THESAURI AND AI

THE ROLE OF SEMANTIC STRUCTURE IN INTERPRETATION

Abstract

The incorporation of Artificial Intelligence-based language models into information search, retrieval, and representation activities has broadened discussions on semantics, cultural diversity, and conceptual mediation in Knowledge Organization Systems. This study seeks to analyze, from a Knowledge Organization perspective, differences in the processes of conceptual interpretation and subject representation produced by a language model when operating with and without access to the semantic structure of a culturally situated thesaurus, taking as its empirical domain the Amazonian cultural heritage represented in the Yãkoana – Thesaurus of the Cultural Heritage of Amazonas. The qualitative, exploratory, and analytical study conducted an experiment with GPT-5 under three conditions: interpretation of terms without access to the thesaurus, interpretation mediated by its semantic structure, and subject representation of texts through descriptor selection. Reflexive thematic analysis identified three analytical dimensions: conceptual preservation, preservation of sociocultural context, and mobilization of the vocabulary’s semantic relations. The results indicate that the availability of the semantic structure was associated with more clearly delimited and contextualized interpretations and provided terminological and relational references for subject representation. They also showed that the conformity of descriptors with the vocabulary does not, in itself, ensure their thematic relevance to the content of the documents. It is concluded that controlled vocabularies can function as structures of semantic mediation in language model-based applications, guiding interpretation and representation processes without fully determining the responses produced.