Learning Accurate, Compact, and Interpretable Tree Annotation

TitleLearning Accurate, Compact, and Interpretable Tree Annotation
Publication TypeConference Paper
Year of Publication2006
AuthorsPetrov, S., Barrett L., Thibaux R., & Klein D.
Published inProceedings of the 44th Annual Meeting of the Association of Computational Linguistics (ACL)
Other Numbers3264
Abstract

We present an automatic approach to tree annotationin which basic nonterminal symbols are alternatelysplit and merged to maximize the likelihoodof a training treebank. Starting with a simple Xbargrammar, we learn a new grammar whose nonterminalsare subsymbols of the original nonterminals.In contrast with previous work, we are ableto split various terminals to different degrees, as appropriateto the actual complexity in the data. Ourgrammars automatically learn the kinds of linguisticdistinctions exhibited in previous work on manualtree annotation. On the other hand, our grammarsare much more compact and substantially more accuratethan previous work on automatic annotation.Despite its simplicity, our best grammar achieves

URLhttp://www.icsi.berkeley.edu/pubs/ai/accuratecompacttree06.pdf
Bibliographic Notes

Proceedings of the 44th Annual Meeting of the Association of Computational Linguistics (ACL), Sydney, Australia

Abbreviated Authors

S. Petrov, L. Barrett, R. Thibeaux, and D. Klein

ICSI Research Group

AI

ICSI Publication Type

Article in conference proceedings