Alternatives to the combinatorial paradigm of linguistic theory based on domain general principles of human cognition

Abstract

It is argued that the principles needed to explain linguistic behavior are domain-general and based on the impact that specific experiences have on the mental organization and representation of language. This organization must be sensitive to both specific information and generalized patterns. In addition, knowledge of language is highly sensitive to frequency of use: frequently-used linguistic sequences become more frequent, more accessible and better integrated. The evidence adduced is mainly from phonology and morphology and addresses the issue of gradience and specificity found in postulated units, categories, and dichotomies such as regular and irregular, but the points apply to all levels of linguistic analysis including the syntactic, semantic, and discourse levels. Appropriate models for representing such phenomena are considered, including exemplar models and connectionist models, which are evolving to achieve a better fit with linguistic data. The major criticism of connectionist models often raised from within the combinatorial paradigm of much existing linguistic theory – that they do not capture ‘free combination’ to the extent that rule-based systems do, is regarded as a strength rather than a weakness. Recent connectionist models exhibit greater productivity and systematicity than earlier variants, but still show less uniformity of generalization than combinatorial models do. The remaining non-uniformity that the connectionist models show is appropriate, given that such non-uniformity is the rule in language structure and language behavior.