An efficient phrase-to-phrase alignment model for arbitrarily long phrase and large corpora.pdf

EAMT 2005 Conference Proceedings 1 An Efficient Phrase-to-Phrase Alignment Model for Arbitrarily Long Phrase and Large Corpora Ying Zhang Stephan Vogel Language Technologies Institute School of Computer Science Carnegie Mellon University {joy+,vogel+}@ Abstract. Most statistical machine translation (SMT) systems use phrase-to-phrase translations to capture local context information, leading to better lexical choices and more reliable word reordering. Long phrases capture more contexts than short phrases and result in better translation qualities. On the other hand, the increasing amount of bilingual data poses serious problems for storing all possible phrases. In this paper, we describe a novel phrase-to-phrase alignment model which allows for arbitrarily long phrases and works for very large bilingual corpora. This model is very efficient in both time and space and the resulting translations are better than the state-of-the-art systems. 1. Introduction In recent years, various phrase-to-phrase translation models (Och 1999; Marcu Wong 2002; Koehn 2003; Zhang 2003) have shown great advantages over the word-based systems (Brown 1990). We believe that longer phrases encapsulate more contexts of the words and the translation qualities are expected to be higher than that of short phrases. Unfortunately, given the increasing volume of the parallel bilingual data for some major languages such as Arabic and Chinese, storing and loading all possible phrase translations from the training corpus becomes more and more expensive by means of space and time in computation. To keep the phrasal translation model of a reasonable size, some models (Koehn 2003) and (Zhang 2003) limit the length of the phrases to be no more than 3 words while others (Vogel 2003) sub- samples the training corpus based on the testing data to down-scale the problem. In this paper, we introduce a new strategy to cope with this problem. Instead of aligning the phr