Thai Sentence Paraphrasing from the Lexical Resource

Copyright 2012 by Krittaporn Phucharasupa and Ponrudee Netisopakul 26th Pacific Asia Conference on Language,Information and Computation pages 381–390 Thai Sentence Paraphrasing from the Lexical Resource Krittaporn Phucharasupa and Ponrudee Netisopakul Faculty of Information Technology King Mongkut’s Institute of Technology Ladkrabang Bangkok, Thailand phucharasupa@hotmail.com, ponrudee@it.kmitl.ac.th Abstract Paraphrase generation in any language has gained much attention and importance in the study of Natural Language Processing. Therefore, the focus of this paper is on Thai language paraphrase generation for the sentence level. Six sentence paraphrasing techniques for Thai are proposed and illustratively explained. In addition, the Thai–sentence Paraphrase Generation (TPG) system is designed using a lexical resource based system subsequently entitled the Thai Lexical Conceptual Structure with Thai Lexicalized Tree Adjoining Grammar (TLCS–TLTAG) Resource. 1 Introduction For any language, putting the same content in different ways can indicate the richness of the language culture. Since the language is one of the major communication tools in every society, the ability to paraphrase what we want to say or write can also imply the society’s civilization. Paraphrasing techniques for the sentence level and others in several languages have been examined and suggested during the past several years (Stede, 1996; Dras, 1999; Barzilay and Lee, 2003; Pang et al., 2003; Qiu et al., 2006; Ellsworth and Janin, 2007; Zhao et al., 2009; Madnani and Dorr, 2010). These paraphrasing techniques were enormously used in several areas of Natural Language Processing such as Question Answering (Duboue and Carroll, 2006), Machine Translation (Shimohata, 2004; Barreiro, 2008), Summary Evaluation (Zhou et al., 2006) and Textual Entailment Recognition (Marsi et al., 2007; Malakasiotis, 2011). In Thai language, its writing structure contains no space between words and no full stops between sentences. This could be potential problems in doing research pertaining to Thai computational paraphrasing. Nevertheless, the construction and patterns of Thai sentences have been partially investigated by a number of renown Thai linguists (Vongsantivanit, 1983; Kanchanacheeva, 1996; Thonglor, 2007; Songsilp, 2008; Settanyakan, 2011). Some researchers classified Thai verbs, identified their arguments, as well as recognized their corresponding thematic roles (Wongsiri, 1981; Sungkhavon, 1984; Panthumetha, 2010). To be able to work on Thai sentence paraphrasing, previous research regarding constructing and paraphrasing sentences in other languages was essential and therefore surveyed (Shimohata, 2004; Barreiro, 2008; Dorr, 1994; Kozlowski et al., 2003; Fujita, 2005). It was subsequently adjusted by (Phucharasupa and Netisopakul, 2011) to fit Thai language more appropriately. Thai sentence paraphrase patterns were categorized into fourteen groups, some of which will be explained and used as examples in this research. To achieve the goal of automatic paraphrase generation, two critical considerations must be addressed. One is that an appropriate semantic structure of the original sentence must be designed so that it facilitates the automatic system to easily generate paraphrases. The other is that the algorithm must be able to generate syntactically correct paraphrases of the original sentence and these paraphrases must faithfully preserve its original meaning. 381 Copyright 2012 by Krittaporn Phucharasupa and Ponrudee Netisopakul 26 th Pacific Asia Conference on Language, Information …

Content truncated to 500 words