SMT Applied to the Patent Domain. Perspectives of Hybridisation with GF and Rule-based Translation Paradigms.

Rule-based translation systems are usually adequate for close languages and/or restricted domains. MOLTO aims to widen this scope by hybridising a GF system with a statistical one. The domain of application, patents, can be considered a quasi-open domain, that is, a GF grammar cannot cover the whole language and statistical methods must go for coverage and robustness.

The first part of this talk introduces the Patents Case Study, the nature of the data, and its use within a GF translation system and a SMT system. Some preliminary results for these systems are shown. The second part describes various hybridisation methods that will be applied within the project and also similar approaches that we are carrying out for other language pairs.