TMX Markup: A Challenge When Adapting SMT to the Localisation Environment

TMX Markup: A Challenge When Adapting SMT to the Localisation Environment Jinhua Du1, Johann Roturier2 and Andy Way1 1. CNGL, School of Computing, Dublin City University, Dublin, Ireland 2. Symantec, Ballycoolin Business Park, Blanchardstown, Dublin 15, Ireland jdu@computing.dcu.ie, johann roturier@symantec.com away@computing.dcu.ie Abstract Translation memory (TM) plays an impor- tant role in localisation workflows and is used as an efficient and fundamental tool to carry out translation. In recent years, statistical machine translation (SMT) tech- niques have been rapidly developed, and the translation quality and speed have been significantly improved as well. However, when applying SMT technique to facilitate post-editing in the localisation industry, we need to adapt SMT to the TM data which is formatted with special mark-up. In this paper, we explore some issues when adapt- ing SMT to Symantec formatted TM data. Three different methods are proposed to handle the Translation Memory eXchange (TMX) markup and a comparative study is carried out between them. Furthermore, we also compare the TMX-based SMT systems with a customised SYSTRAN sys- tem through human evaluation and auto- matic evaluation metrics. The experimen- tal results conducted on the French and En- glish language pair show that the SMT can perform well using TMX as input format either during training or at runtime. 1 Introduction Translation memory (TM) plays an important role in the localisation industry (Planas and Furuse, 2000; Garcia, 2005). TM is an effective way to en- able the translation of segments (sentences, para- graphs, or phrases) of documents by searching for similar segments in a database and retrieving the suggested matches with a fuzzy match score. c⃝ 2010 European Association for Machine Translation. In the localisation process, TM systems are a fundamental tool in automatic translation work- flows. We argue that there are three main reasons: 1) a TM is easy to build; 2) it complies with in- dustry standards and so is efficient to store, share and re-use; 3) localisation is generally limited to a specific domain using a TM provides a fast and rel- atively good translation for specific domains. Even though SMT has been significantly developed and the translation quality has been highly improved in recent years from the academic viewpoint (Brown et al., 1993; Yamada and Knight, 2001; Koehn et al., 2003; Och and Ney, 2003; Chiang, 2005; Koehn et al., 2007), a lot of work must be done to produce high-quality translations when it is ap- plied into industry application. As a result, SMT has a long way to go to establish itself as an ac- ceptable component in computer-assisted transla- tion (CAT) environments such as a commercial post-editing environment, at least when compared to TM systems or rule-based MT systems. Much research progress has been achieved by combining TM with rule-based or SMT tech- niques. Roturier (2009) used a hybrid approach combining a customised SYSTRAN system with TM to refine the MT output in Symantec’s lo- calisation workflow. DeCamp (2009) and Simard and Isabelle (2009) proposed a way of integrat- ing a phrase-based SMT (PB-SMT) …

Content truncated to 500 words