Humboldt-University of Berlin led a team of 10 academic and industrial partners in TraMOOC for the development of high-quality, modern Machine Translation for Big Data in the diverse domain of Massive Open Online Courses (MOOCs) for 11 language pairs, from English to German, Italian, Portuguese, Dutch, Bulgarian, Greek, Polish, Czech, Croatian, Russian and Chinese.
Dublin City University was responsible for the ongoing explicit evaluation of the MT systems developed within TraMOOC for 11 language pairs. This was based on a combination of multi-faceted human evaluation techniques and a range of well-established automatic metrics, including a comparison of statistical and neural MT engines, to track the improvement of the state-of-the-art TraMOOC MT systems customized to the educational domain.
Ionian University collected the infrastructure required for developing the English-Greek machine translation (MT) solutions. Furthermore, they implemented and monitored crowdsourcing tasks for the development of resources that contributed to quality MT for the 11 target language pairs.
Deluxe Media provided linguistic expertise and quality assurance for all the crowdsourcing tasks performed during the course of this project while their focus was mainly devoted on machine translation evaluation activities.
Tilburg University has focused on collecting English to Dutch data used to train machine translation systems. This data was gathered through crowdsourcing processes. Several experiments have been conducted to investigate which are the best settings of the crowdsourcing platform for this task.
University of Edinburgh created high-quality machine translation systems for 11 language pairs, from English to German, Italian, Portuguese, Dutch, Bulgarian, Greek, Polish, Czech, Croatian, Russian and Chinese. The translation systems are based on cutting-edge neural translation technology and adapted to the MOOC domain.
Radboud University developed a method that detects and compares topical information elements in source and target documents. They studied several methods such as unsupervised topic modeling, sentiment analysis, wikification and multilingual word sense disambiguation and entity linking. The latter method was implemented as an open source tool and web service for implicit translation evaluation.
EASN TIS was responsible for the project’s dissemination and exploitation activities acting as a dissemination multiplier, spreading project related information as well as simultaneously identifying further users for the project foreground ensuring the diffusion of the knowledge incubated within the project to a wide audience from all sectors.
Knowledge 4 All was responsible for the exploitation and marketing of the results of the TraMOOC project in commercial settings. Furthermore, they established an industrial showcase to illustrate potentials of the TraMOOC project from an industrial perspective while they worked by parallel research projects and standardisation bodies were analysed to identify on-the fly synergies.