Cultural Awareness is Represented but Not Decoded: Tracing Mythological Knowledge across 18 Open-Source LLMs

2026-08-03

AI Summary

This study addresses the challenge that open-source large language models struggle to accurately recognize and generate deity names from non-Western mythologies, such as Finnish, Slavic, Egyptian, or Chinese traditions. By employing linear probing, logit lens analysis, activation patching, and output extraction across 18 open-source models, the work systematically investigates how cross-cultural mythological knowledge is internally represented and decoded. The findings reveal that cultural bias predominantly arises during the decoding phase rather than in internal representations, and that the language of the prompt exerts a gating effect on decoding accuracy. Residual stream representations effectively distinguish entities across cultures, substantially outperforming baselines relying solely on name strings. The paper contributes the first entity-level probe–output decomposition framework and a benchmark for cross-cultural mythological entity annotation.

Abstract

Open-source LLMs reliably name Zeus, Jupiter, and Thor, but recover their counterparts in less-represented traditions like Finnish, Slavic, Egyptian, or Chinese mythology far less consistently. We ask where inside the model this cultural default is produced. On a parallel cross-cultural substrate of Thompson-motif entities, we instrument 18 open-source LLMs from 8 architecture families with linear probing, logit lens, activation patching, and output extraction. The residual stream cleanly distinguishes cultures, well above a name-string baseline, yet the decoder collapses culturally-specific tokens onto dominant-tradition ones. The failure is at readout, not at representation. Asking the same question in the target culture's native language versus English produces failures that cluster within language but decouple across language: the decoder is gated on prompt language. We release a per-entity (probe, output) decomposition framework, a citation-anchored cross-cultural ground truth, a within- versus cross-mode correlation test for language-conditioned readout, and per-entity predictions for all 18 models.

Problem

• cultural awareness
• mythological knowledge
• large language models
• cross-cultural representation
• decoder bias

Innovation

• cultural awareness
• linear probing
• decoder bias
• cross-cultural representation
• language-conditioned readout