A lattice-based framework for enhancing statistical parsers with information from unlabeled corpora
Michaela Atterer, Hinrich Schütze · 2006
Great strides have been made in building statistical parsers trained on annotated corpora such as the Penn tree-bank. However, recently performance improvements have leveled off. New information sources need to be considered to make further progress in parsing. In this paper, we propose a new method of using unlabeled corpora for improving syntactic disambiguation. The method is tested on the problem of relative clause attachment with encouraging results.