On Dimensional Linguistic Properties of the Word Embedding Space

TL;DR

This paper investigates the geometric and syntactic properties of word embeddings, revealing that variance explained by principal components does not correlate with syntactic information and that variance-based post-processing can be counterproductive.

Contribution

It provides novel insights into the limitations of variance-based analysis and post-processing of word embeddings, highlighting the importance of non-isotropic geometry.

Findings

Variance explained by principal components does not correlate with syntactic information.

Variance-based post-processing can harm performance in NLP tasks.

Non-isotropic geometry may be crucial for effective word embeddings.

Abstract

Word embeddings have become a staple of several natural language processing tasks, yet much remains to be understood about their properties. In this work, we analyze word embeddings in terms of their principal components and arrive at a number of novel and counterintuitive observations. In particular, we characterize the utility of variance explained by the principal components as a proxy for downstream performance. Furthermore, through syntactic probing of the principal embedding space, we show that the syntactic information captured by a principal component does not correlate with the amount of variance it explains. Consequently, we investigate the limitations of variance based embedding post-processing and demonstrate that such post-processing is counter-productive in sentence classification and machine translation tasks. Finally, we offer a few precautionary guidelines on applying variance based embedding post-processing and explain why non-isotropic geometry might be integral to word embedding performance.