Bilingual Spoken Language Comprehension: A Computational and Empirical Study of Lexical- Semantic Processing in Bilinguals.

摘要

During speech comprehension, bilinguals co-activate both of their languages, resulting in cross-linguistic competition at multiple levels of processing. This dissertation explores the nature of interactivity in the bilingual language system through both computational modeling and empirical research. In Study 1, a computational model of bilingual language processing, the Bilingual Language Interaction Network for Comprehension of Speech (BLINCS), is presented, which posits a high degree of interactivity within and across languages in bilinguals. BLINCS successfully captures natural processes of bilingual spoken language comprehension, such as within- and between-language lexical competition, and can make novel predictions about language processing, using a dynamic and highly flexible computational architecture. Study 2 explored the extent of bilingual language interaction by investigating whether a bilingual's two languages are co-activated even when the input maps to only a single language. English-Spanish bilinguals performing an English-only task heard the word "duck" and were found to look more at an image of a shovel, because the Spanish translation for "duck" (pato) is phonologically similar to the Spanish translation for "shovel" (pala), a process known as covert co-activation. The presence of covert co-activation provides compelling evidence of a highly interactive language system, and raises the question of how covert co-activation occurs in bilinguals. Study 3 investigates the underlying processes of covert competition by comparing two possible explanatory accounts: direct connections between translation equivalents and feedback from semantic representations. English-Spanish bilinguals and English monolinguals completed a visual discrimination task while listening to task-irrelevant spoken words. The spoken words were related to a distractor image by overt phonology, semantics, or covert phonology. Results indicate that semantic access (and subsequent feedback) is unlikely to exclusively account for covert co-activation and suggest that direct connections between translation equivalents play an important role in cross-linguistic interaction in bilinguals. The computational and empirical investigations presented in this dissertation have important consequences for the structure of the language system, and for the mechanisms by which humans process spoken language.