Detecting hate speech and offensive language is essential for maintaining a safe and respectful digital environment. This study examines the limitations of state-of-the-art large language models (LLMs) in identifying offensive content within systematically perturbed data, with a focus on Chinese, a language particularly susceptible to such perturbations. We introduce ToxiCloakCN, an enhanced dataset derived from ToxiCN, augmented with homophonic substitutions and emoji transformations, to test the robustness of LLMs against these cloaking perturbations. Our findings reveal that existing models significantly underperform in detecting offensive content when these perturbations are applied. We provide an in-depth analysis of how different types of offensive content are affected by these perturbations and explore the alignment between human and model explanations of offensiveness. Our work highlights the urgent need for more advanced techniques in offensive language detection to combat the evolving tactics used to evade detection mechanisms.

Role-playing agents (RPAs), powered by large language models, have emerged as a flourishing field of applications. However, a key challenge lies in assessing whether RPAs accurately reproduce the personas of target characters, namely their character fidelity. Existing methods mainly focus on the knowledge and linguistic patterns of characters. This paper, instead, introduces a novel perspective to evaluate the personality fidelity of RPAs with psychological scales. Overcoming drawbacks of previous self-report assessments on RPAs, we propose InCharacter, namely **In**terviewing **Character** agents for personality tests. Experiments include various types of RPAs and LLMs, covering 32 distinct characters on 14 widely used psychological scales. The results validate the effectiveness of InCharacter in measuring RPA personalities. Then, with InCharacter, we show that state-of-the-art RPAs exhibit personalities highly aligned with the human-perceived personalities of the characters, achieving an accuracy up to 80.7%.

Lexical-syntactic flexibility, in the form of conversion (or zero-derivation) is a hallmark of English morphology. In conversion, a word with one part of speech is placed in a non-prototypical context, where it is coerced to behave as if it had a different part of speech. However, while this process affects a large part of the English lexicon, little work has been done to establish the degree to which language models capture this type of generalization. This paper reports the first study on the behavior of large language models with reference to conversion. We design a task for testing lexical-syntactic flexibility—the degree to which models can generalize over words in a construction with a non-prototypical part of speech. This task is situated within a natural language inference paradigm. We test the abilities of five language models—two proprietary models (GPT-3.5 and GPT-4), three open source model (Mistral 7B, Falcon 40B, and Llama 2 70B). We find that GPT-4 performs best on the task, followed by GPT-3.5, but that the open source language models are also able to perform it and that the 7-billion parameter Mistral displays as little difference between its baseline performance on the natural language inference task and the non-prototypical syntactic category task, as the massive GPT-4.

The spread of disinformation and propagandistic content poses a threat to societal harmony, undermining informed decision-making and trust in reliable sources. Online platforms often serve as breeding grounds for such content, and malicious actors exploit the vulnerabilities of audiences to shape public opinion. Although there have been research efforts aimed at the automatic identification of disinformation and propaganda in social media content, there remain challenges in terms of performance. The ArAIEval shared task aims to further research on these particular issues within the context of the Arabic language. In this paper, we discuss our participation in these shared tasks. We competed in subtasks 1A and 2A, where our submitted system secured positions 9th and 10th, respectively. Our experiments consist of fine-tuning transformer models and using zero- and few-shot learning with GPT-4.