Langston,
Psychology of Language, Notes 4 -- Speech Production
I. Goals.
A. Production errors.
B. A production model.
II. Production errors. We need to rethink the way
we’re studying language. Instead of thinking about how you
understand
language, we’re going to look at how you produce language. The
main
way to tell what’s happening in production is to look at the kinds of
errors
that people make.
A. Some of the more common types of errors:
1. Shift: One speech segment moves from its appropriate
location “That’s so she’ll be ready in case she decide to hits it.”
2. Exchange: Two units change place “Fancy getting your
model renosed.”
3. Anticipation: A segment is produced too early “Go ahead,
bake my bike.”
4. Perseveration: A segment gets repeated “He pulled a
pantrum.”
5. Addition: Add a segment from somewhere “I didn’t explain
this clarefully enough.”
6. Deletion: A segment gets left out “I’ll just get up
and mutter _intelligibly.”
7. Substitution: Substitute a related segment for the
intended
segment “At low speeds it’s too light.”
8. Blends: Two possible segments become blended into one
output “That child is looking to be spaddled (spanked/paddled).”
B. Where do errors occur? All levels of production
(phonological,
syntactic, lexical insertion...). But, you don’t see mixes of
levels.
So, for example, phonological errors usually occur in syntactically
correct
sentences.
C. Common properties of errors:
1. Elements that interact come from similar environments.
Beginnings change with beginnings, ends with ends, etc.
2. Elements that interact tend to be similar. Consonants
don’t replace vowels.
3. Even when errors produce novel linguistic items (spaddled)
they’re consistent with the rules of the language (in this case
spelling
rules).
4. Errors will have the same stress pattern as the thing they’re
replacing.
D. Where do errors come from?
1. Freud: Errors are Freudian slips. It’s a way to
look at what someone is really thinking. You have a lot of stuff
going on in your unconscious, and some of it’s bad and being
repressed.
But, repressed stuff sneaks out in production errors. For
example,
there’s a cartoon with a king and a queen, and the king is saying “Good
morning beheaded... er, beloved.” The slip is that she’s about to
be beheaded...
Experimental evidence: Take male subjects and put them in a room
with a scantily clad female experimenter. Errors like saying
“fine
body” instead of “bine foddy” when reading out loud under time pressure
indicate that errors are based on impulses from the Id.
2. ReverseSpeech: To fit with the unconscious message being
uttered in reverse, the forward message will sometimes need to be off a
bit. More on this later.
3. Psycholinguistics: The errors come from the complexity
of producing language. A lot of stuff is happening, it’s
happening
fast, and it’s happening in a system with limited capacity. All
of
that makes errors likely. What is the production system?
Top
III. A production model. The basic model has four
parts:
A. Conceptualize: This buys into the language of thought
idea (that thoughts are in mentalese, not language). You have to
have a thought before you can begin to produce an utterance.
B. Form a linguistic plan:
1. Identify meaning: It’s kind of like looking in a
dictionary
backwards. You have a thought you want to express, and you look
for
definitions that match the thought.
2. Select syntactic structure. We will have a rule in our
phrase structure grammar that says S -> NP + VP (a sentence is a
noun phrase
plus a verb phrase). The real grammar has many possible sentence
types. Choose one here and build the tree (like Chomsky’s toy
grammar).
3. Generate an intonation contour. Fans of Seinfeld will
know that if Jerry’s not sure if he’s invited to a party and the host
says
to Elaine “Why would Jerry bring something?” it’s not the same
as
“Why would Jerry bring something?” In the one case,
Jerry’s
probably not invited. In the other case, he’s invited, just not
expected
to bring anything. This step is where you lay out stress
patterns.
4. Insert content words: Put in the words.
5. Form affixes and function words.
6. Specify phonetic segments: Figure out how it’s
pronounced.
7. Edit: Look at the “darn bore” task. People are
much more likely to say “barn door” after the biasing context than
“bart
doard.” That’s because “barn door” makes words (even if it’s not
the intended utterance), but “bart doard” doesn’t. So, people
probably
have an editing function at the end of the plan to check for basic
errors.
Two things to say about this model:
1. Errors support dividing the plan into these particular
steps.
For example, if you say “Stop beating your brick against a head wall”
it’s
a problem at only stage four (insertion). Everything else seems
to
have gone off OK.
2. Errors support this particular sequence. For example,
if you say “It certainly run outs fast” and pronounce the ‘s’ in “outs”
as /s/, it indicates that the sound part came after the affix part
(because
if the sound came first, the ‘s’ would sound like /z/, since that’s how
it would sound in “runs”).
C. Implement the plan: Once you make a plan, you have to
say what you came to say. What do I want to say about this?
That planning and production seem to go in cycles. If you graph
time
on the x-axis, and amount produced on the y-axis, it looks like people
plan a while, say that much, plan some more, say some more, etc.
Why? Because it’s a hard task, which probably maxes out working
memory
all the time. You have to do it in cycles.
D. Self-monitoring: Obviously, some errors get by the
editing
step. What happens when errors get out? Stop and make a
self-repair.
Three steps:
1. Self-interruption: Signal that you’ve spotted the error
and stop. A procedure to get a lot of this is to have people
describe
colored diagrams (like the one below). They have to describe them
so someone else can draw them. People make a lot of mistakes, and
have to correct themselves a lot.
a. Some statistics about when people interrupt themselves:
Interrupt within a word: 18% of the time “We can go straight on
to
the ye-, er pink.” Interrupt after the word: 51% “Straight
on to green- to red.” Interrupt at some later point: 31%
“and
from green left to pink, er from blue left to pink.”
b. How do they interrupt themselves? Utter an editing
expression
to indicate trouble (generally “uh”).
c. Self-repair: Fix the problem.
Top
Psychology of Language Notes 4
Will Langston
Back to Langston's Psychology of Language
Page