Translation practice in the workplace and Machine Translation

Translation practice in the workplace and Machine Translation Nikiforos Karamanis Saturnino Luz Department of Computer Science Trinity College Dublin Ireland firstName.lastName@cs.tcd.ie Gavin Doherty Abstract Localisation has long been regarded as an appropriate domain for the deployment of MT. This paper reports the results of a qualitative study which investigated the ac- tivities of professional translators working in a Language Services Provider. We anal- yse the observed work practices and high- light issues related to translation efficiency, quality assessment and teamwork. These issues may impact on the successful de- ployment of MT in a localisation setting but have not been considered extensively by research assessing the appropriateness of MT for localisation as yet. 1 Introduction Localisation has long been regarded as an ap- propriate domain for the deployment of MT. The “Best Practice Guide” issued by the Localiza- tion Industry Standards Association (Dillinger and Lommel, 2004) includes four case studies of suc- cessful uses of MT in localisation settings. How- ever, progress has been slow and other reports by localisation experts are still referring to the role of MT as “emerging” (Esselink, 2003; Wittner and Goldschmidt, 2007; Yanishevsky, 2009). Additional insights have been shared with the MT community by providers of commercial MT services (Groves, 2008) or large organisations (Ro- turier, 2009) which deploy MT to pre-translate content that is then passed on to professional trans- lators to be post-edited. Although these reports in- clude some high-level feedback from the transla- tors, they do not provide a description of the ac- tivities that translators engage in on a day-to-day c⃝ 2010 European Association for Machine Translation. basis. Moreover, these accounts mostly express the perspective of other stakeholders such as the client or the MT developer who assume different roles within the localisation workflow. This per- spective is reflected on attempts to “educate trans- lators about MT” (Groves, 2008, p.16). Efforts to deploy MT in localisation are mostly motivated by the expectation that it can help in- crease throughput. That the output of the MT en- gine is of substantial quality is seen as a prereq- uisite to achieving this aim. As a result, a lot of emphasis is placed on assessing MT quality, mainly in terms of fluency and adequacy [as e.g. in Turian et al. (2003)], a practice also adopted in some competitive evaluations of MT engines. Al- though such judgements of fluency and adequacy are sometimes provided by professional transla- tors, the task that the judges are asked to perform and the conditions under which they perform it for the purposes of such evaluation exercises have lit- tle resemblance with their actual working environ- ment and practice. In industry settings, a reported best practice is to assess MT quality on the basis of judgements provided by members of the organisation which deploys the MT engine rather than by the pro- fessional translators who post-edit the MT output. Roturier (2009), for example, reports that linguists who own the MT resources judge whether a ma- chine translated sentence can be understood …

Content truncated to 500 words