Study Of Translation Selection Of Polyseme Base On Semantic Language

The problem of the translation selection of polyseme is a basic problem in the study of natural language processing. Affected by collocation relations and context, it is difficult for this problem to be processed with computer. To properly process this problem, we attempt to classify translation selection of polyseme, and propose different approaches to cope with each class. The theory of semantic language and multi-language machine translation method based on SL were proposed by academician Gao Qingshi in 2003. This theory simplifies the translation steps in machine translation, and reduces the number of translation system among N languages to N sets. What's more important is that it provides a way to precisely represent the phenomena of natural languages.Based on the theory of semantic language, this thesis makes a classification of translation selection of polyseme, and carries on descriptions of each type by semantic element. The functional relations in the corpus are summarized and a database of functional relations is constructed. This database certainly improves the accuracy of translation by testing in a small system. The thesis also introduces the mathematic representation of extended functional relations, and constructs a contextual pattern database to process the problem in a more complex context. We take "yes" as an example to perform the identification of its context, because "yes" has a more obvious context-dependable word. In the process, we use link grammar parser, which provides an accurate grammatical relationship among words and a reliable basis for the identification of sentence patterns and the analysis of sentence structures, ensuring the accuracy of the identification of context.We take a close test of the system for "yes" in the bilingual corpus of "Family Album, U.S.A." and "Olympics'. The recall rate and accuracy rate are respectively 90.7% and 88.2%. In the two open test of the system in "English 900" and eleven famous plays, the recall rate and accuracy rate are respectively 88.7%, 86.7% and 83.3%, 83.7%. These results imply that it is effective to cope with translation selection of extended functional relations using match of context pattern.