Learning and Inference for Hierarchically Split PCFGs

Google Tech Talks
February, 28 2008

ABSTRACT

Treebank parsing can be seen as the search for an optimally refined grammar consistent with a coarse training treebank. We describe a method in which a minimal grammar is hierarchically refined using EM to give accurate, compact grammars. The resulting grammars are extremely compact compared to other high-performance parsers, yet the parser gives the best published accuracies on several languages, as well as the best generative parsing numbers in English. In addition, we give an associated coarse-to-fine inference scheme which vastly improves inference time with no loss in test set accuracy.

Slides: http://www.eecs.berkeley.edu/~petrov/data/google_talk.ppt

Speaker: Slav Petrov
Slav Petrov is a Ph.D. Candidate at University of California Berkeley Dept of Computer Science, where he is also a research assistant working with Dan Klein and Jitendra Malik on inducing latent structure for perception problems in vision and language.

Category:

People & Blogs

Tags:

• google
• techtalks
• techtalk
• engedu
• talk
• talks
• googletechtalks
• education

[prashnrao]

Thank you for uploading this talk! I wish Google uploads a lot more on Computational Linguistics.

[mablacn]

Very inspiring talk. Nice methods, great results.