On Quasi-Modularity Of Language Processing

The dispute between serial processing and parallel processing has been perennial. Over the last quarter of century, serial model has been influential, but recently, the popularity has been given to parallel models with the emergence of connectionism and its related theory, PDP (parallel distributed processing). Each model has its unique strength as well as weakness in its interpretability. Serial processing takes the computer as its metaphor. Parallel processing is neurally inspired because it takes the brain as the metaphor. Serial processing has modularity as its cornerstone, while parallel processing has connectionism as its theoretical foundation. This paper examines the biological foundations of language and language processing by focusing mainly on speech perception and production because of speech's superiority over written language. It points out the importance of memory in language processing, the components of language processing, and possible information processing schemas. In addition it provides a detailed overview of various serial models in comparison to parallel models from different dimensions of language processing, for example, from the angle of lexical access, and from that of speech production. The researches concerning predictions projected by these two models are reviewed to show respectively each model's interpretability.The issue of what part each model plays in language processing and whether one is preferred over the other looms large in Psycholinguistics. No theoretical framework has been proposed to compromise these two models abroad. Quasi-modularity of language processing has been proposed by Chen (2005) in China to accommodate both serial and parallel processing. Through examining each model's theoretical foundation respectively, this paper takes the view that there exist both serial and parallel processing in the language module, and that each model has its unique strength and weakness.A comprehensive comparison is given of serial vs. parallel model from the perspective of lexical access by means of data collecting and theoretical analysis. It is shown that Quasi-Modularity is deficient in these aspects: (1) Quasi- Modularity is unidirectional. (2) Quasi-Modularity can't account for non-word accessing. (3) Quasi-Modularity cannot account for some speech errors. (4) Quasi-Modularity is fuzzy. Lastly, language processing paradigm should contain other strategies such as bottom-up and top-down, and serial and parallel processing. Viewed from another angle, there may be controlled vs. autonomous processes. Language processing paradigm should exhibit the properties of fastness, high rate of correctness, context-sensitivity and negotiability.Further research lies in the direction of ascertaining how these two models interact to achieve the linguistic performance. Research work done in the domain of the computer science, especially the work on parallel computers, might shed light on the ancient issue of language processing. As the cognitive science and other related scientific fields develop, the mystery of language processing will be unraveled.