Informativity expectations in on-line reference comprehension

Abstract:

It has long been claimed that modifier phrases serve the discourse function of distinguishing between multiple discourse referents identified by the head noun, and that discourse context effects on syntactic ambiguity resolution reflect this knowledge (Crain & Steedman, 1985).  More recent work has observed discourse context effects on resolving temporary referential indeterminacies of prenominal adjectives (Sedivy et al., 1999).  However, this discourse function may be not a general linguistic property of modification, but rather, an indirect one reflecting expectations of optimal informativity of linguistic expressions with respect to referent identification.

Sedivy (2001) provided evidence for this view by showing that the presence of a discourse-available contrasting object exerted an effect on the on-line processing of referential expressions containing prenominal adjectives only when the adjective denoted a property that was typically unlikely to be produced in describing the target object in isolation (e.g., material and scalar properties).  For properties that are frequently redundantly encoded (e.g., with color adjectives), no such effect of discourse contrast was observed.

The current paper presents further evidence that default expectations pertaining to the amount of information provided in referring to an object impact on-line language processing in discourse contexts.  Experiment 1 shows that discourse contrast effects are also observed with unmodified bare nouns that are more specific than typical descriptions for an object (i.e., subordinate-level expressions such as pump, rather than shoe).  Eye movements to a visual display in response to spoken instructions such as "Click on the pump" showed fewer eye movements to a lexical cohort competitor (e.g., "pumpkin") shortly after the onset of the critical noun when a contrasting object of the same basic-level category as the target (e.g., sneaker) was present in the display than when it was not.  In addition, eye movements to the contrasting object after target identification were dramatically more frequent than the baseline (for unrelated distractor items), suggesting that subjects took this object to be relevant for the interpretation of the instructions.  The fixations to the target object were both much more frequent, and occurred substantially later than the fixations that have been observed for objects that stand is some general semantic relationship to target words (see Yee & Sedivy, 2001).

Experiments 2 and 3 were designed to determine more conclusively whether previous findings that color adjectives don't exhibit discourse contrast effects can be attributed to informativity expectations, rather a generalization about the lexical class of color adjectives.  An elicited production study determined that color adjectives are rarely used to describe objects when their color is predictable from the category of the object (e.g., yellow banana).  Hence, it was predicted that in a comprehension task, instructions containing a color adjective for such objects would be more informative than would be typically expected.  Therefore, in the presence of a contrasting object (e.g., a green banana), the use of the color adjective should trigger the inference that "yellow" is used to distinguish between the target and contrasting objects (yellow vs. green bananas) rather than referring to some other yellow object in the display.  Preliminary eye movement data suggest that this is indeed the case, providing further evidence for sensitivity to informational expectations.

References

Crain, S., & Steedman, M. (1985).  On not being led up the garden path: The use of context by the psychological parser. In D. Dowty, L. Karttunnen, & A. Zwicky (Eds.), Natural Language Parsing.  Cambridge, UK: Cambridge University Press.

Sedivy, J. (2001).  Evidence for Gricean mechanisms in on-line language processing.  Paper presented at the 14th annual CUNY Conference on Human Sentence Processing. Philadelphia, PA.

Sedivy, J., Tanenhaus, M., Chambers, C., & Carlson, G. (1999).  Achieving incremental semantic interpretation through contextual representation.  Cognition, 71, 109-147.

Yee, E., & Sedivy, J. (2001).  Using eye movements to study the time-course of semantic priming in lexical access.  Paper presented at the 14th annual CUNY Conference on Human Sentence Processing. Philadelphia, PA.