United States

Counter Narrative (CN) generation via Large Language Models (LLMs) offers a scalable ap proach to combating hate speech by produc ing targeted responses that challenge harmful content. However, existing methods typically require costly post-training or fine-tuning to improve narrative diversity and quality. We introduce a fine-tuning-free prompt optimiza tion technique that enhances Counter Narrative effectiveness without additional model train ing, making it both resource-efficient and read ily deployable. We conduct extensive evalua tion on hate speech datasets spanning English and Tamil, employing both reference-based metrics and rubric-based LLM-as-a-judge scor ing to capture multiple dimensions of nar rative quality. Experiments across multiple LLMs demonstrate that our approach consis tently outperforms vanilla prompting baselines, exhibits strong transferability across models, and adapts seamlessly to new evaluation met rics—requiring no architectural or procedu ral changes. Our findings suggest that care fully optimized prompting strategies can match or exceed the performance of more resource intensive approaches, offering a practical path toward scalable hate speech intervention.

ACL 2026

NEUNI@LT-EDI 2026: Counter Narrative Generation on Homophobic and Transphobic Comments

counter narratives

prompt optimization

prompt engineering

responsible ai

large language models

hate speech detection

It is my great pleasure and honor to welcome you to the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), held in San Diego, California, USA, July 2–7, 2026. ACL 2026 continues our fields 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.

It is exciting to see the many developments in the field reflected at our conference, such as the advancement of large language models (born out of our community) into many application directions. This is not without challenges, partly due to unsolved problems, and partly due to misuse of the technology. All this is reflected well at ACL 2026. The conference also reflects that the field is not only about advancing technology, but also the science of fundamentally understanding language and how it is use by many different speaker communities. <br>
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) <br><br><br>
<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing')">Conference Handbook</button></html>
<br><br><br><br>

To access this event page, you need to log in with the **email address you registered with**. <br>Access credentials will be sent to your email from Underline -  subject line "Welcome to ACL 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.aclweb.org/registration/) first.

Registration Required

The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) will take place in San Diego, California, from July 2 - 7, 2026.

workshop paper

The problem of harmful online discourse against the LGBTQ+ community is still a con cern on social media platforms. Although hate speech detection is a well-explored area, the task of constructive counter-narrative genera tion is still an emerging field of research, es pecially in the multilingual and low-resource settings. Counter-narratives are designed to counter harmful discourse with respectful and empathetic responses, as opposed to mere content deletion. In this paper, the model proposes a zero-shot multilingual system for counter-narrative generation in English and Tamil. The proposed system employs the pretrained google/flan-t5-base transformer model guided by rubric-aligned prompts to en courage politeness, contextual relevance, and non-toxic response generation. The system operates in a zero-shot setting without task specific fine-tuning and uses beam search de coding for controlled response generation. On the English test data, the system scored an over all score of 70.33 per cent with a contextual coherence score of 81.82 per cent. On the Tamil test data, the system scored an overall score of 33.57 per cent with significantly lower scores on coherence and quality. These find ings indicate that structured prompting can fa cilitate safe and coherent generation in English, but also underscore the challenges of zero-shot multilingual models in low-resource language scenarios.

RspectNLP@LT-EDI 2026: Rubric-Driven Prompting for Safe Multilingual Counter Narrative Generation

Social media is now an important platform for communication and interaction. At the same time, the amount of abusive and harmful con tent online has also increased. Offensive lan guage and hate speech are making these plat forms less safe and less welcoming for users. Many of these contents include homophobic and transphobic remarks aimed at the LGBT+ community. Such behaviour damages healthy discussions and can negatively affect users. For this reason, it is important to detect these con tents early so they can be flagged and removed to maintain a healthy online well-being. The is sue becomes more difficult when harmful mes sages appear in popular formats like memes. Memes are widely used by younger users to communicate online. Because they combine images and text, detecting offensive meaning becomes challenging. In this work, we attempt to address this problem. We develop a method to identify such content using the meme dataset released for the LT-EDI 2026 challenge and se cured rank 5 in the shared task. We propose a Zero-shot learning based method employ ing two LLMs (Qwen2.5-VL-3B-Instruct and Meta-Llama-3-8B-Instruct) to generate descrip tions and classify such memes. We achieved a macro F1-score of 0.55 for the English lan guage meme.

SAJI_English@LT-EDI 2026: Detection of Homophobia and Transphobia in Internet Memes Using Zero-Shot Learning

Automated detection of LGBTQ+ phobia is critical for digital safety. We participated in the LT-EDI@ACL 2026 shared task for multimodal meme classification across En glish, Hindi, and Chinese. Our methodol ogy leverages a late-fusion multimodal archi tecture combining XLM-RoBERTa for textual features and ResNet-50 for visual representa tion. By utilizing weighted cross-entropy to address extreme class imbalance, we achieved Rank 3 in Chinese (F1: 0.7371), Rank 4 in English (F1: 0.6121), and Rank 7 in Hindi (F1: 0.1616). This paper details our system description and provides a critical analysis of the ”Hindi Failure,” where class imbalance led to a lack of model convergence.

Susmitha@LT-EDI 2026: Detecting LGBTQ+ Phobia in Multilingual Memes via Joint Representation

We present our approach to LT-EDI@ACL 2026 on counter-narrative generation for ho mophobic and transphobic comments. Gener ating high-quality counter-narratives in multi lingual and low-resource settings remains chal lenging, particularly when data imbalance and script variation affect model performance. To address these issues, we explore multiple mod eling strategies built around Gemma 3 12B with QLoRA fine-tuning, including data rebalanc ing and alternative input strategies for Tamil. Our findings show that task-specific fine-tuning combined with native-script Tamil produces more stable and higher-quality outputs than large few-shot prompts or transliteration-based inputs. On the official leaderboard, our system ranks second in English with an overall score of 86.35% and sixth in Tamil with 63.77%, highlighting both the effectiveness of targeted fine-tuning and the challenges of low-resource counter-narrative generation.

SigJBS@LT-EDI 2026: QLoRA-Tuned Homophobic and Transphobic Counter Narrative Generation

This paper describes the system developed by TeamV for the LT-EDI 2026 Shared Task on Counter-Narrative Generation on Homophobic & Transphobic Comments. The shared task comprises two subtasks: (1) Hate Speech Span Detection in English, Tamil, and Hindi, and (2) Counter-Narrative Generation in English and Tamil. Our system leverages the reasoning and multilingual capabilities of a large proprietary language model (Qwen3-Max) through rigor ous few-shot in-context learning (ICL) and ro bust post-processing mechanisms. Our submit ted system demonstrated state-of-the-art perfor mance on the official CodaBench leaderboard. In Task 1, our approach achieved 1st Place across all three languages, securing macro F1 scores of 0.5338 in English, 0.5272 in Tamil, and 0.5478 in Hindi. For Task 2, our generated counter-narratives ranked 1st globally in En glish with an overall average score of 87.47% and 5th in Tamil. We present our prompting methodology, robust span-matching pipeline, detailed official results, and an analysis of the model’s performance across diverse languages.

TeamV at LT-EDI 2026: Multilingual Hate Speech Span Detection and Counter-Narrative Generation via Few-Shot In-Context Learning

The "law of conformity," the finding that frequent words are semantically stable, has been treated as a broad regularity of language change. We show it does not hold for Korean. Using diachronic word embeddings trained on historical corpora spanning 500 years (15th–20th centuries), we find a robust positive correlation between frequency and semantic shift: high-frequency Korean words change more, not less. The pattern survives six robustness controls and is validated against an English replication. Partial correlation analysis reveals that the role of polysemy in mediating the frequency–change relationship is not fixed but depends on time resolution and corpus homogeneity. We connect the reversal to frequency-driven reductive processes, including grammaticalization, semantic bleaching, and domain shift, that are especially productive in Korean. The frequency–change relationship is not a fixed regularity but varies with language typology and analytical conditions.

Frequency Accelerates Semantic Change: Evidence from 500 Years of Korean

This study proposes a quantitative framework for profiling LLM dispositions as stable, model-specific regularities in output under repeated, controlled elicitation. Using a structured narrative constraint-selection task administered across six frontier models and three instruction types, we operationalize disposition through two dimensions: "consistency", measured as cross-replication selection overlap via Jaccard similarity, and "diversity", measured as dispersion across options via the inverse Simpson index. We further introduce Narrative Landscape, a PCA-based visualization that maps each model's selection profile into a shared space for direct comparison. Results reveal a clear rigidity-exploration spectrum across model families and show that instruction types shift the geometry of selection spaces even when scalar metrics appear similar, indicating that comparable scores can mask qualitatively distinct selection topologies.

Narrative Landscape: Mapping Narrative Dispositions Across LLMs

We present a new publicly available corpus of 100,502 movie reviews from Kazakhstan collected from kino.kz, spanning 2001–2025 and covering 4,943 unique film titles. The dataset is multilingual, consisting mainly of Russian reviews alongside Kazakh and code-switched texts. Reviews are manually annotated for language and sentiment polarity, and 11,309 reviews additionally contain explicit user-provided ratings. We define two sentiment tasks—three-way polarity classification and five-class score classification—and benchmark classical BoW/TF–IDF baselines against multilingual transformer models (mBERT, XLM-RoBERTa, RemBERT). Experimental results show that transformer models consistently outperform classical baselines on polarity classification, while score classification remains challenging under leakage-controlled evaluation due to severe class imbalance and subtle distinctions between adjacent rating levels.

100,000+ Movie Reviews from Kazakhstan: Russian, Kazakh, and Code-Switched Texts

The Kṛṣṇa Yajurveda survives in multiple recensions that share substantial ritual content, yet the degree and distribution of textual overlap across recensions have never been quantified systematically. This paper presents a computational analysis of text reuse across three recensions---the Maitrāyaṇī Saṃhitā (MS), the Kāṭhaka Saṃhitā (KS), and the Taittirīya Saṃhitā (TS)---for two ritual sections (Agnyupasthāna and Punarādhāna), using ICoMa (Intertextuality Collation Machine), a new web-based multi-algorithm collation tool. Five independent similarity algorithms consistently rank MS--KS as the most closely related pair, corroborating the philological consensus. Crucially, the two ritual sections exhibit strikingly different reuse profiles: Punarādhāna shows near-identical MS--KS overlap (up to 93.5%) with sharp divergence from TS, while Agnyupasthāna displays moderate, broadly distributed similarity across all three pairs. These contrasting patterns provide quantitative evidence that different ritual categories followed distinct paths of textual transmission within the Yajurvedic tradition. ICoMa and the experimental data are freely available.

Quantifying Text Reuse Across Three Kṛṣṇa Yajurveda Recensions: Using Multi-Algorithm Computational Collation

Ancient and endangered languages pose a unique challenge for NLP: their datasets are inherently scarce, difficult to expand, and built from formulaic corpora---making data-quality issues especially consequential yet rarely audited. Motivated by the need to understand what current NMT can realistically achieve for such languages, we investigate hieroglyphic-to-German translation, where a recent study reported 61.5 BLEU using fine-tuned M2M-100. Our reproduction yields only 37.0 BLEU with the released model. Investigating this gap, we find \textbf{32\% of test targets appear identically in training} (16/50; 50\% under 8-gram overlap at 70\% threshold). This contamination inflates scores dramatically: contaminated samples achieve up to 83.8 BLEU / 0.924 COMET-22 versus 30.9--39.2 BLEU / 0.622--0.676 COMET-22 on clean samples across five model configurations spanning two architectures. Document-level decontamination reduces contaminated BLEU by only 4.6 points because 8/16 targets persist via other source documents---target-level deduplication is required. We release a decontaminated 34-sample test set and establish corrected baselines (30.9--39.2 BLEU), providing a realistic assessment of NMT capability for this endangered writing system.

Downloads

Next from ACL 2026

RspectNLP@LT-EDI 2026: Rubric-Driven Prompting for Safe Multilingual Counter Narrative Generation

Similar lecture

JusticeBots@LT-EDI 2026: Prompt-Based Counter-Narrative Generation for Homophobia and Transphobia Comments

.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 2026

RspectNLP@LT-EDI 2026: Rubric-Driven Prompting for Safe Multilingual Counter Narrative Generation

Similar lecture

JusticeBots@LT-EDI 2026: Prompt-Based Counter-Narrative Generation for Homophobia and Transphobia Comments

Downloads