eye movements and spoken language comprehension: effects of visual context on syntactic ambiguity...
DESCRIPTION
Eye Movements and Spoken Language Comprehension: effects of visual context on syntactic ambiguity resolution. Spivey et al. (2002) Psych 526 Eun-Kyung Lee. Research Question. Does referential context affect initial parsing of syntactically ambiguous sentences? - PowerPoint PPT PresentationTRANSCRIPT
Eye Movements and Spoken Language Comprehension: effects of visual context on syntactic ambiguity resolution
Spivey et al. (2002)
Psych 526Eun-Kyung Lee
Research Question
Does referential context affect initial parsing of syntactically ambiguous sentences? When referential context supports less preferred
syntactic structure, could it eliminate processing difficulty in early phrases?
The role of nonlinguistic factors in sentence processing
Garden path model vs. Constraint-based model
Previous Research[1]Limitations
Reading time measures restricted to measuring processing difficulty No information about what is being processed how
Misleading notion of referential context Not just equated with the preceding linguistic
context Salient information in the environment, the set of
presuppositions shared by discourse participants
Previous Research[2]Two Paradigms
Language-as-action Interactive settings Real-world referents Clear behavioral goals Offline methods
Language-as-product Online measures
(Response measures time-locked to the linguistic input) Decontextualized input Not goal-directed
Current Study
Combines the two paradigms Communication task, well-defined context, clear
behavioral goal (Language-as-action) On-line measure of eye-movement (Language-as-
product)
Target Sentence
A temporarily ambiguous prepositional phrase Put the apple on the towel in the box
Preference for a goal argument over an optional adjunct
Syntactically simpler (Frazier 1987) General preference for arguments over adjuncts (Abney 1989) Linguistic presupposition of uniqueness associated with a defin
ite noun phrase (Crain & Steedman 1985)
Ambiguous region Disambiguating region
Linguistic Presupposition& Referential Context
When there is a single entity in the context Modification is redundant favor argument analysis
When there is more than one entity in the context Referential indeterminacy is created Modification is required to establish a unique referent
Multiple-referent contexts eliminate processing difficulty for the otherwise less-preferred modification analysis (Crain & Steedman 1985, Altmann & Steedman 1988)
What if there is no referential indeterminacy in multiple-referent contexts?
Experiment 1
Method
6 participants Listen to a spoken instruction read out from a script Move objects in a visual workspace following the instructio
n Lightweight headband-mounted eyetracker to monitor the
participant’s attentional shifts 3 types of context (one-referent, two-referent, three-and-o
ne referent context) with ambiguous and unambiguous instructions
Put the apple on the towel in the box Put the box that’s on the towel in the box
18 experimental, 90 filler instructions in 36 trials (or instruction triplets)
Example of an instruction set
Look at the crossPut the apple on the towel in the boxNow put the pencil on the other towelNow put it in the box
Critical instructions were always the first instruction in the set
3 types of Visual Context [1]
One-referent context Single referent If there is a garden path
effect, more looks to the empty towel for “on the towel” in the ambiguous instruction compared to the unambiguous instruction
3 types of Visual Context[2]
Two-referent context Multiple referents (eliciting
referential indeterminacy) whether referential
context eliminates garden path effect
If a referential account is correct, looks to the incorrect goal should be eliminated in the ambiguous instruction
modifier interpretation
3 types of Visual Context[3]
Three-and-one-referent context Multiple referents (eliciting
no referential indeterminacy) Whether linguistic
presuppositions with definite NPs are used on-line in resolving syntactic ambiguity
If yes, looks to the incorrect goal should be eliminated in the ambiguous instruction modifier interpretation
Results[1]
Distractor Object
Incorrect Goal
Results[2]
One-referent context More frequent saccade (55%) out of the target referent region and
into the incorrect goal region in the ambiguous instruction Two-referent context
Rare looks at the incorrect goal (14%) in the ambiguous instruction No difference between the ambiguous and unambiguous
instructions Three-and-one referent context
No significant difference in looks at the incorrect goal between the ambiguous (0%) and unambiguous instructions (22%)
The decision to modify the noun phrase is not purely due to the presence or absence of referential indeterminacy
Reflects on-line access to specific presuppositions associated with definiteness and modification
Results[3]
Referential contexts influence an initial interpretation of ambiguous sentences
However,
Possible confounding effects by some intonational patterns
Experiment 2
Method
The same stimuli and instructions as Experiment 1, but with prerecorded instructions
6 participants Ambiguous instructions were digitally converted from
the unambiguous versions by editing out “that’s” e.g. Put the apple that’s on the towel in the box
What about the prosodic cues in the critical regions?
Results[1]
Parallel results with those of experiment 1
Results[2]Combined Analysis of Exp 1,2
One-Referent
Incorrect Goal > Correct Goal
Garden Path Effect in the ambi. condition
Results[3]Combined analysis of Exp 1,2
Two-Referent
fixation to the distractor referent due to
Referential indeterminacy
No difference b/w ambi. and unambi.
conditions
Results[4]Combined analysis of Exp 1,2
Three & OneReferent
Fewer fixation to distractor reference
Only a few fixation toIncorrect Instrument
No difference b/w ambi. and unambi.
conditions
Summary
Referential contexts play an initial role in parsing (even when the verb takes an obligatory verb argument)
The online use of linguistically coded presuppositions even in the absence of referential indeterminacy (Three & one reference context)
Supports a constraint-based model of parsing
Thank you!