Pay Attention to the Robustness of Chinese Minority Language Models! Syllable-level Textual Adversarial Attack on Tibetan Script
Xi Cao, Dolma Dawa, Nuo Qun, Trashi Nyima
TL;DR
The paper introduces TSAttacker, a Tibetan syllable-level black-box adversarial attack for Tibetan-script NLP, using syllable cosine distance and a scoring mechanism to select substitutions. It formalizes the attack on K-class classifiers and demonstrates its effectiveness across six CINO-based models and three downstream tasks, with high attack success and low perturbation cost. Ablation shows that increasing the allowable syllable-distance $d_{max}$ boosts attack strength, while robustness gaps remain across models, underscoring the vulnerability of Chinese minority language PLMs. By releasing code and emphasizing minority-language security, the work provides a foundation for future robustness defenses and evaluation in Tibetan NLP and related languages.
Abstract
The textual adversarial attack refers to an attack method in which the attacker adds imperceptible perturbations to the original texts by elaborate design so that the NLP (natural language processing) model produces false judgments. This method is also used to evaluate the robustness of NLP models. Currently, most of the research in this field focuses on English, and there is also a certain amount of research on Chinese. However, to the best of our knowledge, there is little research targeting Chinese minority languages. Textual adversarial attacks are a new challenge for the information processing of Chinese minority languages. In response to this situation, we propose a Tibetan syllable-level black-box textual adversarial attack called TSAttacker based on syllable cosine distance and scoring mechanism. And then, we conduct TSAttacker on six models generated by fine-tuning two PLMs (pre-trained language models) for three downstream tasks. The experiment results show that TSAttacker is effective and generates high-quality adversarial samples. In addition, the robustness of the involved models still has much room for improvement.