Robustness to Modification with Shared Words in Paraphrase Identification

TL;DR

This paper investigates the robustness of paraphrase identification models against modifications involving shared words, revealing significant vulnerabilities and demonstrating that adversarial training can improve robustness.

Contribution

The study introduces a novel method to test model robustness through shared word modifications and shows how adversarial training can mitigate identified vulnerabilities.

Findings

Models exhibit significant performance drops on modified examples.

Shared word modifications can cause incorrect predictions.

Adversarial training improves robustness against such modifications.

Abstract

Revealing the robustness issues of natural language processing models and improving their robustness is important to their performance under difficult situations. In this paper, we study the robustness of paraphrase identification models from a new perspective -- via modification with shared words, and we show that the models have significant robustness issues when facing such modifications. To modify an example consisting of a sentence pair, we either replace some words shared by both sentences or introduce new shared words. We aim to construct a valid new example such that a target model makes a wrong prediction. To find a modification solution, we use beam search constrained by heuristic rules, and we leverage a BERT masked language model for generating substitution words compatible with the context. Experiments show that the performance of the target models has a dramatic drop on the modified examples, thereby revealing the robustness issue. We also show that adversarial training can mitigate this issue.