@MISC{Cheng05pltigparsing, author = {Winston Cheng and et al.}, title = {PLTIG Parsing}, year = {2005} }
Bookmark
OpenURL
Abstract
We describe the TIG formalism and its advantages over CFGs and TAGs. We then give an algorithm for training PLTIGs following Hwa (2001). We train on a partially supervised corpus by repeated parsing with a lexicalized normal form. The training algorithm is EM-style, and the parsing algorithm is a generalization of CKY. Finally, we look toward the possibility of extending this method into a synchronous form.