Comparing LLM-Assisted and Human-Authored Privacy Requirements Catalogs: An Exploratory Bilingual Case Study (DATASET)

This Zenodo record provides the replication package for an exploratory bilingual case study comparing privacy requirements catalogs generated under the evaluated RequiCreator configuration with the English and Spanish versions of the human-authored PDP2019 reference catalog. The package includes (i) the RequiCreator-generated privacy requirements catalogs, (ii) the two language versions of the PDP2019 reference catalog, and (iii) the complete evaluation outputs (per-catalog results and SPSS Statistics projects) organized across four analysis branches: Ambiguity, Consistency, Readability, and Redundancy. The evaluation workflow is structured by branch, each containing analysis outputs (results/) and the corresponding statistical artifacts (spss_statistics/). Additionally, the Consistency branch includes traceability outputs produced by RequiTrace, linking requirements to their normative sources as part of the semantic-alignment and traceability analysis. Package contents (high-level): ai_catalogs/: RequiCreator-generated JSON privacy requirements catalogs. human_catalogs/: English and Spanish JSON versions of the human-authored PDP2019 reference catalog. validation/: Branch-wise validation outputs for ambiguity, consistency (incl. traceability), readability, and redundancy; includes per-run results and SPSS projects.