Inclusive and Sustainable Machine Translation
Machine translation (MT), the task of automatically translating text in one language, into text in another language using a computer system, has undergone many shifts since its inception in the late 1950s. The latest of which, neural machine translation (NMT), has reached unprecedented translation qualities at almost human-level performance, of course for some use-cases and under certain conditions.
-
dr. Dimitar Shterionov
Principal Investigator
MT has become an indispensable tool for professional translators (to assist in the translation workflow), for commercial users, e.g., e-commerce companies (to make their content quickly available in multiple languages), to every-day users (to access information unrestricted by the language in which it is produced).
User- and use-case-centric MT
We focus on the specific user requirements as well as the domain, the language, the style, etc. of a specific use-case. Through smart data analysis, selection and processing and optimized models we investigate faster and better tools that are tuned towards users and use-cases.
Inclusive Language Technology
Current MT and NLP models exacerbate bias and may produce inaccurate or sometimes even offensive outputs. We explore bias-related language phenomena and develop techniques to mitigate bias is an important research direction we are undertaking.
Sustainable and Environmentally Conscious AI
We study the environmental impact of language technology and ways to cut down the use of compute resources without losing quality. Can we reduce, reuse and reorganize for a less intrusive technology?
Project
SignON Is a 3-year H2020 project that aims to facilitate the exchange of information between deaf, hard of hearing and hearing individuals across Europe, by conducting state of the art research and developing a mobile solution for automatic translation between sign and spoken (written and spoken) languages.
It is an unprecedented project, not only because an effective communication service between the target groups has not been previously developed, but also because of the user-centric and open research and innovation approach which revolves around a co-creation workflow.
Partners
- European Union of the Deaf
- Aglatech14
- Microsoft Ireland
- FINCONS Group
- MAC
- Instituut voor de Nederlandse Taal
- VRT
- Vlaams Gebarentaal Centrum
- Netherlands Gebarentaal Centrum
- De Nederlandse TaalUnie
- ORLAN
- Acapela-group
- Dublin City University
- The ADAPT Centre
- Trinity College Dublin
- University College Dublin
- University of the Basque Country
- Pompeu Fabra University
- Technical University Dublin
- Radboud University
- KU Leuven
- University of Naples “L'Orientale”
- University of Turku
- University of Surrey
- COMTEC