We believe that MT can only be truly successful if developers do not work in isolation, but in cooperation with translators. Are there any such cooperative development models (designers/developers AND translators working as a team) and – if so – how are translators remunerated for their work in such a context?

  • Published: 3 ans ago on 13 janvier 2018


  1. Aljoscha Burchardt dit :

    We at DFKI have been cooperating with translators and LSPs in several MT projects, where LSPs were either project partners or subcontractors. Together with them, we have developed pricing models for post-editing, quality ranking, etc. back in 2011 when all this was quite avant-garde. The paper I have mentioned in the post on quality “Towards a Systematic and Human-Informed Paradigm for High-Quality Machine Translation” (https://www.dfki.de/lt/publication_show.php?id=8367) shows several ways in which translators can provide informed feedback during the development of translation engines. For me personally, this cooperation is the only way of designing useful tools that assist people in their daily business like other tools (Translation Memories, Word Processors, etc.). Still, many researchers are reluctant as they believe that this brings is too much of « human factor » into the mathematically « clean » development lab. What they do not see is that their normal approach of using reference translations (see my post on quality) or A-B tests « suffers » even well from the natural fact that language is not precise. There are usually several more or less correct ways of translating one idea and there is a certain vagueness even if one asks different human experts. Don’t get me wrong, I am a Computational Linguist and I very much like it. This is makes language and communication so interesting, efficient, and magical.