Words in minds and machines : a computational characterization of Chinese mental lexicon

Abstract

Distributional semantic models induced from text corpora play an increasingly central role in characterizing human semantic cognition. Within this general trend, the last decades have witnessed widespread interests in applying these models to conceptualize psycholinguistic properties and to estimate normative ratings on various semantic dimensions. In this thesis, I took advantage of distributional semantic models to address the central question of how we mentally represent lexical semantic information. With a special focus on Chinese, I investigated a number of research topics including (i) compound compositionality, (ii) lexical ambiguity, and (iii) the extrapolation of semantic rating norms (i.e., valence, arousal, and concreteness) that could reflect individual differences. On the theoretical side, findings in these studies have unraveled the mental operations to access compound meaning, as well as the organizational principles of Chinese characters within semantic memory. On the methodological side, the proposed computational models are theoretically sound and empirically supported, which substantially extend lexical resources for Chinese psycholinguistic study. Altogether, the four studies that make up the thesis have highlighted the potential of using distributional semantic models to characterize Chinese mental lexicon. Predictions made by these models can open new research avenues in the domain of conceptual representation and semantic processing.