Morpho-syntactic analysis for reordering in statistical machine translation

Morpho-Syntactic Analysis for Reordering in Statistical Machine Translation Sonja Nießen and Hermann Ney Lehrstuhl f¨ur Informatik VI Computer Science Department RWTH Aachen – University of Technology D-52056 Aachen, Germany Email: niessen@cs.rwth-aachen.de Abstract In the framework of statistical machine translation (SMT), correspondences between the words in the source and the target language are learned from bilingual corpora on the basis of so-called alignment models. Among other things these are meant to capture the differences in word order in different languages. In this paper we show that SMT can take advantage of the explicit introduction of some linguistic knowledge about the sentence structure in the languages under consideration. In contrast to previous publications dealing with the incorporation of morphological and syntactic information into SMT, we focus on two aspects of reordering for the language pair German and English, namely question inversion and detachable German verb prefixes. The results of systematic experiments are reported and demonstrate the applicability of the approach to both translation directions on a German-English corpus. Keywords Statistical Machine Translation, Reordering, Morpho-Syntactic Information, Question Inversion, Seperable Pre- fixes 1 Introduction In this paper, we address the question of how mor- phological and syntactic analysis can help statistical machine translation (SMT). Although there has been a number of publications dealing with morphological and syntactic analysis in general and its application to machine translation in particular, there have only been a few which incorporate information from this anal- ysis in the process of statistical machine translation (Brown et al., 1992; Nießen and Ney, 2000). In our approach, we introduce transformations to both the source and target string. In our experi- ments the considered languages are German and En- glish. Systematic evaluations demonstrate that lin- guistic knowledge can improve translation results. We concentrate on transformations which aim at “harmo- nizing” the word order in corresponding sentences. Our experiences with various tasks and language pairs show that difference in word order is one of the main sources of errors in machine translation, if not the most dominant problem in this field. The presentation fo- cusses on the following aspects: Separated German verb prefixes: Some German verbs consist of a main part and a detachable prefix which can be shifted to the end of the clause. In (Nießen and Ney, 2000), we have suggested to prepend separated prefixes to the main part and reported results for the translation direction from German to English. Here, we extend this method and apply it to the inverse direction from English to German which implies some additional complications. Question Inversion: In many languages, the sen- tence structure of interrogative sentences differs from the structure in declarative sentences. We propose a method of harmonizing the word order in both sentence types. For the language pair En- glish and French, this procedure was suggested by (Brown et al., 1992). In contrast to them, we investigate question inversion for German, which is more complicated in comparison to French be- cause of the less restricted word order. Unlike (Brown et al., 1992), we …

Content truncated to 500 words