Rule-based Breton to French machine translation

Rule-based Breton to French machine translation Francis M. Tyers Departament de Llenguatges i Sistemes Inform`atics Universitat d’Alacant E-03070 Alacant ftyers@dlsi.ua.es Abstract This paper describes a rule-based machine translation system from Breton to French intended for producing gisting translations. The paper presents a summary of the ongo- ing development of the system, along with an evaluation of two versions, and some reflection on the use of MT systems for lesser-resourced or minority languages. 1 Introduction This paper describes the development of a “gist- ing” machine translation system between Breton and French.1 The first section will give a general overview of the two languages in question and de- scribe the aims for the current system. The subse- quent sections will describe some of the develop- ment work, the current status, an evaluation, and some prospects for future development. Breton is a Celtic language of the Brythonic branch which is today largely spoken in Brittany in the north-west of France. Historically it was spo- ken to different degrees throughout Brittany, but has been losing territory to French since the 12th century, most rapidly in the last 100 years. The language is classed as a language in “serious dan- ger of extinction” by the UNESCO Red Book on Endangered Languages (Salminen, 1999). Although both Breton and French both belong to the large Indo-European language group, they are from widely different families, Celtic and Ro- mance respectively, with a number of substantial differences between them. c⃝ 2010 European Association for Machine Translation. 1The system can be tried out online at http: //www.ofis-bzh.org/bzh/ressources_ linguistiques/index-troerofis.php Speakers of minority or regional languages typ- ically differ from the majority in being bilingual, speaking both their language, and the language of the majority. In contrast, a majority language speaker does not usually speak the minority or re- gional language. This has some implications for the requirements society will put on machine trans- lation systems. Applications of machine translation system can be divided in two main groups: assimilation, that is, to enable a user to understand what the text is about; and dissemination, that is, to help in the task of translating a text to be published. The require- ments of either group of applications is different. Assimilation may be possible even when the text is far from being grammatically correct; how- ever, for dissemination, the effort needed to correct (post-edit) the text must not be so high that it is preferable to translate it manually from scratch. A majority to minority language system will mainly be used for dissemination purposes; it must therefore be such that post-editing the output is faster than translating from scratch. Intelligibil- ity is secondary, and only important if it helps the post-editor. A minority to majority language system will, however be mainly used for assimilation, for in- stance, to answer vital questions such as “what are they writing about me in the minority language newspaper?”. Therefore, the main goal is intelli- gibility. The system in this paper was indeed developed with this second objective in …

Content truncated to 500 words