Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inference

TL;DR

This paper investigates whether natural language inference models rely on fallible syntactic heuristics by introducing a controlled dataset, HANS, revealing that even advanced models like BERT perform poorly on heuristic-challenging examples.

Contribution

The paper introduces HANS, a dataset designed to diagnose heuristic reliance in NLI models, and demonstrates that current models depend on these heuristics, highlighting areas for improvement.

Findings

Models perform poorly on HANS, indicating reliance on heuristics.

BERT and other models fail on heuristic-challenging examples.

HANS can be used to measure progress in NLI system robustness.

Abstract

A machine learning system can score well on a given test set by relying on heuristics that are effective for frequent example types but break down in more challenging cases. We study this issue within natural language inference (NLI), the task of determining whether one sentence entails another. We hypothesize that statistical NLI models may adopt three fallible syntactic heuristics: the lexical overlap heuristic, the subsequence heuristic, and the constituent heuristic. To determine whether models have adopted these heuristics, we introduce a controlled evaluation set called HANS (Heuristic Analysis for NLI Systems), which contains many examples where the heuristics fail. We find that models trained on MNLI, including BERT, a state-of-the-art model, perform very poorly on HANS, suggesting that they have indeed adopted these heuristics. We conclude that there is substantial room for improvement in NLI systems, and that the HANS dataset can motivate and measure progress in this area