Abstract
We present a comparative study of transition-, graph- and PCFG-based models aimed at illuminating more precisely the likely contribution of CFGs in improving Chinese dependency parsing accuracy, especially by combining heterogeneous models. Inspired by the impact of a constituency grammar on dependency parsing, we propose several strategies to acquire pseudo CFGs only from dependency annotations. Compared to linguistic grammars learned from rich phrase-structure treebanks, well designed pseudo grammars achieve similar parsing accuracy and have equivalent contributions to parser ensemble. Moreover, pseudo grammars increase the diversity of base models; therefore, together with all other models, further improve system combination. Based on automatic POS tagging, our final model achieves a UAS of 87.23%, resulting in a significant improvement of the state of the art.
Highlights
Popular approaches to dependency parsing can be divided into two classes: grammar-free and grammar-based
In order to exploit the diversity gain, we address the issue of parser combination
The main reason is that the reparsing algorithm is a graph-based one, which performs worse with regard to the prediction of a whole sentence
Summary
Popular approaches to dependency parsing can be divided into two classes: grammar-free and grammar-based. Data-driven, grammar-free approaches make essential use of machine learning from linguistic annotations in order to parse new sentences Such approaches, e.g. transition-based (Nivre, 2008) and graph-based (McDonald, 2006; Torres Martins et al, 2009) have attracted the most attention in recent years. The mainstream work on recent dependency parsing focuses on data-driven approaches that automatically learn to produce dependency graphs for sentences solely from a hand-crafted dependency treebank The advantage of such models is that they are ported to any language in which labeled linguistic resources exist. Hatori et al (2011) combined both and obtained a state-of-the-art supervised parsing result
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have