Semiautomatic acquisition of translation templates from monolingual unannotated corpora

We propose a new approach which can semiautomatically acquire translation templates from the unannotated Chinese spoken language corpora in the domain of travel information accessing. In the approach, we introduce two elements into the unsupervised agglomerative clustering method, which are called extended contexts and cohesion degree. With these two elements, the similarity and the degree of cohesion of two entities in the corpora may be described more exactly. In our approach the semantic and phrasal structures are firstly acquired from the unannotated corpus, and then the translation templates are manually built based on the semantic and phrasal structures. The preliminary results of our experiment show that the approach can get a higher performance than the method merely using the local contexts and mutual information.