Feb. 17, 11:45, Wei Ji Ma. Optimality and Probabilistic Computation in Visual Categorization

Bruce Tesar and Dr. Kristen Syrett

Linguistics & Cognitive Science, Rutgers

Researching perception and language: Insights from computational and experimental linguistics
A longstanding problem in linguistics is how listeners convert "messy" perceptual input into useful phonological information, which maps to meaningful syntactic, semantic, and/or conceptual information. The interface between perception and language is particularly suited to interdisciplinary work. Our two presenters approach this topic from two different interdisciplinary perspectives, demonstrating the benefit of collaborations between linguists, computer scientists, and psychologists. Bruce Tesar, (computational linguist): When learners "hear" human language, they must learn to relate the phones they hear to the appropriate mental representations of sounds. This requires that the learner simultaneously infer the map between underlying representations and surface phones, and the correct lexical representations of morphemes (in terms of the underlying sound representations). Despite the truly immense size of the space of possibilities the learner must sift through, it is possible to do so efficiently if that immense space is structured in a particular linguistically motivated way. Kristen Syrett (experimental linguist): For decades, linguists have asked whether ambiguous sentences involving quantifiers (e.g., All the men didn't go) are consistently disambiguated by speakers using intonation. Such a role of intonation would not only provide further evidence for an intimate connection between sound and meaning, but would also mean there is evidence detectable on the surface for the language learner to distinguish between the intended interpretations. Occasional attempts to identify a systematic prosodic contour have largely come up empty handed. I will present evidence from a production study that there are acoustic and prosodic features linked to the interpretation of such sentences, but they are variable within and across speakers, can be rather subtle, and do not simply reduce to the difference of a rising or falling sentence-final contour. Even in light of these findings, I will show from a perception study that listeners can successfully recruit these auditory cues to assign the correct interpretation. These findings indicate that linguistic studies involving sentence-level ambiguity-including acquisition studies-should carefully control for features of the discourse context favoring sentence interpretation and the prosody with which the key sentences are delivered.