On integrating hybrid and rule-based components for patent MT with several levels of output

We present a methodology integrating hybrid and rule-based components for speeding up the development of a patent MT system. The methodology is suitable for highly inflecting languages and described on the example of translating patent claims from Russian into English. Based on different combinations of hybrid and rule-based components the system performs shallow or/and deep parsing and provides for several complementary levels of output, (i) translation of terminology, that only involves shallow MT procedures, and (ii) full translation that is based on both shallow and deep parsing integrated either automatically, or in an interactive environment. Full translation of the patent claim is output in two formats, a legal one sentence format and a better readable set of simple sentences. To control the quality of claim translation by better understanding the input, the system also outputs a SL claim decomposed into simple sentences.