proposes a tree matching model, whose parameters include the size of tree fragments, their translation probability, and context similarity of examples, which is defined as the similarity of the surrounding phrases of a translation example and an input phrase