Emotivations: Data and Materials for an Inductive Study of the Lexical Structure of Reversal Theory Emotional Families

This deposit provides the data and materials needed to inspect and reproduce a study mapping everyday emotional expressions onto the sixteen emotional families of Reversal Theory (RT), coded B01–B16. Large language models are used as analytical instruments under strict anti-circularity safeguards (the system that generates material never judges it in the same session, and no generation or clustering prompt names a target emotion); a non-LLM sentence-embedding pipeline (SBERT → UMAP → HDBSCAN) serves as an independent computational anchor. The deposit contains several complementary components: an English inductive study on an AI-generated corpus (398 expressions) with three-system clustering and blind RT mapping; a lexical taxonomy of the sixteen families and its validation; a blind human-coding study by four named raters; a replication on 525 real, sourced public statements; and a single-model, native-French replication using Mistral, with its own non-LLM anchor. Headline results: 14–15 of the 16 families recovered in English (mean cluster purity 91.3%); the non-LLM anchor recovers a neighbouring structure with no language model; human raters fall within the expected range of agreement; and the French replication shows 16/16 recognition and 11/16 emergence of the same families (lexicon purity 81.6%). A README and a codebook document every folder and file; an ethics note covers the named raters (with consent) and the sourcing of public statements. Prepared in connection with a paper submitted to the Reversal Theory Society Conference (Cambridge, Trinity Hall, 2026).