Analyzing discourse topics and topic keywords

Discourse topic is an intractable and inherently subjective notion making analysis problematic. This paper overcomes some of the problems by treating topic as a fuzzy concept and views discourse topics as sets of topic keywords. The study examines the identification of topic boundaries and topic keywords by informants and by four methods of analyzing topics — topical structure analysis, given-new progression, lexical analysis, and topic-based analysis. Comparing the findings from these four methods against those from the informants, it was found that given-new progression is the most valid method for identifying topic boundaries, and topic-based analysis is the most valid for identifying topic keywords. There are also notable differences in the types of keywords identified and the bases for identifying keywords between the methods and the informants.