Self-Adapting Machine Translation with Multi-Approach Language Technology
The rising demand for machine translation (MT) and the resulting implementations have shown that workflows and system architectures can become very complex - both scientifically and practically. Particularly, the setup of combination or hybrid systems making use of different MT engines running in parallel has kept many language service providers from introducing MT technology. To address this issue, the taraXÜ consortium started this highly ambitious R&D project in 2010. The project has by now resulted in the first advanced machine translation environment that deploys amongst other thing
- Corpus and translation management (running different rule or template based and statistical machine translation engines in parallel)
- Automated source- and target-language quality assessment
- Automatic selection of the best translation through innovative machine learning methods
- A web-interface allowing human translators to correct machine translation output and assess its quality by ranking, post-editing, error-analysis, etc.
The taraXÜ workflow is schematically shown below:
A key ingredient of taraXÜ is a realistic evaluation scenario that takes into account real-world needs of the industry partners rather than being based on relatively arbitrary machine accessible criteria. It will serve as a much more reliable basis for scientific progress than the currently available and widely used procedures.
The TaraXÜ consortium consists of leading partners of Berlin-based research and industry.
Project financed by TSB Technologiestiftung Berlin – Zukunftsfonds Berlin, co-financed by the European Union – European fund for regional development.
