DOCUMENT RIME

ID 240 512 CS 007 499

AUTHOR Allen, Joleth

TITLE Inferences A Research Review.
PUB DATE [80]
N OTE 31p.
PUB TYPE Information Analyses (070)

E MS PRICE 11101 Plus Postage. PC Not Available from ERRS.

DESCRIPTORS *Age Differences; *Cognitive Development; *Cognitive
Processes; Literature Reviews; Reading Ability;
*Reading Comprehension; *Reading Research; *Reading
Skills; Research Methodology
IDENTIFIERS *Inferences; Theory Development

ABSTRACT
Since E.L. Thorndike's landmark 1917 study of the
complexity of reading comprehension, inferential research has
generally focused oa either inference as a developmental process or
the nature of inferences made during reading. In his 1930 study, R.
M. Tyler established that inferior* could be objectively measured. S.
G. Paris conducted several studies in the 1970s showing that readers'
age and ability level influenced their inferenciag skills, and
indicating that young childres'are capable of producing inferences
but usually do not do so spontaneously. Examinations of the nature of
inferences have not been as conclusive as developmental studies. S.
T. Goets's 1977 study of high school students found no main effects
for the importance or salience of the reading material. A 1979 study
by P. D. Pearson, J. Hansen, and C. Gordon revealed that prior
knowledge has a simple effect for Inferable questions but no effect
for explicitly stated questions. In a major contribution on the
nature of inferencing, R. Tierney, C. A. Bridge, and M. Coma
discovered that good readers made more causal and conditional
connections between propositions while poor readers more often
overgeseralised. One point of unanimity in reading research is that
inferential ability is vital to mature readers. (MM)

***********************************************************************
* Reproductions supplied by EMS are the best that can be made *
* from the original document. *
***********************************************************************
 TAIL 011PAI1ERINT Of SONCA116111
NATIONAL INSTEIVIE Of ESUCATION
EDUCATIONAL RESOURCES INFORMATION
CENTER iEluCI
)dhs dOMMOI hog limn ispeamod
amtoml Item Ow porton WOOWSION
044111640 4
4001 AO** basil 0i00, a) 4m
MMAKUMOWft
Poems of vow at opinions stood in *adieu
wont do not notwtototty musuom Wheel NO
moon Of pokey
INFERENCE: A RESEARCH REVIEW

JoBeth Allen

Assistant Professor

Kansas State University

"PERMISSION TO REPRODUCE MS
MATERIAL IN MICROFICHE ONLY
HAS SEEN GRANTED BY

JoBeth Anal

TO THE EDUCATIONAL RESOURCES

INFORMATION CENTER (ERIC}."
 Inference: A Research Review

As interest in the topic of the unot3ervable part of the reading

process, comprehension, has grown, several new areas of reading research

have emerged. The present paper will examine one of these factors,

inference, which was an almost unheard of topic before 1970. Calfee and

Drum (1979) note that

Comprehension is a complicated matter; it can be virtually

synonomous with thinking. Trying to analyze the process
of comprehension is an interesting challenge. We propose
here two broad categories of comprehension tasks -- transliteral
and inferential.

While many authors agree that there is such a skill as drawing in-

ferences, agreement on a definition is more difficult. A basic definition,

representing many writers who do not deal with hierarchies of inference,

comes from Strange (1980).

A question/answer relationship is textualll implicit if both

question, and answer are derived 4rom the text, but there
is no expressed grammatical cue tying the questions to the
answer (p. 395).

Frederiksen (1979) has developed much more specific and elaborate

definitions about inference and types of inferences. His broad definition,

however, adds specificity to Strange's concept.

The fifth and highest level of processing operates entirely

on propositions, generating new propositions that are given,
for example, from prior discourse, from discourse context,
or from previously acquired knowledge about the world. Any
such processes will be referred to here as inferences.

In elaborating on the nature of inferences, TrabIsso (1900) has

defined two ways in which inferences are made. The first involves text
 2

connecting: finding relationships between events in the text. The second

is slot-filling: adding missing information to enable text connecting.

Trabosso reiterates the difficulty of studying inference raported by many

researches; "...the making of an inference is a highly automatic and

largely unconscious process...(but) is not a simple or obvious process"

(p. 3). In spite of difficulties with observations, measurement and

even definitions, inference has been the topic of much research in the

past ten years. That research will be analyzed in this paper.

In 1917, Edward L. Thorndike conducted a landmark study in reading

comprehension. The study was significant because it looked at the com-

plexity of comprehension for the first time; up to that point reading had

been considered to be "...a rather simple compounding of habits"

(Thorndike, 1917, p. 425). Thorndike gave two hundred sixth-grade pupils

a series of simpl: paragraphs and asked them to write the answers to

questions about them. The questions included examples of main idea,

factual recall and inference. Thorndike was topressou with the extensive

variety of answers to his questions, and found that the responses did

not fit neatly into categories of errors. Rather, he drew conclusions

about the nature of *correct reading" based on the analysis of responses

(Thorndike, 1917):

In correct reading (1) each word produces a correct

meaning (2) each such element of meaning is given a correct
weight in
i comparibon with the others, and (3) the resulting
ideas are examined and validated to make sure that they
satisfy the mental set or adjustment or purpose for whose
sake the reading was done (p. 428).

He went on to describe certain errors as failures to perform one of

these operations. Thorndike concluded that correct reading was very

closely tied to the reasoning process; thus, his title in reporting the

4
 3

research is "Reading as Reasoning: A Study of Mistakes in Paragraph

Reading."

Before leaving the Thorndike study, it is interesting to note that the

three factors Thorndike describes might be viewed by present theorists as

representative of the interactive nature of reading. Thorndike found that

each word must produce a correct meaning (text factor), which the reader

must weigh correctly ih relation to other words (reader factor), and then

validate according to the purpose for reading (process factor).

The reasoning factor identified by Thorndike les supported in several

studies designed to analyze factors of comprehension. Feder (1938) factor

analyzed the responses of 700 college students on "comprehension rmaturity"

tests. He identified two distinct levels of response: reading for infer-

motion and reading for inference. Comprehension factors were examined by

F. B. Davis in two separate studies. The first study (Davis, 1944),

conducted in 1941 with college freshmen, concluded that the two most

important factors in reading were memory for word meanings and reasoning

in reading (combining the categories of weaving ideas and drawing infe.ences).

The second study was a much more carefully controlled and planned in-

spection of the reading process. In it, Davis (1968) administered a

series of nine tests to 1,100 twelfth-grade students. He analyzed eight

possible factors of comprehension to determine what percent of the unique

variance each contributed. His categories and findings are summarized

in Table 1. The percentages are averages from within-day and across-day

maerices.
 4

Table 1

DAVIS CATEGORIES AND AVERAGE PERCENTAGES OF COMPREHENSION FACTORS

Categories Percent Unique Variance

1. recalling word meaeings 32

2. drawing inferences about word meanings 3.5

3. factual recall 10

4. weaving together ideas 5

5. drawing inferences from context 20.5

6. recognizing author's purpose, tone, 11

etc.

7. identifying author's technique 5.5

8. following structure of passage 13.5

Davis again interpreted his data to support word meanings and drawing

inferences as the two factors accounting for over half of the unique

valiance.

Spearritt (1972) used Davis' 1968 study and applied the statistical

analysis of maximum likelihood. He found support for the word knowledge

factor, but felt that three other factors, drawing inferences from context,

recognizing a writer's purpose, and following the structure of a passage,

were so closely related that they should be considered as a single factor:

reasoning.

Although these studies used categories identified as inference ability,

the first.study to focus exclusively on inference was conducted in 1930 by

a professor at Ohio State University (Tyler, 1930). He was concerned with

the objective measurement of all educational goals. His department had

adopter! the goal that each student should have "An ability to draw inferences

6
from facts, that is, to propose hypotheses." To measure students' ability

to infer, Tyler tested sixty-six students in a beginning zoology course.

The students took two tests in the same day, one requiring production of

an inference, the second requiring recognition of an inference. Students

read a series of factual statements; they were instructed to "Write...

the one inference which seems to you most reasonable in attempting to

interpret the facts" (Tyler, 1930, p. 476). A sample paragraph and logical

inference were provided. Papers were collected, and the students were

given the same items in multiple choice form. One might expect a rather

large interference between the inference generated and the multiple-choice

items. However, Tyler found a correlation of only .38. His interpretation

of this data was that "the ability to select the most reasonable inference

from a given list is not the same as the ability to propose an original

inference " (p. 477-8), Tyler did not consider the possibility that

inferences drawn in the first measure might have interferred with the ability

of students to select a correct inference.

Tyler deleted the multiple choice format in favor of the generated

inference, and developed a procedure for evaluating such inferences. He

sorted inferences into five "equal-appearing intervals," which he had each

member of the education faculty use for sorting samples of student inferences.

He found three instructors to be as reliable as the whole department in

determining "the quality of the original inferences...proposed by the

student" (p. 479). He further validated 'he procedure with new faculty

members, finding a correlation of their judgement with that of existing

members' to be .96. In another analysis, Tyler found that his method of

testing inference had, very low correlation (r..29) with a test of factual

7
 6

knowledge over the same material. He concluded that inference was not only

a separate skill, but that it could be measured and "the merit of an in-

ference is not the purely subjective quality often claimed" (p. 479).

Inference did not appear again in the literature as a separate :on-

cern until the seventies. The only mention of inferential ability was in

studies such as Davis' (1944), where it was considered as part of the hier-

archy of reading skills. Even these studies dealt with .nature readers,

rather than with children learning the basics of the reading process. As

researchers began to look at young readers, they first focused on more

observable behaviors such as word recognition for lists of words, nonsense

words, syllables and even letters. Rate was easily measured, as were eye

movements and other physical characteristics of the reader. It was not

until the behavioral wave of research subsided that the complexities of

coaprehension were examined. Reading research was given added impetus from

such fields as phycholinguistics, artificial language and psychology. The

seventies produced intense discussion and theory about the nature of the

reading process; the theories are presently being tested through research.

The studies evolving from this atmosphere pertinent to inference are re-

ported here.

Several interesting studies of children's inferential skills were con-

ducted by Paris (Paris & Upton, 1974; Paris, 1975; Paris & Lindauer, 1976).

Tierney and Cunningham (1980) summarize the first study by noting that "Paris

(1975) found young children are less able to relate their own background ex-

periences in the process of inferencing" (p. 46). This .onclusion contrasts

with the opinion of those who feel that young children do not have sufficient
 7

background experiences to make the needed connection. Paris pursued this

factor in his next study (Paris & Upton, 1974). 'Paragraphs were read to

twelve subjects at each of six grade levels, kindergarten through fifth

grade. Pupils were asked to respond to four literal and four inferential

questions. The most interesting conclusion, apart from anticipated main

affects for grade and question type (p. < .01), was that younger children

seemed to be capable of drawing inferences. What separated them in ability

from older children was that they did not do so spontaneously.

In order to explore this conclusion, Paris and Lindauer (1976) con-

ducted an experiment using subjects from first, third and fifth grades.

A total of forty-eight students read sentencewhere an instrument was

either explicit (The man dug a hole with a shovel) or implicit (The man

dug a hole.) They were then tested on the four implicit and four explicit

sentences by being given the cue (which had been stated in only half of

the sentences) and asked to recall the sentence. Analysis showed a signi-

ficant difference between grades one and three, and one and five, but not

between three and five, indicating that the development of the skill

occurred in the primary grades. By fifth grade, students performed

equally well on explicit and implicit instrument sentences. Paris and

Lindauer performed a second experiment using sixteea each kindergarten,

second and fourth graders. In this design, all sentences implied an in-

strument, but did not state it. Questions were asked using as recall cues

explicit information (the stated subject, verb or object of the sentence)

or implicit (the instrument). Again, younger children did not do as well

on implicit-cued sentences as on explicit ones.

9
 8

The third experiment reported by Paris and Lindauer attempted to

remedy what they hypothesized as a lack of spontaneous production rather

than an inability to deal with inferential material. The researchers had

ten first graders act out the implicit sentences. The children were in-

structed to perform the actions for "The man dug a hole." The results

were dramatic. Overall recall was sigificantly improved. More importantly,

the difference between explicitly cued and implicitly cued sentences was

eradicated, with subjects recalling 72% of the explicit-cue sentences and

70% of the implicit-cue sentences. Paris and Lindauer concluded that young

children were capable of drawing inferences, but that they had no spon-

taneous strategy. This observation has been supported in other areas of

learning, such as memory for lists (Smirnov A Zinchenko, 1969). It is

unfortunate that Paris and Lindauer did not use a large enough sample to

report their results in terms of significance; and that no one has sought

to replicate this study using controls. The implications for teaching

inference skirls seem promising.

A number of studies have dealt with variations of the developmental

process identified by Paris. Iiildyard (1976) had sixteen first, third and

fifth graders listen to passages, end sixteen third and fifth graders read

passages silently. Measure were take, of implications (defined as ex-

plicit inferences requiring no world knowledge) and for inferences, both

those related to world knowledge and contradictory to world knowledge.

A *marked developmental difference' was found for implications and for

those inferences that contradicted previous world knowledge. Reading

facilitated implication; listening facilitated inference related to world

knowledge.

10
 9

Handler and Johnson (1977), in heir classic study of story structure

and its effect on recall, examined the types of inferences children made in

terms of their appropriateness. Their subjects were first and fourth grade

students, and adults. The data in Table 8, Proportion of Subjects Producing

Additions During Recall, Total Number of Additions, and Proportion Occuring

in each Classification, is of particular interest. All subjects were

approximately equal in the proportion of "reasonable elaborations" (1st -

.36; 4th - .31; adult - .35). There were developmental trends in two other

categories, however. Emphatic or redundant additions which were consistent

with the original text increased with age (1st - .15; 4th - .41; adult - .49).

Irrelevant or structural "fillers" which were not text-based decreased ith

age (1st - .40; 4th - .29; adult -.16). Wandler and Johnson interpreted these.

data in terms of young children's need to fill in missing story structure;

it seems equally important to note that the skill of making appropriate

inferences undergoes developmental changes in the years between first and

fourth grades.

In an unpublished study done in 1977, Stein (1978) examined first and

fifth graders' interpretations and recall of stories with very discrepant

information (e.g., a mean, selfish fox helps a weak bear catch and cook a

fish). When asked for exact recall, first graders recalled only the story

line, not the discrepant description of the fox. Fifth graders recalled

both the story and the description; some even made some inferences about

the fox's motives. When asked to "fill in anything that is missing," first

graders again gave fairly verbatim recall, with no mention of incongruities.

Fifth graders made inferences as to the reasons for the incongruous action,
 10

and sought to provide resolution to the story. While her report contained

no statistical data, these findings have been replicated in other studies

(Stein & Glenn, 1975).

In one such study, Nezworski, Steioland Trabasso (1979) looked for

differences between kindergarten and third-grade students, testing a total

of 144 children. Their material was written in two forms; the control

passage presented a negative protagonist, while the added information

passage gave an explanation of why the protanogist acted as reported,

makinj that person less negative. Results of the questioning of each

type of passage are reported in Table 2.

Table 2

"1C2ORTION OF CORRECT INFERENCES MADE TO PROBE QUESTIONS

Inference type Control Added information

K 3 K 3

causal .36 .38 .64 .87(*)

consequence .33 .33 .82 .96(*)

reaction .89 .97t9 .88 .97(*)

(I significant at the(.06 level

It appears that the low proportion of correct responses in the causal

and consequence categories on the control passage indicates a uniform

difficulty in der mg with ambiguity for first and third graders. However,

the older students were better able to make use of added information in

resolving ambiguity.

A developm'ntal difference at an even younger loel wis reported by

Poulsen, Kintsch, Kintsch,and Premack (1979). They found that four and
r

six year old children can both make up stories from pictures presented in

logical order. However, when the pictures were presented in random order,

the younger children reverted to simply labeling each picture. The older

children tried to make sense of the pictures as presented, and made in-

ferences about thoughts and emotions in order to tie a story line together.

This finding is consistent with Stein's 1977 study in its developmental

difference, but a perplexing discrepency exists. The "older" students

in the Poulsen, et al. study were six; the younger children in Stein's

study were first grade, probably six or seven years old. The age that

Stein reported as unable to resolve incongruity showed itself to be quite

able to do so in the picture story telling task presented by Poulsen et al.

One possible explanation is that in the Stein study, the story was read to

subjects; therefore, they did not have the text for reference. Subjects

in the Poulsen et al. investigation had concrete representation of the in-

congruity, and could ponder and process with more ease. Also, the required

task would tend to facilitte inference in the story-telling condition.

Telling a story, for a child who understands the construct of a story

(which both four and six year old children in the study did),- requires

the logical connection of ideas and events. In the case of the incon-

gruous pictures (those in scrambled order), inferences were required to

create a cohesive story. In contrast, the Stein task was only loosely

focused on resolving ambiguity, with instructions to fill in missing

information. It is quite possible that with specific inference probes,

these younger children could have produced logical inferences. The

work of Paris and Lindauer (1976) supports this hypothesis.

13
 12

Another major category of inferential research, besides its develop-

mental nature, has to do with the nature of inferences made. Much of the

current examination of inference is based on schema theory proposed by

reading theorists (Anderson, 1977; Rumelhart & Ortony, 1977; Spiro, 1977).

Various systems for analyzing inferences have been devised, such as Collins,

Brown and Larkin's (1977) Progressive Refinement Model, which *entails

examining inference holistically, within the framework of the reader's

construction, evaluation, ano perhaps revision of a scenario for the

text" (Tierney, Vaughn, & Bridge, 1979). Frederiksen's Taxonomy of Text -

Based Inferences has been used in several research designs as a system for

categorizing types of inferences (Frederiksen, 1977; Frederiksen, Frederiksen,

Humphrey, & Otteten, 1978). An excellent, highly usable taxonomy was

proposed by Warren, Nicholas and Trabasso (1879). The following group

of studies adds to these taxonomies empirical data on the nature of

inference.

Goetz (1977) conducted a study of 184 subjects divided into a ninth/

tenth grade group and an eleventh/twelfth grade group. Groups were

further separated into high and low verbal ability sections. The material

read was of two types: an important version and an unimportant version.

One story, for example, presented the reader with the information that a

woman was late to the airport because she stopped to talk to her son. In

the unimportant version, the narrative describes the outcome of this

diversion to be less time spent in the airport. In the important version,

the talk causes her to miss her flight, which crashes; thus the time with

her son saved her life. Other text variables included salience and
 13

explicitness, described by Goetz in the following passage:

Salience of the premise was varied by changing the amount of detail

or emphaasis given to the material that cued the specific target in-
ference. Explicitness of the target was varied in order to provide
a control conditiorifirwhich the target had been explicitly stated
(p. 45).

There were six sets of eight passages with various combinations of im-

portance, salience and explicitness, each from 500 to 520 words in

length. The experimenter carefully matched these fiction passages,

and arranged them to avoid an order affect.

Subjects were asked five questions about each passage. The first was

always an inference question, the last an "importance" question. Inter-

spersed were basic premise questions and questions used to control for

general memory. Subjects also rated each question as 1) an exact quote,

2) paraphrase, 3) directly implied, or 4) consistent with story. An

interesting design strategy had the students perform a word recognition

test after reading the passage and before answering the questions; this

test served as a buffer for short-term memory as well as a control for

readability of the passages (albeit a limited control).

Main effects were reported for grade level and verbal ability. There

was no main effect for importance; however, there was an interaction be-

tween importance and explicitly stated items. Goetz reports this inter-

action:

Simple effects analysis of this interaction revealed that the

important passages produced significantly higher scores than
the unimportant passages for the implied condition (Important
a .81, Unimportant .74) p 1C.01. Thus the importance of the
target inference had a significant effect on the probability
that the inference would be made.

Salience was also involved in an interaction; low verbal ability students

15
 14

inferred significantly more highly salient passages. Students seemed

avert that they were making inferences, since they rated significantly

more explicitly stated propositions as having been stated. In summary,

it appears that inferences are more likely to be made when it is important

that they be made.

A study designed to examine when and how inferences are made was

conducted by Spiro and Esposito (1977). They required twenty educational

psychology majors to read 250 word vignettes and answer questions. In

the passages, A *Practically implied" B, and variable C lessened the

extent of that implied relationship. One group of students read the

passages with C coming after the A-B relationship; the other group read

C before A -B. The control group had no C statement. Seven questions

1
were asked for each passage, with the following distribution:

1) asked if B had been mentioned

2) asked a yes/no question of B

3)-6) other questions about the passage, designed to mask the intent

of the study (not analyzed)

7) C information.

Subjects further rated the certainty of their responses on a six point

scale.

There were significantly (p < .015) more errors when C was read after

A-B than when C occurred before. In the C-after condition, subjects often

reported that the inference was not in the text (B-mention errors) or that

the opposite was true (B- incorrect errors). Spiro and Esposito concluded

that "predictable information is superficially processed and not stably

 15

represented in memory." While it appears likely from this experiment that

readers are more likely to make errors after the inference has been made,

then contradicted, design limitations lessen the generalizability of the

findings. The authors report sample sizes of only four students in each

of the experimental gra,'Ps, and eight in the control croup. While they

report that group differences accounted for 23.7% of the reported variance,

they did not specify whether this analysis was of the experimental versus

control, between the two experimentals, or among all three. Four remains

an almost in-onsequential sample size. Perhaps an even more damanging

problem surfaces in examining the instructions given the students and the

time allowed for reading. Subjects were instructed to read and understand

each passage, and asked not to look back. However, they were given nine

minutes for each 250 word passage. Sonic must have filled the eight minutes

with rereading, if only out of boredom. Others might have processed the

incongruity of the C-after condition, and reread for resolution.

Schreiner and Shannon (1NO) examined "The Recall of Explicit and

Implied Propositions by Good and Poor Readers Using Three Types of Assess

ment Procedures." They defined good readers as eighth grade students

reading at a mean grade equivalent of 10.0, based on the California

Achievement Test; poor readers were from the same grade reading at a

mean grade equivalent of 5.0. The subjects were randomly assigned to

code groups FR (free recall), AR (aided recall) or MC (multiple choice).

Three "natural expository passages" were read; they were reported as

having readabilities of 3.0, 4.0 and 5.0 on the Fry graph (although the

Fry does not measure to tenths).

17
 16

Hain effects were found for ability, explicitness and question type.

Good readers performed better than poor readers, with means of 2.85 and

1.45 respectively. Explicit propositions were recalled a mean number of

times 2.47; the mean for implicit items was 1.59. Means for multiple

choice, aided recall and free recall were 2.32, 1.40 and .80, respectively.

The authors reported surprise that even the high group recalled only 52%

of the total propositions; however, this finding is consistent with other

research (Tierney, Bridge, & Cera, 1978-79).

A major limitation of the study is the classification of good and

poor readers, and subsequent testing based on that classification. It

was not reported whether the scores from the CAT were overall reading

score, vocabulary, or comprehension section. If comprehension was a

large part of the determinant score, then the poor performance on com-

prehension questions was to be expected. However, a matter for greater

concern is the readability level of the passages used. All passages were

at an independent level for the good readers. The poor readers, on the

other hand, may have been reading at instructional and even frustration

levels, since their average reading ability was 5.0. This failure to

match reading ability with passage difficulty is a crucial limitation.

The rule of prior knowledge on a reader; ability to make inferences

was the focus of a study conducted by Pearson, Hansen, and Gordon (1979).

Their population was "slightly above average" second graders, differen-

tiated by strong and weak schemata for knowledge about spiders. Recall

was measured by "whi" questions, which were judged to be more ecologically

valid than frame recall. The two groups did not differ significantly on

18
 17

average reading ability scores, as measured by the Metropolitan Achievement

Test, or on IQ (weak Schema mean IQ = 114.8, strong group a 120.4). The

subjects were placed in schema groups based on their scores on an eight-

question test of knowledge about spiders. Students answering five, six,

or seven (mean = 5,8) were designated as having strong schemata. Students

answering four questions were dropped from the study; the remaining groups

were significantly different at the .001 level.

To these well- defined groups, Pearson et al. presented a basal

selection (Lyons and Carnahan, 1972) rewritten to include extra information

about spiders, and a narrative line. The story had a readability level of

2.8 on the Spache Readability Formula. Students read the passage once,

silently, and could not refer to it during questioning. Examiners did not

help with unknown words. Explicit and inferential knowledge were measured

during questioning.

Results indicated a simple effect for prior knowledge on inferrable

questions (p .025). There was not a significant effect for prior

knowledge on explicitly stated questions. Main effects for strength of

prior knowledge and question type, both significant at the .01 level,

are reported in Table 3.

Table 3

EFFECTS OF PRIOR KNOWLEDGE AND QUESTION TYPE

explicit questions implicit questions total

strong
schema 4.7 2.8 7.5

weak
schema 3.8 1.0 4.8

average 4.25 1.9

19
 18

In their discussion of the results, Pearson et al. note that the

findings are not consistent with the prediction that prior knowledge would

affect both inferential and literal recall. However, it seems that the

lack of significant differences on the explicit material is actually a

validation of the reading ability/intelligence control so carefully built

into the design. The study showed that both groups learned stated facts

equally well. The more difficult process of combining facts with prior

knowledge (inferencing) did prove to be dependent on level of prior

knowledge.

One of the most widely referred to studies of the nature of inference

was performed by Connie Bridge as dissertation research, and reported by

Tierney, Bridge,and Cera (1978-79). The study was concerned with the

extent and type of inferential processing of written, non-narrative

discourse. Subjects were thirty-six third grade readers, divided into

groups of good and poor readers on the basis of Stanford Achievement Test

scores and teacher judgement. Children read orally a passage about

dinosaurs, read a buffer passage to control for short-term memory, and

responded to free and probed recall. Probe questions were based on free

recall reg es, and avoided leading readers to insights they had not

generated. Frederiksen's system of text analysis for Text-Based In-

ferences was used to categorize responses; his eight classes of inferential

operations were the basis for comparison of types of inference between

the two groups.

Major findings included the well-documented point that good readers

answer significantly more questions than poor readers, whether the questions

are analyzed as a whole or for explicit and inferential types separately.

20
 19

Good readers tend to do everything in the reading task better than poor

readers; that is why their SAT scores and teachers' opinions single them

out as good readers. There was no significant difference for good or

poor readers in the proportion of inferred to explicit recalls. (There

is a discrepency in the reporting of this result. In the abstract of

Bridge's dissertation, it was stated that approximately 45% of the re-

calls were inference. In the Tierney, et al. report, the proportion of

inferred to explicit recalls was statistically reported as .51 for all

readers.) The third major finding was that during free recall, both

groups recalled about twice as elm explicit propositions as inferred.

In the probed condition, more inferences than literal recalls were

recorded.

Possibly the most important contribution this study makes is about

the nature of inferences made by good and poor readers. The discussion

section explains:

...good readers generated more causal and conditional connections

between propositions, while poor readers tended to substitute more
general concepts for the specific terms used in the passage. The
good readers' use of causal and conditional inferences seemed to
add to the organization and cohesiveness of their recalls. The
poor readers' use of superordinated concepts seems to result in
a loss of accuracy and specificity (p. 564).

This fis44ng indicates the need for more specific criteria for

evaluating inferences. The Warren et al. system is more accurate in

assigning inferences on the basis of correct inferences, as well as

on their basis in text. Warren et al. (1979) point out the necessity

of applying a "relevancy criterion" to certain categories.

There are several problems affecting the generalizability of the

study. There was no matching of groups, nor reporting of group means

21
 20

for such relevant data as intelligence or verbal ability. The stalardized

test vac not reported; groups were assigned according to percentiles (good

. above sixty-eighth percentile, poor . below thirty-eighth percentile;

these are not equivalent). It was reported that the experimenters dropped

those who scored below 84% word recognition on the experimental material.

That is a terribly low word recognition criterionOt is most likely that

many of the poor readers were reading at a frustration level. Did their

necessary attention to word recognition and analysis interfere significantly

with their ability to comprehend? Bid the fact that the passage was read

orally interfere with the comprehension of either group?

In discussing the findings, the authors hypothesized that "maybe with

probing the good readers were willing to extend their recalls beyond the

explicit." It also seems likely that students, conditioned by school,

were tuned into the "right answer syndrome" which emphasizes the recall

of literal information (Durkin, 1978 -79). There may be another factor

influencing the reversed proportion of explicit/inferred recalls i. the

probed condition. It is likely that the questioners asked questions

posed to elicit inferences, since that was the focus of the study. It

is also logical to assume that the good readers, especially, had already

recalled most of the explicit information they were capable of relembering;

inference was the next level of processing (robe, 1g65).

In 1979, Tierney and Bridge performed further analysis on Bridge's

data They examined Frederiksen's Text-Based Inference categories, and

found no statistical differences across categories. lhey then reexamined

the responses of nine poor and nine good readers according to five

additional categories. These included thrte of Frederlksen's Functional

22
 21

Contexts of Inference: generalization, deletion and integration. Also

included were vanDijk's macro-rules for extension and connection. No

significant differences were found for generalization, extension or

connection. Results were significant for the finding that poor readers

deleted more propositions than good readers (p .05), and that good

readers integrated more propositions than poor readers (1)4(.01).

Tierney and Bridge conclude that "the ability to integrate propositions

is essentially what differentiates 'control' of inferred processes by

good and poor readers" (p. 32). While this seems to be an oversimplifi-

cation, even an overgeneralization from the data, it is certainly a

variable worth investigating. Further research should differentiate

between correct and incorrect (or appropriate and inappropriate)

inference categories to simplify interpretation of data.

A final category of investigation concerning inference is the more

pragmatic research which asks how inferential ability can be improved.

This research will be presented briefly, since it is peripherally related

but not the central focus of this paper. Nardelli (1957) failed to find

significant gains ia sixth grade pupils who participated in a short unit

of instruction on drawing inferences. Wong (1980) found that learning -

disabled students were better able to draw inferences when prompted

recalls were used; however, no gain in actual ability was measured.

Tierney and Cunningham (198D) report that an unpublished doctoral

dissertation (Cordon, 1979) compared tree treatments on increasing

inferencing. Treatment one sought to build prior knowledge and awareness

of text structure; the second taught inferencing strategies directly;

the third treatment was actually a control utilizing a "language-related"

23
 22

curriculum. Better transfer was reported for direct instruction, although

this was not the result hypothesized. Although their results with college

students may not generalize, Steingart and Clock (1979) found imagery

to have a more facilitating effect than mere repetition for making in-

ferences. Second graders studied by Hansen and Pearson (1980) made more

inferences, and the learning transferred better, when they were psrt of a

group where inferential questions were asked regularly. This group was

superior to a *strategy" group that was trained in the process of in-

ferencing from their own experiences.

This paper has been an attempt to analyze two main branches of inferen-

tial research, inference as a developmental process, and the nature of in-

ferences made during reading. Inference has been identified as a unique

component of comprehension (Davis, 1944 & 1968; Spearritt, 1972). Tyler

(1930) establisived that inference was a skill that could be objectively

measured. When several different ages and ability levels of readers were

examined, differing patterns of inference emerged. Paris conducted

several experiments showing that young children are capable of producing

inferences, but that they usually do not do so spontaneously (Paris &

Upton, 1974; Paris, 1975; Paris & Lindauer, 1976).

While almost every study of inferende has shown that older children

spontaneously produce more inferences (as do better readers), and that

most readers give more literal recalls than inferential ones, other

significant differences have been noted. Many of these differences deal

with type or quality of inference as a developmental variable. Hildyard

(1976) found that there was a "marked developmental difference" in the

ability to draw inferences that contradicted previous world knowledge.

24
 23

Younger children produced more irrelevant inferences, while older readers

tended to have more appropriate additions in Handler and Johnson's 1977

study. Third graders proved to make better use of added information in

drawing inferences to solve ambiguity than did kindergarteners (Nezworski,

et al.). This younger age g -moo however, proved capable of producing

inferences to solve ambiguity when presented with scrambled-order pictures,

a task that proved too difficult for four year old subjects (Paulsen, et al.).

Examinations of the nature of inferences made by readers of connected

discourse have nt been as conclusive as the developmental studies. With

the hypotheses that importance, salience and explicitness would affect the

reader, Goetz (1977) found no main effects for importance or salience. It

did appear, from reported interactions, that inferences are more likely to

be made when they are important to the 'tory; salience was reported to

enhance the inferencing of students with low verbal ability. Inference

appeared to be superficially processed, without stability in memory, in a

study by Spiro and Esposito (1977). Prior knowledge seems to have a.

facilitating affect on inferential processing (Pearson, et al.). In

a major contribution on the nature of inferencing, Tierney, et al. (1978-

79) discovered that good readers made more causal and conditional connections

between propositions;. poor readers more often overgeneralized, Tierney and

Bridge (1979) later reported that the key to good reading was the ability

to integrate propositions.

Here and there, in bits and pieces, inference is being established as

a measurable, definable skill. Perhaps the only point of unanimity is

that inferentia' ability is vital in the mature reader. whose who are

devoting their energies to. producing mature readers have accepted the
 24

challenge presented by Thorndike over half a century ago: It is not a

small or ubwortby task to learn 'what the book says" (Thorndikes 1917,

p. 434).
 References

Anderson, R. C. The notion of schemata and the educational enterprise.

In R. C. Anderson, R. J. Spiro, & W. E. Montague (Eds.), Schooling and

the acquisition of knowledge. Hillsdale, N. J.: Erlbaum, 1977.

Calfee, R. C. & Drum, P. A. How the researcher can help the reading

teacher with classroom assessment. In L. B. Resnick & P. A. Weaver

(Eds.), Theory and practice of early reading (Vol. 1). Hillsdale, N.J.:

Erlbaum, 1979.

Collins, A., Brown, J. S., & Larkin, X. M. Inference in text understanding

(Tech. Rep. No. 40). Urbana, Ill.: University of Illinois, Center for

the Study of Reading, 1977.

Davis, F. B. Fundamental factors of comprehension in reading. Psycho-

metrika, 1944, 9, 185-197.

Davis, F. B. Research in comprehension in reading. Reading Research

Quarterly, Summer, 1968, 2, 499-545.

Durkin, D. What classroom observations reveal about reading comprehension

instruction. Reading Research Quarterj, 1978-79, XIV, 481-533.

Feder, D. D. Comprehension maturity tests - a new technique in mental

measurements. Journal of Educational Psychology, November 1938,

XXIX, 597-606.

Frederiksen, C. H. Discourse comprehension and early reading, In

1. B. Resnick & P. A. Weaver (Eds.), Theory and practice of early

reading (Vol. 1). Hillsdale, N.J.: Erlbaum, 1979.

Frederiksen, C. H. Inference and the structure of children's discourse.

Quoted in J. C. Harste & R. F. Carey (Eds.) New perspectives on

comprehension. 1979 (3, US ISSN 0193-4759).

27
 Frederiksen, C. H., Frederiksen, J. D., Humphrey, F. M., & Ottesen, J.

Discourse inference: Adapting to the inferential demands of school

texts. Quoted in J. C. Harste & R. F. Carey (Eds.) New perspectives

on comprehension. 1979 (3, US ISSN 0193-4759).

Goetz, E. T. Inferences in the comprehension of and memory for text,

Urbana, Ill.: University of Illinois, Center for the Study of Reading,

1977. (ERIC Document and Reproduction Service No ED 150 548),

Gordon, C. J. The effects of instruction in metacognition and inferencing

on children's comprehension abilities, Unpublished doctoral disser-

tation, University of Minnesota, 1979. Quoted in RS 4% Tierney &

J. W. Cunningham. Research on teaching reading comPrehensfo4(Tech,

Rep. No 187). Urbana, Ill.: University of Illtncis, Center for

the Study of Reading, November 1980.

Hansen, J., & Pearson, P, 0, The effects of inference traini9lanOractice

on ,bung children's comprehension, April 1980. (ERIC Document Repro-

duction Service No. ED 186 839),

Hildyard, Alt Children's abilities to produce inferences from written

and oral material. Unpublished doctoral dissertation, Vniyersity of

Toronto, 1976.

Handler, Z. M., & Johnson, N. St Remembrance of things parsed; Story

structure and recall. Cognitive Psychology" 1977, 1 111-151'1

Nardelli, R. R. Some aspects of creative reading, Journal 'of Educational,

Research, March 1957, L, 495-508.

Nezworski, T., Stein, N. L & Trabasso, T. Story structure versus

content effects on children's recall and eyaluative inferences, June

1979. (ERIC Document Reproduction Seryice No, ED 172 187),

Paris, S. G. Integration and inference in children's comprehension and

merory. In F. Restle et al, (Ed:,) Cognitive Theory (Vol, 1)%

New York: John Wiley and Sons, 1975.

Paris, S. G., & Lindauer, B. K. The role of inference in children's

comprehension and memory for sentences. Cognitive psychology, 1976,

8, 217-227.

Paris, S. G., & Upton, L. B. Children's comprehension of implicit and

explicit information in paragraphs, 1974. (ERIC Document Reproduction

Service No. ED 099 104)

Pearson, P. D., Hansen, J. & Gordon, C. The effect of background knowledge

on young children's comprehension of explicit and implicit information,

March 1979. (ERIC Document Reproduction Service No. ED 169 521)

Poulsen, D., Kintsch, E., Kintsch, V., & Premack, D. Children's

comprehension and memory for stories. Journal of Experimental Child

Psychology, December 1979, 21, 379-403.

Rumelbart, D. E., & (*tom, A. The representation of knowledge in memory.

In R. C. Anderson et al. (Eds), Schooling and the acquisition of,

knowledge. Hillsdale, N. J.: Erlbaum, 1977.

Schreiner, R., & Shannon, P. The recall of explicit and implied propo-

sitions by good and bad readers using three types of assessment

procedures. Paper presented at the Annual IRA Convention, St. Louis,

May 1980.

Smirnov, A. A., & Zinchenko, P. I. Problems in the psychology of memory.

In M. Cole & I.N6ltzman (Eds.), A handbook of contemporary Soviet

psychology. New York: Basic Books, 1969.

Spearritt, D. Identification of subskills of reading comprehension by

maximum likelihood factor analysis. Reading Research Quarterly, 1972,

VIII, fl, 92-111.

Spiro, R. J. Remembering information from text: Theoretical and empirical

issues concerning the state of schema reconstruction hypothesis. In

R. C. Anderson et al. (Eds), Schooling and the acquisition of knowledge.

Hillsdale, N.J.: Erlbaum, 1977.

29
Spiro, R. J., & Esposito, J. J. Superficial processing of explicit inferences

in text, December 1977. (ERIC Document Reproduction Service No. ED

150 545).

Stein, N. W. How children understand stories: A developmental analysis,

March 1978. (ERIC Document Reproduction Service No. 153 205).

Stein, N. L., & Glenn, C. G. A developmental study of children's recall

of story material. Paper presented to Biennial Meeting for the

Society for Research in Child Development, Denver, April 1975.

Steingart, S. K., & Glock, M. D. Imagery and the recall of connected discourse.

Reading Research Quarterly, 1979, IL dl, 66-82.

Strange, M. Instructional implications of a conceptual theory of reading

comprehension. The Reading Teacher, January 1980, 21, 14, 391-397.

Thorndike, E. L. Reading as reasoning: A study of mistakes in paragraph

reading. The Journal of Educational Psychology, 1917, VIII, 16,

323-332.

Tierney, R., & Bridge, C. A. The functions cf inferences: an extended

examination of discourse comprehension. In N, T, Kamil & A. J,

Moe (Eds.), Reading research: Studies and applications. Clemson, S.C.:

Reading Conference, 1979, 129-133.

Tierney, R., Bridge, C. A., & Cera, M. The discourse prccessing

operations of children. Reading Research Quarterly, 1978-79, 1/V,

14, 539-573.

Tierney, R., & Cunningham, J. W. Research on teaching comprehension,

November 1980. (ERIC Document Reproduction Service No. ED 195 946),

Tierney, R., Vaughn, J. 1., & Bridge, C. A, Toward Understanding com-

prehension: An examination of systems for analyzing inferences,

In J. C. Horste & R. F. Carey (Eds.) New perspectives on

comprehension. Monograph in Teaching and Learning, 1979 (3, US ISSN

0193-4759)
Trabosso, T. On the making of inferences during reading and their

assessment, January 1980. (ERIC Document Reproduction Service No.

ED 181 429).

Tyler, R. W. Measuring the ability to infer. Educational Research

Bulletin, November 1930, IX, 475-80.

Warren, W., Nicholas, D., & Trabasso, T. Event chains and inferences

in understanding narratives. In R. Freedle (Ed.) New directions in

discourse processing. Norwood, N.J.: Ablex, 1979.

Wong, B. Y. L. Activating the inactive learner: Use of questions /

prompts to enhance comprehension and retention of implied information

in learning disabled children. Learning Disabilities Quarterly,

Winter 1980, 3, #1,