Personal Translator at WMT2011

Proceedings of the 6th Workshop on Statistical Machine Translation, pages 303–308, Edinburgh, Scotland, UK, July 30–31, 2011. c⃝2011 Association for Computational Linguistics Personal Translator at WMT2011 - A rule-based MTsystem with hybrid components - Vera Aleksić Gregor Thurmair Linguatec Gmbh Linguatec Gmbh Gottfried-Keller-Str. 12 Gottfried-Keller-Str. 12 Munich, Germany Munich, Germany v.aleksic@linguatec.de g.thurmair@linguatec.de Abstract This paper presents the Linguatec submission to the WMT 2011 sixth workshop on statistical machine translation. It describes the architecture of our machine translation system ‘Personal Translator’ (hereinafter also referred to as PT), developed by Linguatec, which is a rule-based translation system, enriched by statistical approaches. We participate for the German-English translation direction. For the current submission we have chosen the latest commercial version of the system, PT14. The translation quality improvement for the submission was done mainly by lexicon tuning: detection of unknown words, extracting of possible translations, partly from the wmt11 training corpora, and enlarging the lexicon by manually coding the chosen transfer candidates. 1 Introduction The origin of the PT technology dates back to the 80’s when a translation system based on logic programming and slot grammars was developed by Michael McCord at IBM T.J. Watson Research Center. In many years of development the translation engine has been driven forward and enhanced. Most recently we have added statistical approaches for tasks such as erroneous input correction, subject area recognition and word disambiguation. Today ‘Personal Translator’ is one of the leading programs in the translation technology field. It is a commercial MT system whose product range includes 7 language pairs, i.e. 14 translation directions, for single users and networks. Linguatec is a leading provider of language-technology software for office use in Germany. In addition to machine translation, we develop and provide commercial products in the fields of speech recognition and speech synthesis. Linguatec is the only company to have won the European Information Technology Prize three times. 2 System fundamentals Personal Translator is implemented as a modular system which basically consists of the following components: • the grammar, written in Prolog, based on the concept of slot grammar • the lexicon, administrated in the data base internally called TransLexis • additional morphological analysers written partly in C and C++ • hybrid (rule-based and statistical) methods for word disambiguation, subject area recognition and spell-checking • a range of pre- and post-processing components such as format converters for 303 html, pdf, doc, txt and rtf formats, sentence splitter, tokeniser, lemmatizer. As Personal Translator is a commercial system, aiming at providing a complete translator work bench and creating added value for users, it integrates a wide range of advanced features such as: • Translation memory system for management, creation, analysis and maintenance of TMs, as well as large system modules, containing tens of thousands of sentence pairs • Translation project management tool, enabling the user to save and administer all important translation settings and project relevant options • Text to speech functionality to support editing and learning processes such as text revision/correction in the language(s) mastered by the …

Content truncated to 500 words