Enhancing statistical machine translation with bilingual terminology in a CAT environment

Enhancing Statistical Machine Translation with Bilingual Terminology in a CAT Environment Mihael Arcan mihael.arcan@insight-centre.com Insight Centre for Data Analytics, National University of Ireland, Galway, Ireland Marco Turchi turchi@fbk.eu Sara Tonelli satonelli@fbk.eu FBK - Fondazione Bruno Kessler, Via Sommarive 18, 38123 Trento, Italy Paul Buitelaar paul.buitelaar@insight-centre.com Insight Centre for Data Analytics, National University of Ireland, Galway, Ireland Abstract In this paper, we address the problem of extracting and integrating bilingual terminology into a Statistical Machine Translation (SMT) system for a Computer Aided Translation (CAT) tool scenario. We develop a framework that, taking as input a small amount of parallel in-domain data, gathers domain-specific bilingual terms and injects them in an SMT system to enhance the translation productivity. Therefore, we investigate several strategies to extract and align bilingual terminology, and to embed it into the SMT. We compare two embedding methods that can be easily used at run-time without altering the normal activity of an SMT system: XML markup and the cache-based model. We tested our framework on two different domains showing improvements up to 15% BLEU score points. 1 Introduction Recent studies (Federico et al., 2012; L¨aubli et al., 2013; Green et al., 2013) have shown sig- nificant productivity gains when human translators post-edit machine translation output rather than translating documents from scratch. This evidence has raised interest in the integration of machine translation systems within CAT software. In this context, an important open issue is how to support translators with domain-specific information when dealing with highly spe- cific texts, i.e. manuals coming from different domains (information technology (IT), legal, agriculture, etc.). Translation tools such as Google Translate,1 Bing Translator2 or open source SMT systems such as Moses (Koehn et al., 2007) trained on generic data are the most common solutions, but they often result in unsatisfactory translations. A valuable alternative to support professional translators is represented by online terminology resources, e.g. IATE,3 which are continuously updated and can be easily queried. However, the manual use of these services can be very time demanding when working with a CAT tool. For these reasons, the automatic identification and integration of bilingual domain-specific terms into an SMT system is a crucial step towards increasing translation quality of high-specific texts in a CAT environment. This also reduces translators’ initial overload when dealing with different domains, because termi- nological lists are managed directly by the SMT system and no additional human intervention for retrieving domain-specific terminology is required. 1 http://translate.google.com/ 2 http://www.bing.com/translator 3 Inter-Active Terminology for Europe, http://iate.europa.eu/ Al-Onaizan & Simard (Eds.) Proceedings of AMTA 2014, vol. 1: MT Researchers Vancouver, BC © The Authors 54 In this paper, we propose a framework for extracting bilingual terms from parallel data and using them to enhance the performance of an SMT system embedded in a CAT tool. We focus on a real scenario, where a large translation project is split across different translators and each translator post-edits daily a limited amount of sentences provided by the SMT system. Our approach takes advantage of such post-edited data …

Content truncated to 500 words