Deep grammar Formalism such as Tree Adjoining Grammar(TAG) is widely used in Natural Language Generation(NLG). However, crafting the grammar not only requires expert knowledge of both the grammar formalism and the target application domain of NLG, but also needs a lot of human labor. In this paper, we propose a semi-automatic approach to extract wellformed Lexicalized Tree Adjoining Grammar(LTAG) anchored with proper semantics from a parallel corpus. The parallel cor-pus consists of sentences

more »

... d the correspondent semantics. This approach requires much less human effort. The result shows that our approach achieves comparable performance with the state-of-the-art. Meanwhile, the grammar extracted in this paper has better linguistic interpretation than previous work.