COMPS: Conceptual Minimal Pair Sentences for testing Robust Property Knowledge and its Inheritance in Pre-trained Language Models

TL;DR

This paper introduces COMPS, a dataset of minimal pair sentences designed to evaluate pre-trained language models' abilities to attribute properties to concepts and demonstrate property inheritance, revealing strengths and limitations in their reasoning capabilities.

Contribution

The paper presents COMPS, a novel benchmark for testing property attribution and inheritance in PLMs, and analyzes 22 models to assess their reasoning robustness.

Findings

PLMs easily distinguish trivial property differences.

Models struggle with nuanced concept-property relations.

Performance drops significantly with distracting information.

Abstract

A characteristic feature of human semantic cognition is its ability to not only store and retrieve the properties of concepts observed through experience, but to also facilitate the inheritance of properties (can breathe) from superordinate concepts (animal) to their subordinates (dog) -- i.e. demonstrate property inheritance. In this paper, we present COMPS, a collection of minimal pair sentences that jointly tests pre-trained language models (PLMs) on their ability to attribute properties to concepts and their ability to demonstrate property inheritance behavior. Analyses of 22 different PLMs on COMPS reveal that they can easily distinguish between concepts on the basis of a property when they are trivially different, but find it relatively difficult when concepts are related on the basis of nuanced knowledge representations. Furthermore, we find that PLMs can demonstrate behavior consistent with property inheritance to a great extent, but fail in the presence of distracting information, which decreases the performance of many models, sometimes even below chance. This lack of robustness in demonstrating simple reasoning raises important questions about PLMs' capacity to make correct inferences even when they appear to possess the prerequisite knowledge.