Jennifer E. Arnold, Maria Fagnano & Michael K. Tanenhaus
University of Rochester
Spontaneous speech is rarely fluent, resulting in hesitations, fillers ("um" / "uh"), repeated or repaired words, or pronouncing "the" as /thiy/ (Fox Tree & Clark, 1997). Yet these features are generally considered to not affect the core processes of language comprehension. While disfluencies have been argued to signal that the speaker is having difficulty (Clark & Wasow, 1998; Fox Tree & Clark, 1997), this metalinguistic knowledge has not been linked to specific language comprehension phenomena.
A corpus analysis showed that speakers are disfluent more often when referring to entities that are new (rather than given) in the discourse. If listeners are sensitive to this correlation, disfluencies at the start of a noun phrase should lead them to focus on objects that are visible but have not yet been mentioned.
Eye movements of 24 native speakers of English were recorded as they listened to pairs of instructions to move objects on a computer screen (Table 1). Each display contained 4 colored pictures (Rossion & Purtois, 2001), including two cohorts (e.g., camel/candle). We investigated the time course of referent identification for the first noun in the second instruction, manipulating whether: 1) the critical NP was fluent (the camel) or disfluent (thiy, uh, camel), and 2) the referent was discourse-new, or was given but unfocused, having just been mentioned as the goal of the first instruction. All NPs were accented.
Disfluent NPs should lead to faster target looks in the new condition, and increased cohort competition in the given condition. By contrast, fluent, accented NPs provide an initial bias toward the given but nonfocused object (Dahan et al., in press), so we expected fluent NPs to lead to faster target looks in the given condition and more cohort competition in the new condition. Results showed precisely this interaction, beginning 200 msec after the onset of the head noun ("ca-"). Prior to the noun, there was also a preference for new objects in the disfluent condition and given objects in the fluent condition, emerging 200 msec after the determiner (the/thiy), which provided the first information about fluency.
Thus, comprehenders immediately use information provided by disfluencies. This may stem from use of purely distributional information about disfluencies and discourse status, or may result from inferring that the speaker is having difficulty in lexical retrieval (which would be less likely for a just-mentioned referent). Regardless, information about fluency affects the earliest moments of reference resolution.
Table 1: Sample instructions (target NP is underlined)
Given (Discourse-Old) Context: Put the grapes below the candle.
Discourse-new Context: Put the grapes below the camel.
a. fluent (accented): Now put the candle below the salt shaker.
b. disfluent: Now put thiy, uh, CANDLE below the salt shaker.
Clark, H. H., & Wasow, T. (1998). Repeating Words in Spontaneous Speech. Cognitive Psychology, 37, 201-242.
Dahan, D., Tanenhaus, M. K. & Chambers, C. G. (in press). Accent and reference resolution in spoken-language comprehension. Journal of Memory and Language.
Fox Tree, J. E., & Clark, H. H. (1997). Pronouncing "the" as "thee" to signal problems in speaking. Cognition, 62, 151-167.
Rossion, B., & Pourtois, G. (2001). Revisiting Snodgrass and Vanderwart's object database: Color and texture improve object recognition. Paper presented at the 1st Vision Science Conference, Sarasota, Florida, May, 2001.