Two-Level Clausal Segmentation System Based on Semantic Information for Automatic Summarization

This paper suggests a method that breaks a long sentence into simple sentences using not only syntactical information but also a sense of a word. Unlike uniformed sentences, common sentences are often ambiguous and frequently omit particles thus they should be divided into simple sentences at semantic level. This study shows the method that analyzes the dependency structure between predicate and complement using the information on sense from a subcategorization dictionary and thesaurus. Then it proposes a Two-Level Clausal Segmentation System (TLCSS) that includes other elements in the dependency structure gradually based on the syntactic information and grammatical knowledge.