Syntax Augmented Machine Translation via Chart Parsing

Proceedings of the Workshop on Statistical Machine Translation, pages 138–141, New York City, June 2006. c⃝2006 Association for Computational Linguistics Syntax Augmented Machine Translation via Chart Parsing Andreas Zollmann and Ashish Venugopal School of Computer Science Carnegie Mellon University {zollmann,ashishv}@cs.cmu.edu Abstract We present translation results on the shared task ”Exploiting Parallel Texts for Statistical Machine Translation” gener- ated by a chart parsing decoder operating on phrase tables augmented and general- ized with target language syntactic cate- gories. We use a target language parser to generate parse trees for each sentence on the target side of the bilingual train- ing corpus, matching them with phrase table lattices built for the corresponding source sentence. Considering phrases that correspond to syntactic categories in the parse trees we develop techniques to aug- ment (declare a syntactically motivated category for a phrase pair) and general- ize (form mixed terminal and nonterminal phrases) the phrase table into a synchro- nous bilingual grammar. We present re- sults on the French-to-English task for this workshop, representing significant im- provements over the workshop’s baseline system. Our translation system is avail- able open-source under the GNU General Public License. 1 Introduction Recent work in machine translation has evolved from the traditional word (Brown et al., 1993) and phrase based (Koehn et al., 2003a) models to in- clude hierarchical phrase models (Chiang, 2005) and bilingual synchronous grammars (Melamed, 2004). These advances are motivated by the desire to in- tegrate richer knowledge sources within the transla- tion process with the explicit goal of producing more fluent translations in the target language. The hi- erarchical translation operations introduced in these methods call for extensions to the traditional beam decoder (Koehn et al., 2003a). In this work we introduce techniques to generate syntactically mo- tivated generalized phrases and discuss issues in chart parser based decoding in the statistical ma- chine translation environment. (Chiang, 2005) generates synchronous context- free grammar (SynCFG) rules from an existing phrase translation table. These rules can be viewed as phrase pairs with mixed lexical and non-terminal entries, where non-terminal entries (occurring as pairs in the source and target side) represent place- holders for inserting additional phrases pairs (which again may contain nonterminals) at decoding time. While (Chiang, 2005) uses only two nonterminal symbols in his grammar, we introduce multiple syn- tactic categories, taking advantage of a target lan- guage parser for this information. While (Yamada and Knight, 2002) represent syntactical information in the decoding process through a series of transfor- mation operations, we operate directly at the phrase level. In addition to the benefits that come from a more structured hierarchical rule set, we believe that these restrictions serve as a syntax driven lan- guage model that can guide the decoding process, as n-gram context based language models do in tra- ditional decoding. In the following sections, we describe our phrase annotation and generalization process followed by the design and pruning deci- sions in our chart parser. We give results on the French-English Europarl data and conclude with prospects for …

Content truncated to 500 words