Files
Abstract
This dissertation develops a cognition-driven computational approach to investigate variation and historical change of basic-level semantic categories. While prior research on language change has started to incorporate cognitive factors influencing large-scale language dynamics, the two primary approaches—distributional and dynamical models—have been pursued predominantly independently, leading to different research focal points and frequently incompatible predictions. This dissertation proposes a cohesive approach by creating both a distributional and a dynamical model to examine how semantic cognition and semantic evolution may reciprocally interact, in particular, by shaping semantic knowledge in the domains of color, kinship, smell, and shape. In the first computational chapter (Chapter II), I develop a two-dimensional embedding model that maps semantic typicality and contextuality of basic-level categories in hyperbolic coordinates, named the "typicality-contextuality" model (Model I). This model is predicated on the conditional entropy of word co-occurrence weighted by collocation diversity and conditional probability ranking. Model I is empirically evaluated in a domain-specific manner. The model examines the interplay between semantic cognition and relatively long-term (hundreds to thousands of years) semantic shifts in literary ancient Chinese in the domains of color, smell, and shape-based classifiers (Chapter III).
In the second computational chapter (Chapter IV), I present the "asynchronicity-singularity" model (Model II), a two-dimensional metacognitive dynamical-systems framework designed to investigate the potential interaction between metacognitive feedback on semantic knowledge and diachronic shifts of semantic categories. This model is applied to diachronic change data from Google Ngram in modern written American English, exploring the interplay between semantic cognition and relatively shorter-term (decades to hundreds of years) semantic changes in the realms of color, kinship, and smell (Chapter V). In Chapter VI, the two models are further extended to include other semantic or linguistic domains such as plural suffixes, pragmatic knowledge of place-naming, and metacognitive awareness of semantic categories within political attitudes, religions, and professions.
Together, this dissertation underscores the importance of linking cognitive mechanisms of semantic categorization with large-scale socio-historical language variations and shifts, and of evaluating such linkages with domain-oriented focuses. More broadly, this work contributes to the understanding of how language cognition and language change intersect with cognition, culture, and society, a key research topic in many branches of contemporary cognitive, computational, and social sciences.