Bu, Fan, Zheng Wang, Siyi Wang, and Ziyao Liu. 2025. “An Investigation into Value Misalignment in LLM-Generated Texts for Cultural Heritage.” IEEE Transactions on Emerging Topics in Computational Intelligence, 1–15. https://doi.org/10.1109/TETCI.2025.3597289.

Notes

Misalignment types

• Detail Inaccuracy
• Cultural misunderstanding
• Knowledge gap
• Premature certainty
• Cultural reductionism
• Historical bias

In-text annotations

"RQ1: Misalignment Type" (Page 7)

"Detail inaccuracy (VM1): One of the common issues in LLM-generated texts is the frequent occurrence of inaccuracies in details such as timelines, locations, characters, or causal relationships between events." (Page 7)

"Cultural misunderstanding (VM2): LLMs often demonstrate a lack of understanding when explaining or describing concepts, symbols, or phenomena related to cultural heritage." (Page 8)

"Knowledge gap (VM3): The knowledge gap refers to the tendency of LLMs to either hallucinate, i.e., fabricate information to generate a seemingly plausible answer, or refuse answers, i.e., admit their lack of knowledge, when queried about specific concepts of cultural heritage, cultural practices, or meanings." (Page 8)

"Premature certainty (VM4): LLMs often fail to account for unresolved or controversial topics related to cultural heritage within the academic community, providing overly definitive answers that oversimplify complex subjects." (Page 8)

"Cultural reductionism (VM5): Cultural reductionism refers to simplifications of complex cultural phenomena by reducing them to a single characteristic or description, thereby overlooking their inherent diversity and complexity." (Page 8)

"Historical bias (VM6): Historical bias refers to emphasizing certain aspects while downplaying or omitting others." (Page 8)

"Selective narration (VM7): Our observations indicate that LLM-generated texts often reflect dominant historical narratives while marginalizing alternative perspectives or contested interpretations." (Page 9)