Multi-Granularity Tibetan Textual Adversarial Attack Method Based on Masked Language Model

TL;DR

This paper introduces TSTricker, a multi-granularity adversarial attack method for Tibetan text using masked language models, significantly challenging Chinese minority language models and improving robustness evaluation.

Contribution

It presents a novel Tibetan-specific adversarial attack approach leveraging masked language models, addressing the gap in minority language NLP security research.

Findings

Reduces model accuracy by over 28.70%

Changes predictions in more than 90.60% of samples

Outperforms baseline attack methods

Abstract

In social media, neural network models have been applied to hate speech detection, sentiment analysis, etc., but neural network models are susceptible to adversarial attacks. For instance, in a text classification task, the attacker elaborately introduces perturbations to the original texts that hardly alter the original semantics in order to trick the model into making different predictions. By studying textual adversarial attack methods, the robustness of language models can be evaluated and then improved. Currently, most of the research in this field focuses on English, and there is also a certain amount of research on Chinese. However, there is little research targeting Chinese minority languages. With the rapid development of artificial intelligence technology and the emergence of Chinese minority language models, textual adversarial attacks become a new challenge for the information processing of Chinese minority languages. In response to this situation, we propose a multi-granularity Tibetan textual adversarial attack method based on masked language models called TSTricker. We utilize the masked language models to generate candidate substitution syllables or words, adopt the scoring mechanism to determine the substitution order, and then conduct the attack method on several fine-tuned victim models. The experimental results show that TSTricker reduces the accuracy of the classification models by more than 28.70% and makes the classification models change the predictions of more than 90.60% of the samples, which has an evidently higher attack effect than the baseline method.