The Computing of Discourse Focus

This dissertation is a comprehensive and in-depth study of the notion of focus with two of its salient characteristics being the interdisciplinary approach adopted and the range of query covered. The research is largely motivated by the realization that focus is a notion figuring crucially in both core linguistic studies such as phonetics/phonology, syntax, semantics, and discourse analysis, and peripheral linguistic areas such as psycholinguistics, cognitive linguistics and in particular computational linguistics.

The major contributions of the study can be summarized as follows. Firstly after a critical review of the various previous definitional accounts, it is concluded that the notion of focus is best defined as a cognito-pragmatic one and accordingly the constructs of knowledge store and discourse model are postulated before a working definition of focus is proposed. Secondly, on the basis of this definition, the process of focus determination in discourse is captured in the form of a formal algorithm, namely, Focus Determination Algorithm (FDA), which, being programmable and operationable, can be integrated into the discourse modeling system. This formalization is enabled by the constructs of knowledge store and discourse model introduced before and substantiated by research findings from social psychology and neuropsychology. Thirdly, the linguistic, in particular the prosodic, means of realizing focus is discussed and for this purpose data from a range of languages are cited. Similar to the determination of focus, the process of how accent is assigned on the basis of focus is also captured in a semi-formalized manner by virtue of a procedural algorithm, namely, an Accent Determination Procedure (ADP).

As just mentioned, the research is marked by its interdisciplinary perspective; it incorporates impetuses, insights and inputs from core linguistics (specifically phonology and discourse analysis), computational linguistics, social psychology, and neuropsychology. What is particularly noteworthy is its orientation towards and potential significance for natural language processing, especially the generation of natural speech by enabling the integration of discourse-level information such as focus into the prosody generation model.