Research On Term Automatic Translation Technology Based On NP Tree For English Patent Documentation

Along with the increase of foreign exchanges and introduction of the advanced technology, people have more chances to contact with patent documentation. Those who engage in the technique documentation translation often encounter the translation problems of patent documentation. Term translation is an important part of patent documentation translation. In recent years, great progresses have been made, which provides a strong force for patent documentation translations.Aiming at the characteristic of term noun phrase (NP) in patent documentation, a method based on statistics and rules is proposed to carry out automatical translation of term NPs (noun phrases). A complete NP translation flow is formed by NP tree database establishment, NP translation pattern generalization, search for the most similar NP tree based on HowNet and automatic evaluation of translation results.The work of this thesis mainly includes:First, a method of improved mutual information is proposed to select the best word translation. The selection of words in the translation is not independently carried out. The context is helpful for correctly translating words in the context. On the basis of the improvement of the characteristics of the existing mutual information formula, the best translation can be selected by combining the translation model with mutual information.Second, NP translation patterns are adopted to adjust the word order. According to the characteristic of NPs, some NP translation patterns are generalized as common NP translation patterns after syntactic analysis, which is used as the basis of adjusting translation order of complex NPs.Third, a method based on HowNet of looking for the most similar NP tree is presented. The existing word similarity formula is extened, and a NP semantic similarity calculating method based on HowNet is adopted to search for the most similar NP tree.Fourth, a complete system for term translation is built. Automatic evaluation technology of machine translation is used to evaluate the translation results and carry out comparing experiments with the statistical machine translation system (Pharaoh) based on the terms.The experimental results show that the term translation method based on NP tree is better than the method of statistical machine translation (Pharaoh). For the work of next step, more NP translation patterns and linguistic knowledge can be added to improve translation quality.