A comparative evaluation and analysis of three generations of Distributional Semantic Models

TL;DR

This paper compares three generations of Distributional Semantic Models, revealing that static models often outperform contextualized ones in semantic tasks and highlighting factors influencing their behavior through statistical and neuroscientific analyses.

Contribution

It provides a comprehensive evaluation of static and contextualized DSMs, challenging assumptions about the superiority of predict models and analyzing semantic space representations.

Findings

Static DSMs outperform contextualized models in many semantic tasks.

Predict models' supposed superiority is less evident and not universal.

RSA uncovers differences based on frequency and part-of-speech of words.

Abstract

Distributional semantics has deeply changed in the last decades. First, predict models stole the thunder from traditional count ones, and more recently both of them were replaced in many NLP applications by contextualized vectors produced by Transformer neural language models. Although an extensive body of research has been devoted to Distributional Semantic Model (DSM) evaluation, we still lack a thorough comparison with respect to tested models, semantic tasks, and benchmark datasets. Moreover, previous work has mostly focused on task-driven evaluation, instead of exploring the differences between the way models represent the lexical semantic space. In this paper, we perform a comprehensive evaluation of type distributional vectors, either produced by static DSMs or obtained by averaging the contextualized vectors generated by BERT. First of all, we investigate the performance of embeddings in several semantic tasks, carrying out an in-depth statistical analysis to identify the major factors influencing the behavior of DSMs. The results show that i.) the alleged superiority of predict based models is more apparent than real, and surely not ubiquitous and ii.) static DSMs surpass contextualized representations in most out-of-context semantic tasks and datasets. Furthermore, we borrow from cognitive neuroscience the methodology of Representational Similarity Analysis (RSA) to inspect the semantic spaces generated by distributional models. RSA reveals important differences related to the frequency and part-of-speech of lexical items.