arXiv subjects

Manuel Tonneau

Publications and source records attributed to Manuel Tonneau.

13 recordsLinked to original sources

arXiv2026

LLMs Mirror Country-Specific Gender Patterns If Asked, but Skew Male When Generating Media in Local Languages

Large language models (LLMs) are increasingly used to generate media, but whether their content perpetuates gender stereotypes is unknown: standard benchmarks rely on selection-based formats rather than long-form generation, and surveyed baselines for local gender associations are scarce outside the West. We collect gender associations for 22 occupational and domestic roles from 695 respondents across the United States, India, Kenya, and Nigeria, and evaluate eight LLMs under two regimes: direct questioning and media generation. Models track the surveyed associations under direct questioning but skew substantially more male under media generation in major local-language cells, consistent with the male bias documented in human-produced media. Outside the US, the shift is much smaller and non-significant under English prompting, so English-only or country-agnostic evaluation would miss this bias in the languages where these models are most deployed. Instruction prompting reduces the shift directionally, but trades off against alignment with the surveyed associations. Evaluating LLM gender bias for global deployment therefore requires generation-format testing, local-language prompting, and locally-collected human baselines.

cs.CL↗

arXiv2026

Bye Bye Perspective API: Lessons for Building and Governing Measurement Infrastructure

Perspective API closes at the end of 2026, removing the de facto standard for toxicity measurement and exposing researchers' dependence on a tool they did not control. Drawing on this case, we argue that a research field must build and govern its own measurement infrastructure rather than borrow it. Surveying 241 papers that use or study Perspective, we show what depending on it cost the research community: claims reaching past what the tool could support, results that shifted when its model was silently retrained, and disparities researchers could measure but not explain. These failures were amplified throughout the LLM lifecycle, where Perspective supplied the labels, filtered the corpora, and graded the systems trained on each, rewarding errors rather than catching them. To keep the instrument open to study after shutdown, we release Perspective scores for 5.9 million text snippets from 77 datasets. We further specify ten requirements for measurement infrastructure a field owns, and argue that what blocks such infrastructure is not technical capability but the value the field places on infrastructure work.

cs.CL↗

arXiv2026

The Enforcement and Feasibility of Hate Speech Moderation

Online hate speech is associated with harms ranging from deteriorating mental health to violence, yet how consistently platforms moderate hate, and whether enforcement is feasible at scale, remain poorly understood. We audit hate speech moderation on Twitter (now X) using 540,000 tweets annotated by trained native speakers, representative of a full day on the platform. Five months after posting, 80% of hateful tweets, including violent ones, remained online. Removal was only marginally more likely than for non-hateful tweets, far below scams or adult content, and insensitive to severity and reach. Automated detection could not reliably classify hate but ranked it highly, enabling human triage. Simulating this workflow, current staffing curbed little exposure, yet substantial reductions proved financially feasible, far below applicable regulatory fines. Persistent hate reflects resource allocation, not technical limits.

cs.CY↗

arXiv2026

Different Demographic Cues Yield Inconsistent Conclusions About LLM Personalization and Bias

Demographic cue-based evaluation is widely used to study how large language models (LLMs) adapt their responses to signaled demographic attributes within and across groups. This approach typically relies on a single cue (e.g., names) as a proxy for group membership, implicitly treating different cues as interchangeable operationalizations of a single underlying identity-conditioned behavior. We test this assumption in realistic advice-seeking interactions spanning 14.8 million prompts, focusing on race and gender in a U.S. context. We find that cues for the same group induce only partially overlapping changes in model responses, yielding inconsistent conclusions about personalization, while bias conclusions are unstable, with both magnitude and direction of group differences varying across cues. We further show that these inconsistencies reflect differences in cue-group association strength and linguistic features bundled within cues that shape model responses. Together, our findings suggest that demographic conditioning in LLMs is not a cue-invariant category-level parameter but depends fundamentally on how identity is cued, reflecting responses to linguistic signals rather than stable demographic categories. We therefore call for multi-cue, mechanism-aware evaluations as a foundation for robust and interpretable claims about demographic variation in LLM responses.

cs.CL↗

arXiv2026

Celebrity messages reduce online hate and limit its spread

Online hate spreads rapidly, yet little is known about whether preventive and scalable strategies can curb it. We conducted the largest randomized controlled trial of hate speech prevention to date: a 20-week messaging campaign on X in Nigeria targeting ethnic hate. 73,136 users who had previously engaged with hate speech were randomly assigned to receive prosocial video messages from Nigerian celebrities. The campaign reduced hate content by 2.5% to 5.5% during treatment, with about 75% of the reduction persisting over the following four months. Reaching a larger share of a user's audience reduced amplification of that user's hate posts among both treated and untreated users, cutting hate reposts by over 50% for the most exposed accounts. Scalable messaging can limit online hate without removing content.

cs.SI↗

arXiv2025

Effect of Static vs. Conversational AI-Generated Messages on Colorectal Cancer Screening Intent: a Randomized Controlled Trial

Large language model (LLM) chatbots show increasing promise in persuasive communication. Yet their real-world utility remains uncertain, particularly in clinical settings where sustained conversations are difficult to scale. In a pre-registered randomized controlled trial, we enrolled 915 U.S. adults (ages 45-75) who had never completed colorectal cancer (CRC) screening. Participants were randomized to: (1) no message control, (2) expert-written patient materials, (3) single AI-generated message, or (4) a motivational interviewing chatbot. All participants were required to remain in their assigned condition for at least three minutes. Both AI arms tailored content using participant's self-reported demographics including age and gender. Both AI interventions significantly increased stool test intentions by over 12 points (12.9-13.8/100), compared to a 7.5 gain for expert materials (p<.001 for all comparisons). While the AI arms outperformed the no message control for colonoscopy intent, neither showed improvement xover expert materials. Notably, for both outcomes, the chatbot did not outperform the single AI message in boosting intent despite participants spending ~3.5 minutes more on average engaging with it. These findings suggest concise, demographically tailored AI messages may offer a more scalable and clinically viable path to health behavior change than more complex conversational agents and generic time intensive expert-written materials. Moreover, LLMs appear more persuasive for lesser-known and less-invasive screening approaches like stool testing, but may be less effective for entrenched preferences like colonoscopy. Future work should examine which facets of personalization drive behavior change, whether integrating structural supports can translate these modest intent gains into completed screenings, and which health behaviors are most responsive to AI-supported guidance.

cs.CY↗

arXiv2025

Large Language Model Chatbot Conversations vs Public Health Materials and Parental HPV Vaccination Intentions: A Randomized Clinical Trial

Health care systems are increasingly considering large language model (LLM)-based chatbots for vaccine communication, but evidence that they improve durable, behaviorally relevant outcomes beyond existing health materials is limited. This randomized clinical trial tested whether brief, multiturn LLM chatbot interactions increased parental intention to vaccinate children against human papillomavirus (HPV) compared with no intervention and government public health materials, and whether effects persisted. Parents in the US, Canada, and UK were recruited online from March 3 to May 25, 2025, with follow-up at 15 and 45 days. Eligible participants were adults with at least one HPV vaccine-eligible child who was unvaccinated or whose vaccination status was unknown. Participants were randomized to no-message control, country-matched government materials with at least 3 minutes of exposure, or a 3-minute GPT-4o chatbot interaction using either a default persuasive style or a shorter conversational style. The primary outcome was self-reported likelihood of vaccinating the child against HPV within 12 months, measured immediately after intervention on a 0-100 scale. Follow-up outcomes included vaccination intent and self-reported vaccination at 15 and 45 days. In total, 1297 participants were randomized (mean age 42.84 years; 72.1% female). Compared with no intervention, public health materials increased immediate vaccination intent (Cohen d = 0.53; 95% CI, 0.36-0.70), as did the default chatbot (d = 0.48; 95% CI, 0.30-0.65) and conversational chatbot (d = 0.33; 95% CI, 0.17-0.49). At 45 days, neither chatbot increased intent relative to controls, whereas public health materials maintained modest effects. No intervention increased self-reported vaccination uptake. Findings suggest well-designed public health materials may match or exceed short LLM chatbot conversations for HPV vaccine promotion.

cs.CY↗

arXiv2025

When Claims Evolve: Evaluating and Enhancing the Robustness of Embedding Models Against Misinformation Edits

Online misinformation remains a critical challenge, and fact-checkers increasingly rely on claim matching systems that use sentence embedding models to retrieve relevant fact-checks. However, as users interact with claims online, they often introduce edits, and it remains unclear whether current embedding models used in retrieval are robust to such edits. To investigate this, we introduce a perturbation framework that generates valid and natural claim variations, enabling us to assess the robustness of a wide-range of sentence embedding models in a multi-stage retrieval pipeline and evaluate the effectiveness of various mitigation approaches. Our evaluation reveals that standard embedding models exhibit notable performance drops on edited claims, while LLM-distilled embedding models offer improved robustness at a higher computational cost. Although a strong reranker helps to reduce the performance drop, it cannot fully compensate for first-stage retrieval gaps. To address these retrieval gaps, we evaluate train- and inference-time mitigation approaches, demonstrating that they can improve in-domain robustness by up to 17 percentage points and boost out-of-domain generalization by 10 percentage points. Overall, our findings provide practical improvements to claim-matching systems, enabling more reliable fact-checking of evolving misinformation. Code and data are available at https://github.com/JabezNzomo99/claim-matching-robustness.

cs.CL↗

arXiv2024

HateDay: Insights from a Global Hate Speech Dataset Representative of a Day on Twitter

To address the global challenge of online hate speech, prior research has developed detection models to flag such content on social media. However, due to systematic biases in evaluation datasets, the real-world effectiveness of these models remains unclear, particularly across geographies. We introduce HateDay, the first global hate speech dataset representative of social media settings, constructed from a random sample of all tweets posted on September 21, 2022 and covering eight languages and four English-speaking countries. Using HateDay, we uncover substantial variation in the prevalence and composition of hate speech across languages and regions. We show that evaluations on academic datasets greatly overestimate real-world detection performance, which we find is very low, especially for non-European languages. Our analysis identifies key drivers of this gap, including models' difficulty to distinguish hate from offensive speech and a mismatch between the target groups emphasized in academic datasets and those most frequently targeted in real-world settings. We argue that poor model performance makes public models ill-suited for automatic hate speech moderation and find that high moderation rates are only achievable with substantial human oversight. Our results underscore the need to evaluate detection systems on data that reflects the complexity and diversity of real-world social media.

cs.CL↗

arXiv2024

Indian-BhED: A Dataset for Measuring India-Centric Biases in Large Language Models

Large Language Models (LLMs), now used daily by millions, can encode societal biases, exposing their users to representational harms. A large body of scholarship on LLM bias exists but it predominantly adopts a Western-centric frame and attends comparatively less to bias levels and potential harms in the Global South. In this paper, we quantify stereotypical bias in popular LLMs according to an Indian-centric frame through Indian-BhED, a first of its kind dataset, containing stereotypical and anti-stereotypical examples in the context of caste and religious stereotypes in India. We find that the majority of LLMs tested have a strong propensity to output stereotypes in the Indian context, especially when compared to axes of bias traditionally studied in the Western context, such as gender and race. Notably, we find that GPT-2, GPT-2 Large, and GPT 3.5 have a particularly high propensity for preferring stereotypical outputs as a percent of all sentences for the axes of caste (63-79%) and religion (69-72%). We finally investigate potential causes for such harmful behaviour in LLMs, and posit intervention techniques to reduce both stereotypical and anti-stereotypical biases. The findings of this work highlight the need for including more diverse voices when researching fairness in AI and evaluating LLMs.

cs.CL↗

arXiv2024

NaijaHate: Evaluating Hate Speech Detection on Nigerian Twitter Using Representative Data

To address the global issue of online hate, hate speech detection (HSD) systems are typically developed on datasets from the United States, thereby failing to generalize to English dialects from the Majority World. Furthermore, HSD models are often evaluated on non-representative samples, raising concerns about overestimating model performance in real-world settings. In this work, we introduce NaijaHate, the first dataset annotated for HSD which contains a representative sample of Nigerian tweets. We demonstrate that HSD evaluated on biased datasets traditionally used in the literature consistently overestimates real-world performance by at least two-fold. We then propose NaijaXLM-T, a pretrained model tailored to the Nigerian Twitter context, and establish the key role played by domain-adaptive pretraining and finetuning in maximizing HSD performance. Finally, owing to the modest performance of HSD systems in real-world conditions, we find that content moderators would need to review about ten thousand Nigerian tweets flagged as hateful daily to moderate 60% of all hateful content, highlighting the challenges of moderating hate speech at scale as social media usage continues to grow globally. Taken together, these results pave the way towards robust HSD systems and a better protection of social media users from hateful content in low-resource settings.

cs.CL↗

arXiv2024

From Languages to Geographies: Towards Evaluating Cultural Bias in Hate Speech Datasets

Perceptions of hate can vary greatly across cultural contexts. Hate speech (HS) datasets, however, have traditionally been developed by language. This hides potential cultural biases, as one language may be spoken in different countries home to different cultures. In this work, we evaluate cultural bias in HS datasets by leveraging two interrelated cultural proxies: language and geography. We conduct a systematic survey of HS datasets in eight languages and confirm past findings on their English-language bias, but also show that this bias has been steadily decreasing in the past few years. For three geographically-widespread languages -- English, Arabic and Spanish -- we then leverage geographical metadata from tweets to approximate geo-cultural contexts by pairing language and country information. We find that HS datasets for these languages exhibit a strong geo-cultural bias, largely overrepresenting a handful of countries (e.g., US and UK for English) relative to their prominence in both the broader social media population and the general population speaking these languages. Based on these findings, we formulate recommendations for the creation of future HS datasets.

cs.CL↗

arXiv2022

Multilingual Detection of Personal Employment Status on Twitter

Detecting disclosures of individuals' employment status on social media can provide valuable information to match job seekers with suitable vacancies, offer social protection, or measure labor market flows. However, identifying such personal disclosures is a challenging task due to their rarity in a sea of social media content and the variety of linguistic forms used to describe them. Here, we examine three Active Learning (AL) strategies in real-world settings of extreme class imbalance, and identify five types of disclosures about individuals' employment status (e.g. job loss) in three languages using BERT-based classification models. Our findings show that, even under extreme imbalance settings, a small number of AL iterations is sufficient to obtain large and significant gains in precision, recall, and diversity of results compared to a supervised baseline with the same number of labels. We also find that no AL strategy consistently outperforms the rest. Qualitative analysis suggests that AL helps focus the attention mechanism of BERT on core terms and adjust the boundaries of semantic expansion, highlighting the importance of interpretable models to provide greater control and visibility into this dynamic learning process.

cs.CL↗