WHO:	BCS Lunch Talk: Sarah Brown-Schmidt , BCS Graduate student
TOPIC:	Speakers continuously update what they plan to say: Evidence from eye movements during unrestricted conversation
ABSTRACT:	nderstanding how production constrains comprehension, and vice versa, will likely require investigations of interactive conversation, where participants are both speakers and addressees. As a first step, we designed an interactive task in which two naïve participants were seated in front of displays containing an identical set of 14 pictures, separated into two domains that looked like 'islands'. At the beginning of each trial, a picture on one participant's screen was highlighted, cueing that participant to tell her partner to click on that object. Some target objects appeared with a cohort competitor on the same screen (e.g., peach/peas) and some targets appeared with a scalar contrast item, e.g., a large and a small peach. We report data from twenty pairs; all mentioned effects are reliable at p<.05. Speakers frequently used pre-nominal scalar adjectives, e.g., "the large peach" when target objects appeared with a contrast member on the same island; when no contrast was present, scalars were rarely used (72% vs. 8%). When the target and the contrast appeared on the same island, yet the speaker failed to use a pre-nominal scalar, disfluencies increased 20% over cases with a pre-nominal scalar. We found the opposite pattern when a scalar contrast was not present; here, extraneous modification was associated with disfluency. Listeners' interpretation of their partner's referring expressions reflected these modification patterns. When speakers mentioned which island the target was on, e.g. "On the top, the large peach", addressees interpreted these expressions with respect to the referential domain indicated by the locative construction and the scalar contrast member, e.g. rarely looking at a size-matched cohort competitor, 'the big peas' when the top island included the big peas, the big peach and the small peach (but the small peas was on the bottom island). Most strikingly, the speaker's eye movements to the contrast member predicted the form of the referring expression when a contrast was present. When size was never mentioned, speakers rarely looked at the contrast member (e.g. the small peach), a markedly different pattern than for NPs with modification, where speakers typically did look at it. For trials with looks to the contrast member, the timing of the speaker's first look to the contrast member predicted the form of the NP. When using pre-nominal modification, speakers first looked to the contrast object approximately 1900ms before the NP onset. For post-nominal repairs, e.g., "The peach…oh…BIG one" speakers first looked at the contrast 1600ms later, just before uttering the NP. Thus, new visual information, encountered during production led to a repair-- a process that requires continuous communication between message formulation and utterance generation, perhaps via monitoring. More generally, our results illustrate how the interplay between eye movements and production can provide insights about the planning process during production.
WHEN:	3/3/2004 12:00:00 PM
WHERE:	Meliora 269

Events Homepage