Combining the Sparsity and Unambiguity Biases for Grammar Induction
Kewei Tu · 2013
In this paper we describe our participating sys-tem for the dependency induction track of the PASCAL Challenge on Grammar Induction. Our system incorporates two types of induc-tive biases: the sparsity bias and the unambi-guity bias. The sparsity bias favors a gram-mar with fewer grammar rules. The unambi-guity bias favors a grammar that leads to un-ambiguous parses, which is motivated by the observation that natural language is remark-ably unambiguous in the sense that the num-ber of plausible parses of a natural language sentence is very small. We introduce our ap-proach to combining these two types of biases and discuss the system implementation. Our experiments show that both types of inductive biases are beneficial to grammar induction. 1