The Research Of Classification Of New Branch Topic Sentences

This paper studied the way of topic change in the view of natural language processing, distinguished two kinds of hierarchical topic-advance base on Chen’s theory of hierarchical topic-advance, and defined the concept of New Branch Topic Sentence. The main works are as follows.First, we labeled the corpus of the novel Fortress Besieged, including220thousand Chinese characters. we use linefeed and indentation to show the Topic-Comment Structure based on the stack model and its extension model Stream Model. We distinguished different types of the Stream Model, such as Topic Postposition Model, New Branch Topic Model and Confluence Model, etc. By discovering the linguistic phenomena that do not conform to the Stream Model, then analyzing and classifying those special linguistic phenomena, we can make the coverage area of the Stream Model as wide as possible.Second, we extracted a total of537sentences which labeled New Branch Topic Sentences from a total of22450punctuation sentences in the whole corpus. We classified different cases that caused the New Branch Topic Sentences, and summarized them into four basic causes:the constraint of sentence-initial, the constraint of special sentence patterns, the elicitation by objects of verbs, and the elicitation by the objects of prepositions.Third, we anatomize the verbs that are able to elicit New Branch Topics in the view of lexical semantics of verbs, and summarize them into seven semantic types of the verbs, such as relationship, attraction, recognition, existential, creation, command, dispose and so on.The classification of New Branch Topic is valuable for advanced research of Generalized Topic Structure. The analysis of New Branch Topic can help people to explore the inherent laws of Chinese topic structure, and provide reference for the research of Chinese grammar and Chinese teaching, as well as language information processing.