Supplementary Material: When the Machine Cannot Feel 'Taliata'

Supplementary material for the paper "When the Machine Cannot Feel 'Taliata': Antonino Contiliano's Human–AI Experiments as Evidence for the Structural Convergence of Large Language Model Creativity," submitted to NL4AI 2026 (9th Workshop on Natural Language for Artificial Intelligence, co-located with AIxIA 2026, Perugia, Italy). This dataset contains the full experimental corpus and analysis code used in the paper: Human corpus (H): 12 antagonistic poems by Peter Genito (4) and Antonino Contiliano (8), covering four thematic axes (language and machine, capital and resistance, embodiment/dialect, digital consensus). LLM corpus (M): 48 poems generated by four large language models (Qwen, ChatGPT, Gemini, Meta AI) on matched thematic prompts, across three independent runs, used to test convergence and homogenization effects in machine-generated poetic text. Analysis code: Python scripts (corpus.py, analyze.py) computing distinct-n lexical diversity, within-prompt cross-model TF-IDF cosine similarity, and within-model self-consistency, reproducing all quantitative results (Sections 4–5) reported in the paper.