The Return of Lexical Dependencies: Neural Lexicalized PCFGs

TL;DR

This paper introduces neural lexicalized PCFG models that effectively induce both constituents and lexical dependencies, overcoming previous sparsity issues and improving grammar induction results.

Contribution

It presents a novel neural approach to lexicalized PCFGs that jointly models constituents and dependencies, addressing sparsity challenges in unsupervised grammar induction.

Findings

Improved grammar induction performance over previous methods

Unified modeling of constituents and dependencies

Effective handling of lexical sparsity

Abstract

In this paper we demonstrate that $\textit{context free grammar (CFG) based methods for grammar induction benefit from modeling lexical dependencies}$. This contrasts to the most popular current methods for grammar induction, which focus on discovering $\textit{either}$ constituents $\textit{or}$ dependencies. Previous approaches to marry these two disparate syntactic formalisms (e.g. lexicalized PCFGs) have been plagued by sparsity, making them unsuitable for unsupervised grammar induction. However, in this work, we present novel neural models of lexicalized PCFGs which allow us to overcome sparsity problems and effectively induce both constituents and dependencies within a single model. Experiments demonstrate that this unified framework results in stronger results on both representations than achieved when modeling either formalism alone. Code is available at https://github.com/neulab/neural-lpcfg.