This Phrase-Based SMT System is Out of Order: Generalised Word Reordering in Machine Translation

Proceedings of the 2006 Australasian Language Technology Workshop (ALTW2006), pages 149–156. This Phrase-Based SMT System is Out of Order: Generalised Word Reordering in Machine Translation Simon Zwarts Centre for Language Technology Macquarie University Sydney, Australia szwarts@ics.mq.edu.au Mark Dras Centre for Language Technology Macquarie University Sydney, Australia madras@ics.mq.edu.au Abstract Many natural language processes have some degree of preprocessing of data: tokenisation, stemming and so on. In the domain of Statisti- cal Machine Translation it has been shown that word reordering as a preprocessing step can help the translation process. Recently, hand-written rules for reordering in German–English translation have shown good results, but this is clearly a labour-intensive and language pair-specific approach. Two possible sources of the observed improvement are that (1) the reordering explicitly matches the syntax of the source language more closely to that of the target language, or that (2) it fits the data bet- ter to the mechanisms of phrasal SMT; but it is not clear which. In this paper, we apply a gen- eral principle based on dependency distance min- imisation to produce reorderings. Our language- independent approach achieves half of the im- provement of a reimplementation of the hand- crafted approach, and suggests that reason (2) is a possible explanation for why that reordering ap- proach works. Help you I can, yes. Jedi Master Yoda 1 Introduction Preprocessing is an essential step in Natural Lan- guage applications. Reordering of words on a sen- tence level as a more extensive step for prepro- cessing has succeeded in improving results in Sta- tistical Machine Translation (SMT). Here, both in the training and in the decoding phase, sentences in the source language are reordered before being processed. This reordering can be done based on rules over word alignment learnt statistically; (Costa-Juss and Donollosa, 2006), for example, describe such a system. In this work an improvement in over- all translation quality in a Spanish-English MT system was achieved by using statistical word classes and a word-based distortion model to re- order words in the source language. Reordering here is purely a statistical process and no syntacti- cal knowledge of the language is used. Xia and McCord (2004) do use syntactical know- ledge; they use pattern learning in their reorder- ing system. In their work in the training phase they parse and align sentences and derive reorder- ing patterns. From the English-French Canadian Hansard they extract 56,000 different transforma- tions for translation. In the decoding phase they use these transformations on the source language. The main focus then is monotonic decoding (that is, decoding while roughly keeping the same order in the target language as in the source language — reordering done within phrases, for example, is an exception). Syntactically motivated rules are also used in re- ordering models. In Collins et al. (2005) six hand- written rules for reordering source sentences are defined. These rules operate on the output of an automatic parser. The reordering rules however are language-pair (German-English) specific and hand-written. We want to extend this idea of …

Content truncated to 500 words