Austria

Large language models (LLMs) have become ubiquitous, thus it is important to understand their risks and limitations, such as their propensity to generate harmful output. This includes smaller LLMs, which are important for settings with constrained compute resources, such as edge devices. Detection of LLM harm typically requires human annotation, which is expensive to collect. This work studies two questions: How do smaller LLMs rank regarding generation of harmful content? How well can larger LLMs annotate harmfulness? We prompt three small LLMs to elicit harmful content of various types, such as discriminatory language, offensive content, privacy invasion, or negative influence, and collect human rankings of their outputs. Then, we compare harm annotation from three state-of-the-art large LLMs with each other and with humans. We find that the smaller models differ with respect to harmfulness. We also find that large LLMs show low to moderate agreement with humans.

ACL 2025

Can LLMs Rank the Harmfulness of Smaller LLMs? We are Not There Yet

## Welcome!
"I am excited to welcome you to this year’s edition of the Conference on Empirical Methods in Natural Language Processing! Importantly, it marks the 30th edition of EMNLP. With over 8,000 submissions, more than 3,000 accepted papers, and thousands of attendees, we have come a long way from that first
workshop, which had 14 accepted papers. As the field looks ahead, Suzhou is the fitting location for celebrating this milestone: rooted in a long literary tradition, yet modern and forward-looking, and home to a large share of the NLP community." 

*Message from the General Chair, Dirk Hovy*

EMNLP 2025

Celebrate 30 Years of EMNLP! 
EMNLP 2025 will be held in Suzhou, China from November 5th to November 9th, 2025.

workshop paper

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair: 
*It is my great pleasure and honor to welcome you to the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), held in beautiful Vienna, Austria, from July 27 to August 1, 2025. ACL2025continues our field’s tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.* 
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to ACL 2025". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you need to log in with the **email address you registered with**. 

Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

The way we relay factual information and the way we present deceptive information as truth differs from the perspective of argumentation. In this paper, we explore whether these differences can be exploited to detect deceptive political news in English. We do this by training a model to detect different kinds of argumentation in online news text. We use sentence embeddings extracted from an argumentation type classification model as features for a deceptive news classifier. This deceptive news classification model leverages the sequence of argumentation types within an article to determine whether it is credible or deceptive. Our approach outperforms other state-of-the-art models while having lower variance. Finally, we use the output of our argumentation model to analyze the differences between credible and deceptive news based on the distribution of argumentation types across the articles. Results of this analysis indicate that credible political news presents statements supported by a variety of argumentation types, while deceptive news relies on anecdotes and testimonial.

Are You Trying to Convince Me or Are You Trying to Deceive Me? Using Argumentation Types to Identify Deceptive News

The recent advancements in Large Language Models(LLMs) have had a significant impact on a wide range of fields, from general domains to specialized areas. However, these advancements have also significantly increased the potential for malicious users to exploit harmful and jailbreak prompts for malicious attacks. Although there have been many efforts to prevent harmful prompts and jailbreak prompts, protecting LLMs from such malicious attacks remains an important and challenging task. In this paper, we propose QGuard, a simple yet effective safety guard method, that utilizes question prompting to block harmful prompts in a zero-shot manner. Our method can defend LLMs not only from text-based harmful prompts but also from multi-modal harmful prompt attacks. Moreover, by diversifying and modifying guard questions, our approach remains robust against the latest harmful prompts without fine-tuning. Experimental results show that our model performs competitively on both text-only and multi-modal harmful datasets. Additionally, by providing an analysis of question prompting, we enable a white-box analysis of user inputs. We believe our method provides valuable insights for real-world LLM services in mitigating security risks associated with harmful prompts.

QGuard:Question-based Zero-shot Guard for Multi-modal LLM Safety

A dogwhistle is a communicative act intended to broadcast a message only understood by a select in-group while going unnoticed by others (out-group). We illustrate that political dogwhistle behavior in a more radical community precedes the occurrence of the dogwhistles in a less radical community, but the reverse does not hold. We study two Swedish online communities -- Flashback and Familjeliv -- which both contain discussions of life and society, with the former having a stronger anti-immigrant subtext. Expressions associated with dogwhistles are substantially more frequent in Flashback than in Familjeliv. We analyze the time series of changes in intensity of three dogwhistle expressions (DWEs), i.e., the strength of association of a DWE and its in-group meaning modeled by Swedish Sentence-BERT, and model the dynamic temporal relationship of intensity in the two communities for the three DWEs using Vector Autoregression (VAR). We show that changes in intensity in Familjeliv are explained by the changes of intensity observed at previous lags in Flashback but not the other way around. This suggests a direction of travel for dogwhistles associated with radical ideologies to less radical contexts.

Who leads? Who follows? Temporal dynamics of political dogwhistles in Swedish online communities

Despite growing interest in automated hate speech detection, most existing approaches overlook the linguistic diversity of online content. Multilingual instruction-tuned large language models such as LLaMA, Aya, Qwen, and BloomZ offer promising capabilities across languages, but their effectiveness in identifying hate speech through zero-shot and few-shot prompting remains underexplored. This work evaluates LLM prompting-based detection across eight non-English languages, utilizing several prompting techniques and comparing them to fine-tuned encoder models. We show that while zero-shot and few-shot prompting lag behind fine-tuned encoder models on most of the real-world evaluation sets, they achieve better generalization on functional tests for hate speech detection. Our study also reveals that prompt design plays a critical role, with each language often requiring customized prompting techniques to maximize performance.

Can Prompting LLMs Unlock Hate Speech Detection across Languages? A Zero-shot and Few-shot Study

Conspiracist narratives posit an omnipotent, evil group causing harm throughout domains. However, modern-day online conspiracism is often more erratic, consisting of loosely connected posts displaying a general anti-establishment attitude pervaded by negative emotions. We gather a dataset of 300 conspiracist and mainstream, Telegram channels in Italian and English and use the automatic extraction of entities and emotion detection to compare structural characteristics of both types of channels. We create a co-occurrence network of entities to analyze how the different types of channels introduce and use them across posts and topics. We find that conspiracist channels are characterized by anger. Moreover, co-occurrence networks of entities appearing in conspiracist channels are more dense. We theorize that this reflects a narrative structure where all actants are pushed into a single domain. Conspiracist channels disproportionately associate the most central group of entities with anger and fear. We do not find evidence that entities in conspiracist narratives occur across more topics. This could indicate an erratic type of online conspiracism where everything can be connected to everything and that is characterized by a high number of entities and high levels of anger.

Multilingual Analysis of narrative characteristics in conspiracist vs mainstream Telegram channels

Hate speech detection is vital for creating safe online environments, as harmful content can drive social polarization. This study explores the impact of enriching text with intent and group tags on machine performance and human moderation workflows. For machine performance, we enriched text with intent and group tags to train hate speech classifiers. Intent tags were the most effective, achieving state-of-the-art F1-score improvements on the IHC, SBIC, and DH datasets, respectively. Cross-dataset evaluations further demonstrated the superior generalization of intent-tagged models compared to other pre-trained approaches. Then, through a user study (N=100), we evaluated seven moderation settings, including intent tags, group tags, model probabilities, and randomized counterparts. Intent annotations significantly improved the accuracy of the moderators, allowing them to outperform machine classifiers by 12.9%. Moderators also rated intent tags as the most useful explanation tool, with a 41% increase in perceived helpfulness over the control group. Our findings demonstrate that intent-based annotations enhance both machine classification performance and human moderation workflows.

Hate Explained: Evaluating NER-Enriched Text in Human and Machine Moderation of Hate Speech

Technology-facilitated gender-based violence has become a global threat to women's political representation and democracy. Understanding how online hate affects its targets is thus paramount. This study analyzes 10 million tweets directed at female candidates in the Brazilian election in 2022 and examines their reactions to online misogyny. Using a self-trained machine learning classifier to detect Portuguese misogynistic tweets and a quantitative analysis of the candidates' tweeting behavior, the study investigates how the number of misogynistic attacks received alters the online activity of the female candidates. The analysis finds that young and left-wing candidates and candidates with higher visibility online received significantly more attacks. Furthermore, the analysis finds that an increase in misogynistic attacks in the previous week is associated with a decrease in female candidates' tweets in the following week. This potentially threatens their equal participation in public opinion formation and silences women's voices in political discourse.

Online Misogyny Against Female Candidates in the 2022 Brazilian Elections: A Threat to Women's Political Representation?

Technology-facilitated violence constitutes a growing threat to inclusive democratic participation, particularly affecting women and marginalized groups. Drawing on a mixed-methods study from Brazil, this research examines how online harassment impacts politically active individuals, with a focus on gendered experiences and differential outcomes. A quantitative survey of 597 politically active individuals and seven in-depth interviews with female politicians were analyzed to assess the psychological and political consequences of technology-facilitated violence. More than 70% of survey respondents reported harassment within the past four weeks, with women disproportionately affected by gendered types of abuse. Regression results demonstrate that years of activism, education, and occupation significantly shape both exposure to harassment and coping responses. While some interviewees reported emotional exhaustion and disengagement, others described increased motivation and political resolve. These findings offer critical insights into how technology-facilitated violence shapes political behavior and underscore the need for platform accountability and institutional support structures.

Technology-Facilitated Violence and Political Participation: Evidence from a Mixed-Methods Study in Brazil

This work in progress presents an early-stage, conceptual exploration of how civil society actors can be meaningfully involved in the development of an AI-assisted tool for monitoring anti-democratic movements on Telegram. Using conspiracy theory classification as a use case, we outline and compare two concrete approaches to enabling such collaboration.

Civil Society in the Loop: Feedback-Driven Adaptation of (L)LM-Assisted Classification in an Open-Source Telegram Monitoring Tool

**(Civil Society Reports)**


Online hate speech propagation is a complex issue, deeply influenced by both the perpetrator and the target's cultural, historical, and societal contexts. Consequently, developing a universally robust hate speech classifier for diverse social media texts remains a challenging and unsolved task. The lack of mechanisms to track the spread and severity of hate speech further complicates the formulation of effective solutions. In response to this, to monitor hate speech in Indonesia during the recent 2024 presidential election, we have employed advanced
Natural Language Processing (NLP) technologies to create an improved hate speech classifier
tailored for a narrower subset of texts; specifically, texts that target vulnerable groups that
have historically been the targets of hate speech in Indonesia.

Downloads

Next from ACL 2025

Are You Trying to Convince Me or Are You Trying to Deceive Me? Using Argumentation Types to Identify Deceptive News

Similar lecture

Do LLMs Align Human Values Regarding Social Biases? Judging and Explaining Social Biases with LLMs

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from ACL 2025

Are You Trying to Convince Me or Are You Trying to Deceive Me? Using Argumentation Types to Identify Deceptive News

Similar lecture

Do LLMs Align Human Values Regarding Social Biases? Judging and Explaining Social Biases with LLMs

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads