Cognitive Psychol-
ogy, 38, 465–494.
McArthur, T., ed. 1992. The Oxford Companion to the English Language.
Oxford: Oxford University Press.

McCracken, G., and Walcutt, C. C. 1963. Basic Reading. Philadelphia:

Lippincott.

McGuinness, C., McGuinness, D., and McGuinness, G. 1996. Phono-

Graphix: A new method for remediating reading difficulties. Annals of Dyslexia,
46, 73–96.
 386

McGuinness, C., and McGuinness, G. 1998. Reading Reflex. Simon and

References

Schuster/Free Press.

McGuinness, D. 1992. Allographs Dictionary. Unpublished manuscript.

McGuinness, D. 1997a. Allographs I: A Linguistic Spelling Program. Sanibel, FL:

SeaGate Press.

McGuinness, D. 1997b. Decoding strategies as predictors of reading skill: A

follow-on study. Annals of Dyslexia, 47, 117–150.
McGuinness, D. 1997c. Why Our Children Can’t Read. New York: Simon and
Schuster/Free Press.

McGuinness, D. 1998a. Allographs II: A Linguistic Spelling Program. Multisyl-

lable Word Building. Sanibel, FL: SeaGate Press.

McGuinness, D. 1998b. Why Children Can’t Read. London: Penguin Books.

McGuinness, D. 2004. Language Development and Learning to Read. Cam-

bridge, MA: MIT Press.

McGuinness, D. Forthcoming. Sound Steps to Reading.

McGuinness, D., McGuinness, C., and Donohue, J. 1995. Phonological

training and the alphabet principle: Evidence for reciprocal causality. Reading
Research Quarterly, 30, 830–852.

McGuinness, D., and Pribram, K. H. 1980. The neuropsychology of atten-

tion: Emotional and motivational controls. In M. C. Wittrock, ed., The Brain
and Educational Psychology. New York: Academic Press.

McKeown, M. G., Beck, I. L., Omanson, R. C., and Pople, M. T. 1985. Some
effects of the nature and frequency of vocabulary instruction on the knowledge
and use of words. Reading Research Quarterly, 20, 522–535.

McNemar, Q. 1949. Psychological Statistics. New York: Wiley.

Meyer, D. E., Schvaneveldt, R. W., and Ruddy, M. G. 1971. Facilitation in
recognizing pairs of words: Evidence of a dependence between retrieval oper-
ations. Journal of Experimental Psychology, 90, 227–234.

Meyer, L. A., Stahl, S. A., Linn, R. L., and Wardrop, J. L. 1994. Effects of
reading storybooks aloud to children. Journal of Educational Research, 88, 69–85.

Michalowski, P. 1996. Mesopotamian cuneiform. In P. T. Daniels and W.

Bright, eds., The World’s Writing Systems, 33–36. Oxford: Oxford University
Press.
 387

Morris, D., and Perney, J. 1984. Developmental spelling as a predictor of first-

References
grade reading achievement. Elementary School Journal, 84, 441–457.

Morris, J. M. 1984. Phonics: From an unsophisticated past to a linguistics-

informed future. In G. Brooks and A. K. Pugh, eds., Studies in the History of
Reading. Reading, England: University of Reading.

Mullis, I. V. S., Campbell, J. R., and Farstrup, A. E. 1993. National Assessment

of Educational Progress 1992: Reading Report Card for the Nation and States.
Washington, DC: Office of Educational Research and Improvement, U.S.
Department of Education.

Muter, V., and Snowling, M. 1997. Grammar and phonology predict spell-
ing in middle childhood. Reading and Writing: An Interdisciplinary Journal, 9,
407–425.

Nagy, W. F., and Herman, P. 1987. Breadth and depth of vocabulary knowl-
edge: Implications for acquisition and instruction. In M. McKeown and
M. Curtis, eds., The Nature of Vocabulary Acquisition, 19–36. Hillsdale, NJ:
Erlbaum.

Nation, K., and Snowling, M. 1997. Assessing reading difficulties: The validity
and utility of current measures of reading skill. British Journal of Educational
Psychology, 67, 359–370.

National Reading Panel. 2000. Report. Washington, DC: National Institute of

Child Health and Human Development.

Nelson, K. 1998. Language in Cognitive Development. Cambridge, England:

Cambridge University Press.

Nisbet, S. D. 1939. Non-dictated spelling tests. British Journal of Educational

Psychology, 9, 29–44.

Nunes, T., Bryant, P., and Bindman, M. 1997. Morphological spelling strat-
egies: Developmental stages and processes. Developmental Psychology, 33, 637–
649.
Olson, R., Forsberg, H., Wise, B., and Rack, J. 1994. Measurement of word
recognition, orthographic, and phonological skills. In G. R. Lyon, ed., Frames
of Reference for the Assessment of Learning Disabilities: New Views on Measurement
Issues, 229–277. Baltimore: Brookes.

Organization for Economic Cooperation and Development. 1995. Literacy,

Economy, and Society. Ottawa: Statistics Canada.
 388

Organization for Economic Cooperation and Development. 1997. Literacy

References

Skills. Ottawa: Statistics Canada.

Palincsar, A. S., and Brown, A. L. 1984. Reciprocal teaching of

comprehension-fostering and comprehension-monitoring activities. Cognition
and Instruction, 1, 117–175.

Pearson, D. 1997. The first grade studies: A personal reflection. Reading Re-
search Quarterly, 32, 428–432.

Perfetti, C. A. 1984. Reading acquisition and beyond: Decoding includes cog-

nition. American Journal of Education, 40–57.

Piaget, J. [1964] 1993. Development and learning. In M. Gauvain and M.

Cole, eds., Readings on the Development of Children, 25–33. New York: Scien-
tific American Books. W. H. Freeman.

Pinel, J. P. J. 1997. Biopsychology. Boston: Allyn and Bacon.

Pintner, R., Rinsland, H. D., and Zubin, J. 1929. The evaluation of self-
administering spelling tests. Journal of Educational Psychology, 20, 107–111.

Pitman, J., and St. John, J. 1969. Alphabets and Reading. London: Pitman.

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., and Patterson, K. 1996.

Understanding normal and impaired word reading: Computational principles
in quasi-regular domains. Psychological Review, 103, 56–115.

Plomin, R., Fulker, D. W., Corley, R., and DeFries, J. C. 1997. Nature, nur-
ture, and cognitive development from 1 to 16 years: A parent-offspring adop-
tion study. Psychological Science, 8, 442–448.

Postman, L. 1975. Verbal learning and memory. Annual Review of Psychology,

26, 291–335.

Pribram, K. H. 1991. Brain and Perception: Holonomy and Structure in Figural

Processing. Hillsdale, NJ: Erlbaum.
Pribram, K. H., and Gill, M. 1976. Freud’s Project Re-assessed. New York: Basic
Books.

Pribram, K. H., and McGuinness, D. 1975. Arousal, activation, and effort in

the control of attention. Psychological Review, 82, 116–149.

Rashotte, C. A., and Torgesen, J. K. 1985. Repeated reading and reading flu-
ency in learning disabled children. Reading Research Quarterly, 20, 180–188.
 389

Rayner, K. 1977. Visual attention in reading: Eye movements reflect cognitive

References
processes. Memory and Cognition, 4, 443–448.

Rayner, K. 1986. Eye movements and the perceptual span in beginning and
skilled readers. Journal of Experimental Child Psychology, 41, 211–236.

Rayner, K. 1998. Eye movements in reading and information processing: 20

years of research. Psychological Bulletin, 124, 372–422.

Read, C. 1971. Pre-school children’s knowledge of English phonology. Har-

vard Educational Review, 41, 1–34.

Read, C. 1986. Children’s Creative Spelling. London: Routledge and Kegan

Paul.

Robbins, C., and Ehri, L. C. 1994. Reading storybooks to kindergartners

helps them learn new vocabulary words. Journal of Educational Psychology, 86,
54–64.

Robinson, A. 1995. The Story of Writing. London: Thames and Hudson.

Rosenshine, B., and Meister, C. 1994. Reciprocal teaching: A review of the

research. Review of Educational Research, 64, 479–530.

Rosenshine, B., Meister, C., and Chapman, S. 1996. Teaching students to

generate questions: A review of the intervention studies. Review of Educational
Research, 66, 181–221.

Samuels, S. J. 1972. The effect of letter-name knowledge on learning to read.

American Educational Research Journal, 9, 65–74.

Samuels, S. J. 1979. The method of repeated readings. The Reading Teacher, 4,

403–408.

Samuels, S. J., Laberge, D., and Bremer, C. D. 1978. Units of word recogni-
tion: Evidence for developmental changes. Journal of Verbal Learning and Ver-
bal Behavior, 17, 715–720.
Samuels, S. J., Miller, N. L., and Eisenberg, P. 1979. Practice effects on the
unit of word recognition. Journal of Educational Psychology, 71, 514–520.

Schatschneider, C., Francis, D. J., Foorman, B. R., Fletcher, J. M., and Mehta,
P. 1999. The dimensionality of phonological awareness: An application of
item response theory. Journal of Educational Psychology, 91, 439–449.

Schmandt-Besserat, D. 1978. The earliest precursors of writing. Scientific

American, 238, 50–59.
 390

Schneider, W., Roth, E., and Ennemoser, M. 2000. Training phonological

References

skills and letter knowledge in children at risk for dyslexia: A comparison of

three kindergarten intervention programs. Journal of Educational Psychology, 92,
284–295.

Scragg, D. G. 1974. A History of English Spelling. Manchester, England: Man-

chester University Press.

Seidenberg, M. S. 1992. Dyslexia in a computational model of word recogni-

tion in reading. In P. B. Gough, L. C. Ehri, and R. Treiman, eds., Reading
Acquisition. Hillsdale, NJ: Erlbaum.

Seidenberg, M. S., and McClelland, J. L. 1989. A distributed, developmental

model of word recognition and naming. Psychological Review, 96, 523–568.

Senechal, M., and Cornell, E. H. 1993. Vocabulary acquisition through shared

reading experiences. Reading Research Quarterly, 28, 361–374.

Shanahan, T. 1984. Nature of the reading-writing relation: An exploratory

multivariate analysis. Journal of Educational Psychology, 76, 466–477.

Share, D. L., Jorm, A. F., Maclean, R., and Matthews, R. 1984. Sources of
individual difference in reading acquisition. Journal of Educational Psychology,
76, 1309–1324.

Siegel, L. S., Share, D., and Geva, E. 1995. Evidence for superior ortho-
graphic skills in dyslexics. Psychological Science, 6, 250–254.

Silberberg, N., Iversen, I., and Goins, J. 1973. Which remedial method works
best? Journal of Learning Disabilities, 6, 18–22.

Smith, A. A. 1999. The simple logic of sound-to-letter mapping: A reversible

code. Unpublished master’s thesis, Massey University, Albany, New Zealand.

Solso, R. L., and Juel, C. 1980. Positional frequency and versatility of bigrams
for two- through nine-letter English words. Behavior Research Methods and
Instrumentation, 12, 297–343.
Stahl, S. A., and Fairbanks, M. M. 1986. The effects of vocabulary instruction:
A model-based meta-analysis. Review of Educational Research, 56, 72–110.

Stahl, S. A., and Miller, P. D. 1989. Whole language and language experience
approaches for beginning reading: A quantitative research synthesis. Review of
Educational Research, 59, 87–116.

Stanovich, K. E., and West, R. 1989. Exposure to print and orthographic

processing. Reading Research Quarterly, 24, 402–433.
 391

Stebbins, L., St. Pierre, R. G., Proper, E. L., Anderson, R. B., and Cerva, T. R.

References
1977. Education as Experimentation: A Planned Variation Model. Vols. IVA–D.
Cambridge, MA: Abt Associates.

Steffler, D. J., Varnhagen, C. K., Treiman, R., and Friesen, C. K. 1998.

There’s more to children’s spelling than the errors they make: Strategic and
automatic processes for one-syllable words. Journal of Educational Psychology,
90, 492–505.

Stuart, M. 1999. Getting ready for reading: Early phoneme awareness and
phonics teaching improves reading and spelling in inner-city second language
learners. British Journal of Educational Psychology, 69, 587–605.

Sumbler, K. 1999. Phonological awareness combined with explicit alphabetic

coding instruction in kindergarten: Classroom observations and evaluation.
Unpublished doctoral dissertation, University of Toronto.

Szeszulski, P. A., and Manis, F. R. 1990. An examination of familiar resem-

blance among subgroups of dyslexics. Annals of Dyslexia, 40, 180–191.
Torgesen, J. K., Wagner, R. K., Rashotte, C. A., Rose, E., Lindamood, P.,
Conway, T., and Garvan, C. 1999. Preventing reading failure in young chil-
dren with phonological processing disabilities: Group and individual responses
to instruction. Journal of Educational Psychology, 91, 579–593.

Treiman, R. 1994. Use of consonant letter names in beginning spelling.

Developmental Psychology, 30, 567–580.

Treiman, R., and Broderick, V. 1998. What’s in a name: Children’s knowl-

edge about the letters in their own names. Journal of Experimental Child Psy-
chology, 70, 97–116.

Treiman, R., Mullennix, J., Bijeljac-Babic, R., and Richmond-Welty, E. D.

1995. The special role of rimes in the description, use, and acquisition of
English orthography. Journal of Educational Psychology, 124, 107–136.

Treiman, R., and Tincoff, R. 1997. The fragility of the alphabetic princi-
ple: Children’s knowledge of letter names can cause them to spell syllabi-
cally rather than alphabetically. Journal of Experimental Child Psychology, 64,
425–451.

Uhry, J. K., and Shepherd, M. J. 1993. Segmentation/spelling instruction as

part of a first-grade reading program: Effects on several measures of reading.
Reading Research Quarterly, 28, 219–233.
 392

Varnhagen, C. K., McCallum, M., and Burstown, M. 1997. Is children’s

References

spelling naturally stage-like? Reading and Writing: An Interdisciplinary Journal,

9, 451–481.

Vellutino, F. R., and Scanlon, D. M. 1987. Phonological coding, phonological

awareness, and reading ability: Evidence from a longitudinal and experimental
study. Merrill-Palmer Quarterly, 33, 321–363.

Vellutino, F. R., Scanlon, D. M., and Tanzman, M. S. 1994. Components of

reading ability: Issues and problems in operationalizing word identification,
phonological coding, and orthographic coding. In G. R. Lyon, ed., Frames of
Reference for the Assessment of Learning Disabilities: New Views on Measurement
Issues, 279–332. Baltimore: Brookes.

Venezky, R. L. 1967. English orthography: Its graphical structure and its

relation to sound. Reading Research Quarterly, 2, 75–105.

Venezky, R. L. 1970. The Structure of English Orthography. The Hague:

Mouton.
Venezky, R. L. 1973. Letter-sound generalizations of first, second, and third-
grade Finnish children. Journal of Educational Psychology, 64, 288–292.

Venezky, R. L. 1995. From orthography to psychology to reading. In V. W.

Berninger, ed., The Varieties of Orthographic Knowledge, Vol. 2: Relationships to
Phonology, Reading and Writing. Boston: Kluwer Academic Publishers.

Venezky, R. L. 1999. The American Way of Spelling. New York: Guilford

Press.

Vihman, M. M. 1993. Variable paths to early word production. Journal of

Phonetics, 21, 61–82.

Wagner, R. K., and Barker, T. B. 1994. The development of orthographic

processing ability. In V. W. Berninger, ed., The Varieties of Orthographic
Knowledge, Vol. 1: Theoretical and Developmental Issues. Boston: Kluwer Aca-
demic Publishers.
Waters, G. S., Bruck, M., and Malus-Abramovitz, M. 1988. The role of lin-
guistic and visual information in spelling: A developmental study. Journal of
Experimental Child Psychology, 45, 400–421.

Webster, N. 1783. A Grammatical Institute of the English Language. Part I.

Facsimile (1968). Menston, England: Scholar Press.
 393

White, T. G., Graves, M. F., and Slater, W. H. 1990. Growth of reading in

References
diverse elementary schools: Decoding and word meaning. Journal of Educa-
tional Psychology, 82, 281–290.

Williams, J. P. 1980. Teaching decoding with an emphasis on phoneme anal-

ysis and phoneme blending. Journal of Educational Psychology, 72, 1–15.

Willows, D. M., and Ryan, E. B. 1986. The development of grammatical

sensitivity and its relationship to early reading achievement. Reading Research
Quarterly, 21, 253–266.
Wimmer, H. 1993. Characteristics of developmental dyslexia in a regular
writing system. Applied Psycholinguistics, 14, 1–33.

Wimmer, H., and Goswami, U. 1994. The influence of orthographic consis-

tency on reading development: Word recognition in English and German
children. Cognition, 51, 91–103.

Wimmer, H., and Landerl, K. 1997. How learning to spell German differs
from learning to spell English. In C. A. Perfetti, L. Rieben, and M. Fayol,
Learning to Spell, 81–96. Mahwah, NJ: Erlbaum.

Witherspoon, A. 1973. Common Errors in English. Totowa, NJ: Littlefield,

Adams.

Yopp, H. K. 1988. The validity and reliability of phonemic awareness tests.

Reading Research Quarterly, 23, 159–177.

Zinna, D. R., Liberman, I. Y., and Shankweiler, D. 1986. Children’s sensitivity

to factors influencing vowel reading. Reading Research Quarterly, 21, 465–479.
 Author Index

Aaron, P. G., 19, 213, 283–286, 341 Bremer, C. D., 342

Adams, G. L., 127 Brennan, F., 184–185
Adams, M. J., 153, 340 Brett, A., 225–226
Ahrens, M., 215 Bright, W., 7
Allred, R. A., 249–250 Broderick, V., 276–277
Aslin, R. N., 7, 23 Brown, A. L., 238
Brown, A. S., 117–118
Ball, E. W., 146, 170–171, 179–181 Brown, G. D. A., 307
Balmuth, M., xviii Brown, I. S., 132
Barker, T. B., 45 Brown, R., 241
Baron, J., 288 Bruck, M., 270–273, 282, 287
Barr, R. C, 115 Bruner, E., 104
Beck, I. L., 146, 229–232 Bryant, P. E., 164–168, 260–262,
Becker, W. C., 104–105 272
Beers, C. S., 257 Burstown, M., 262–264, 275
Beers, J. W., 255, 257 Bus, A. G., 158–160
Bell, N., 238–239 Byrne, B., 176–178
Berninger, V. W., 66–68, 71, 321
Biemiller, A., 342 Calfee, R. C., 68
Bindman, M., 260–262, 272 Campbell, J. R., 1, 212–213
Blachman, B. A., 146, 170–171, 179– Carr, T. H., 110–112, 340
181 Cattell, J. M., 341
Block, C. C., 242–244, 331–333 Chall, J., 80–86, 100–101, 107–109,
Bond, C. L., 147 123
Bond, G. L., 86–102, 107, 123, 131, Chaney, C., 7, 154, 173
249–250, 282 Chang, Kwang-chih, 21
Boronat, C. B., 115, 121 Chapman, S., 235–236, 238, 240
Bradley, L., 114, 164–168, 274 Chomsky, C., 197
Brady, S., 170 Chomsky, N., 288
 396

Civil, M., 20 Ennemoser, M., 171–175

Author Index

Clanchy, M. T., 14 Evans, M. A., 110–112, 340

Clay, M., 182
Clymer, T., 43–44, 86, 267 Fairbanks, M. M., 226–227, 233
Coltheart, M., 287–288, 326, 328 Falkenstein, A., 20
Cooper, J. S., 20 Faulkner, H. J., 207–208, 329
Cornell, E. H., 222–223, 228 Feldman, S., 101, 109
Cossu, G., 2, 155 Felton, R. H., 132
Coulmas, F., 7, 14, 16–18, 20, 22, 24, Fielding-Barnsley, R., 176–178
27 Fischer, F. W., 266–269, 272, 281, 290
Craik, F. I. M., 18 Fisher, R., 18
Cunningham, A. E., 114, 169–170, Flavell, J. H., 254
274–275, 282–283 Flesch, R., 77–78, 80, 86, 108
Foorman, B. R., 45, 135–137
Dahl, P. R., 197 Freud, S., 253–254, 343
Dale, N., xvii, 4, 76–78, 131, 275, Friederici, A. D., 7, 23
325–326 Frith, U., 2, 191, 255–256
Daniels, P. T., 7 Frost, J., 163
Darch, C., 127 Frost, R., 24
Dickinson, D. K., 223–225, 228 Fulwiler, G., 132
DiPersio, R., 205–206
Ditunno, P., 327 Gelb, I., 6–7, 17–18, 23–24
Dixon, M., 119–120 Gentry, J., 263, 275
Donohue, J., 133, 135, 166 Gersten, R., 104–105, 127
Dowhower, S. L., 200–206, 209, Geva, E., 40, 55, 286
329–330 Gill, M., 343
Dykstra, R., 80, 86–104, 107, 109, Gleason, M., 127
123, 131, 249–250, 282 Glushko, R. J., 288–297, 299, 305–
306, 311, 315, 337, 343
Ehri, L. C., 45, 116, 168, 170, 221– Goins, J., 132
223, 228, 251, 255–256, 265, 338, Goodman, K., 108
340–341 Goswami, U. C., 2, 45, 191
Eisenberg, P., 342 Gough, P. B., 213
Eldredge, L., 149 Graham, S., 247–248
Elley, W. B., 225 Graves, M. F., 80, 216
Ellis, A. J., xvii, 56, 76 Greenberg, S. N., 345–347
Ellis, A. W., 182–184 Griffith, P. L., 128, 147–148, 150,
Engel, M., 146–147 231, 321
Engelmann, S., 104, 127 Groff, P., 132
 397

Haddock, M., 166 Johnston, R. S., 139–145, 325–

Author Index
Halle, M., 288 327
Hammill, D. D., 70 Jordan, T. R., 45
Hanna, P. R., 53–55, 64–65, 267 Joshi, R. M., 213–214
Hart, B., 217–222, 331 Juel, C., 213, 297–298
Hatcher, P. J., 182–184 Jusczyk, P. W., 23
Hay, J., 77
Haynes, D. P., 215 Kaminska, Z., 119–120
Healy, A. F., 345 Karlgren, B., 21
Helfgott, J. A., 345 Katz, L., 24
Henderson, E. H., 255, 257, 275 Kirby, J., 150
Henry, M. K., 68–69, 71 Kirsner, K., 205
Herman, P. A., 198–200, 216–217 Klesius, J. P., 128, 147–148, 150,
Ho, Ping-ti, 21 321
Hodges, R. E., 65 Kohen, D., 206–208
Hohn, W. E., 170 Koriat, A., 345–347
Hollingshead, A., 118–119, 205– Kramer, S. N., 26
206 Krevisky, J., 119
Holmes, V. M., 269–270, 282 Kromrey, J. D., 128, 148, 150, 321
Hoover, W. A., 213
Howard, M. P., 133 Laberge, D., 342
Hudson, J. A., 212 Landerl, K., 2, 40, 55, 191
Huey, E. B., 195 Larson, S. C., 70
Huggins, A. W. F., 340 Leinhardt, G., 146–147
Hulme, C., 114, 182–184, 274 Leslie, L., 45
Hurley, M., 225–226 Levy, B. A., 205–208, 329
Liberman, A. M., 6
Ireson, J., 184–185 Liberman, D., 45
Iversen, I., 132 Liberman, I. Y., 6, 153, 266–269,
Ives, S., 114 172, 290, 298–300
Lie, F., 178–179
Jacoby, L. L., 118–119 Lindamood, C. H., 77, 131–135,
Jared, D., 306–311, 313 325–326
Jeffrey, W. E., 277–278 Lindamood, P. C., 77, 131–135,
Jenkins, J. R., 232–233, 331 325–326
Jensen, H., 23 Linfield, J. L., 119
Johnson, S., xvi, 43, 266 Lloyd, S., 138–139, 141, 232, 324–
Johnson-Glenberg, M. C., 238– 327
239 Logan, G. D., 115, 121
 398

Lundberg, I., 163, 166, 172, 182 Mullis, I. V. S., 1, 212–213

Author Index

Lysynchuk, L. M., 240 Muter, V., 249

Mair, V., 21 Nagy, W. F., 216–217

Malus-Abramovitz, M., 270–273, Nation, K., 214
282, 287 Nelson, K., 212, 216
Manis, F. R., 45 Newport, E. L., 7, 23
Mann, V. A., 249, 327 Ng, E., 269–270
Markman, E. M., 234 Nicholls, A., 206–208
Marshall, J. C., 2, 155 Nisbet, S. D., 117
Martinussen, R., 150 Nunes, T., 260–262, 272
Matlock, B. A., 232–233, 331
Mattingly, I. G., 24 Olofsson, A., 163, 166
Mattys, S. L., 7, 23 Olson, R., 45
McArthur, T., 17
McCallum, M., 262–264, 275 Palincsar, A. S., 238
McClelland, J. L., 288, 305–306, 309, Pearson, D., 100–101
311, 315 Perfetti, C. A., 45, 229–230
McCracken, G., 78 Perney, J., 250–263
McGuinness, C., 56, 133, 135, 150, Petersen, O., 163
166, 184, 334 Piaget, J., 253–254, 257
McGuinness, D., xiv, xviii, 12, 22, 24, Pinel, J. P. J., 338
27, 48, 55–65, 69, 74, 78, 114–115, Pintner, R., 117
133–135, 150, 166, 184, 328, 334– Pitman, I., xvi, xvii, 76
335, 343–344 Pitman, J., 77
McGuinness, G., 56, 150, 184, Plaut, D. C., 288, 305–306, 311–315,
334 336–338
McKeown, M. G., 229–232, 331 Plomin, R., 219
McNemar, Q., 79, 93 Postman, L., 18
McRae, K., 306–311, 313 Pressley, M., 333
Meister, C., 235–236, 238, 240 Pribram, K. H., 312, 343–344
Meyer, D. E., 18
Meyer, L. A., 113 Rashotte, C. A., 199–200
Michalowski, P., 20, 26 Rayner, K., 191–192, 208, 235, 343,
Miller, N. L., 342 345
Miller, P. D., 123 Read, C., 252, 255, 275
Monk, A., 114 Rinsland, H. D., 117
Morris, D., 259–263 Risley, T. R., 217–222
Morris, J. M., xviii, 76 Robbins, C., 221–223, 228
 399

Robinson, A., 34 Strawson, C., 288

Author Index
Rosenshine, B., 235–236, 238, 240 Stuart, M., 139–143
Rossini, F., 2, 155 Sumbler, K., 111–113, 141–144, 325,
Roth, E., 171–175 340
Rothlein, L., 225–226 Szeszulski, P. A., 45
Ruddy, M. G., 18
Ryan, E. B., 235 Tanzman, M. S., 45
Thimke, B., 45
Saffran, J. R., 7, 23 Tincoff, R., 117, 276–277
Samuels, S. J., 196–197, 277–278, Torgesen, J. K., 133–135, 146, 199–
342 200
Scanlon, D. M., 45, 115 Treiman, R., 117, 275–277, 300–305,
Schatschneider, C., 165, 173, 188 310
Schmandt-Besserat, D., 19
Schneider, W., 171–175, 179 Uhry, J. K., 70, 116–117, 168
Schvaneveldt, R. W., 18
Scragg, D. G., xviii, 76 van Ijzendoorn, M. H., 158–160
Seidenberg, M. S., 288, 305–315, 336 Varnhagen, C. K., 262–264, 268,
Senechal, M., 222–223 275, 343
Shanahan, T., 250, 282 Vellutino, F. R., 45–46, 115
Shankweiler, D. P., 266–269, 272, Venezky, R. L., 39, 41, 48–52, 55,
290, 298–300 65–70, 288, 290, 321
Shapiro, L. R., 212 Vihman, M. M., 153
Share, D. L., 166, 286
Shepherd, M. J., 70, 116–117, 168 Wagner, R. K., 45
Siegel, L. S., 40, 55, 286 Walcutt, C. C., 78
Silberberg, N., 132 Wall, S., 163, 166
Slater, W. H., 216 Waters, G. S., 270–273, 282, 287
Slocum, T. A., 232–233, 331 Watson, J., 139–145, 325–327
Smith, A. A., 69–70, 116, 334 Webster, N., 47, 55, 78, 82
Smith, M. W., 223–225, 228 Wessels, J. M. I., 7, 23
Snowling, M., 214, 249 West, R. F., 45, 270, 281–283, 328
Solso, R. L., 297–298 White, T. G., 216
St. John, J., 77 Wilce, L. S., 70, 116, 168
Stahl, S. A., 123, 226–227, 233 Williams, J. P., 163, 166
Stanovich, K. E., 45, 114, 270, 274– Williams, K. A., 213–214
275, 280–283, 328 Willows, D. M., 235
Stebbins, L., 104 Wimmer, H., 1, 2, 40, 55, 174, 175,
Steffler, D. J., 264 189–191, 277
 400

Wingo, C., 77
Author Index

Witherspoon, A., 267

Wood, J. R., 213–214

Yopp, H. K., 166, 185–186

Zielonka, P., 147–148

Zinna, D. R., 298–300
Zubin, J., 117
 Subject Index

Advanced spelling code, 55–65, 70, theories of ‘‘brain activation,’’ 290,

323, 334–335 292, 294, 296, 306, 309, 342
Affix, 267 unconscious analysis in reading, 206
Allophone, 42
Alphabet charts, 32 Classroom observations, 83–85, 101,
Alphabetical order, 28, 29 109–115, 323, 331
Alphabet principle, 83, 176–177, 276 Clay tablets, 20
Articulation (speech-motor) training, Coarticulation, 155, 190, 344
131, 178, 325–326 Code overlaps, 46–47, 63, 83
Artificial transparent alphabet, xvi, Codes, 252
56, 75–76, 323 as writing systems, 11–13, 17, 39, 49
Attention Cognitive models of reading/spelling,
in the classroom, 138 terms and expressions. See also
and memory, 115, 121 Spelling
and phoneme awareness, 155 consistent/inconsistent neighbors,
Attention span (brain), 343 292–295, 299, 301, 306–307, 312
consistent/inconsistent rimes, 293–
Basic code, 56, 59, 75–77, 82, 111, 295, 302, 304, 306–308, 310
321–324 exception words, 292–295
programs for teaching, 130–145 friends/enemies, 301, 304, 307, 312
spelling chart, ix orthographic consistency, 297–298,
Binomial test, 164, 221 302, 304, 306
Brain orthographic redundancy, 297
conscious analysis in reading, 190, orthographic rime, 292–297, 299–
192–193, 248, 343–347 300, 304, 307–309, 311–312, 315,
neuro-anatomy and neural 338
pathways, 338–339 orthographic versatility, 297–298
processing print, 280, 288, 296, 305, phonological rhyme, 292, 296, 308–
315, 336, 338–347 309, 312
 402

Cognitive models of reading/spelling, DISTAR (direct instruction method),

Subject Index

terms and expressions (cont.) 104–105, 127

regular words, 292–295, 338 Down’s syndrome and reading, 2,
word body, 291, 295, 338 155
word family, 292, 299 Dyslexia, 2, 3, 6
word frequency effect, 292, 303
Color-coded text, 136 Effect size
Compounding languages, 62 definition, 124–125
Computer models of reading, 287– formula, 124
288, 296, 305–316, 336–339 Encoding, 37–39, 252
attractor network model, 311–315, English language structure, 55–56
337 English spelling classification systems
dual-route model, 287–289, 307, Webster, 47–48
338 Venezky, 48–52
parallel distributed processing Hanna et al., 53–55
model, 288–289, 305, 336, 344 McGuinness, 55–65
recurrent connections, 311–312 English spelling system. See also
Computer training studies, 159 Spelling
Consonants Anglo-Saxon spellings, 55, 59, 62
blends, clusters, 9, 61, 83, 176, 313, defined, 41
318 Greek-based spellings in, 54–55, 62,
voiced, 155 64–65, 68
Cooperative Research Program, 86– Latin-based spellings in, 54–55, 62–
103 64, 334–335
Copying/writing as learning tool, multisyllable words and, 54, 62
112, 114, 118, 156, 185, 323, Norman French spellings in, 55, 59,
335 62, 64
predictable patterns of, 51, 61
Dale, Nellie, 76–77, 325–326 problems with, 46–47, 65
Deaf children and reading/spelling, spelling alternatives, 57–60, 63,
19, 283–286 334
Decoding, xv, 37–39, 48, 113, 164, structural features, 59–62
211, 252 ERIC database, 194–195
Decoding strategy, 57, 114–115, 269, Experimenter-designed tests, 161
328 Eye movements and reading, 191–
Demographic data, 88, 99 193, 208, 343
Diacritic mark, 2, 31, 54 context effects, 192, 196–197
Digraphs, 2, 44, 45, 50, 77, 83, 180, functional span, 192
287, 299, 301, 313, 320, 322, perceptual span, 192
324 syntax effects, 192–193
 403

Freud’s stage model, 253–254 Longitudinal studies, 103–105, 260,

Subject Index
Functional illiteracy 265, 327
countries and, 1, 6, 322
rate of, 1, 5 Many-word problem, xiv, 248, 279–
Function words in reading, 344–347 316
Mapping systems, 11–12, 15, 320
Head Start, 103, 217, 223 Memorization
Homophones, 16 letters as aid, 170
limits to, 18–19, 22, 26, 34, 74, 318
International reading surveys, 1, Memory
349–353 promoted by, 114
recall, 37, 70, 116, 232, 247, 273
Jolly Phonics actions, 138–139, 141, recognition, 37, 70, 116, 247, 273
325–326 rote visual in reading, 279, 283,
284–287, 289
Kindergarten, origin of, 171 sight-word reading and, 251
Meta-analysis, 124–125
LAC test of phoneme awareness, 169 Missing letter effect, 345
Languages, impact on writing system Montessori, Maria, 37, 275
Akkadian, 20 Morphology, 62
Arabic, 19 Moveable alphabet, 131
Aramaic, 30
Chinese, 21, 56 Naming speed, 277, 327, 341–342
Egyptian, 19 National Assessment of Educational
German, 3 Progress (NAEP), 1, 212–213, 236
Greek, 34 National Reading Panel, viii, xiii, xiv,
Hebrew, 19 xviii, 70, 73–75, 107, 247, 317–
Indian, xv, 27–28 320, 322–334
Japanese, xv, 22 comprehension training, 215, 234,
Sumerian, 20 237, 331
Language development, 153 fluency training, 193–196, 328
Language types and writing systems phoneme awareness training, 153–
Hamito-Semitic, 19, 23, 31 188
Indo-European, 27, 30 reading instruction, 121–152, 323–
Tonal languages, 21 324
Letter names and learning to read, vocabulary training, 226, 228, 330
184, 251, 323
Letter-sound correspondences train- OECD (1995, 1997), 1, 349–353
ing, 166, 170, 175, 178–179, 182 Opaque alphabet, xvi, 3, 13, 39–40,
Ligature as diacritic, 27 46, 75, 159, 247, 250
 404

Paired-associate learning, 18, 19, 74, visual phonics, 130, 147, 149, 151,
Subject Index

277 319, 320–323, 336

Paleography, 6, 11–13, 23 whole-to-part phonics, 130, 320
Parental involvement, 138 Phonics and speech-motor training,
Phoneme awareness 131, 152
blending/segmenting, 111, 116, 136, Phonics teaching approaches
141, 154, 156, 161–163, 166–167, analytic or intrinsic phonics, 81, 85,
170–171, 173, 176–179, 187, 129, 139, 319, 324
323–324, 326, 335 basal-reader phonics, 129, 146, 148,
computer training in, 160 320–321
decoding and, 279, 280–281, 283 embedded phonics, 133, 136
explicit awareness, 153–155 explicit phonics, 129, 147
skills in, 148 linguistic phonics, 129–130, 146,
training of, 112, 134, 153–188, 318, 151–152, 155–156, 173, 178,
327–328 187–188, 320, 322–323, 325–326,
training with and without letters, 334–335
160–161 synthetic phonics, 81–82, 85, 107,
Phonemes 129–130, 151, 320
chart of, ix systematic phonics, 129, 151
consonants, 51, 54 Phonograms, 46, 83, 287, 301, 320
definition, xv Phonological awareness. See also
schwa vowel, 52, 54 Phoneme awareness
tasks alliteration, 164–166, 173
classification, 161 origin of, 6
discrimination, 161, 166, 169, 186 rhyme identity, 154, 164–166, 173,
identification, 161, 166–168, 173, 186
178, 185–186, 335 syllable analysis, 180
manipulation/deletion, 161, 169, syllable/rhyme segmenting tasks,
170, 173, 186 159, 163–166, 172
sequencing (blending/segmenting), training of, xviii, 109
161, 185–186 Phonological development theory, 6,
transitivity, 176, 335 109, 153–154, 157, 163–164, 186
vowel þr, x, 51, 53 Phonological/phoneme training
vowels, 53–54 programs
Phonics categories Lundberg program, 163, 172, 182,
junk phonics, 130, 320 184
linguistic phonics (see Phonics, Sound Categorization, 164–166,
teaching approaches) 169, 173
multi-sound phonics, 130, 320 Sound Foundations, 176
 405

Phonological units in reading critical thinking strategies, 244

Subject Index
instruction length of training, 236
onset-rime analogy, 126, 129, 141 multiple strategy technique, 238
syllable analysis, 180 National Reading Panel review,
Phonotactics, 7, 23, 34, 55–56, 74, 215, 234, 237
280, 291, 318 reciprocal teaching programs, 235,
Piaget’s stages, 253–254 238–241
Prefixes/suffixes and spelling, 62–63, summary of, 330–333
268 visualizing/verbalizing program,
Project Follow Through, 103–105 238–240
Prototype for reading instruction, 38, Reading frequency (print exposure)
73, 82, 111, 121, 127, 184, 317– impact on accuracy, 282–283
319, 323–324 tests of, 281–286
programs fitting prototype, 131– for training fluency, 194–195, 199,
145, 152 328
Proto-writing, 14 word familiarity and, 344
PsychINFO database, 194–195 Reading instruction
Austria, 175, 210
Reading comprehension California, 5
measures of Canada, 40, 139, 141
cloze test, 214 England, 40, 138, 191
factor analysis, 214 Finland, 39–40
functional illiteracy, 212–213 Germany, 4, 39, 171–173
in-house tests, 235–238 Italy, 39
meta-analysis, 235–237 Norway, 39, 178
NAEP tests, 212, 236 Scotland, 139, 141
standardized tests, 212–214, 236– Sweden, 39
239, 241, 244 Texas, 187
treatment fidelity, 243 Reading methods. See also Phonics,
skills. See also Story comprehension teaching approaches
decoding accuracy and, 211–214, basal readers, 4–5, 77, 80–83, 86,
235 92, 94, 97, 102–103, 108, 128,
oral (listening) comprehension 147, 152, 216, 317
and, 211–214, 235 basal þ phonics, 87, 90–91, 94,
reading fluency and, 211–213, 235 103
vocabulary and, 211 eclectic or balanced, 5, 6, 74, 84,
training methods 111–113, 146
Block comprehension program, language experience, 87, 90–91, 94,
242–244 110–111, 123–124
 406

Reading methods (cont.) Orton-Gillingham, 126, 180

Subject Index

linguistics-based, 81–82, 90–91, 94, Phonic Exercises, 91

103 Project Read, 68–69
look-say, 4, 77, 85, 341 Reading Recovery, 137, 182, 184
phonics, 4, 42, 77–78, 83, 85, 97, Scott-Foresman basal reader, 82, 90,
110–113, 128, 319 147, 181
reading schemes (UK), 4, 128 Sing, Spell, Read, and Write, 147
real books, 4 Speech to Print, 91
sight-word, xviii, 4, 9, 19, 34, 74, Success in Kindergarten, 185
112, 114–115, 181 Webster’s Blue-Backed Speller, 47–
whole language, 4–5, 108, 122–124, 48, 78
136, 152, 147–149, 152, 216, Word Power, 91
220 Reading speed and fluency, 159,
whole word, xviii, 4, 108, 146, 317– 189–211, 159, 328. See also
318 Rereading method for fluency
Reading programs training
DISTAR, 104–105, 127, 180–181 comprehension and, 189, 191, 193,
Fast Phonics First, 139–145 198, 200–201, 203
Ginn basal reader, 82 good versus poor readers, 208
Glossic, 76 National Reading Panel report, 189,
Harper and Row basal reader, 147 193–196
Hay-Wingo, 77 prosody and, 200–203
Houghton-Mifflin basal reader, 147, reading rate, 191–193, 196
149 sight word training and, 196–197
Initial teaching alphabet (i.t.a.), 77, slow readers, 2, 189–193, 210
82, 87, 90, 92–94, 97–99, 103 speaking rate and, 190–191
Jolly Phonics, 111, 113, 138–145, Reading tests, generic
152, 162, 188, 323–324, 327, fluency, 159, 172
334 nonsense word decoding, 131, 148,
Letterland, 185 164, 178
Lindamood Auditory reading comprehension, 110–111,
Discrimination in Depth, 77, 172
131–135, 152, 170, 325–326 word attack, 133, 334
Lippincott, 78, 82–83, 87, 90–91, word identification, 133, 334
94, 97–99, 103, 107, 117, 131– word recognition, 133, 334
132, 188, 323 Reading wars, 78, 86, 317
New Primary Grade Reading Rereading method for fluency
System, 146–147 training
Open Court, 135–137 audiotape and, 200–201, 203
 407

decoding accuracy and, 189–191, consonant-doubling ‘‘rules,’’ 267–

Subject Index
193, 196–201, 203, 205, 207–209 268, 271, 289
proofreading and, 205–208 grapheme-phoneme correspon-
target goal, 196–205 dence ‘‘rules’’ (GPCs), 44–45,
text difficulty and, 207–210 288–290, 306, 308, 344
training studies, 194–210, 328–330 linguistic structure and, 265–266,
transfer effects, 196–206 271, 279
Root words, 54–55, 267–268 morphological structure and, 266–
Rosner test of phoneme awareness, 269, 271, 274, 279, 290
186 orthographic ‘‘rules,’’ 279–280,
290, 340
Sex differences orthographic structure, 279–281,
in reading, 88, 92, 98, 103 291, 297, 305–306, 313, 315, 344
in spelling, 92, 98, 103 orthography defined, 17, 43, 45–
Sight words 46, 48–49, 55
functional sight vocabulary, 340 phoneme-grapheme correspond-
late-stage sight-word reading, 339– ences, 279, 285, 288–290, 326,
343 328
myth of, 9, 34, 343 phonotactic structure and spelling,
as reading method, 288 280, 291
traditional sight word lists, 52 spelling generalizations, 43–44
true sight words, 57–59 spelling ‘‘rules,’’ 43–44, 50, 119,
Silent letters, 57 251, 266, 268, 274
Speech perception, 153, 155 methods based on structure of the
Spelling code, 55–65, 69–70
alternatives, 46–47, 54, 83, 130– methods in schools
131, 139, 151, 270–271, 298, 310, invented spelling, 108, 116–117,
320–321 120, 184, 187, 248, 252
categories of spelling words letter-name spelling, 116–117
rare spellings (exception words), miscellaneous, 120–121
267, 269, 271, 279 research on traditional programs,
regular spelling, 43, 44–45, 56–57, 247–248
247, 266–267, 270–271, 273 rule-based, 43–44, 50, 119
errors, 115–121 predictors
linguistic/visual features of spelling reading skill, 250
code used in cognitive research sex, 249–250
bigram/trigram frequencies, 285, verbal IQ, 249, 268, 282
287, 291, 297–298, 308 probability structure of spelling
checked vowel ‘‘rule,’’ 289 code
 408

Spelling (cont.) specific

Subject Index

context-dependent spellings, 280 British Ability Scales, 183

context-free spellings, 271, 313 California Achievement Test, 120
frequency in print, 50, 53, 57, 284, California Test of Basic Skills, 147
291, 293, 297–298, 301, 307–308, Comprehensive Test for Basic
338 Skills, 120
statistical probability, 46–47, 53, Detroit Test of Learning Aptitude,
57, 59, 120, 284–286, 287, 297– 239
298, 301, 336, 338 Fry Word List, 88–90, 94, 96–98
structural redundancy, 45–47, Gates-MacGinitie, 149, 239, 250
59, 61, 280, 286, 297, 300, 306, Gates Word List, 89–90, 94, 96–
315 97
processing strategies, 264 Gilmore Tests of Accuracy and
letter names, 251, 274, 275–278 Rate, 88–89, 94, 96
letter-sounds, 251, 276–277 Iowa Test of Basic Skills, 229, 243,
visual memory, 268, 269–270 333
reform, xvi McGraw-Hill Basic Study Skills,
research on new teaching methods, 120
65–70 Metropolitan Achievement Test
stage models, 250–265, 272 (MAT), 104–105, 120, 169
stages, 55, 109, 116–117, 248–250, Murphy-Durrell Letter-Name, 89
275 Peabody Comprehension Test,
structural analyses of the spelling 214
code Phoneme Discrimination Test, 89
Hanna et al., 53–55 Pintner-Cunningham IQ Test, 89
McGuinness, 55–65 Schonell Spelling, 185
Venezky, 48–53 Stanford Achievement Test (SAT),
Webster, 47–48 147, 241, 284
systems origin, 40–41 Test of Written Spelling, 70
test types Wechsler Adult Intelligence Scale
spelling dictation, 117–119, 167, (WAIS), 268
335 Wechsler Intelligence Scales for
spelling recognition, 118–121, 268, Children (WISC), 214, 239
272 Wide Range Achievement Tests
Standardized spelling, 43, 46 (WRAT), 104–105, 239, 268,
Standardized tests 281–282, 286
general, 79, 149, 158–159, 160–162, Woodcock Reading Mastery, 69,
171–172, 179, 181, 185, 189, 270, 147, 150, 170, 176, 179, 213–214,
334 281–282, 284
 409

Statistical analysis multiple-choice tests in research,

Subject Index
availability of, 79, 85 221–223, 225, 229
invalid methods of, 91–93, 98 National Reading Panel report,
valid methods of, 94–97 226, 228
Story productive vocabulary and, 222,
comprehension, 224 230–232, 331
grammar, 211, 236 receptive vocabulary and, 223, 230,
recall, 230–231 232
Sumerian schools, 38, 115 standardized tests and, 221, 227,
229
Test construction, 188 successful programs, 229–235
Transparent alphabets, xv, 2–3, 32, teacher-child interaction, 224
39–40, 55, 75–76, 159, 175, 178– Voice-onset latency/response time,
179, 191, 277 293–294, 296, 303, 312
Treatment fidelity, 100
Webster, Noah, 37, 44, 82
Universal education, xviii Word families, 5, 7, 9, 46, 62, 82–83,
126, 318
Vocabulary Word play in teaching, 154
acquisition, 215–219 Writing systems
children’s literature and, 215 ancient
heritability/verbal IQ, 219 Akkadian, 11, 20
oral comprehension, 215 Anglo-Saxon, 39, 40–41, 46
productive vocabulary size, 216, 233 Aramaic, 11, 30
reading and, 215 Assyrian, 24
television and, 215 Babylonian, 11, 13, 24, 29–30
word derivation, 215 Chinese, 18, 20–21, 25–26
training Crete, 11, 27
classroom lessons and, 112–113 Egyptian, 11, 13–16, 18–19, 23,
dangers of whole language, 220 31
deducing meaning from context, Greek, 28, 31, 32, 34, 39, 41
232–233 Hittite, 6, 11, 24
frequency of exposure and, 229– Indian, 27–29
233 Japanese, 21–22, 33
Head Start, 217, 233, 331 Korean, 28–29, 30
listening to stories as method, Mayan, 11, 15–16, 27
221–226 Norman French, 40–41
meta-analysis of classroom Old English, 40
research, 226–227 Old Persian, 28–30
 410

Writing systems (cont.) types (see also Opaque alphabet,

Subject Index

Phoenician, 29, 31–32 Transparent alphabets)

Roman (Latin), 41, 46 alphabet, xv, 6–7, 13, 31–32, 34–
Sumerian, 11, 13–14, 18–20, 25– 35, 75, 83, 154–155, 318
26, 38 consonantal alphabet (consonant
Vietnamese, 24 cueing), xv, 19, 31–32
assumptions about, 32 consonant-vowel diphone (CV
comparative analysis of, 6–8, 11, diphone), xv, 22, 26–32, 34, 74
14–18, 339 logographic, 6, 17–21, 27, 33–34,
definitions of, xiv, 11–18, 33 339
evolutionary theory of, 6–7, 17, 23– meaning-based, 17–18, 20, 74
24, 33 pictographic, 6, 17, 19–20, 27
function of, 14 sound-based, 17, 20
modern syllabary, xiv, 6, 20–21, 23–26, 31–
Arabic, 31 32
Cherokee, 28 whole word, 16, 34, 74
Danish, 159
English, 40–41, 191
European, 32
Finnish, 39–40, 159
German, xv, 40, 159, 172–175,
189, 191, 277
Israeli, 40, 159
Italian, xv, 159
Norwegian, 159, 178
Spanish, xv, 159
Swedish, 159
myths about, 8, 9
scripts
Brahmi script, 28
cuneiform, 30
Han’gul, 29–30
hieroglyphics, 16, 19, 31
hiragana and katakana, 22, 28
kanji, 21–22, 33
Linear A, 11
Linear B, 11, 27, 34
semantic classifiers, 19–21, 32