On the Semantic Knowledge Resources for Information Extraction
Yulin Yuan · Zhongwen xinxi xuebao · 2002
This paper discusses the matter with the semantic knowledge resources for information extraction (briefly, IE) via many examplescome from real Chinese texts. It demonstrates that a workable IE system at least needs following three levels of semantic knowledge as supporting resources: (i) the discourse structure knowledge of real text, by which the IE system can expect the type of information template and the distribution of the key information items; (ii) the argument structure knowledge of key sentences in real text, by which the propagation and inheritance relation between argument constituents and event template, and the anaphorical relation between the pronouns or empty categories and their precedents can be determined; (iii)logic structure knowledge,by which the IE system can decide the logic relation between the logic operators(e. g. ,negative word,quantifier and modal word,etc. )and their bound elements,e. g. ,the scope and focus of negative words,and the grammatical condition under which the negative word is redundant. Finally, it suggests briefly the available theories and methods to investigate the aforementioned semantic knowledge.