You are here

Can we trust machine translated texts? Learning how to estimate the quality of a translated sentence in a computer-assisted translation environment

Event date: 
Wednesday, 11 November, 2015 - 09:30 to 10:30
Location: 
Room 102 (North Building)
Speaker: 
Marco Turchi (FBK)

Tenure Track Evaluation Seminar

Abstract: 

Even if not perfect, machine translation (MT) is now getting reliable enough to support and speed-up human translation. Thanks to this progress, the work of professional translators is gradually shifting from full translation from scratch to MT post-editing, where translated texts are manually revised in a computer-assisted translation (CAT) environment. In this framework, human post-editors started to play a central role and the development of tools capable to support them with on-the-fly and reliable MT quality estimates is one of the emerging trends in the translation community.  This has required a revisitation of previous approaches and a significant paradigm shift to novel and robust techniques geared towards the real users' needs.


In this talk, I will present the work done at FBK on the development of QE systems able to deal with continuos streams of real industrial data and to adapt to different users' behaviours, text domains MT engines and customer needs. Experiments with professional translators interacting with our solutions in their daily life have shown the capability of QE to increase their productivity and enhance their translation experience. Our results opened at the same time interesting research paths towards a better exploitation of users' feedback, and promising solutions for translation service providers aiming to increase their revenues.