Obtaining Word Phrases with Stochastic Inversion Translation Grammars for Phrase-based Statistical Machine Translation

Obtaining Word Phrases with Stochastic Inversion Transduction Grammars for Phrase-based Statistical Machine Translation∗ J.A. S´anchez and J.M. Bened´ı DSIC Universidad Polit´ecnica de Valencia, 46022 Valencia, Spain {jandreu | jbenedi}@dsic.upv.es Abstract Phrase-based statistical translation systems are currently providing excel- lent results in real machine translation tasks. In phrase-based statistical translation systems, the basic translation units are word phrases. An im- portant problem that is related to the estimation of phrase-based statistical models is the obtaining of word phrases from an aligned bilingual training corpus. In this work, we propose obtaining word phrases by means of a Stochastic Inversion Transduction Grammar. Preliminary experiments have been carried out on real tasks and promising results have been obtained. 1 Introduction Machine Translation is a problem that can be addressed by means of statistical tech- niques (Brown, Pietra, Pietra, & Mercer, 1993). In this approach, the process of hu- man language translation is modeled sta- tistically by means of statistical translation models. In order to estimate these statistical translation models, several approaches have been proposed in the literature: finite-state techniques (Bangalore & Riccardi, 2001; Casacuberta & Vidal, 2004); alignment techniques (Brown et al., 1990, 1993; Zens, Och, & Ney, 2002; Vogel et al., 2003; Koehn, 2004; Och & Ney, 2004); and syntax-based techniques (Wu, 1997; Yamada & Knight, 2001). Phrase-based techniques are based on the alignment of word phrases (Marcu & Wong, 2002; Zens et al., 2002; Vogel et al., 2003; Koehn, 2004; Tom´as, Lloret, & Casacuberta, 2005). Phrase-based statisti- cal translation systems are currently provid- ing excellent results in real machine transla- tion tasks. In phrase-based statistical trans- lation systems, the basic translation units ∗This work has been partially supported by the Universidad Polit´ecnica de Valencia with the ILETA project. are word phrases. An important problem that is related to phrase-based statistical translation is to au- tomatically obtain bilingual word phrases from parallel corpora. Several methods have been defined for dealing with this problem (Och & Ney, 2003). In this work, we study a method to obtain word phrases that is based on Stochastic Inversion Transduction Gram- mars that was proposed in (Wu, 1997). Stochastic Inversion Transduction Gram- mars (SITG) can be viewed as a re- stricted Stochastic Context-Free Syntax- Directed Transduction Scheme (Aho & Ull- man, 1972; Maryanski & Thomason, 1979; Casacuberta, 1995). SITGs can be used to carry out a simultaneous parsing of both the input string and the output string. In this work, we propose to apply this idea to obtain aligned word phrases to be used in phrase-based translation systems. Some works along this idea have been proposed elsewhere (Zhang & Gildea, 2005). In Section 2, we review the phrase-based machine translation approach. SITGs are reviewed in Section 3. In Section 4, we present preliminary experiments with two real tasks. 2 Phrase-based Statistical Machine Translation The translation units in a phrase-based statistical translation system are bilingual phrases rather than simple paired words. Several systems that follow this approach have been presented in recent works (Zens et al., 2002; Koehn, 2004; …

Content truncated to 500 words