Project | MMPE

In order to ensure professional human quality level translation results, in many cases, the output of Machine Translation (MT) systems has to be manually post-edited by human experts. The postediting process is carried out within a post-editing (PE) environment, a user-interface, which supports the capture and correction of mistakes, as well as the selection, manipulation, adaptation and recombination of good segments. PE is a complex and challenging task involving considerable cognitive load. To date, PE environments mostly rely on traditional graphical user interfaces (GUIs), involving a computer screen as display and keyboard and mouse as input devices. In this research project we propose the design, development, implementation and extensive road-testing and evaluation of a novel multi-modal post-editing support for machine translation for translation professionals, which extends traditional input techniques of a PE system, such as keyboard and mouse, with novel free-hand and screen gestures, as well as speech and gaze input modalities (and their combinations). The objectives of the research are to increase the usability and the user experience of post-editing Machine Translation and to reduce the overall cognitive load of the translation task, supporting (i) the core post-editing tasks as well as (ii) controlling the PE system and environment. The multimodal PE environments will be integrated with quality prediction (QE) to automatically guide search for useful segments and mistakes, as well as automatic PE via incremental adaptation of MT to PEs to avoid repeat mistakes, in order to achieve the above mentioned objectives. The environments will be road-tested with human translation professionals and trainees and (where possible) within the partner projects in the Paketantrag (Riezler, Frazer, Ney and Waibel) and (where possible) the post-edited data captured will feed into dynamic and incremental MT retraining and update approaches pursued in the partner projects.