Structured and Unstructured Cache Models for SMT Domain Adaptation

Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 155–163, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics Structured and Unstructured Cache Models for SMT Domain Adaptation Annie Louis School of Informatics University of Edinburgh 10 Crichton Street Edinburgh EH8 9AB alouis@inf.ed.ac.uk Bonnie Webber School of Informatics University of Edinburgh 10 Crichton Street Edinburgh EH8 9AB bonnie@inf.ed.ac.uk Abstract We present a French to English transla- tion system for Wikipedia biography ar- ticles. We use training data from out- of-domain corpora and adapt the system for biographies. We propose two forms of domain adaptation. The first biases the system towards words likely in biogra- phies and encourages repetition of words across the document. Since biographies in Wikipedia follow a regular structure, our second model exploits this structure as a sequence of topic segments, where each segment discusses a narrower subtopic of the biography domain. In this structured model, the system is encouraged to use words likely in the current segment’s topic rather than in biographies as a whole. We implement both systems using cache- based translation techniques. We show that a system trained on Europarl and news can be adapted for biographies with 0.5 BLEU score improvement using our mod- els. Further the structure-aware model out- performs the system which treats the entire document as a single segment. 1 Introduction This paper explores domain adaptation of statisti- cal machine translation (SMT) systems to contexts where the target documents have predictable reg- ularity in topic and document structure. Regular- ities can take the form of high rates of word rep- etition across documents, similarities in sentence syntax, similar subtopics and discourse organiza- tion. Domain adaptation for such documents can exploit these similarities. In this paper we focus on topic (lexical) regularities in a domain. We present a system that translates Wikipedia biogra- phies from French to English by adapting a system trained on Europarl and news commentaries. This task is interesting for the following two reasons. Many techniques for SMT domain adaption have focused on rather diverse domains such as us- ing systems trained on Europarl or news to trans- late medical articles (Tiedemann, 2010a), blogs (Su et al., 2012) and transcribed lectures (Federico et al., 2012). The main challenge for such systems is translating out-of-vocabulary words (Carpuat et al., 2012). In contrast, words in biographies are closer to a training corpus of news commentaries and parlimentary proceedings and allow us to ex- amine how well domain adaptation techniques can disambiguate lexical choices. Such an analysis is harder to do on very divergent domains. In addition, biographies have a fairly regu- lar discourse structure: a central entity (person who is the topic of the biography), recurring subtopics such as ‘childhood’, ‘schooling’, ‘ca- reer’ and ‘later life’, and a likely chronological order to these topics. These regularities become more predictable in documents from sources such as Wikipedia. This setting allows us to explore the utility of models which make translation decisions depending on the …

Content truncated to 500 words