The Effect of Speaker-Specific Information in On-Line Pragmatic Inferencing

Sentences frequently convey more information than they explicitly contain. For instance, suppose a speaker uses a noun phrase (NP) modified by an adjective, as in “the tall cup.” In this situation, perceivers infer the existence of two sets of entities in the discourse: (1) a target set corresponding to the literal denotation of the expression (e.g., a tall cup), and (2) a contrast set containing an object matching the noun, but differing by virtue of the property expressed by the adjective (e.g., a short cup). Evidence for the latter inference comes from monitoring perceiver eye-movements as they listen to spoken instructions. In particular, in the presence of a contrasting object, individuals identify target objects faster and make fewer spurious looks to competitors that share the adjectival property (e.g, a tall pitcher) (Sedivy, et al, 1999). These effects can be observed within two hundred ms of the onset of the head noun. Existing evidence suggests that contrastive inferences arise from an understanding between conversational participants that speakers are only as informative as they need to be (Grice, 1975). A simple NP (e.g., “the cup”) would suffice to pick out the intended referent in a context with only a single entity. When a speaker uses a more elaborate form, the perceiver infers that a different state of affairs prevails. The adjective is most easily made informative by attributing to it a distinguishing function. In support of this view, contrastive inferences only arise for adjective types that are not used to label objects in isolation. Scalar and material adjectives are rarely used to label objects in isolation, whereas color adjectives are frequently included in such descriptions. Correspondingly, scalar and material adjectives are interpreted contrastively, but color adjectives are not (Sedivy, 2001). This effect is driven by discourse-level expectations and not the semantic class of the adjective. When color is a highly predictable property of an object (e.g., a yellow banana), color modifiers are rarely used in default labels. In this case, the use of a modifier is seen as overly informative and the modifier is interpreted contrastively.. (Sedivy, in press). Because default descriptions are statically linked with individual referents, the inferences explored in the above studies do not require that perceivers consider the circumstances of the immediate discourse. Perceivers might reflexively infer a contrast upon deviation from a stored default form. In theory, many different factors might be weighed in deciding whether a modified form should be interpreted contrastively. These include the intrinsic properties of a referent, the linguistic context, the reliability of the speaker, the intentions of the speaker, and so on. It is unlikely that all of these are considered in the limited time frame that contextual contrast effects have been observed.