RobustLR: Evaluating Robustness to Logical Perturbation in Deductive Reasoning

TL;DR

RobustLR introduces evaluation datasets to test the robustness of deductive reasoning models like RoBERTa and T5 against logical perturbations, revealing their vulnerabilities and highlighting areas for improvement in understanding logical semantics.

Contribution

The paper presents RobustLR, a new suite of datasets for assessing the logical robustness of language models, exposing their weaknesses in handling minimal logical edits and operators.

Findings

Models lack robustness to logical perturbations.

Difficulty in learning negation and disjunction operators.

Models do not perform consistently across different logical edits.

Abstract

Transformers have been shown to be able to perform deductive reasoning on a logical rulebase containing rules and statements written in English natural language. While the progress is promising, it is currently unclear if these models indeed perform logical reasoning by understanding the underlying logical semantics in the language. To this end, we propose RobustLR, a suite of evaluation datasets that evaluate the robustness of these models to minimal logical edits in rulebases and some standard logical equivalence conditions. In our experiments with RoBERTa and T5, we find that the models trained in prior works do not perform consistently on the different perturbations in RobustLR, thus showing that the models are not robust to the proposed logical perturbations. Further, we find that the models find it especially hard to learn logical negation and disjunction operators. Overall, using our evaluation sets, we demonstrate some shortcomings of the deductive reasoning-based language models, which can eventually help towards designing better models for logical reasoning over natural language. All the datasets and code base have been made publicly available.