Integrating MT at Swiss Post’s Language Service: preliminary results

Integrating MT at Swiss Post’s Language Service: preliminary results Pierrette Bouillon, Sabrina Girletti FTI/TIM, University of Geneva Boulevard du Pont-d’Arve 40 1211 Geneva, Switzerland Pierrette.Bouillon@unige.ch Sabrina.Girletti@unige.ch Martina Bellodi, Beatrice Bircher Swiss Post Ltd Wankdorfallee 4 3030 Bern, Switzerland martina.bellodi@post.ch beatrice.bircher@post.ch Paula Estrella, Jonathan Mutal FaMaF, University of C´ordoba Av. Medina Allende s/n X5000HUA C´ordoba, Argentina pestrella@famaf.unc.edu.ar jdm0113@famaf.unc.edu.ar Abstract This paper presents the preliminary results of an ongoing academia-industry collabo- ration that aims to integrate MT into the workflow of Swiss Post’s Language Ser- vice. We describe the evaluations car- ried out to select an MT tool (commercial or open-source) and assess the suitability of machine translation for post-editing in Swiss Post’s various subject areas and lan- guage pairs. The goal of this first phase is to provide recommendations with regard to the tool, language pair and most suitable domain for implementing MT. 1 Introduction Nowadays, the production environments of many companies incorporate MT for various reasons: it might be upon request of a client, an initiative to add new services to a company’s assets or an at- tempt to cut costs and shorten delivery times. The technology can be developed by a third party or in- house, each solution having its own pros and cons. Swiss Post’s Language Service would like to integrate MT in their workflow in different con- texts, ranging from gisting to professional post- editing, thereby allowing for reduced turnaround times. Hence, in collaboration with the University c⃝ 2018 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. of Geneva and one of its partners, the University of C´ordoba, a preliminary study was carried out to 1) select an MT engine (open source or commer- cial) and 2) determine the language pairs and sub- ject areas for which MT would be most suitable. In particular, we focused on assessing the potential suitability of MT sentences for professional post- editing. The source data used to train and test the dif- ferent systems for the various language pairs are almost parallel, making it possible to compare re- sults across less-studied pairs. In addition, when designing our experimental setting, we chose to put the focus on users, namely Swiss Post’s pro- fessional translators, providing them with specific training before involving them in the evaluation process. We are convinced that when reorganiz- ing the traditional workflow of professional trans- lators, it is important to give them an active role in the change in order to foster acceptance and avoid biased evaluation due to reluctant MT users. The paper is structured as follows: we first de- scribe the available data for the various languages and subject areas (Section 2), then explain how we selected the MT engine (Section 3). We then present how the suitability of the MT for PE was assessed by Swiss Post’s in-house translators (Sec- tion 4) and discuss the results (Section 4.4), before concluding (Section 5). P´erez-Ortiz, S´anchez-Mart´ınez, Espl`a-Gomis, Popovi´c, Rico, Martins, Van den Bogaert, Forcada (eds.) Proceedings of …

Content truncated to 500 words