Exploiting Lexical Conceptual Structure for Paraphrase Generation

Exploiting Lexical Conceptual Structure for Paraphrase Generation Atsushi Fujita1, Kentaro Inui2, and Yuji Matsumoto2 1 Graduate School of Informatics, Kyoto University fujita@pine.kuee.kyoto-u.ac.jp 2 Graduate School of Information Science, Nara Institute of Science and Technology {inui, matsu}@is.naist.jp Abstract. Lexical Conceptual Structure (LCS) represents verbs as semantic structures with a limited number of semantic predicates. This paper attempts to exploit how LCS can be used to explain the regularities underlying lexical and syntactic paraphrases, such as verb alternation, compound word decomposition, and lexical derivation. We propose a paraphrase generation model which trans- forms LCSs of verbs, and then conduct an empirical experiment taking the para- phrasing of Japanese light-verb constructions as an example. Experimental results justify that syntactic and semantic properties of verbs encoded in LCS are useful to semantically constrain the syntactic transformation in paraphrase generation. 1 Introduction Automatic paraphrasing has recently been attracting increasing attention due to its po- tential in a broad range of natural language processing tasks. For example, a system that is capable of simplifying a given text, or showing the user several alternative expres- sions conveying the same content, would be useful for assisting a reader. There are several classes of paraphrase that exhibit a degree of regularity. For exam- ple, paraphrasing associated with verb alternation, lexical derivation, compound word decomposition, and paraphrasing of light-verb constructions (LVC(s)) all fall into such classes. Examples1 (1) and (2) appear to exhibit the same transformation pattern, in which a compound noun is transformed into a verb phrase. Likewise, paraphrases in- volving an LVC as in (3) and (4) (from [4]) have considerable similarities. (1) s. His machine operation is very good. t. He operates the machine very well. (2) s. My son’s bat control is unskillful yet. t. My son controls his bat poorly yet. (3) s. Steven made an attempt to stop playing. t. Steven attempted to stop playing. (4) s. It had a noticeable effect on the trade. t. It noticeably affected the trade. 1 For each example, “s” and “t” denote an original sentence and its paraphrase, respectively. Note that our target language is Japanese. English examples are used for an explanatory purpose. R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 908–919, 2005. c⃝ Springer-Verlag Berlin Heidelberg 2005 Exploiting Lexical Conceptual Structure for Paraphrase Generation 909 However, the regularity we find in these examples is not so simple that it cannot be captured only in syntactic terms. For example, the transformation pattern as in (1) and (2) does not apply to another compound noun “machine translation.” We can also find a range of varieties in paraphrasing of LVCs as we describe in Section 3. In spite of this complexity, the regularity each paraphrase class exhibits were ex- plained by recent advances in lexical semantics, such as the Lexical Conceptual Struc- ture (LCS) [8] and the Generative Lexicon [17]. According to the LCS, for instance, a wide variety of paraphrases including word association within compounds, transitiv- ity alternation, and lexical derivation, were explained by means …

Content truncated to 500 words