Transcript PSY 369: Psycholinguistics - the Department of Psychology at

PSY 369: Psycholinguistics
Language Comprehension:
The role of memory

Center embedded structures


The house burned down.
The house the handyman painted burned down.
He sent a letter about it to Galileo, the great Italian scientist.
Memory and comprehension

Brief summary so far:

What is the role of syntax in comprehension?

Syntax is important for getting the right
interpretation during on-line comprehension
 There is a lot of research examining what factors
influence the on-line construction of syntax

e.g., parsing principles, lexical semantics,
plausibility, discourse context

Today: What is the role of Memory in language
comprehension?
Memory for sentences
Fillenbaum (1966)
 Given:


The window is not closed
Conclusions: Meaning gets
preserved, surface structure
(and syntax) forgotten
Tested:




The window is not closed
The window is closed
<-- surface similar, meaning different
The window is not open <-- surface similar, meaning different
The window is open
<-- surface different, meaning similar
Most common error
Memory for sentences
Think back to the beginning of class. Which of the following
sentences did you read?

Galileo, the great Italian scientist, sent him a letter about it.

He sent Galileo, the great Italian scientist, a letter about it.

He sent a letter about it to Galileo, the great Italian scientist.

A letter about it was sent to Galileo, the great Italian scientist.
Memory for sentences
Sachs (1967, 1974)
 Heard (read):


Tested:





“He sent a letter about it to Galileo, the great Italian scientist.”
Same: He sent a letter about it to Galileo, the great Italian scientist.
Act/Pass: A letter about it was sent to Galileo, the great Italian scientist.
Formal: He sent Galileo, the great Italian scientist, a letter about it.
Meaning: Galileo, the great Italian scientist, sent him a letter about it.
Measured accuracy of detecting changes
Memory for sentences
Percent correct
Sachs (1967, 1974)
Sachs (1967)
semantic
change
active/passive
change
formal change
100
90
80
70
60
50
identical
sentence
0
80
160
Amount of interpolated material
(number of syllables)
Conclusions: Meaning gets preserved, surface structure
(and syntax) forgotten
Just good enough representations

Ferreira and colleagues (Christianson et al 2001)

Garden-path sentence



While Anna dressed the baby played in the crib
While Anna dressed, the baby played in the crib
Did the baby play in the crib?
Did Anna dress the baby?
100% correct
40% correct
Comprehenders don’t always get all of the meaning
right, but get enough to get by
Memory and comprehension

Brief summary so far:

What do we remember about sentences?

Syntax may not be too important later, we
remember the meaning of sentences but not so
much the form (syntax) of the sentence
The house the handyman the teacher hired painted
burned down.

Center
embedded
structures
This sentence
is syntactically
legal.





So
is it soburned
hard to comprehend?
Thewhy
house
down.
One possibility is that there are limitations as a result of our
The
house the handyman painted burned down.
cognitive systems
Memory and comprehension

Center embedded structures

The house burned down.
Memory and comprehension

Center embedded structures


The house burned down.
The house the handyman painted burned down.
Memory and comprehension
This one may be legal, but

Center embedded structures



that doesn’t mean that it is
(easily) comprehensible
The house burned down.
The house the handyman painted burned down.
The house the handyman the teacher hired painted
burned down.

(the handyman that the teacher hired painted the house that burned
down)
Memory and comprehension
The man that the woman that the child hugged kissed laughed.
S

NP
RR
NP
C
VP

NP
VP
V
V
laughed
The man
that
RR
NP
the woman
NP
C
kissed
VP
NP
that
V
The child
hugged
Typically we build right
branching structures
But here there is a big
series of branches
down the center
Memory and comprehension
The man that the woman that the child hugged kissed laughed.

Most readers having trouble figuring out who did what to
whom (called thematic role assignment).
 Easier to assign thematic roles in the two sentences that
form it:



The man that the woman kissed laughed.
The woman that the child hugged kissed the man.
Possible explanation for the trouble:
 Insufficient working memory resources to
retain the intermediate products of
computation made building the complex
syntactic structure
Memory: a brief review
Information
Information ‘flows’ from one memory buffer to the next
Memory: a brief review
Declarative
episodic
semantic
Procedural

Properties

Organization

Multiple subsystems for type of memory


Syntax within the procedural domain (e.g., Ullman, 2004)
Semantics as Associative networks
Memory: a brief review

Properties



rapid access (about 35 milliseconds per item)
limited capacity (7+/- 2 chunks; George Miller, 1956)
fast decay, about 12 seconds (longer if rehearsed or elaborated)
Working Memory

Working memory instead of STM
Working Memory




Working memory instead of STM
Phonological rehearsal mechanism
Phonological store
Very limited capacity
 Rehearsal maintains information
in the store
Working Memory

Working memory instead of STM

Store and manipulate visual and spatial
information
 Directly from perception
 Indirectly from imagery
Working Memory

Working memory instead of STM
Allocate attentional resources to the
subcomponents
 Directs elaboration/manipulation of
information

Measuring memory span

Increasing your STM span

Chunking

Grouping information together into larger units

I’ll read a few more lists of words for you to recall
barn snow tree car rock book key plant dress cup slide lamp
dog cat mouse shoe sock toe couch pillow blanket table desk chair
down flowers the by with chased yellow several girls a river boy
a boy chased several girls with yellow flowers down by the river

Notice that the previous two are the same words, but the syntax
allows for grouping into meaningful ‘chunks’
Measuring memory span

Daneman and Carpenter (1980) Technique:



This technique involves presenting sequences of 2 to 6
sentences, each of 12 to 17 words.
The participant has to read the sentences out loud, and
attempt to remember the last word of each.
Then asked to recall as many last words as possible (in
any order).
Measuring memory span
When at last his eyes opened, there was no gleam of triumph, no shade of anger.
Measuring memory span
The taxi turned up Michigan Avenue where they had a clear view of the lake.
Measuring memory span

Recall the last words
When at last his eyes opened, there was no gleam of triumph, no shade of anger.
The taxi turned up Michigan Avenue where they had a clear view of the lake.
Measuring memory span
I turned my memories over at random like pictures in a photograph album.
Measuring memory span
I will not shock my readers by describing the cold-blooded butchery that followed.
Measuring memory span
He had an odd elongated skull which sat on his shoulder like a pear on a dish.
Measuring memory span
You can check out the books that you need for this course at the local library.
Measuring memory span
The radio station was promoting the concert with free tickets and back stage passes.
Measuring memory span
The professor could be seen on weekends in the backyard garden pulling out weeds.
Measuring memory span

Recall the last words
I turned my memories over at random like pictures in a photograph album.
I will not shock my readers by describing the cold-blooded butchery that followed.
He had an odd elongated skull which sat on his shoulder like a pear on a dish.
You can check out the books that you need for this course at the local library.
The radio station was promoting the concert with free tickets and back stage passes.
The professor could be seen on weekends in the backyard garden pulling out weeds.


Ok for two sentences; Hard at 3 sentences; Very hard
for 4 or more.
Used to classify readers as high and low span
(there is a “speaking span” version too)
Memory and online comprehension
The Capacity Theory of Comprehension

(Just & Carpenter, 1992)

Proposed that individual differences in working memory
capacity should influence how readers comprehend
sentences
Memory and online comprehension
The Capacity Theory of Comprehension

(Just & Carpenter, 1992)
Proposed that individual differences in working memory
capacity should influence how readers comprehend
sentences

The animacy of the first noun may constrain the possible
interpretation of the sentence
Semantically Unconstrained:
The defendant examined by the lawyer shocked the jury.
The defendant that was examined by the lawyer shocked the jury.
Semantically Constrained (so should be faster if animacy can be used)
The evidence examined by the lawyer shocked the jury.
The evidence that was examined by the lawyer shocked the jury.
“that was” disambiguates these
sentences

Memory and online comprehension
Just & Carpenter (1992)
600
550
The defendant examined by the
lawyer shocked the jury.
500
msec
Just the ambiguous sentences
animate NP
Inanimate NP
450
400
The evidence examined by the
lawyer shocked the jury.
350
300
Low-span
High-span
High span readers could use the semantic information
to resolve the ambiguity
Memory and online comprehension
King and Just (1991)

Verbs which could provide strong pragmatic cues as to which of
the two potential actors in the sentence was the agent:
Embedded clause
The robber that the fireman rescued stole the jewelry.


Two possible agents:


the robber
the fireman

Two verbs, which is the
main verb of the
sentence?:


rescued
stole
Memory and online comprehension
King and Just (1991)

Verbs which could provide strong pragmatic cues as to which of
the two potential actors in the sentence was the agent:
Strong bias  .The robber that the fireman rescued stole the jewelry.


No bias



The robber that the fireman rescued watched the program.
The robber that the fireman detested stole the jewelry.
The robber that the fireman detested watched the program.
Can bias which Noun goes with which Verb pragmatically (or not)
Questions:
 Can speakers use this information?
 Does memory have an impact?
Memory and online comprehension
King and Just (1991)

Verbs which could provide strong pragmatic cues as to which of
the two potential actors in the sentence was the agent:
Embedded relative verb




Main verb
The robber that the fireman rescued stole the jewelry.
The robber that the fireman rescued watched the program.
The robber that the fireman detested stole the jewelry.
The robber that the fireman detested watched the program.
Memory and online comprehension
King and Just (1991)
Comprehension accuracy


Verbs which could provide strong pragmatic cues as to which of
the two potential actors in the sentence was the agent:
Method
90
H
80 L
H
L

H
H
Data
70
L
L
60
both
Results



Word-by-word moving window
procedure
relV
main
V
neither


% correct on a T/F
comprehension question
when relative clause is tested
Reading times by region
Low-capacity subjects had lower comprehension overall & slower reading in
syntactically difficult regions
High-capacity subjects did NOT improve with pragmatic info
Low-capacity subjects did improve with pragmatic info
Memory and online comprehension
Garnsey, Pearlmutter, Pirog (2003)
The professor (who was) confronted by the student was
not ready for an argument.
The professor (had) confronted the student but was
not ready for an argument.
Question:
Do readers differ specifically in how quickly they can use
disambiguating words to rule out incorrect alternatives?
Memory and online comprehension
Garnsey, Pearlmutter, Pirog (2003)
Eye fixations were analyzed separately
- By whether preview of “by” while still fixating on verb likely
If last fix was here,
trial not used
The professor confronted by the student was not ready to …
If last fix was here,
trial coded as
Preview Unlikely
If last fix was here,
trial coded as
Preview Likely
Memory and online comprehension
Readers who score high on the Reading Span test
- Make better use of a peripherally visible disambiguating word to
quickly rule out a preferred but incorrect interpretation
Memory and online comprehension




Just & Carpenter (1992) - high span readers used
semantic information early, but low span readers didn’t
King & Just (1991) - high span readers did not use
pragmatic information to resolve ambiguity, but low span
readers did
Garnsey, Pearlmutter, Pirog (2003) - span differences may
also depend on where the eye lands (which determines
what kind of preview readers get)
What information is used to resolve syntactic
ambiguities depends on individuals working memory
capacity (but see Walters and Caplan (1996) for alternative
view)
Memory and comprehension

Brief summary so far:

What do we remember about sentences?

Syntax may not be too important later, we
remember the meaning of sentences but not so
much the form (syntax) of the sentence

What is the role of syntax in comprehension?

Syntax is important for getting the right
interpretation during on-line comprehension

Memory capacity may play an important role in
determining what kinds of information we can use
to comprehend sentences
Memory and comprehension

There is room for about 15 more mins. Of material (for
next semester)