Word Representations, Tree Models and Syntactic Functions

TL;DR

This paper introduces a method for learning word representations using unsupervised tree-structured hidden Markov models that incorporate syntactic functions, leading to improved NLP task performance.

Contribution

It formalizes unsupervised learning of tree-based word representations with syntactic functions, demonstrating their effectiveness in NLP tasks and analyzing the benefits over sequential models.

Findings

Syntactic functions improve word representation quality.

Enhanced performance in named entity recognition and semantic frame identification.

Tree models' advantages over sequential models are not always clear.

Abstract

Word representations induced from models with discrete latent variables (e.g.\ HMMs) have been shown to be beneficial in many NLP applications. In this work, we exploit labeled syntactic dependency trees and formalize the induction problem as unsupervised learning of tree-structured hidden Markov models. Syntactic functions are used as additional observed variables in the model, influencing both transition and emission components. Such syntactic information can potentially lead to capturing more fine-grain and functional distinctions between words, which, in turn, may be desirable in many NLP applications. We evaluate the word representations on two tasks -- named entity recognition and semantic frame identification. We observe improvements from exploiting syntactic function information in both cases, and the results rivaling those of state-of-the-art representation learning methods. Additionally, we revisit the relationship between sequential and unlabeled-tree models and find that the advantage of the latter is not self-evident.