Large Language Models (LLMs) have become essential for offensive language detection, yet their ability to handle annotation disagreement remains underexplored. Disagreement samples, which arise from subjective interpretations, pose a unique challenge due to their ambiguous nature. Understanding how LLMs process these cases, particularly their confidence levels, can offer insight into their alignment with human annotators. This study systematically evaluates the performance of multiple LLMs in detecting offensive language at varying levels of annotation agreement. We analyze binary classification accuracy, examine the relationship between model confidence and human disagreement, and explore how disagreement samples influence model decision-making during few-shot learning and instruction fine-tuning. Our findings reveal that LLMs struggle with low-agreement samples, often exhibiting overconfidence in these ambiguous cases. However, utilizing disagreement samples in training improves both detection accuracy and model alignment with human judgment. These insights provide a foundation for enhancing LLM-based offensive language detection in real-world moderation tasks.

The proliferation of hate speech has caused significant harm to society. The intensity and directionality of hate are closely tied to the target and argument it is associated with. However, research on hate speech detection in Chinese has lagged behind, and existing datasets lack span-level fine-grained annotations. Furthermore, the lack of research on Chinese hateful slang poses a significant challenge. In this paper, we provide two valuable fine-grained Chinese hate speech detection research resources. First, we construct a Span-level Target-Aware Toxicity Extraction dataset (STATE ToxiCN), which is the first span-level Chinese hate speech dataset. Secondly, we evaluate the span-level hate speech detection performance of existing models using STATE ToxiCN. Finally, we conduct the first study on Chinese hateful slang and evaluate the ability of LLMs to understand hate semantics. Our work contributes valuable resources and insights to advance span-level hate speech detection in Chinese.

The proliferation of hate speech has caused significant harm to society. The intensity and directionality of hate are closely tied to the target and argument it is associated with. However, research on hate speech detection in Chinese has lagged behind, and existing datasets lack span-level fine-grained annotations. Furthermore, the lack of research on Chinese hateful slang poses a significant challenge. In this paper, we provide two valuable fine-grained Chinese hate speech detection research resources. First, we construct a Span-level Target-Aware Toxicity Extraction dataset (STATE ToxiCN), which is the first span-level Chinese hate speech dataset. Secondly, we evaluate the span-level hate speech detection performance of existing models using STATE ToxiCN. Finally, we conduct the first study on Chinese hateful slang and evaluate the ability of LLMs to understand hate semantics. Our work contributes valuable resources and insights to advance span-level hate speech detection in Chinese.

Humor recognition aims to identify whether a specific speaker’s text is humorous. Current methods for humor recognition mainly suffer from two limitations: (1) they solely focus on one aspect of humor commonalities, ignoring the multifaceted nature of humor; and (2) they typically overlook the critical role of speaker individuality, which is essential for a comprehensive understanding of humor expressions. To bridge these gaps, we introduce the Commonality and Individuality Incorporated Network for Humor Recognition (CIHR), a novel model designed to enhance humor recognition by integrating multifaceted humor commonalities with the distinctive individuality of speakers. The CIHR features a Humor Commonality Analysis module that explores various perspectives of multifaceted humor commonality within user texts, and a Speaker Individuality Extraction module that captures both static and dynamic aspects of a speaker’s profile to accurately model their distinctive individuality. Additionally, Static and Dynamic Fusion modules are introduced to effectively incorporate the humor commonality with speaker’s individuality in the humor recognition process. Extensive experiments demonstrate the effectiveness of CIHR, underscoring the importance of concurrently addressing both multifaceted humor commonality and distinctive speaker individuality in humor recognition.

Disclaimer: Samples in this paper may be harmful and cause discomfort! Patronizing and condescending language (PCL) is a form of speech directed at vulnerable groups. As an essential branch of toxic language, this type of language exacerbates conflicts and confrontations among Internet communities and detrimentally impacts disadvantaged groups. Traditional pre-trained language models (PLMs) perform poorly in detecting PCL due to its implicit toxicity traits like hypocrisy and false sympathy. With the rise of large language models (LLMs), we can harness their rich emotional semantics to establish a paradigm for exploring implicit toxicity. In this paper, we introduce PclGPT, a comprehensive LLM benchmark designed specifically for PCL. We collect, annotate, and integrate the Pcl-PT/SFT dataset, and then develop a bilingual PclGPT-EN/CN model group through a comprehensive pre-training and supervised fine-tuning staircase process to facilitate implicit toxic detection. Group detection results and fine-grained detection from PclGPT and other models reveal significant variations in the degree of bias in PCL towards different vulnerable groups, necessitating increased societal attention to protect them.

The widespread dissemination of toxic online posts is increasingly damaging to society. However, research on detecting toxic language in Chinese has lagged significantly due to limited datasets. Existing datasets suffer from a lack of fine-grained annotations, such as the toxic type and expressions with indirect toxicity. These fine-grained annotations are crucial factors for accurately detecting the toxicity of posts involved with lexical knowledge, which has been a challenge for researchers. To tackle this problem, we facilitate the fine-grained detection of Chinese toxic language by building a new dataset with benchmark results. First, we devised Monitor Toxic Frame, a hierarchical taxonomy to analyze the toxic type and expressions. Then, we built a fine-grained dataset ToxiCN, including both direct and indirect toxic samples. ToxiCN is based on an insulting vocabulary containing implicit profanity. We further propose a benchmark model, Toxic Knowledge Enhancement (TKE), by incorporating lexical features to detect toxic language. We demonstrate the usability of ToxiCN and the effectiveness of TKE based on a systematic quantitative and qualitative analysis.

This paper describes our system used in the SemEval-2023 Task 9 Multilingual Tweet Intimacy Analysis. There are two key challenges in this task: the complexity of multilingual and zero-shot cross-lingual learning, and the difficulty of semantic mining of tweet intimacy. To solve the above problems, our system extracts contextual representations from the pretrained language models, XLM-T, and employs various optimization methods, including adversarial training, data augmentation, ordinal regression loss and special training strategy. Our system ranked 14th out of 54 participating teams on the leaderboard and ranked 10th on predicting languages not in the training data. Our code is available on Github.

Sexism is an injustice afflicting women and has become a common form of oppression in social media. In recent years, the automatic detection of sexist instances has been utilized to combat this oppression. The Subtask A of SemEval-2023 Task 10, Explainable Detection of Online Sexism, aims to detect whether an English-language post is sexist. In this paper, we describe our system for the competition. The structure of the classification model is based on RoBERTa, and we further pre-train it on the domain corpus. For fine-tuning, we adopt Unsupervised Data Augmentation (UDA), a semi-supervised learning approach, to improve the robustness of the system. Specifically, we employ Easy Data Augmentation (EDA) method as the noising operation for consistency training. We train multiple models based on different hyperparameter settings and adopt the majority voting method to predict the labels of test entries. Our proposed system achieves a Macro-F1 score of 0.8352 and a ranking of 41/84 on the leaderboard of Subtask A.