UALL 2004

PSYCHOLINGUISTICS

Lecture 10

Language Processing

1
 Introduction
• We engage in language processing every day of our
lives.This processing takes place when we watch
television, listen to the radio, read a passing billboard,
write a letter, or have a conversation.

• We might sometimes be aware that we are searching for

a word, composing a sentence, or straining to
understand someone else, but we are never aware of
the actual mechanisms and operations involved in
producing and understanding language.

• Psycholinguistics seek to understand how language is

done (O’Grady, 1997).
2
 Levels of Representation
• You are walking through the streets of London and you
encounter the sad sight of a homeless person in a
doorway.
• In an accent which resembles London Cockney, the
person says to you:

Got any change

gɒt/(ʔ)ˈɛni ʧeɪnʤ

The symbol ʔ represents a ‘glottal stop’ – a brief blockage of air at the back of the back of the
throat.
3
• Here are the levels of
representation through
which a listener might
conceivably need to
proceed.

4
• It is unlikely that the process we have just examined
represents what actually happens. There are a number
of reason. Among them are:

5
 In addition…

• This is a bottom-up account based entirely upon linguistic

data.

• So far, we have taken no account of the possible effects of

context (e.g. world knowledge or previous experience of
homeless people).

6
 Serial vs. Parallel Processing
• Evidence suggests that a listener begins to process an
utterance about 200 milliseconds after the speaker has
begun to speak.

• So, we process an utterance while it is happening. This

means that the kind of serial process shown in Figure
B6.1 does not happen.

• Instead, it seems that we process linguistic information in

parallel as shown in Figure B6.2.

7
• Let us assume that the listener
is operating at the phonological
level, attributing a standard
from to [ʧeɪnʤ], the last part of
the utterance.
• By this point, they might have
reached the segmentational
level with GOTANY and be
inserting possible word
boundaries.
• They might have reached the
syntactic level with GOT and be
working out its grammatical
role.
• So, different parts of the
utterance are being processed
at different levels – all at the
same time.
8
9
 The Role of Context
• The sequence in both serial and parallel processing
operates in one direction – bottom – up – and it is
based entirely on what the listener hears or the reader
sees.
• But it is possible that our analysis of the input is
influenced by external factors in a top-down way.

World Paralinguistic
knowledge evidence

An expectation
based on
Analogy
words already
uttered
Language
Processing

10
 From Concept to Expression
• The Speech Chain - The different forms in which a
spoken message exists in its progress from the mind of
the speaker to the mind of the listener.

• Speakers will have some notion, concepts, or message

that they wish to convey (Fodor, 1975).

• Speech Production – the process by which a speaker

turns a mental concept into a spoken utterance.

11
 From Concept to Expression
• Two kinds of data in the construction of speech
production models – SPEECH ERRORS and SPEECH
DISFLUENCIES.

• These date have provided evidence for the units used in

generating speech and for the stages that lie between the
message the speaker wishes to convey and its spoken
expression.

12
 Sources of Data for Models of
Speech Production
1. Speech error
• All of us have experienced, either as speakers or hearer,
utterances that seem to have gotten mixed up on their
way out.
Intended Utterance Actual Utterance
You have missed all my history. You have hissed all my mystery lectures.
You have wasted the whole term. You have tasted the whole worm.
You are occupying my pew You are occupewing my pie
The dear old Queen The queer old dean

• Such errors in production, called speech errors or slips

of the tongue, occur regularly in normal conversation.
13
 • Examples in the previous table
are all attributed to the Reverend
William A. Spooner, the warden
of New College, Oxford
University, England, from 1903 –
1924.

• Because he became “famous” for

producing such errors, they are
Rev.William Spooner was well known often called Spoonerisms.
for his slips of the tongue.

14
• A sufferer of dysgraphia, which is a form of dyslexia -
disturbance of the clear distinction of the sounds of
words - confusion between closely related phonemes.

• Notable for absent-mindedness.

• Liable to mix up the syllables in a spoken phrase, with

unintentionally comic effect (genuine mistakes).

• Such errors may be funny to the listener or

embarrassing to the speaker.

15
2. Disfluencies
• Many utterances are characterized by hesitations,
repetitions, false starts, and “filler” words such as um,
well, or you know (sometimes called filled pauses).

• Such disfluencies are actually more common than we

think – we tend not to notice them.

• Goldman – Eisler (1968) suggests that hesitations occur

roughly every five words when people describe pictures.

16
• Such lapses in fluent speech production actually provide
valuable insights into the units of speech production and
permit us to evaluate how much of speech is mentally
planned in advance of its production.

• Pausing phenomena - this strategy is used to pre-plan

utterances.

17
 The Units in
Speech
Production

Phonemic Phonetic
The Syllable Stress
Segments Features
• Phonemic Segments: There are FOUR (4) kinds of errors at this level:

1. Anticipation errors
• In anticipation errors, sounds which come later in the utterance
inappropriately appear earlier than intended.
 Consonant anticipation

 Vowel anticipation

18
2. Preservation Errors
• In preservation errors, a sound produced early in an utterance reappears
in an incorrect location later in the utterance.
 Consonant preservation

 Vowel preservation

3. Phoneme Exchange
• In phoneme exchange errors, phonemes are reversed.

4. Phoneme Deletion
• In phoneme deletion errors, phonemes are deleted.

19
 What Speech Error Data Suggest About
The Process of Speech Production
• The lexicon is organized both semantically and
phonologically.
– In word substitution errors and word blends, words
involved are semantically or phonologically similar or
both.

– Fay and Cutler (1977), Cutler and Fay (1982), and

Hurford (1981) show that in such speech errors the
target word and substituted word share significantly
similar initial segments, stress placement,
morphological structure and phonological form.

20
What Speech Error Data Suggest About
The Process of Speech Production
– Examples;
• That’s a horse of another color  … a horse of
another race (semantic substitution)

• White Anglo-Saxon Protestant  …prostitute

(phonological substitution)

• Edited/annotated  editated (semantic blend)

• Arrested and prosecuted  arrested and
persecuted (phonological/semantic)

21
• Affixes behave differently from contents words in slips of
the tongue/speech errors.

– Garrett (1976, 1984, 1988) points out that affix

morphemes and minor sentence elements – adverbs,
intensifiers, determiners may be “moved”/“shifted”.

– The major category stems and words (nouns, verbs,

and adjectives) tend to be involved in exchange
errors.

22
What Speech Error Data Suggest About
The Process of Speech Production
– Examples;
• Did you stay up very late last night?  Did you
stay up late very last night?

• That would be the same as adding ten  …as add

tenning

• If she wantS (-/s/) to come here  if she want to

comeS (/-z/) here

23
 The Garrett Model
• In 1975, Garrett proposed a speech production model
based on speech error data.

• This model has provided a major framework for further

research in the speech production field.

– Garrett’s (1975) Model

24
Garrett’s 1975 Model

25
 Garrett’s 1975 Model

• Has three levels

– A conceptual level
– A language-specific sentence level
– A motor level of articulatory control

• A pronunciation-oriented representation occurs at the

positional level.

• The phonologization of grammatical morphemes takes

place at the level of phonetic representation.

• This level is required to account for errors which show

apparent alteration so that the eventual output conforms to
the regular phonological constraints of the language. 26
27
 Dell’s Model
• Dell’s spreading activation model of speech production
(Dell, 1986) is such a connectionist model.

• Words are organized into networks with connections

between units based on semantic and phonological
relatedness.

• The activation of a concept spreads activation to those

lexical items sharing semantic features with the
thought to be conveyed.

28
 Dell’s Model
– Example : Some swimmers sink
– The speaker’s thought, “swimmers,” activates,
among other things, a class of nouns, nominal
affixes – plural

– Activation of other lexical items - Selection of

swimmer also to some extent activates drown
and swim, in addition to sink.

– Activation of swimmer and sink also activates

aspects of their grammatical usage, as well
as their phonological forms.
29
• Activation is presumed
to be bidirectional.

• It is theoretically
possible to have
interactions between
semantic and
phonological
representations , leading
to slips that share both
phonological and
semantics properties
with the intended output.

• Dell(1995) shows how a

lexical network could
slip from the intended
output present to the
unintended production of
prevent.
30
 Writing and Reading Processes
THE WRITING BUFFER
• You have 5 minutes to write an essay. Stop after each minute of
writing. When you stop, answer the following questions.

• To what extent were you aware of dealing with low level issues
(spelling, letter order, vocabulary, grammar)?
• To what extent were you preoccupied with higher level issues
31
(ordering ideas, planning arguments)?
• It seems that, when we are in the process of writing, we
need to store strings of words in a kind of buffer in the
mind.

• It is not possible to write without some kind of forward

planning.

• Consider how Charles Dickens came to write the line :

• He could not have written the first part without

anticipating the second.

• We have to assume that he was ready with the

continuation: ‘it was the worst of times’ in his mental
32
buffer.
The Stages of Writing
(Brown, McDonald, Brown and Carr, 1988)

33
 The Stages of Writing
1. The amount of processing at the formulation stage is one of the
differences between a skilled and a less skilled writer.

2. The skilled writer takes account both of task demands (‘What do

I want to achieve? Who is my reader?’) and of text demands
(‘What genre and text type is appropriate?’).

3. Within formulation, planning involves setting goals, thinking of

ideas to represent and organizing these ideas so that they are
coherently structured.

4. Translating transforms these abstract concepts into linguistics

form – selecting appropriate vocabulary and syntactic structure
34
 The Stages of Writing
5. It is then that a buffer is needed to store the products of
translating.
6. The next stages, programming and execution, convert the
phonological code of the buffer into motor instructions to
the appropriate muscles, and carry out those instructions.
7. Monitoring, the third stage, is a further mark of skill. A good
writer does not lead simply produce a text but review it both
while and after writing
8. Reading may lead to revisions to the form of the text (for
example, spelling correction), while editing may involve
rethinking decisions made at the formulation stage.

35
 Eye Movements in Reading
• Here are some assumptions that are sometimes made
about reading as a process – especially in the ‘speed
reading’ literature. Do you agree or disagree with
them?
 Efficient readers do not need to read all the words in a text.
They predict many words from the context in which they
appear.
 Efficient readers make large sweeps with their eyes as they
read along lines of text.
 Words can be identified by their overall shapes. So longer
words often take the same time to read as short ones.
 A slow reader is one whose eyes do not move fast enough
from left to right. Increases in reading speed can be achieved
without loss of comprehension.
36
• Reading involves a series of rapid eye movements
(known as saccades) along the line of print or writing,
followed by periods of fixation when the eye rests
upon a point in the text.

• A saccade typically last from 20 – 30 msec while a

fixation can last 150 – 500 msec and sometimes
longer. At the end of a line, the reader makes a return
sweep on to the following line.

• By comparing the eye movements of skilled and less

skilled readers, we can get a clearer idea of what
makes for efficient reading.

• We can also get an idea of what aspects of a text

cause processing problems. 37
• The dots above the text mark the fixation points and the figures show how
long in msec each fixation lasted.
• The saccades move in a left-to-right direction except where an arrowhead
indicates a regression (with fixations shown on a higher line). 38