A Parallel Architecture perspective on language processing.

This article sketches the Parallel Architecture, an approach to the structure of grammar that contrasts with mainstream generative grammar (MGG) in that (a) it treats phonology, syntax, and semantics as independent generative components whose structures are linked by interface rules; (b) it uses a parallel constraint-based formalism that is nondirectional; (c) it treats words and rules alike as pieces of linguistic structure stored in long-term memory. In addition to the theoretical advantages offered by the Parallel Architecture, it lends itself to a direct interpretation in processing terms, in which pieces of structure stored in long-term memory are assembled in working memory, and alternative structures are in competition. The resulting model of processing is compared both with processing models derived from MGG and with lexically driven connectionist architectures.