Transcript Verb bias and memory constraints in language production

Verb bias and memory
constraints in language
production
Amit Almor, Elaine Andersen, Daniel Kempler, Maryellen
MacDonald*, UnJa Hayes, Houri Hintiryan
University of Southern California
* University of Wisconsin Madison
Introduction
What causes choice of sentence structure during
language production? Several possible factors:
1. Sentence Form Frequency
Example: Prepositional dative (PD) is more common
than no preposition double object (DO).
– PD: I gave a book to them.
– DO: I gave them a book.
Another example: Passive sentences are rarer than
actives.
2. Lexical constraints--verbs vary in their
acceptability in different constructions.
• Double Object (DO) example:
– I gave them a book.
– * I donated them a book.
• Passive example:
– The proposal was considered by the contractor.
– ? The page was turned by the teacher.
Verb bias could interact with construction
frequency--rare structures may get easier with
helpful verb biases.
Another lexical constraint--Thematic fit between
nouns and verb arguments.
Example: “teacher” is a better agent than patient for
the verb “turn”, and “page” is a better patient than
agent.
The teacher turned the page.
? The page turned the teacher.
3. Discourse constraints--Topics are likely to be
grammatical subjects.
If the topic is a better patient than agent for the verb
the sentence is more likely to be passive.
Active Response
Tell me about
the teacher.
The teacher
turned the page.
Passive Response
Tell me about
the page.
The page was
turned by the
teacher.
4. Working memory--when available resources are
limited, harder constructions may be selectively
dispreferred.
Thus, resource limitation can also affect the extent
to which other constraints affect production.
Aging provides a good test bed because:
1. Lexical knowledge remains largely intact.
2. Aging is known to impair global discourse
processing in comprehension and production.
3. Aging reduces working memory.
Experiment 1: Tested lexical constraints & working
memory in young & elderly dative production.
Experiment 2: Lexical, discourse, & working
memory constraints in young & elderly passive
production.
Experiment 1
Questions:
• Do both young and elderly people produce sentence
form in line with lexical biases?
• Does working memory performance play a role in
the production of less-frequent sentence forms?
Participants:
24 young and 24 elderly (mean ages 20 vs. 78).
Working memory: was assessed with two tasks:
Month Ordering and Digit Ordering. (Repeat in order
a list of months or digits that is presented out of order.)
Mean performance in the two tasks was lower for the
elderly subjects, t (46) = 2.60, p < .02.
Items:
• 20 Verbs that take two objects but vary in bias for
prepositional dative vs. double object.
• Each verb was combined with a noun that
constituted a good theme.
• The pronouns “we” and “them” were always the
subject and patient/recipient (to minimize irrelevant
lexical effects and to reduce biases related to the
length of the noun objects).
Pre-study rating task: Lexical biases were assessed
via a rating task.
• Rate on a scale of 1 to 7 which version is better:
1 2 3 4 5 6 7 We sent letters to
We sent them letters.
them.
• 141 young and 47 old raters (age > 60) who did not
participate in the production experiment.
• There was high agreement between the young and
the elderly raters, r = .96, p < .001.
Production Task:
Make a sentence
using the words
on the screen
• The order of the words on the screen was
counterbalanced across different items and different
lists.
Results: DO production rates
Elderly subjects produced fewer DOs, t (46) = 2.76, p < .01.
DO production and lexical biases:
• DO production was linked to the bias ratings for
both young and elderly subjects.
Correlations
Young
r = .96, p < .001
Elderly
r = .88, p < .001
DO production and working memory:
• Young subjects showed no reliable correlation.
• Elderly subjects with higher working memory
scores curiously produced fewer DOs than those
with lower scores.
Correlations
Young
r = .18, n.s.
Elderly
r = -.51, p < .01
Experiment 1 Conclusions:
• Elderly subjects produced fewer infrequent forms
than young subjects.
• Both young and elderly subjects showed great
sensitivity to lexical constraints in the choice of
sentence form.
• Working memory was reliably linked to production
performance in elderly but not young subjects.
Experiment 2
Questions:
• Do both young and elderly people produce sentence
form in line with discourse as well as lexical
biases?
• Does working memory performance interact with
discourse constraints in the production of lessfrequent sentence forms?
Participants:
• Same as Experiment 1.
Items:
• 20 Verbs that varied in their passive bias.
• Each verb was matched with a good agent and a
good patient.
• For example, the verb considered was matched with
the intended agent contractor and the intended
patient proposal.
• Lexical biases were assessed via surveys similarly to
Experiment 1. Again, there was high agreement
between young and elderly raters, r = .9, p < .001.
Passive Production Task:
Tell me about the
proposal.
• Discourse constraint: The experimenter’s prompt
always topicalized the patient noun.
• In filler items, the prompted entity was always the
agent noun.
Results: Passive production rates
Elderly subjects produced fewer passives, t (46) = 4.13, p
< .001.
Passive production and lexical biases:
• Passive production was linked to the bias ratings for
only the elderly subjects but not the young subjects.
Correlations
Young
r = .26, n.s.
Elderly
r = .67, p < .001
Passive production and working memory:
• Young subjects showed no evidence of correlation.
• Elderly subjects with higher working memory
scores produced more passives than those with
lower scores.
Correlations
Young
r = .09, n.s.
Elderly
r = .44, p < .04
Experiment 2 Conclusions:
• As in Experiment 1, elderly subjects produced
fewer infrequent forms than young subjects.
• Only the elderly subjects showed sensitivity to
lexical constraints in the choice of sentence form.
• Working memory was reliably linked to production
performance in elderly but not young subjects.
Conclusions
• Although elderly people have well preserved lexical
biases which affect their production, they
nevertheless produce fewer infrequent forms than
young people.
• Even when discourse constraints overcome lexical
constraints in young people’s production, elderly
people’s production is still strongly attuned to local
lexical constraints.
• The sensitivity of elderly production to discourse
constraints increases with increased working
memory performance.
This research was supported by NIA grant AG11773-05.
To receive the paper or a copy of this poster, please write
your email below: