Scholarlink

This study explores the alignment of cultural values in various language models through a unique application of Hofstede’s cultural dimensions. Utilizing a set of one-shot questions derived from Hofstede’s framework, we conducted an empirical investigation into how different artificial intelligence systems, including GPT-4o, mini GPT, YandexGPT, GigaChat, Claude Sonnet, and Gemini, conceptualize and express cultural values. Our findings reveal distinct alignments of these models with various cultural profiles: for instance, YandexGPT and GigaChat show collectivist tendencies, while Gemini and Claude Sonnet lean toward individualistic Western values. The study’s novelty lies in its inclusion of underrepresented Russian models and a multilingual approach across five culturally diverse languages. This work contributes to the understanding of AI’s sociocultural alignment and offers practical implications for global deployment of AI systems. Academically, it provides an interdisciplinary bridge between computer science and cultural sociology; practically, it supports AI developers in designing culturally sensitive technologies.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Operationalizing Cultural Values in Language Models: A Comparative Analysis Using Hofstede’s Dimensions

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Alexander Didenko,
  • Alexander Balakir,
  • Vladislav Zapilikhin,
  • Anna Shabanova

摘要

This study explores the alignment of cultural values in various language models through a unique application of Hofstede’s cultural dimensions. Utilizing a set of one-shot questions derived from Hofstede’s framework, we conducted an empirical investigation into how different artificial intelligence systems, including GPT-4o, mini GPT, YandexGPT, GigaChat, Claude Sonnet, and Gemini, conceptualize and express cultural values. Our findings reveal distinct alignments of these models with various cultural profiles: for instance, YandexGPT and GigaChat show collectivist tendencies, while Gemini and Claude Sonnet lean toward individualistic Western values. The study’s novelty lies in its inclusion of underrepresented Russian models and a multilingual approach across five culturally diverse languages. This work contributes to the understanding of AI’s sociocultural alignment and offers practical implications for global deployment of AI systems. Academically, it provides an interdisciplinary bridge between computer science and cultural sociology; practically, it supports AI developers in designing culturally sensitive technologies.