Auto-extracting Paraphrases of Letter-word Phrases in Live Texts

Auto-extracting Paraphrases of Letter-word Phrases in Live Texts1 Zezhi ZHENG Dep. of Chinese Language & Literature, Xiamen University, Xiamen 361005 Zhengzz@xmu.edu.cn Abstract: In this paper we will discuss the Auto-extraction of paraphrases of letter-word phrases in live Chinese texts. The paper discusses the modes of conventional dictionaries firstly, and then gives the principles of paraphrase of letter-word phrases; with an analysis of the examples of letter-word phrases paraphrases secondly, and then gives their formalized denotations and presents an auto-recognizing algorithm for bilingual synonymous letter-word phrases; lastly, based on the labeled result of our auto-labeling software of letter-word phrase, uses the vector space distance to extract the paraphrase of letter-word phrases in live Chinese texts. Key words: Phrases; Letter-word phrase; Auto-extract 1 introduction Traditional dictionaries have the veracious meaning expression, and rational, normative language, and better systematic semantic systems. The paraphrases in dictionaries are generalized from the collection of example sentences which are gathered from the fields that seemed most feasible. Therefore compiling a dictionary always cost a great deal manpower, material resources and time. The renewed period of traditional dictionaries has not adapted to the increasing and changing of new words and new meanings, so using computers to help the compile of dictionaries is brought forward. About the extraction for the definition, [zhang Yan, 2003] instructed a definition extraction means for Chinese terms, which based on Chinese syntactic parsing. They segmented and tagged the corpora of computer domain with part-of-speech firstly, then used two parsers to gain structures and phrases of sentences. They summarized the structures characteristics of term definitions and automatically extracted the patterns of definitions. Finally they gave an algorithm to define a new term according to their knowledge database. Their algorithm to define a term needs concept semantic knowledge and other knowledge, and segmented and tagged texts. In 2004, XU Yong et al presented an experiment of web based term definition retrieval system. For a given term, the system used the algorithm based on term definition patterns with indicated words to extract its definition. The two means of definition extraction are all for given terms, and have available to some extent. We think that definitions are for terms, the paraphrases are suitable for letter-word phrases (hereinafter we will use “LWP” to denote “letter-word phrase”) in live Chinese texts in general field. Therefore in this paper we’ll discuss the principles of paraphrases of LWPs firstly, and then generalize their formalized patterns based on the collection of paraphrases of LWPs. Secondly, we’ll analyse the especial paraphrases, bilingual synonymous LWPs, and present an algorithm to obtain this kind paraphrases. Thirdly, we’ll label the LWPs using our auto-labeling software, and then put forward an algorithm to extract the paraphrases of LWPs. 2 The principles of our paraphrase of letter-word phrases LWPs, especially those prevailed in general media, many of them are terms or proper noun, and most readers have no acquaintance with them, so they want to know the meaning of them. With their 1 Sustentation fund: science fund (code: 60475022), …

Content truncated to 500 words