Using Supertags as Source Language Context in SMT

Proceedings of the 13th Annual Conference of the EAMT, pages 234–241, Barcelona, May 2009 Using Supertags as Source Language Context in SMT Rejwanul Haque†, Sudip Kumar Naskar†, Yanjun Ma* and Andy Way†* CNGL†/NCLT* School of Computing Dublin City University Dublin 9, Ireland {rhaque, snaskar, yma, away}@computing.dcu.ie Abstract Recent research has shown that Phrase- Based Statistical Machine Translation (PB-SMT) systems can benefit from two enhancements: (i) using words and POS tags as context-informed features on the source side; and (ii) incorporating lexical syntactic descriptions in the form of su- pertags on the target side. In this work we present a novel PB-SMT model that combines these two aspects by using su- pertags as source language context- informed features. These features enable us to exploit source similarity in addition to target similarity, as modelled by the language model. In our experiments two kinds of supertags are employed: those from Lexicalized Tree-Adjoining Gram- mar and Combinatory Categorial Gram- mar. We use a memory-based classifica- tion framework that enables the estima- tion of these features while avoiding problems of sparseness. Despite the dif- ferences between these two approaches, the supertaggers give similar improve- ments. We evaluate the performance of our approach on an English-to-Chinese translation task using a state-of-the-art phrase-based SMT system, and report an improvement of 7.88% BLEU score in translation quality when adding supertags as context-informed features. 1 Introduction In log-linear phrase-based SMT, the probability P(eI 1|fJ 1) of a target phrase eI 1 given a source phrase fJ 1 is modelled as a log-linear combination of features which normally consist of a finite set © 2009 European Association for Machine Translation. of translational features, and a language model (Och and Ney, 2002). The usual translational features involved in those models express de- pendencies between the source and target phrases, but not dependencies between the phrases in the source language themselves. Stroppa et al. (2007) were the first to show that incorporating source language context using neighbouring words and part-of-speech tags had the potential to improve translation quality. In a separate strand of research, Hassan et al. (2006, 2007, 2008) showed that incorporating lexical syntactic descriptions in the form of su- pertags in the target language model and on the target side of the translation model could im- prove significantly on state-of-the-art approaches to MT. Despite the significance of this work, it is currently not possible to develop a fully su- pertagged PB-SMT system given that supertag- gers exist only for English. In this paper, we begin to explore whether such a system could indeed generate improve- ments across all PB-SMT system components. Our novel approach combines the methods of (Stroppa et al., 2007) and (Hassan et al., 2006, 2007, 2008; Hassan, 2009) in one model. We extend a standard PB-SMT system with syntactic descriptions on the source side. Crucially, the kind of lexical descriptions that we employ are those that are commonly devised within lexicon- driven approaches to linguistic syntax, namely Lexicalized Tree-Adjoining Grammar (LTAG: Joshi and Schabes, 1992; Bangalore and …

Content truncated to 500 words