Translating from Morphologically Complex Languages: A Paraphrase-Based Approach

Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, pages 1298–1307, Portland, Oregon, June 19-24, 2011. c⃝2011 Association for Computational Linguistics Translating from Morphologically Complex Languages: A Paraphrase-Based Approach Preslav Nakov Department of Computer Science National University of Singapore 13 Computing Drive Singapore 117417 nakov@comp.nus.edu.sg Hwee Tou Ng Department of Computer Science National University of Singapore 13 Computing Drive Singapore 117417 nght@comp.nus.edu.sg Abstract We propose a novel approach to translating from a morphologically complex language. Unlike previous research, which has targeted word inflections and concatenations, we fo- cus on the pairwise relationship between mor- phologically related words, which we treat as potential paraphrases and handle using para- phrasing techniques at the word, phrase, and sentence level. An important advantage of this framework is that it can cope with deriva- tional morphology, which has so far remained largely beyond the capabilities of statistical machine translation systems. Our experiments translating from Malay, whose morphology is mostly derivational, into English show signif- icant improvements over rivaling approaches based on five automatic evaluation measures (for 320,000 sentence pairs; 9.5 million En- glish word tokens). 1 Introduction Traditionally, statistical machine translation (SMT) models have assumed that the word should be the ba- sic token-unit of translation, thus ignoring any word- internal morphological structure. This assumption can be traced back to the first word-based models of IBM (Brown et al., 1993), which were initially pro- posed for two languages with limited morphology: French and English. While several significantly im- proved models have been developed since then, in- cluding phrase-based (Koehn et al., 2003), hierarchi- cal (Chiang, 2005), treelet (Quirk et al., 2005), and syntactic (Galley et al., 2004) models, they all pre- served the assumption that words should be atomic. Ignoring morphology was fine as long as the main research interest remained focused on languages with limited (e.g., English, French, Spanish) or min- imal (e.g., Chinese) morphology. Since the attention shifted to languages like Arabic, however, the im- portance of morphology became obvious and sev- eral approaches to handle it have been proposed. Depending on the particular language of interest, researchers have paid attention to word inflections and clitics, e.g., for Arabic, Finnish, and Turkish, or to noun compounds, e.g., for German. However, derivational morphology has not been specifically targeted so far. In this paper, we propose a paraphrase-based ap- proach to translating from a morphologically com- plex language. Unlike previous research, we focus on the pairwise relationship between morphologi- cally related wordforms, which we treat as poten- tial paraphrases, and which we handle using para- phrasing techniques at various levels: word, phrase, and sentence level. An important advantage of this framework is that it can cope with various kinds of morphological wordforms, including derivational ones. We demonstrate its potential on Malay, whose morphology is mostly derivational. The remainder of the paper is organized as fol- lows: Section 2 gives an overview of Malay mor- phology, Section 3 introduces our paraphrase-based approach to translating from morphologically com- plex languages, Section 4 describes our …

Content truncated to 500 words