Patent documentation – comparison of two MT strategies

This paper focuses on two matters: A comparison of how two different MT strategies manage translating the text type of patent documentation and a survey of what is needed to transform a MT research prototype system to a translation application for patent texts. The two MT strategies is represented by PaTrans a transfer and rule based system being used for more than 15 years by the translation agency Lingtech A/S and SpaTrans a SMT system based on the Pharaoh framework. The SMT systems are characterised by shorter development time and low development cost compared to rulebased systems. The distinctive text type of patents pose special demands for machine translation and these aspects are discussed based on linguistic observations with focus on the users point of view. Two main demands are automatic pre processing of the documents and implementation of a module which in a flexible and user-friendly manner offers the opportunity to extend the lexical coverage of the system. These demands and the comparison of the two MT strategies are discussed on the basis of proofread patents.