AlignCultura: Towards Culturally Aligned Large Language Models?

Gautam Siddharth Kashyap, Mark Dras, Usman Naseem · 2026

Read AlignCultura: Towards Culturally Aligned Large Language Models? by Gautam Siddharth Kashyap, Mark Dras, Usman Naseem (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-19016.txt.

Introduction

Kashyap, Dras, and Naseem build Align-Cultura for cultural alignment in language models. No benchmark tests cultural alignment against UNESCO diversity principles under the HHH paradigm. Stage one builds CULTURAX, an English set grounded in the UNESCO taxonomy, reclassifying prompts and curbing leakage with SimHash. Stage two benchmarks it on general, tuned, and open weight models. Tuned models improve joint HHH by 4 to 6 percent, cut failures by 18 percent, and hold leakage to 0.3 percent while gaining 10 to 12 percent in efficiency.

Catalog record

CREATORGautam Siddharth Kashyap, Mark Dras, Usman Naseem
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37460
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.19016

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

AlignCultura: Towards Culturally Aligned Large Language Models? Gautam Siddharth Kashyap, Mark Dras, Usman Naseem Abstract Cultural alignment in Large Language Models (LLMs) is essential for producing contextually aware, respectful, and trustworthy outputs. Without it, models risk generating stereotyped, insensitive, or misleading responses that fail to reflect cultural diversity w.r.t Helpful, Harmless, and Honest (HHH) paradigm. Existing benchmarks represent early steps toward cultural alignment; yet, no benchmarks currently enables systematic evaluation of cultural alignment in line with UNESCO’s 1 1 1 A globally recognized taxonomy– diverse cultural– diverse regions. principles of cultural diversity w.r.t HHH paradigm. Therefore, to address this gap, we built AlignCultura 2 2 2 Data is available at: https://github.com/gskgautam/AlignCultura , two-stage pipeline for cultural alignment. Stage I constructs CulturaX , the HHH-English dataset grounded in the UNESCO cultural taxonomy, through Query Construction , which reclassifies prompts, expands underrepresented domains (or labels), and prevents data leakage with SimHash. Then, Response Generation pairs prompts with culturally grounded responses via two-stage rejection sampling. The final dataset contains 1,500 samples spanning 30 subdomains of tangible and intangible cultural forms. Stage II benchmarks CulturaX on general-purpose models, culturally fine-tuned models, and open-weight LLMs (Qwen3-8B and DeepSeek-R1-Distill-Qwen-7B). Empirically, culturally fine-tuned models improve joint HHH by 4%–6%, reduce cultural failures by 18%, achieve 10%–12% efficiency gains, and limit leakage to 0.3%. Query: Should alcohol be served at all family gatherings? Instruction: Produce a response considering Helpful-Harmless-Honest (HHH) principles under cultural norms. Without Culturally Grounded HHH: Alcohol should either always be served or always avoided at family gatherings, regardless of cultural, religious, or social context. With Culturally Grounded HHH: Practices around alcohol vary across cultures and families. A respectful approach considers cultural, religious, and personal preferences, ensuring inclusive and comfortable participation. Figure 1: Illustration of the role of cultural diversity w.r.t HHH. For the same prompt, responses without tend to be rigid or universalized, whereas responses with tend to be context-sensitive and inclusive guidance. Figure 2: UNESCO Framework for Cultural Statistics (UFCS) taxonomy, outlining 9 high-level cultural domains and 46 subdomains of tangible and intangible cultural forms. 1 Introduction Cultural alignment in Large Language Models (LLMs) is crucial for producing contextually aware, respectful, and trustworthy outputs. Without it, models risk generating stereotyped, insensitive, or misleading responses that fail to reflect cultural diversity. According to UNESCO, cultural diversity are central to equitable knowledge exchange UNESCO 2009 —principles ( w.r.t Helpful, Harmless, and Honest (HHH) 3 3 3 According to Askell et al. 2021 , Helpfulness requires the model to provide responses that meaningfully addresses the user’s query; Harmlessness requires avoiding misleading or harmful responses; and Honesty requires factual accuracy and transparency in responses. paradigm Kashyap et al. 2026 ; Kashyap et al. 2025 –see Figure 1 ) that LLMs must uphold to remain globally inclusive and ethically grounded. Existing benchmarks such as CAReDiO Yao et al. 2025 , CIVICS Pistilli et al. 2024 , CVC Wu et al. 2025 , DIWALI Sahoo et al. 2025 , CulturalBench Chiu et al. 2025 and the Community Alignment Dataset Zhang et al. 2025 represent early steps toward cultural alignment but remain limited—focusing on single domain, or omitting systematic HHH evaluation. Furthermore, prominent datasets— Alpaca (Helpful) Taori et al. 2023 , BeaverTails (Harmless) Ji et al. 2023 , and TruthfulQA (Honesty) Lin et al. 2022 —address individual HHH dimensions but overlook the cultural foundations that fails to reflect cultural diversity. Figure 3: Overview of the AlignCultura pipeline. Stage I constructs CulturaX through two modules: (i) Query Construction , and (ii) Response Generation via a two-stage rejection sampling process— Candidate Filtering and Feedback-Guided Resampling . Stage II benchmarks general-purpose, culturally fine-tuned, and open-weight LLMs on CulturaX . Therefore, to address this gap, we built AlignCultura, two-stage pipeline for cultural alignment. Stage I constructs CulturaX , the HHH-English dataset for cultural alignment–through Query Construction –where prompts are drawn from Cultural Kaleidoscope Banerjee et al. 2025 4 4 4 We select Cultural Kaleidoscope as it provides broad, systematically curated coverage of cultural norms than earlier cultural resources (e.g., CAReDiO Yao et al. 2025 , CIVICS Pistilli et al. 2024 , CVC Wu et al. 2025 , DIWALI Sahoo et al. 2025 ), which are limited to single cultures. While not originally designed for HHH alignment or UNESCO domains, it offers a richer foundation. , and reclassified into taxonomies defined by the UNESCO Framework for Cultural Statistics (UFCS) 5 5 5 https://unesdoc.unesco.org/ark:/48223/pf0000395490 (see Figure 2 )—covering both tangible (e.g., artifacts, monuments, recorded works) and intangible (e.g., traditions, practices, transmitted knowledge) forms of culture Grammalidis et al. 2016 . Classification is performed using Mistral-7B-Instruct-v0.3 6 6 6 https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3 Naseem et al. 2026 ; AI 2023 , which maps prompts into cultural domains (or labels) Tsoumakas et al. 2010 ; the 9 domains each encompass 46 subdomains. To balance underrepresented domains, Llama-3.1-8B-Instruct 7 7 7 https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct Meta AI 2024 is applied for query expansion (via SimHash fingerprint). Furthermore, in Response Generation , these prompts are paired with culturally grounded responses generated by LLM, filtered through a two-stage rejection sampling process to enforce HHH quality. Stage II establishes the systematic HHH evaluation framework for cultural alignment by benchmarking general-purpose models, culturally fine-tuned models, and open-weight LLMs (Qwen3-8B and DeepSeek-R1-Distill-Qwen-7B) on CulturaX . In summary, our contributions are twofold: • Construction of CulturaX , the HHH-English dataset for cultural alignment, with 1500 samples spanning 9 domains and 30 subdomains, alongside a systematic HHH benchmarking framework. • Empirically, culturally fine-tuned models improve joint HHH by 4%–6%, reduce cultural failures by 18%, achieve 10%–12% efficiency gains, and limit leakage to 0.3%. 2 Related Works General-Purpose Alignment. As outlined in Section 1 , much of the alignment literature has relied on single-dimension datasets, e.g., RAHF Liu et al. 2024 for Helpfulness and Aligner Ji et al. 2024 for Harmlessness and Honesty. More recently, multi-dimension works such as MARL-Focal Tekin et al. 2025 , TrinityX Kashyap et al. 2025 , and H 3 Fusion Tekin et al. 2026 attempt joint optimization across the HHH paradigm Naseem et al. 2025 . While these works demonstrate the feasibility–they remain general-purpose and lack grounding in specific knowledge domains. Cultural-Specific Alignment. Several works have sought to adapt LLMs to cultural contexts as discussed in Section 1 such as mitigating cultural bias in multilingual models Weidinger et al. 2021 , aligning models to culturally diverse safety preferences Ganguli et al. 2022 , or fine-tuning dialogue systems to reflect cultural norms Pujari and Goldwasser 2024 . However, these works remain piecemeal–they often target specific cultural groups, or emphasize safety over balanced HHH paradigm. Furthermore, CDEval Wang et al. 2024 argues that fixed alignment dimensions can be misleading in culturally pluralistic settings, where values and norms are inherently diverse and sometimes conflicting Naseem 2026 . Our work does not claim to resolve cultural pluralism ; instead, we treat HHH as culturally mediated dimensions whose interpretation depends on contextual norms—a different directions than CDEval Wang et al. 2024 . 3 Methodology Overview of the Pipeline. AlignCultura comprises two stages (see Figure 3 ) that operationalize the motivating difference illustrated in Figure 1 by enabling systematic evaluation w.r.t the HHH paradigm. Stage I constructs CulturaX via culturally grounded query construction and response generation with quality-controlled filtering, while Stage II establishes a benchmarking

Read the complete text