A common architecture for expressing linguistic theories: With illustrations from Chinese languages, cognitive grammar, and software engineering.

摘要

This dissertation presents a common architecture for expressing linguistic theories (CAFE-LT), defines a CAFE-LT modeled cognitive grammar (CMCG), and implements corpus analysis software from that basis. Taken as a whole this study addresses three primary concerns facing the practicing linguist. Those concerns may be stated together as the desire to develop a means of comparing different linguistic theories, modeling polyglot language users, and simplifying linguistic software development tasks. Modern linguistics, the modern Mandarin, Taiwanese, and English languages, and the established techniques of software engineering are the study's source material.; CAFE-LT is suitable for any linguistic approach that accepts three principles: that observed language consists of linear strings, that linguistic structure can be described with meronomic relationships, and that stored linguistic are further defined in the least restrictive manner; thus structures may be represented as in generative grammar or as finite state models. The set of linguistic features is also variable and may include statistical weights or semantic concepts. CAFE-LT is in short a meta-language providing a unified basis for describing divergent theories, a platform for quantified theory comparison, and a framework from which software models can be rapidly produced.; CMCG is provided as an example formalism expressed in CAFE-LT. Based on cognitive grammar (Langacker 1991, 1987), CMCG restates that field's central themes in an idiom conducive to algorithmic processing. Among the introduced innovations are a unified formalism for idioms and structures, allowing a full realization of an idiomaticity cline (Fillmore 1997, Langacker 1987).; The dissertation proceeds by introducing a base set of guidelines for developing useful modeling frameworks. Those guidelines are then applied to linguistics, resulting in the CAFE-LT system. CMCG is then developed and a CMCG reference base is instantiated for Mandarin determiner phrases. Finally CAFE-LT software, tailored for CMCG-based analysis, is applied to a Mandarin research corpus as a demonstration of CAFE-LT and CMCG's efficacy. The conclusion discusses broader use of the systems. The appendices provide a wealth of structured data describing and resulting from the corpus analysis that are useful on their own basis.