Multi-Engine Machine Translation with an Open-Source SMT Decoder

Proceedings of the Second Workshop on Statistical Machine Translation, pages 193–196, Prague, June 2007. c⃝2007 Association for Computational Linguistics Multi-Engine Machine Translation with an Open-Source Decoder for Statistical Machine Translation Yu Chen1, Andreas Eisele1,2, Christian Federmann2, Eva Hasler3, Michael Jellinghaus1, Silke Theison1 (authors listed in alphabetical order) 1: Saarland University, Saarbr¨ucken, Germany 2: DFKI GmbH, Saarbr¨ucken, Germany 3: University of Cologne, Germany Abstract We describe an architecture that allows to combine statistical machine translation (SMT) with rule-based machine translation (RBMT) in a multi-engine setup. We use a variant of standard SMT technology to align translations from one or more RBMT sys- tems with the source text. We incorporate phrases extracted from these alignments into the phrase table of the SMT system and use the open-source decoder Moses to find good combinations of phrases from SMT training data with the phrases derived from RBMT. First experiments based on this hybrid archi- tecture achieve promising results. 1 Introduction Recent work on statistical machine translation has led to significant progress in coverage and quality of translation technology, but so far, most of this work focuses on translation into English, where relatively simple morphological structure and abundance of monolingual training data helped to compensate for the relative lack of linguistic sophistication of the underlying models. As SMT systems are trained on massive amounts of data, they are typically quite good at capturing implicit knowledge contained in co-occurrence statistics, which can serve as a shal- low replacement for the world knowledge that would be required for the resolution of ambiguities and the insertion of information that happens to be missing in the source text but is required to generate well- formed text in the target language. Already before, decades of work went into the im- plementation of MT systems (typically rule-based) for frequently used language pairs1, and these sys- tems quite often contain a wealth of linguistic knowledge about the languages involved, such as fairly complete mechanisms for morphological and syntactic analysis and generation, as well as a large number of bilingual lexical entries spanning many application domains. It is an interesting challenge to combine the differ- ent types of knowledge into integrated systems that could then exploit both explicit linguistic knowledge contained in the rules of one or several conventional MT system(s) and implicit knowledge that can be extracted from large amounts of text. The recently started EuroMatrix2 project will ex- plore this integration of rule-based and statistical knowledge sources, and one of the approaches to be investigated is the combination of existing rule- based MT systems into a multi-engine architecture. The work described in this paper is one of the first incarnations of such a multi-engine architec- ture within the project, and a careful analysis of the results will guide us in the choice of further steps within the project. 2 Architectures for multi-engine MT Combinations of MT systems into multi-engine ar- chitectures have a long tradition, starting perhaps with (Frederking and Nirenburg, 1994). Multi- engine systems can be roughly divided into simple …

Content truncated to 500 words