2026· Proceedings of the 1st Workshop on Multilinguality in the Era of Large Language Models (MeLLM 2026)· pp. 1-13· 0 citations· 47 references
TL;DR
The MSAG is formalized as a framework for analyzing systematic weaknesses in multilingual safety annotation pipelines, identifying four sources of bias: language coverage gaps, dialect representation gaps, cultural semantic gaps, and annotation guideline gaps.
Abstract
Large Language Models are increasingly used as safety infrastructure for detecting harmful online content and moderating social media across multiple languages. Yet their effectiveness remains uneven across linguistic communities. This disparity reflects not only disparities in training data availability but also structural problems in annotation design. We argue that a central source of multilingual safety failure lies in the annotation gap underlying existing hate speech datasets. Most annotation guidelines and safety benchmarks are developed for English and standard language varieties, overlooking dialectal variation and culturally embedded forms of hostility. Using Arabic dialectal discourse as a case study, we show how harmful speech expressed through dialects, sarcasm, code-switching, and culturally specific expressions often remains undetected by current annotation schemes. We introduce the concept of the Multilingual Safety Annotation Gap (MSAG), identifying four sources of bias: language coverage gaps, dialect representation gaps, cultural semantic gaps, and annotation guideline gaps. We discuss implications for LLM safety alignment and outline directions for culturally grounded multilingual annotation. This paper is primarily a conceptual and methodological position paper; rather than introducing a new benchmark or empirical evaluation, we aim to formalize the MSAG as a framework for analyzing systematic weaknesses in multilingual safety annotation pipelines.
Current safety alignment training for Large Language Models (LLMs) are heavily English-centric. When such safety filters fail for non-English languages, the consequences are immediate and user-facing: voice assistants and spoken dialogue systems may produce stereotype-reinforcing outputs, bypassing the standard English-focused safety alignments and propagating harmful bias to non-English speaking communities. For spoken language technologies deployed across India's linguistically diverse population, this represents a critical failure mode. To address this cross-lingual gap, we introduce INCLUDE (Indian Cultural Lens for Understanding and Detecting Embedded Biases), a multilingual evaluation benchmark designed to quantify Indian-centric socio-cultural biases. INCLUDE consists of 2,604 prompts spanning six prompt languages: English, Hindi, Bengali, Marathi, Tamil, and Hinglish (Hindi-English code-mix). We evaluate ten open- and closed-source LLMs against this benchmark, analyzing 14,988 bias scores. Our statistical results reveal two key findings. First, Bengali yielded the highest average bias score in open-source models. Second, English demonstrated a notable reversal, producing the lowest bias in open-source models but the highest bias in closed-source models.
This work investigates cross-lingual safety transfer in four African languages, Twi, Hausa, Amharic, and Swahili, using LoDNA, a new safety dataset that pairs literal translations with culturally localized prompts to demonstrate superficial safety alignment.
Abigail Oppong, P SAM SAHIL, Tadesse Destaw Belay et al.· 0 citations
SurakshaEval is introduced, a novel safety benchmark composed of human-written prompts spanning real-world scenarios, explicitly designed for ten major Indian languages - Assamese, Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Punjabi, Tamil, and Telugu - along with English.
Debopriyo Banerjee, K. R. Kavitha, Angana Borah et al.· 0 citations
Large language models (LLMs) demonstrate strong Helpfulness, Harmlessness, and Honesty (3H) alignment in English-centric settings, but these gains transfer poorly to low-resource languages due to cultural mismatches. Existing multilingual 3H benchmarks rely predominantly on automated translation or LLM based synthesis, propagating source-language biases while sacrificing local relevance. To address this gap, we introduce Pak3H1, the first human-validated, culturally contextualized Urdu benchmark suite for 3H alignment, comprising PakAlpaca (helpfulness), PakBeaverTails (harmlessness), and PakTruthfulQA (honesty). Our multi-stage pipeline integrates manual cultural adaptation and dictionary-guided post editing to prioritize native speaker judgment, ensuring both semantic fidelity and contextual authenticity. Zero-shot evaluations across multiple open and proprietary LLM architectures reveal systematic cross-lingual alignment gaps: helpfulness win rates decline under localized contexts, harmlessness guardrails break down against regional safety risks, and composite honesty metrics degrade substantially due to localized factual constraints. These findings expose structural limitations in current alignment approaches, underscoring the necessity of human-guided localization for equitable multilingual evaluation.
Abdullah Hashmat, Usman Naseem, Agha Ali Raza· 0 citations
India is a vast nation of over 1.4 billion people, varied by hundreds of diverse and locally specific traditions and cultures and 22 officially recognized languages. Large language models (LLMs) are now being deployed on a massive scale throughout the mainland as well as in remote villages. However, the common benchmarks - MMLU, BIG-Bench, and TruthfulQA are almost exclusively English- and Western-centric. They do not identify those safety, fairness, and accuracy failures unique to the Indian context. That is the gap Inspect India Evals seeks to fill. It is an open-source framework built on top of UK AISI's Inspect AI platform. It has six benchmarks: Multilingual MMLU across sixteen Indian languages, BharatBBQ (our adaptation of BBQ for Indian social bias), a safety evaluation for Digital Public Infrastructure, a multilingual safety test using harmful prompts in Indian languages, a multi-turn jailbreak resistance test, and an Indian cultural knowledge benchmark scored using LLM-as-judge rubrics. In this study, we tested five open-weight models ranging from 8B to 32B parameters. Sarvam-M 24B and Gemma 2 27B came out on top, both scoring 80% on the composite India Fairness Index, with Sarvam-M even beating larger 32B models on Indian cultural knowledge and DPI safety compliance. All models scored 100% refusal on Multilingual Safety, whereas DPI safety varied from 20% to 100%. The framework is public. It's built to work with the UK AISI registry. Anyone can reproduce or extend this work.
Abhishek Kumar Singh, S. Nag, Sachita et al.· 0 citations
As the adoption of large language models (LLMs) grows in Arabic-speaking regions, ensuring their safety and cultural alignment is increasingly critical. However, Arabic LLM safety remains underexplored, especially in adversarial evaluation settings. We introduce the Arabic Safety Index (ASAS), the first fully human-curated Arabic benchmark for redteaming LLMs. ASAS contains 801 prompts spanning 8 safety categories and 8 attack strategies, with ideal responses in Modern Standard Arabic (MSA). We conduct a redteaming evaluation across seven leading models with Arabic capabilities, including GPT-4o, Claude 3.7 Sonnet, and regional models such as ALLaM and FANAR. Human annotators rate responses using a structured 4-point safety scale, revealing that most models fail to defend against 50% of unsafe prompts. Our findings highlight major safety gaps in high-harm categories such as weapons and illicit substances, with direct and obfuscation-based attacks proving most effective. The results also show that language alignment does not readily transfer across languages, and that automated safety judges (e.g., GPT-4o) perform poorly compared to human annotators. ASAS provides a culturally grounded benchmark and redteaming protocol to drive progress in Arabic LLM safety.
Fidaa Abed, Haidar Khan, M Saiful Bari et al.· 0 citations