Toward Comprehensive Cultural Alignment in LLMs: An Interdisciplinary Framework for Cultural Evaluation

AAAI 2026 Workshop AIGOV Submission40 Authors

Keywords: cultural alignment, large language models, evaluation framework, cultural governance

TL;DR: Current cultural alignment evaluations rely on reductive survey metrics. We present a multi-dimensional framework translating theories from anthropology, sociology, and media studies into concrete evaluation tasks and benchmarks for LLMs.

Abstract: Evaluating cultural alignment in large language models (LLMs) requires a nuanced, multidisciplinary understanding of what “culture” entails. In this paper, we draw from anthropology, sociology, political science, and media theory to propose a broader conceptualization of culture as symbolic meaning-making, embodied social practice, negotiated discourse, institutional structure, strategic repertoire, algorithmic mediation, and power-laden representation. We map these conceptual perspectives to representative datasets and propose corresponding evaluation strategies to assess how LLMs handle context sensitivity, temporal change, behavioral norms, and epistemic asymmetries. By grounding evaluation in interdisciplinary cultural theories, we provide a structured framework for the design and development of comprehensive cultural alignment benchmarks for LLMs.

Submission Number: 40