TSCheater: Generating High-Quality Tibetan Adversarial Texts via Visual Similarity
This addresses the problem of adversarial robustness for Tibetan language models, which is critical for preserving its ancient literature and strategic importance, though it is incremental as it builds on existing adversarial text generation techniques.
The authors tackled the vulnerability of language models to adversarial attacks in Tibetan by proposing TSCheater, a method that leverages visual similarity of syllables to generate high-quality adversarial texts, achieving superior performance in attack effectiveness, perturbation magnitude, and human acceptance compared to existing methods.
Language models based on deep neural networks are vulnerable to textual adversarial attacks. While rich-resource languages like English are receiving focused attention, Tibetan, a cross-border language, is gradually being studied due to its abundant ancient literature and critical language strategy. Currently, there are several Tibetan adversarial text generation methods, but they do not fully consider the textual features of Tibetan script and overestimate the quality of generated adversarial texts. To address this issue, we propose a novel Tibetan adversarial text generation method called TSCheater, which considers the characteristic of Tibetan encoding and the feature that visually similar syllables have similar semantics. This method can also be transferred to other abugidas, such as Devanagari script. We utilize a self-constructed Tibetan syllable visual similarity database called TSVSDB to generate substitution candidates and adopt a greedy algorithm-based scoring mechanism to determine substitution order. After that, we conduct the method on eight victim language models. Experimentally, TSCheater outperforms existing methods in attack effectiveness, perturbation magnitude, semantic similarity, visual similarity, and human acceptance. Finally, we construct the first Tibetan adversarial robustness evaluation benchmark called AdvTS, which is generated by existing methods and proofread by humans.