Modeling language acquisition and representation: connectionist networks

Abstract

Connectionism, Parallel Distributed Processing (PDP), or neural networks have had a profound impact on cognitive sciences in the last two decades. Language, as one of the central human cognitive components, has received in-depth treatments since the beginning of connectionist research. The acquisition of the English past tense (Rumelhart & McClelland, 1986), the recognition of speech (McClelland & Elman, 1986), and the processing of sentences (McClelland & Kawamoto, 1986) are among the earliest domains of connectionist research in the original PDP models. Connectionism has since been applied to the study of many domains involving language, including language acquisition, normal and impaired word reading, and language organization in the brain (e.g. Elman et al., 1996; Plaut et al., 1996; Miikkulainen, 1997; Small et al., 1995). Unfortunately, connectionist models or modeling have had very limited influences on Chinese psycholinguistics as a whole. To date, there are very few connectionist models that are designed specifically to account for the processing or representation of the Chinese language. This lack of interaction between connectionism and Chinese psycholinguistics is lamentable. On the other hand, however, this lack opens new avenues for research. In this chapter, I present research from our laboratories that explores the issue of linguistic representations and acquisition in connectionist networks, with particular reference to Chinese, in both monolingual and bilingual contexts. Connectionism: an overview Connectionist representation and learning A close parallel has been drawn between the human mind and the digital computer in the classical conceptualization of human cognition.