Improving Phrase-Based Statistical Machine Translation with Morpho-Syntactic Analysis and Transformation

Improving Phrase-Based SMT with Morpho-Syntactic Analysis and Transformation Thai Phuong Nguyen and Akira Shimazu School of Information Science Japan Advanced Institute of Science and Technology 1-1, Asahidai, Nomi, Ishikawa, 923-1292, Japan {thai,shimazu}@jaist.ac.jp Abstract This paper presents our study of exploit- ing morpho-syntactic information for phrase-based statistical machine transla- tion (SMT). For morphological transfor- mation, we use hand-crafted transformational rules. For syntactic transformation, we propose a transforma- tional model based on Bayes’ formula. The model is trained using a bilingual corpus and a broad coverage parser of the source language. The morphological and syntactic transformations are used in the preprocessing phase of a SMT system. This preprocessing method is applicable to language pairs in which the target lan- guage is poor in resources. We applied the proposed method to translation from Eng- lish to Vietnamese. Our experiments showed a BLEU-score improvement of more than 3.28% in comparison with Pharaoh, a state-of-the-art phrase-based SMT system. 1 Introduction In the field of statistical machine translation (SMT), several phrase-based SMT models (Och et al., 1999; Marcu and Wong, 2002; Koehn et al., 2003) have achieved the state-of-the-art performance. These models have a number of advantages in comparison with the original IBM SMT models (Brown et al., 1993) such as word choice, idio- matic expression recognition, and local restructur- ing. These advantages are the result of moving from words to phrases as the basic unit of transla- tion. Although phrase-based SMT systems have been successful, they have some potential limitations when it comes to modeling word-order differences between languages. The reason is that the phrase- based systems make little or only indirect use of syntactic information. In other words, they are still “non-linguistic”. That is, in phrase-based systems tokens are treated as words, phrases can be any sequence of tokens (and are not necessarily phrases in any syntactic sense), and reordering models are based solely on movement distance (Och and Ney, 2004; Koehn et al., 2003) but not on the phrase content. Another limitation is the sparse data prob- lem, because acquiring bitext is difficult and ex- pensive. Since in phrase-based SMT differently inflected forms of the same word are often treated as different words, the problem is more serious when one or both of the source and target lan- guages is an inflectional language. In this paper, we describe our study for improv- ing SMT by using linguistic analysis to attack these two problems. The word order problem is solved by parsing source sentences and then trans- forming them into the target language structure. After this step, the resulting source sentences are closer in word-order to the target language than the original sentences. We propose a transformational model based on the Bayes formula. The model’s knowledge is learned from bitext in which the source text has been parsed. The sparse data prob- lem is solved by splitting the stem and the inflec- tional suffix of a word during translation. These 138 Proceedings of the 7th Conference of the Association for Machine Translation in the …

Content truncated to 500 words