Generation of Summaries that Appropriately and Adequately Express the Contents of Original Documents Using Word-Association Knowledge

Generation of Summaries that Appropriately and Adequately Express the Contents of Original Documents Using Word-Association Knowledge Kazuki Takigawa a, Masaki Murata b, Masaaki Tsuchida c, Stijn De Saeger c, Kazuhide Yamamoto a, and Kentaro Torisawa c aDeparment of Electrical Engineering,Nagaoka University of Technology 1603-1, Kamitomioka, Nagaoka City, Niigata, 940-2188, Japan {takigawa,yamamoto}@jnlp.org bGraduate School of Engineering, Tottori University 4-101, Koyama-Minami, Tottori, 680-8550, Japan murata@ike.tottori-u.ac.jp cNational Institute of Information and Communications Technology 3-5, Hikaridai, Seika-cho, Soraku-gun, Kyoto, 619-0289, Japan {m-tsuchida,stijn,torisawa}@nict.go.jp Abstract. In this study, our purpose was to make a short summary for sentences. For ex- ample, we aimed to make a short summary “terror” for sentences “A bomb went off. Some people were killed. This was triggered by rebel campaign.” In this study, we proposed a new method that generates summaries that can appropriately and adequately express the contents of their respective original documents using word-association knowledge. In this method, we assumed that a good summary comprises words that can express the contents of the original document and does not contain words that are unable to express the contents of the original document. Using statistical tests, we confirmed that the use of elements in our method was beneficial. Our method obtained 0.75 as the ratio where the top 10 summaries for each docu- ment include a correct summary and 0.45 as the mean reciprocal rank (MRR) in the “lenient” case of experiments. Keywords: Summarization, Word-Association Knowledge, Precision, Recall, Generation 1 Introduction Summarization is one of the important techniques in natural language processing more so because of the development of internet-based technologies and the existence of many documents and many kinds of information on the Web (Hovy and Mareu, 1988; Mani and Maybury, 1999; Barzilay and Elhada, 1997; Goldstein et al., 1999; Marcu, 2000; Ker and Chen, 2000; Hongyan, 2000; Radev et al., 2001; Barzilay and Lee, 2004; Radev et al., 2004; Kato et al., 2005). In this study, our purpose was to make a short summary for sentences. For example, we aimed to make a short summary “terror” for sentences “A bomb went off. Some people were killed. This was triggered by rebel campaign.” In this paper, we have proposed a new method that generates summaries that appropriately and adequately express the contents of the original documents using word-association knowledge. In our proposed method, a system judges that a candidate summary that satisfies the following criteria as much as possible is a suitable summary: (i) The contents of the original document are conveyed by the candidate. (ii) The content that is not described in the original document is not conveyed by the candidate. In this study, we use co-occurring words as word-association knowledge. By using co- occurring words, a summary can be generated containing words that do not appear in the original document (summarization by paraphrasing). In this aspect, our method is completely different Copyright 2010 by Kazuki Takigawa, Masaki Murata, Masaaki Tsuchida, Stijn De Saeger, Kazuhide Yamamoto, and Kentaro Torisawa PACLIC 24 Proceedings 693 from the existing methods that extract …

Content truncated to 500 words