Bi-layer Template Based Machine Translation

Machine Translation(MT) is growing with the development of network and the appearance of a large mount of multi-language texts, but it is still a developing research. This thesis intends to improve the performance of the Example Based Machine Translation(EBMT) and hopes to provide a new way for the future research.Traditional Example-Based Machine Translation is based on large-scale exemplars, and it has shortcomings such as low matching rate when accurate matching method is used and poor translation quality when fuzzy matching method is used. And the method with mono-layer translation templates is weak in phrase translating. Therefore, this thesis proposes one method which automatically extracts bi-layer translation templates from sentence-aligned bilingual corpus. Those acquired translation templates can obtain higher matching rate than traditional EBMT, while keeping good translation quality.In detail, the following tentative efforts are included in this thesis:1. Bring forward the bi-layer translation template architecture, one is sentence template layer, the other is chunk template layer, and give a series of specific definitions.2. Automatically extract bi-layer translation templates from the sentence-aligned bilingual corpus using the aligned parts of each sentence pair's parse tree. The existing phrase rule can be added to the chunk template library manually.3. Develop a prototype of machine translation system using the extracted bi-layer translation template library.