Neural Bi-Lexicalized PCFG Induction

TL;DR

This paper introduces a neural approach to lexicalized PCFGs that models bilexical dependencies directly, improving parsing accuracy and efficiency without strong independence assumptions.

Contribution

It proposes a novel neural parameterization of L-PCFGs that captures bilexical dependencies, reducing complexity and enhancing performance.

Findings

Improved unsupervised parsing accuracy on WSJ dataset

Faster training and inference times

Effective modeling of bilexical dependencies

Abstract

Neural lexicalized PCFGs (L-PCFGs) have been shown effective in grammar induction. However, to reduce computational complexity, they make a strong independence assumption on the generation of the child word and thus bilexical dependencies are ignored. In this paper, we propose an approach to parameterize L-PCFGs without making implausible independence assumptions. Our approach directly models bilexical dependencies and meanwhile reduces both learning and representation complexities of L-PCFGs. Experimental results on the English WSJ dataset confirm the effectiveness of our approach in improving both running speed and unsupervised parsing performance.