Jared M. Novick & John C. Trueswell
University of Pennsylvania
We introduce a new lexical priming technique designed for use during uninterrupted spoken language comprehension. Subjects' eye movements were recorded as they heard a male voice giving instructions to move objects on a table. Target sentences contained a prepositional phrase attachment ambiguity (e.g., "Now, I'd like you to feel the frog with the feather"), where "with the feather" could indicate an instrument or a noun phrase modifier. At the onset of the target verb, a prime verb of matched duration was uttered by a female and mixed into the digitized audio-track. Two types of primes were compared: Instrument-biased primes (e.g., "hit") which strongly prefer to take Instrument roles; and Modifier-biased primes (e.g., "move") which rarely take Instrument roles.
Referential scenes were designed to support an instrument interpretation: One-Referent scenes, containing, e.g., a single frog holding a small feather; a horse wearing a bow; a large feather; and an unrelated object. Looks to the potential instrument (the large feather) and hand-actions involving this object (i.e., picking it up and using it to feel the frog) can be taken as on-line and off-line measures respectively of listeners' commitment to an instrument interpretation.
Although subjects reported that they rarely identified prime words, the structural preferences of these primes influenced parsing processes. In particular, Modifier-biased primes, as compared to Instrument-biased primes, reliably reduced the proportion of time listeners spent looking at the potential instrument upon hearing "with the feather" (both p's<.05). In addition, hand-actions involving the potential instrument were less for Modifier-biased primes (34%) than Instrument-biased primes (44%), being marginally significant by subjects only (p=.06). Fine-grained eye movement analyses showed that the effect of prime type emerges reliably 600 ms after the onset of "feather" (both p's<.05). A second experiment was conducted with scenes that supported a modifier analysis, i.e., Two-Referent scenes, replacing the horse with another frog. As expected, these scenes eliminated almost all instrument actions, and showed a small but potentially spurious influence of prime-type on the amount of time listeners spent inspecting the potential instrument.
The findings indicate that auditory verb recognition activates semantic roles that are tied to specific syntactic forms. Priming was obtained despite using instrument objects that were poor instruments for the prime verbs, suggesting that the overlap in verb argument structure between targets and primes was the source of these effects. Thus, primes activated abstract information either about their preferred event structure and/or their preferred syntactic complements. We will discuss some post-hoc analyses of the data which support event structure preferences playing an important role in the priming effects. Studies are being prepared that further disentangle syntactic and semantic contributions by comparing primes with similar syntactic preferences but different role preferences (e.g., Instrument vs. Manner PPs). We hope to present preliminary results of this at the conference. Finally, we will discuss implications for exposure-based models of comprehension, which support an important role for lexical-distribution cues in on-line parsing decisions.