Comparing and integrating Tree Adjoining Grammars

Grammars are core elements of many NLP applications. Grammars can be developed in two ways: built by hand or extracted from corpora. In this paper, we compare a handcrajted grammar with a Treebank grammar. We contend that recognizing substructures of the grammars' basic units is necessary tures and semantic information which are rarely represented in the corpora. lt would be ideal if we could combine the strengths of both types of grammar. As a first step towards addressing this issue, in this paper we compare a hand-crafted grammar with a Treebank grammar and propose a way of integrating them to produce new grammars. Two grammars not only because it allows grammars to be compared at a higher level, but also because 2. it provides the building blocks f or consistent and efficient integration of the grammars. The two LTAGs that we compare are the XTAG English grammar (XTAG-Group, 1995) and a grammar extracted from Penn English Treebank. The XTAG grammar has 1004 tree templates.1 The Treebank grammar that we use in this paper is extracted from the Penn English Treebank II (Marcus et al., 1994) using the extraction algorithm described in (Xia, 1999}. The extracted grammar has 3072 templates.