2026-10-01T15:59:40Z https://dcpapers.dublincore.org/oai

oai:dcpapers.dublincore.org:952647402 2026-09-10 dcmi-2026 openaire

When Meaning Gets Lost in Translation: Evaluating Semantic Alignment of LLM-Generated Dublin Core Metadata for Korean Cultural Heritage Choi, Yiseul Cultural heritage metadata is not merely descriptive — it encodes the values, identity, and memory of communities. As large language models (LLMs) are increasingly applied to automate Dublin Core metadata generation, a critical question emerges: can AI-generated metadata preserve the cultural meaning that human catalogers carefully construct? This poster presents a research-in-progress study investigating the semantic gap between LLM-generated and expert-curated Dublin Core records for Korean cultural heritage objects. We introduce a two-dimensional meaning alignment evaluation framework and a concept we term semantic flattening — the systematic erasure of culturally specific meaning under the pressure of globally dominant classification defaults. A research design is outlined to examine how addressing semantic flattening contributes to realizing the DCMI 2026 vision of Meaning-Driven AI: Using Metadata to Align Systems with Human Values. Dublin Core Metadata Initiative 2026-09-10 info:eu-repo/semantics/conferenceObject Text application/pdf text/html https://doi.org/10.23106/dcmi.952647402 https://dcpapers.dublincore.org/article/952647402 Dublin Core Metadata Initiative Conference Proceedings eng info:eu-repo/semantics/openAccess https://creativecommons.org/licenses/by/4.0/