archives

Unsupervised Learning of Probabilistic Context-Free Grammar using Iterative Biclustering (Extended Version)


Home 

About 

Browse 

Search 

Register 

Subscriptions 

Deposit Papers 

Help
    

Tu, Kewei and Honavar, Vasant (2008) Unsupervised Learning of Probabilistic Context-Free Grammar using Iterative Biclustering (Extended Version). Technical Report 572, Computer Science, Iowa State University.

Full text available as:Adobe PDF

There is a later version of this eprint available: Click here to view it.

Abstract

This paper presents an unsupervised algorithm that learns a probabilistic context-free grammar from positive sample sentences only. The algorithm iteratively learns new grammar rules by doing biclustering in a table that enumerates all the symbol pairs in the training corpus. We explain our algorithm in a Bayesian framework, showing that it tries to find grammar rules that maximize the posterior of the grammar given the training corpus. Positive results were obtained in experiments.

Keywords:Probabilistic context-free grammar, grammar induction, grammar learning
Subjects:Computing Methodologies: ARTIFICIAL INTELLIGENCE: Learning (K.3.2)
ID code:00000575
Deposited by:Kewei Tu on 30 May 2008

Available Versions of This Paper



Contact site administrator at: ssg@cs.iastate.edu