Dr. Bob McMurray
Research on language tends to simplify the problem by focusing either on the fast processes responsible for real-time comprehension and production, or the slow processes that underlie acquisition. However, in doing so, we may be miss novel solutions to old problems. This talk explicitly focuses on the interaction of processes across short and long to assess a range of problems in word recognition and word learning. I start by examining the classic problem of mapping acoustic cues onto categories. A series of eye-tracking studies with adults demonstrate that when this problem is framed in terms of online, temporal processes, lexical processes are highly sensitive to fine-grained detail that should be lost early in processing. This begs the question of where this gradiency comes from. Computational work suggests that it derives basic statistical learning processes that underlie phonological development, but only when these mechanisms are buttressed with online competition. This is supported by work with language impaired adolescents suggesting that their perceptual deficits may arise from a deficit in competition. Statistical learning and competition also offer a solution to the problem of cue-integration. Here I describe behavioral studies and computation work showing that lexical processes may be critical for integrating asynchronous cues over time, but that amount of weight to given each cue is a product of developmental history. All of this suggests that speech perception is fundamentally molded by lexical processes, and work from my lab (among others) demonstrates that even by 14 months children may have difficulty employing phonological contrast in word learning situations. Thus, I have begun working on word learning. I will present a computational account of early word learning that demonstrates that slow statistical cooccurrence learning combined with fast competition processes can learn word/object mappings in extremely cluttered environments, and account for children’s online recognition of familiar words, as well as their ability to find the referent of novel words. This led to the realization that many of the specialized processes that we typically attribute children’s rapid learning to, may be merely the product of online competition and inhibition processes layered on top of slow learning. If this is the case, what accounts for the acceleration in word learning seen during the second year of life (e.g. the vocabulary explosion)? Thus, I end with a recent series of analyses showing that acceleration is mathematically guaranteed as long as word learning proceeds in parallel, and words vary in difficulty.
Thus, the interaction of simple mechanisms at two time-scales may shed new light on old problems like the problem of invariance in speech, impairments like SLI, and Quine’s problem and the vocabulary explosion in word learning.
| 
