Lukas Edman, Daryna Dementieva, Alexander Frasercs.CL
Large language models (LLMs) demonstrate impressive performance across a wide range of general NLP tasks; however, their effectiveness in sensitive domains, such as hate speech detection, remains less clear. Prior studies comparing prompted LLMs with state-of-the-art encoder-based models (e.g., BERT variants (Roy et al., 2023; Dönmez et al., 2024)) have shown only marginal gains, suggesting that LLMs may not excel in hate speech detection or mitigation. In this work, we revisit this question through the lens of instruction tuning. By thoroughly unifying 36 English hate speech datasets spanning multiple labeling schemes, we fine-tune a generalist LLM, based on Qwen3 (Qwen Team, 2025), specifically for hate speech mitigation. Our results demonstrate not only state-of-the-art performance on in-domain benchmarks but also substantial improvements in cross-domain and cross-lingual generalization--areas where encoder-based specialist classifiers often struggle.
Urdu, the world's tenth most spoken language with 246 million speakers, remains almost entirely absent from mainstream LLM safety evaluation and nine years of WOAH proceedings. To investigate whether this absence has measurable consequences for content moderation reliability, five large language models, GPT-4o, Claude Sonnet 4.5, Gemini 2.5 Flash, Qwen-2.5, and Llama-3.1, were tested across six datasets spanning Nastaliq Urdu, Roman Urdu, English, and code-switched Urdu-English. Across the five Urdu-script datasets, label instability between original-script and English-translation classification ranged from 15.9% (Gemini 2.5 Flash) to 31.6% (Qwen-2.5), with a 'Missed-in-Urdu' rate, content flagged as harmful in English translation but passed as normal in the original script, ranging from 2.4% to 9.9% (median 4.3%). A complete enumeration of all 205 papers across nine ALW/WOAH editions via the ACL Anthology API confirms zero dedicated Urdu papers across the entire period. Results indicate that current LLMs provide uneven safety assurance across Urdu's script varieties, with smaller open-weight models showing substantially higher instability and missed-harm rates than frontier closed models.
Stephen Meisenbacher, Vlad Garbuz, Chirill Donos +5cs.CL
Hate speech is a real and timely threat that affects a large portion of online users, especially youth and minority groups. While building reliable and robust automatic hate speech detection (HSD) systems is paramount, we argue that this must also be balanced with the individual right to privacy. Exploring the intersection of HSD and privacy, we demonstrate that HSD systems might unintentionally achieve performance at the cost of encoding authorship, posing a threat to privacy. Building on these findings, we establish the notion of a privacy-HSD trade-off, which demands a careful balance. We benchmark a series of text privatization methods, as well as our newly proposed domain-specific AgnoSpeech technique, showing that balancing privacy and HSD is difficult but feasible. The findings make a strong case for more research on the trade-offs between privacy and HSD, both of which have tangible implications for the safeguarding of online participation.
Due to the widespread accessibility of the internet and social media, toxic and hateful con-tent has grown exponentially, causing significant distress and negative societal impacts. Ro-man Urdu, a low-resource language used in Pakistan and among Urdu-speaking communities worldwide, presents additional challenges because of its informal grammar, inconsistent sen-tence structures, and multiple variations in word spellings. This research aims to identify the most effective techniques for hate speech classification in such low-resource settings with limited data. To address this, the study investigates and compares the latest approaches, in-cluding prompt tuning, parameter-efficient fine-tuning (PEFT) using LoRA, and prompt en-gineering, under various experimental configurations. To achieve this objective, four exper-iments were designed. The first experiment involved direct inferencing with LLMs without any fine-tuning, to evaluate how well these models understand Roman Urdu in a zero-shot setting, especially given limited data. The second experiment utilized parameter-efficient fine-tuning (PEFT) with LoRA, which updates only a small subset of parameters, thereby reducing computational cost. The third experiment explored prompt tuning with both mixed and manually crafted prompts, using very small sets of training examples relative to the entire dataset, making it computationally efficient as well. Finally, the fourth experiment applied prompt engineering through zero-shot and few-shot learning, relying solely on care-fully designed instruction prompts for classification without further training.
Toneema Zubair, Muhammad Junaid Asif, Faisal Kamiran +2cs.AI cs.CL
It is challenging to detect hate speech in Low Resource Languages (LRLs) because of the absence of annotated data, the informality of its language structure, and the lack of standardized grammar. A good example of such a challenge is Roman Urdu which is broadly used by South Asians on social media and has a high variation while lacking contextually consistent spellings. The objective of this paper is to conduct a comprehensive assessment of Large Language Models (LLMs) for Hate Speech Detection (HSD) in Roman Urdu script and fine-tune these models using the Parameter-Efficient Fine-Tuning (PEFT) method called Low-Rank Adaptation (LoRA). To evaluate zero-shot inference, we benchmarked it against PEFT on different transformer models, including Mistral, LLaMA, Falcon, and multilingual BERT. Experiments are conducted on the PURUTT (Parallel Urdu and Roman Urdu Corpus for Toxic Comments and Transliteration) dataset with over 72,000 annotated comments. The results suggest that zero shot models perform moderately (F1 = 0.56), but updating a small fraction of the model trainable parameters improves the classification performance significantly (F1 > 0.93). Our results have shown that PEFT delivers outstanding performance alongside excellent computational efficiency, making it highly suitable for low-resource language processing tasks.
Francesco Benedetti, Antonio Pellicani, Gianvito Pio +1cs.SI cs.AI
Online social networks increasingly expose people to users who propagate discriminatory, hateful, and violent content. Young users, in particular, are vulnerable to exposure to such content, which can have harmful psychological and social repercussions. Given the massive scale of today's social networks, in terms of both published content and number of users, there is an urgent need for effective systems to aid Law Enforcement Agencies (LEAs) in identifying and addressing users that disseminate malicious content. In this work we introduce IMMENSE, a machine learning-based method for detecting malicious social network users. Our approach adopts a hybrid classification strategy that integrates three perspectives: the semantics of the users' published content, their social relationships and their spatial information. Such contextual perspectives potentially enhance classification performance beyond text-only analysis. Importantly, IMMENSE employs an inductive learning approach, enabling it to classify previously unseen users or entire new networks without the need for costly and time-consuming model retraining procedures. Experiments carried out on a real-world Twitter/X dataset showed the superiority of IMMENSE against five state of the art competitors, confirming the benefits of its hybrid approach for effective deployment in social network monitoring systems.
State-backed influence operations are routinely measured as high-prevalence sources of ``hate'' and ``toxicity.'' We argue those rates rest on a measurement error: the detectors behind them are validated to catch a broader definition inclusive of hostility or divisiveness aimed at an out-group, and so over-attribute hate to content better described as partisan or geopolitical invective. Across 25.08M tweets from seven government-attributed campaigns in the Twitter Information Operations archive (8,275 accounts), we separate hate from the other forms of divisiveness. We first validate a two-prompt LLM-based detector, matching human labels at Cohen's $κ=0.82$, to identify the broader hostility; we then develop an auditable rule, agreeing with an expert at $κ=0.52$, to further classify this content (5,457 posts) into three sub-categories. About 50.1% are identity-based attacks on people, whereas 30.4% are partisan attacks and 19.5% invective against states and their foreign policy. Reporting all of it as hate therefore overstates hate roughly twofold; only 18.7% is both identity-based and dehumanizing or inciting. Six of seven campaigns sort into three regimes that a single ``hate'' rate flattens, namely identity hate (RU-op and IRA, both Russia-attributed), geopolitical invective (both Iran operations), and partisan divisiveness (both Venezuela operations). We call the shared product $manufactured divisiveness$. The line to separate these constructs itself remains unsettled: on the hardest cases three independent human experts agree only moderately (pairwise $κ=0.37$--$0.50$), and the best of nineteen LLM models tops out at $κ=0.601$ against the experts' majority. Our findings can help redefine the study of hate in the context of influence campaigns and broader online discourse.
The spread of hate speech (HS) across different social media platforms (SMPs) poses a major concern for online safety and ethical moderation. Automatic detection of HS remains a challenging task, especially in under-resourced languages like Bangla, due to cultural context, implicit expressions, and informal linguistic patterns. This study aimed to expose the crisis of Bangla HS detection systems by diagnosing how and why benchmark-trained models fail to identify implicit, context-dependent HS. Six architectures (FastText + CNN, FastText + LSTM, FastText + BiLSTM, BanglaBERT, BanglaBERT + CNN, and BanglaBERT + BiLSTM) were trained on benchmark datasets (about 75,000 posts) and a merged multi-source dataset (about 120,000 posts), then externally validated on an annotated dataset (about 200 posts) collected from Facebook, Twitter, and YouTube, labeled as HS and non-HS, where HS was further categorized as explicit and implicit. BanglaBERT achieved an F1-score of 91.4% on benchmark datasets but declined to 75.3% on the external set and 63.4% for implicit HS involving sarcasm and emojis. The accuracy of FastText + CNN dropped from 78.0% to 51.2% under similar conditions. Emoji-aware preprocessing improved implicit HS detection by up to 12%, whereas emoji removal caused a notable decline in performance (F1: 0.75 to 0.63). Frequent misclassifications in politically charged or satirical comments revealed over-policing risks. This study not only exposes the generalization crisis due to implicit, culturally embedded, and emoji-laden expressions but also underscores the need for developing adaptive, emoji-aware, and culturally grounded frameworks that ensure ethical moderation while preserving freedom of expression. Findings of this study provide insights for researchers, SMPs, and policymakers to design more context-sensitive HS detection systems for low-resource languages.
Sara Yakoubi, Ikram Khalfallah, Kenza Khelkhal +1cs.CL
The widespread adoption of social media platforms has transformed online communication by enabling users to exchange information and opinions instantly. However, these platforms have also facilitated the dissemination of abusive and hateful content, posing major social, psychological, and ethical challenges. Hate speech can incite discrimination, harassment, and violence against individuals or communities based on attributes such as ethnicity, religion, gender, nationality, or political affiliation. Consequently, automatic hate speech detection has become a major research topic in natural language processing (NLP) and an essential component of content moderation systems. This paper investigates automatic hate speech detection in the Algerian Arabic dialect (Darija) on social media. This task remains challenging because of the dialect's linguistic diversity, characterized by the coexistence of Arabic, French, and Arabizi (Arabic written using the Latin alphabet). We compare four categories of text classification approaches: (1) traditional machine learning models using TF-IDF features, (2) deep learning models based on recurrent neural networks, (3) Transformer-based language models, including DziriBERT and multilingual BERT, and (4) a novel hybrid architecture, FAD-SA-GRU, which combines semantic representations from DZ FastText, DZ AraVec, and DziriBERT through multi-embedding fusion, followed by a self-attention-enhanced GRU encoder. Experiments on an annotated dataset of Algerian Darija social media comments for binary hate speech classification show that FAD-SA-GRU outperforms all baselines, achieving 93.2% accuracy, 93.4% precision, 91.0% recall, 92.1% F1-score, and 97.0% ROC-AUC. Results demonstrate the effectiveness of combining complementary embedding representations with attention-based sequence modeling for robust hate speech detection in low-resource dialectal Arabic.
Unlike explicit attacks with obvious profanity, implicit hate speech hides malice within seemingly compliant expressions through metaphors and contextual hints, making its detection in online content review challenging. While existing PLM- or LLM-based methods perform well, they typically apply a single reasoning process to all samples. This overlooks fine-grained linguistic nuances and causes unnecessary computation for simpler cases. We observe that online hate speech is not monolithic but manifests in varied forms. We therefore define three fine-grained categories: Shallow, Targeted, and Context-Dependent. Accordingly, we propose Fine-grained Adaptive Implicit Hate speech Detection (FAID), a novel framework that first performs fine-grained classification and then adapts to specific categories. Specifically, for Shallow samples with surface-identifiable intents, the framework adopts lightweight prompt-tuning for rapid classification; for Targeted comments that bind malicious intent to concealed targets, we design knowledge augmentation to iteratively refine the model and reveal hidden targets; for Context-Dependent comments lacking background information, we utilize an agentic framework that automatically generates prompts to evolve context, infer missing background information and identify ambiguous malicious intents. This adaptive architecture focuses computational resources on complex implicit samples while avoiding redundant reasoning for shallow samples. Experiments on four benchmark datasets demonstrate that FAID significantly outperforms SOTA baselines.
Karina Halevy, Julia Mendelsohn, Chan Young Park +2cs.CL
Addressing hate and violence in society requires timely detection of hateful events from public reporting, but automated identification of hateful events remains underexplored. We introduce the task of hateful event detection and investigate the ability of AI systems, specifically large language models (LLMs), to discover and classify reports of antisemitic events with fine-grained labels. We evaluate OpenAI's GPT-4o and Meta's Llama-3.2-3B-Instruct on multiple expert-annotated datasets containing antisemitic event descriptions from news articles, civil society reports, and official records. We show that LLMs, particularly GPT-4o, have potential for this task, but substantial improvement is needed. Providing clear term definitions and in-context examples in prompts can improve performance: definitions are most helpful for rhetoric-oriented events (e.g. classical antisemitic tropes), while examples help label action-oriented events (e.g. physical assault). A case study of college newspapers demonstrates that LLMs can help surface relevant real-world events, supporting early monitoring and intervention. Overall, our findings highlight both opportunities and critical gaps in AI's ability to recognize complex harms and underscore the need for collaborative efforts among AI developers, policymakers, and civil society to design models, implement robust evaluation, and develop policy frameworks for defining and combating hate efficiently and effectively.
Somaiyeh Dehghan, Gökçe Uludoğan, Mehmet Umut Şen +3cs.CL cs.AI
Online hate speech has been linked to a global rise in violence against minorities, including incidents such as mass shootings, lynchings, and ethnic cleansing. Societies grappling with this issue, particularly when hate speech targets specific groups based on religion, race, ethnicity, culture, nationality, or migration status, face the challenge of balancing freedom of expression with the need for effective content moderation on widely used online platforms. In response to this challenge, we introduce a comprehensive hate speech dataset covering five distinct topics in Turkish: refugees, the Israel-Palestine conflict, anti-Greek sentiment in Turkey, ethnic or religious communities (Alevis, Armenians, Arabs, Jews, and Kurds), and LGBTI+, alongside one topic in Arabic (refugees). In addition, we develop state-of-the-art BERT-based models to address multiple dimensions of hate speech analysis, including hate category classification, hate intensity prediction, target identification, and hate speech span detection, enabling a comprehensive understanding of hateful content in online discourse.
Muhammad Jawad Chowdhury, Adiba Hasan, Ishrak Hossain +2cs.CL cs.AI
Memes have turned out to be a powerful tool through which individuals share their ideas concerning contemporary social and political problems. Their anonymity, as well as their ability to go viral, make them a powerful medium for spreading hate. It remains very difficult to identify such complex and context-dependent hate speech. Although they display excellent performance on multimodal tasks, vision-language models (VLMs) tend to ignore context, irony, and other subtle cues that play a key role in identifying hateful memes. In this work, we present a qualitative analysis of four state-of-the-art VLMs: LLaVA-7B, Qwen-VL, GPT-4o mini, and Claude 3 Haiku. We evaluate these models under zero-shot and few-shot prompting to examine how contextual framing influences their outputs. Our analysis goes beyond simple classification accuracy and focuses on a qualitative evaluation of the models' generated justifications, providing a more in-depth understanding of their thought processes and constraints when dealing with hateful memes.
Hate speech annotation pipelines routinely collapse annotator disagreement into majority vote labels before training. We show that this aggregation is not neutral: 42.6% of all annotator disagreement in HateXplain concentrates specifically at the hate/offensive boundary, a pattern consistent with annotators applying different thresholds for where hate begins (chi-squared = 135.199, df = 2, p < 0.0001). Both a hard-label BERT model (Model A) and a soft-label model (Model B) drop 22 percentage points in accuracy from agreed posts (~80%) to disagreement posts (~58%), confirmed at p < 0.0001. A per-annotator multi-head model (Model C) widens this gap further to 28 points while collapsing offensive disagreement accuracy to 0.245. Critically, Model A expresses significantly higher confidence on boundary case errors than Model C (0.710 vs. 0.495, p < 0.0001), meaning standard evaluation metrics will not detect the failure. Three downstream interventions of increasing sophistication all fail to recover boundary accuracy. We argue the problem is structural. Majority vote presents a contested judgment as ground truth, and models inherit that false certainty. The intervention must be upstream in annotation design.
Muhammad Deedahwar Mazhar Qureshi, Sannaan Khan, Muhammad Atif Qureshi +1cs.CL cs.AI
Online hate against Muslim communities often appears in culturally coded, multilingual forms that evade conventional AI moderation. Such systems, though accurate, remain opaque and risk bias, over-censorship, or under-moderation, particularly when detached from sociocultural context. We propose a \emph{training-time} explainability framework that aligns model reasoning with human-annotated rationales, improving both classification performance and interpretability. Our approach is evaluated on HateXplain (English) and BullySent (Hinglish), reflecting the prevalence of anti-Muslim hate across both languages. Using LIME, Integrated Gradients, Grad X Input, and attention, we assess accuracy, explanation quality, and cross-method agreement. Results show that gradient- and attention-based regularization improve F-scores, enhance plausibility and faithfulness, and capture culturally specific cues for detecting implicit anti-Muslim hate, offering a path toward multilingual, culturally aware content moderation.
Classifying implicit hate speech remains a challenge, as intent is often masked through insinuation and context rather than explicit slurs. Prior supervised contrastive approaches improve in-domain detection but can overfit surface cues and struggle to transfer across datasets. We propose ImpSH, a triplet-based framework that aligns posts with implied statements when available and uses context-bounded semi-hard negatives to focus learning on near confusions. We also examine AugSH, which forms positives via data augmentation. In controlled evaluations on IHC, SBIC, and DynaHate with BERT and HateBERT, ImpSH is a viable alternative to standard supervised contrastive baselines and often improves cross-domain performance under matched preprocessing and tuning budgets. Representation analysis using alignment and uniformity indicates tighter positive pairs with balanced global spread, and qualitative nearest-neighbor case studies illustrate typical false negatives under domain shift. These results demonstrate that aligning posts with their implied statements via context-bounded mining provides a more stable, bijective-like mapping to related insinuations, overcoming the volatility inherent in traditional clustering-based representation learning.
Chinese discriminatory-language detection is challenging because harmful intent is often implicit and context-dependent. We propose MAAM (Myopia--Astigmatism Anchor Mechanism), a lightweight, model-agnostic framework inspired by functional visual blur: rather than preserving every token equally, MAAM retains discrimination-relevant semantic anchors and calibrates them with C--I--S contextual priors (Contextual Tone, Group Identity, and Stance Polarity). We also introduce ChLGBT, to our knowledge the first Chinese LGBT-focused discriminatory-language dataset, with 8,120 manually annotated samples and three ordinal labels: explicit bias, implicit bias, and emotional intensity. Across strong encoder baselines, MAAM improves all three prediction dimensions, with consistent gains in accuracy, F1, Brier score, and expected calibration error. Compared with frontier LLM baselines under zero-shot and few-shot prompting protocols, MAAM remains competitive while offering stronger compactness and stability. These results suggest that interpretable anchor preservation and contextual calibration provide a practical alternative to heavier model scaling for Chinese discriminatory-language assessment.
The analysis of internet memes in the Nepali language is complicated by frequent code-mixing and a lack of established baseline resources. While memes inherently combine visual and textual elements, this study focuses on a text-centric approach by extracting embedded text using an OCR layer and modeling it with Transformer-based architectures. We evaluate six distinct models and investigate the comparative effectiveness of Hard and Soft Voting ensemble strategies across two tasks: binary hate speech detection and three-class sentiment analysis. Experimental results show that a standalone decoder-only model achieved the highest performance for binary classification, whereas the Soft Voting ensemble performed best for the multi-class sentiment task, yielding a 15.8% relative improvement in Macro F1-score over the strongest standalone baseline. These findings suggest that ensemble strategies behave differently across binary and multi-class tasks, highlighting the importance of selecting aggregation methods suited to the classification objective.
Hate speech detection is inherently subjective: people from different demographic groups perceive the same content very differently. Collecting enough annotations from multiple demographic groups is costly and difficult to scale. Persona-conditioned Large Language Models (models prompted to adopt a specific demographic identity) have been proposed as a way to simulate diverse perspectives at scale. But do they actually reflect how different groups disagree? We evaluate three aspects of human social judgement: (i) whether personas from different groups disagree in human-like ways (inter-group disagreement), (ii) whether they become more sensitive when content targets their own identity (in-group sensitivity), and (iii) whether they can accurately predict how another group would react (vicarious prediction). Our results show that no model consistently captures all three dimensions, and performance is highly model-dependent and does not emerge reliably from minimal identity prompts alone. However, vicarious prompting with Llama 3.1 yields the highest cross-group agreement in most demographic axes and provides the closest overall approximation to human disagreement patterns, indicating that this configuration may provide a more reliable setting for automatic annotation aligned with human judgements.
This paper compares a PyCaret AutoML branch and a CNN-BiLSTM branch for binary hate speech detection on Indonesian Twitter using the HS label from the corpus of Ibrohim and Budi. Both branches share the same preprocessing pipeline so that the comparison reflects modelling differences rather than inconsistent data preparation. The conventional branch uses TF-IDF with a lexicon-based abusive-word count, whereas the neural branch learns dense token representations and captures both local phrase patterns and bidirectional context. The benchmark is built from the released 13,130-row annotation table, whose HS label yields a 58:42 class ratio. On the held-out split, CNN-BiLSTM achieves the best result with 83.8% accuracy, 79.8% precision, 82.7% recall, and 81.2% F1-score. Within the PyCaret branch, Random Forest is the strongest conventional model with 77.2% accuracy and 77.0% F1-score. The neural branch therefore improves accuracy by 6.6 points and F1-score by 4.2 points. Exploratory corpus analysis, learning curves, and confusion matrices show that the dataset is short-text, moderately imbalanced, and still difficult because many decisions depend on local lexical cues plus short contextual composition. The study concludes that PyCaret AutoML is an effective conventional benchmarking framework, whereas CNN-BiLSTM is the stronger end model for the reported benchmark setting.
Information disorder is a challenging phenomenon that affects society at large. This phenomenon entails the diffusion of misleading, misinforming, and hateful content online. In different contexts, one aspect of the problem may prevail, but overall, this is a broad problem that requires comprehensive solutions. While each dimension of the problem (hate speech, disinformation, misinformation, etc.) requires in-depth analysis, in this paper, we look into the possibility of argument structure to provide relevant information to link these different areas of the problem. In particular, we focus on the WSF-ARG+ dataset, which consists of white supremacy forum messages annotated in terms of argument structure (premises and conclusion). There, we leverage the checkworthiness and hatefulness annotations of the argument components to obtain insights into the hatefulness of the whole message. Our results show promising insights (up to 96% F1), indicating the possibility of extending this direction in the future to tackle hateful content identification and information disorder countering.
Ivo Bueno, Lea Hirlimann, Enkelejda Kasnecics.CL cs.AI
Detecting hate speech in memes is challenging due to their multimodal nature and subtle, culturally grounded cues such as sarcasm and context. While recent vision-language models (VLMs) enable joint reasoning over text and images, end-to-end prompting can be brittle, as a single prediction must resolve target, stance, implicitness, and irony. These challenges are amplified in multilingual settings. We propose a prompted weak supervision (PWS) approach that decomposes meme understanding into targeted, question-based labeling functions with constrained answer options for homophobia and transphobia detection in the LT-EDI 2026 shared task. Using a quantized Qwen3-VLM to extract features by answering targeted questions, our method outperforms direct VLM classification, with substantial gains for Chinese and Hindi, ranking 1st in English, 2nd in Chinese, and 3rd in Hindi. Iterative refinement via error-driven LF expansion and feature pruning reduces redundancy and improves generalization. Our results highlight the effectiveness of prompted weak supervision for multilingual multimodal hate speech detection.