Translation Templates' Acquisition Based On Partial Parsing

EBMT ( Example-Based Machine Translation) systems are based on large scale example corpus in traditional, having the defect of low precision of matching. Translation template can solve the problem of data sparsity, large storage space and low matching precision of examples. Therefore, this thesis proposes one method which automatically extracts bi-level translation templates from sentence-aligned bilingual corpus, we propose its definition and matching algorithm. Finally, We proposed the matching algorithm as well.Firstly, we analyze some important techniques of machine translation, especially the EBMT method. Secondly, we summarize the meaning of translation template, propose our definition and research strategy. Finally, we accomplish the extraction of bi-level translation templates.The key point of EBMT method is translation templates' extraction and match algorithm. We used partial parsing in the process of extraction, chunking in general. The advantage of partial parsing is the robust result that can decrease the ambiguity of parsing. Concretely, partial parsing on the sentence-aligned bilingual corpus and get the chunking results, then we set the frame of the sentence, after extraction of phrase templates, we can get the sentence template. The matching module searches the most matching template for input sentence in the database, with the information of the chunking result and frame structure. The templates matching algorithm gets the searching result by using the key word in the sentence frame and its chunking template.