Applicability Analysis of Corpus-derived Paraphrases toward Example-based Paraphrasing

Applicability Analysis of Corpus-derived Paraphrases toward Example-based Paraphrasing Kiyonori Ohtaket tATR Spoken Language Translation Research Laboratories 2-2-2 Hikaridai, Seika-cho, Soraku-gun, Kyoto 619-0288 Japan kiyonori.ohtake@ atr.co.jp Kazuhide Yamamotott 4:Dep artment of Electrical Engineering, Nagaoka University of Technology 1603-1 Kamitomioka, Nagaoka, Niigata 940-2188 Japan yamamoto@fw.ipsj.or.jp Abstract Two kinds of paraphrases extracted from a bilingual parallel corpus were analyzed. One is from an adjectival predicate sentence to a non-adjectival one. The other is from a passive form to a non-passive form. The ability to extract paraphrases is strongly desired for paraphrasing studies. Although extracting paraphrases from multi-lingual parallel corpora is possible, the type of paraphrases extracted is unknown. We discov- ered what types of examples can be obtained, and what types of paraphrasing will be available for the two kinds of paraphrases. 1 Introduction Paraphrasing is a very important form of processing for natural language processing (NLP). A characteristic property of natural language is that various expressions can exist to express a single concept. So far, ad hoc paraphrasing methods have been applied to many NLP applica- tions, such as machine translation, information retrieval, automatic summarization, and so on. Nowadays, however, many researchers recognize automatic paraphrasing as a very important technique for many areas of NLP, and work in this field has recently become more active. We view paraphrasing as a translation where the source language and the target language are the same. Thus, we can take a similar approach to that of machine translation for paraphrasing. In this study, we use an example-based approach to control paraphrasing. So far, a number of new methods for paraphrase extraction have emerged (e.g., (Lin and Pan- tel, 2001; Yamamoto, 2002; Barzilay and McKeown, 2001; Barzilay and Lee, 2003; Shinyama and Sekine, 2003; Pang et al., 2003)). There are several extracting methods: for example, one extracts paraphrases from a monolingual corpus, and another extracts from a multilingual parallel corpus. However, most of these methods have focused on efficiency or how many para- phrases would be extracted, and there have been few discussions on the types of paraphrases that were extracted. Meanwhile, a rule-based approach seems reasonable in terms of the application of paraphras- ing (Takahashi et al., 2001). Therefore, an investigation of how many paraphrasing examples are valuable, or difficult-to-write paraphrasing rules that cover the extracted examples are im- portant. In this paper, we thus discuss the types of paraphrases that are extracted from a bilingual parallel corpus. We analyze paraphrasing examples extracted from a Japanese-English parallel corpus of travel conversations. The corpus is made for Japanese speakers traveling to English-speaking countries. We can extract paraphrases from a parallel corpus (Barzilay and McKeown, 2001). 380 The basic idea of this method is that one expression can be translated into several expressions, and the translated expressions are paraphrases of each other. In this paper, we introduce a method that extracts paraphrasing examples from a Japanese- English parallel corpus. Then, we analyze the extracted examples and discuss several issues about them. The purpose of this study is to …

Content truncated to 500 words