Transcript Slide 1

Back to Usual Issues
• So, evidence supports both Parallelism &
Interaction of multiple within-sentence
constraints
• And shows Verb Bias effects in sentences that
should not require any reanalysis
– But Verb Bias effects could be part of 1st stage, since may
be a kind of syntactic knowledge
– And N+V plausibility effects are only testable in sentences
with Clause structure, so could be due to reanalysis
• What about constraints from outside sentence?
3/5/08
Psyc / Ling / Comm 525 Spring08
Discourse Context
• What are Non-minimally Attached PPs?
– They modify the NP they follow
– When does an NP need modification?
– When it wouldn’t be clear in context who or what it refers to
• If you see/hear The doctor examined the patient with ...
– In a context where there are several patients in the waiting room
– You may expect what follows to tell you which patient is meant
– i.e., context may lead you to expect a Non-minimal Attachment
3/5/08
Psyc / Ling / Comm 525 Spring08
Altmann & Steedman
(1988)
Contexts:
A burglar broke into a bank carrying some dynamite. He
planned to blow open a safe. Once inside, he saw that there
was ...
• 2-safe context:
... a safe with a new lock and a safe with an old lock.
• 1-safe context:
... a safe with a new lock and a strongbox with an old lock.
• Target Sentences:
MA: The burglar blew open the safe with the dynamite and ...
NMA: The burglar blew open the safe with the new lock and ...
3/5/08
Psyc / Ling / Comm 525 Spring08
• Prediction
– If people use discourse context immediately,
– they'll read new lock faster than dynamite after the 2-safe context
– because they need new lock to know which safe is meant
• Results (Moving-Window RT)
– People did read new lock faster than dynamite after 2-safe context
– i.e., the Non-Minimally Attached sentence was easier than the Minimally
Attached sentence
• So, suggests people don’t always try the simpler structure 1st
– What they try first seems to depend on the discourse context
– BUT, these results have been notoriously difficult to replicate!
• What if the context comes from the world rather than the
discourse?
– If that influences parsing, the system is clearly very interactive
3/5/08
Psyc / Ling / Comm 525 Spring08
Tanenhaus, Spivey-Knowlton, Eberhard &
Sedivy (Science, 1995)
• Head-mounted eyetracker
• Track eyes while people look at visual scenes &
hear spoken instructions
– They usually look at objects before reaching for them
• So eye movements can show what they’re thinking of
reaching for, based on how they’re understanding a
sentence so far
3/5/08
Psyc / Ling / Comm 525 Spring08
Procedures
• People looked at displays of 4 objects while hearing
instructions like:
– Put the apple on the towel in the box.
– Ambiguous between:
• Put the apple on the towel in the box. towel = destination (MA)
• Put the apple on the towel in the box. towel = modifier (NMA)
• The visual scene had either 1 or 2 apples present
– If visual context influences parsing, then when there are 2 apples:
• They should think PP modifies it, to pick out which apple
• So they should not think the towel is a destination & thus
should not look at it (much)
• - i.e., They should prefer the Non-Minimal Attachment
3/5/08
Psyc / Ling / Comm 525 Spring08
Typical Result in 1-Ref Condition
Only 1 apple
in display
Ambiguous version
Unambiguous version
3/5/08
Psyc / Ling / Comm 525 Spring08
Typical Result in 2-Ref Condition
2 apples
in display
Ambiguous version
Unambiguous version
3/5/08
Psyc / Ling / Comm 525 Spring08
Overall Results
3/5/08
Psyc / Ling / Comm 525 Spring08
Interpretation
• When there were 2 apples, people rarely looked at
the empty towel
– Suggests rarely considered it as a possible destination
• Thus, that they did not always try MA first?
– So, non-linguistic visual context immediately constrained
interpretation
– So, sentence comprehension processing is Interactive
• Whether you believe this interpretation
– Depends on whether you think people would move their
eyes to the empty towel if they briefly mis-parsed the
sentence & thought it was a destination
???
3/5/08
Psyc / Ling / Comm 525 Spring08
• The next section has slides from a talk of mine
about a study with Reduced Relatives, showing
some individual difference effects. I will talk
about this study when we talk about memory
and language processing, so don’t worry if you
can’t figure it out from just the slides.
3/5/08
Psyc / Ling / Comm 525 Spring08
Individual Differences in the
Effect of Yet-to-be-Fixated
Words during Sentence Reading
With thanks to:
Elizabeth Myers
Neal Pearlmutter
Kate Pirog
Mike Tanenhaus
John Trueswell
Gary Wolverton
3/5/08
Psyc / Ling / Comm 525 Spring08
My Very Favorite Example of a
Garden-Path Sentence
“A University student charged with threatening to kill
the President via email was arrested Thursday,
following issuance of a complaint and warrant,
officials said.”
- Daily Illini, 2/27/94
3/5/08
Psyc / Ling / Comm 525 Spring08
“A University student charged with threatening to kill
the President via email
X
Y
3/5/08
Psyc / Ling / Comm 525 Spring08
You may not have noticed that there was
another opportunity for garden-pathing in
this example
who was
“A University student charged with threatening
to kill the President ^… ”
“Reduced Relative”
3/5/08
Psyc / Ling / Comm 525 Spring08
The Most Famous English
Garden-Path Sentence
The horse raced past the barn fell.
What makes this one so much harder?
1. Horses are good racers
2. The word past is no help after raced
vs A University student charged with
3/5/08
Psyc / Ling / Comm 525 Spring08
Dual Purkinje Eyetracker
For the font size and distance we use,
spatial resolution is ~1/4 character
3/5/08
Psyc / Ling / Comm 525 Spring08
First-pass
ReadingTime
= Sum of all
fixations in a
region before
leaving the
region
- Used as a
measure of
“initial
processing”
3/5/08
Psyc / Ling / Comm 525 Spring08
Example Stimulus Set for All
of the Studies that Follow
Animate = Good Agent
The witness (who was) examined by the lawyer turned out to be unreliable.
Inanimate = Bad Agent = Possible Passive Patient
The evidence (that was) examined by the lawyer turned out to be unreliable.
Scoring Regions:
The evidence / (that was) / examined / by the lawyer / turned out /…
Disambiguating Region
3/5/08
Psyc / Ling / Comm 525 Spring08
3/5/08
Psyc / Ling / Comm 525 Spring08
3/5/08
Psyc / Ling / Comm 525 Spring08
We thought there were some problems with the materials:
e.g.,
The trash smelled by the dog was laying on the sidewalk.
The questions asked about the murder could not be answered.
The stories told about the incident were a great source of concern.
3/5/08
Psyc / Ling / Comm 525 Spring08
3/5/08
Psyc / Ling / Comm 525 Spring08
3/5/08
Psyc / Ling / Comm 525 Spring08
Daneman & Carpenter’s (1980)
Reading Span Test
At the conclusion of the musicians' performance, the enthusiastic
crowd applauded.
Without any hesitation, he plunged into the difficult mathematics
assignment blindly.
The devastating effects of the flood were not fully realized until
months later.
When I got to the big tobacco field I saw that it had not suffered
much.
3/5/08
Psyc / Ling / Comm 525 Spring08
What does the test measure?
- Verbal working memory capacity?
- Ability to focus attention?
- Skill with language and reading?
-...?
For some purposes, it doesn’t matter
- As long as it reliably identifies subgroups of people
- Who can then be tested for differences in other tasks
- Which may help us figure out what the test measures
3/5/08
Psyc / Ling / Comm 525 Spring08
3/5/08
Psyc / Ling / Comm 525 Spring08
3/5/08
Psyc / Ling / Comm 525 Spring08
Parafoveal Preview in Reading
Burgess (1991)
- Self-paced moving window reading time study
- Varied window size from single to several words
- Found an animacy effect only when people saw the
disambiguating preposition “by” chunked together
with the verb preceding it
e.g., The evidence examined by the lawyer turned out …
Ferreira & Clifton used a 40-character wide display
- Sometimes the preposition was on the next line
3/5/08
Psyc / Ling / Comm 525 Spring08
We created sentences with verbs at least 8 characters long
- So we could analyze data contingent on whether or not
people were likely to get preview of “by” while still
fixating on verb
If last fix before
“by” was here,
trial not used
The professor confronted by the student was not ready to …
3/5/08
If last fix
before “by”
was here,
trial coded as
“Preview Unlikely”
If last fix before
“by” was here,
trial coded as
“Preview Likely”
Psyc / Ling / Comm 525 Spring08
48 item sets with 4 versions each:
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.
All animate
– Want people biased toward main clause interpretations
- So sentences are fairly hard, so there’s a better chance
to see how much preview of “by” can help
3/5/08
Psyc / Ling / Comm 525 Spring08
Participants
- 23 high-span (>=3.5, 13 female)
- 20 low-span (<=3.0, 10 female)
(More would be good, since breaking it down by both span
& critical fixation locations & dropping some trials)
3/5/08
Psyc / Ling / Comm 525 Spring08
3/5/08
Psyc / Ling / Comm 525 Spring08
3/5/08
Psyc / Ling / Comm 525 Spring08
3/5/08
Psyc / Ling / Comm 525 Spring08
3/5/08
Psyc / Ling / Comm 525 Spring08
3/5/08
Psyc / Ling / Comm 525 Spring08
- People are differently abled when it comes to
making rapid use of cues that can help in
understanding hard sentences
- In Trueswell et al. (1994),
- We probably just happened to get a high
proportion of high-span subjects
- And something about the display &/or materials
happened to make it easy to get preview of “by”
- Presenting sentences one word at a time in ERP
studies (among others) may yield unrepresentative
results especially for highly skilled readers
3/5/08
Psyc / Ling / Comm 525 Spring08