Five‑Tails Multi‑Turn Visual Reasoning

- Overview: Five‑Tails Multi‑Turn Visual Reasoning (5T‑MVR) evaluates five prompting conditions — BASELINE, OBSERVE, ANALYZE, EVALUATE, INTERPRET — across 5 dialogue turns with an image stimulus. This deposit provides scripts to generate JSON logs via OpenAI API, analyze all turns, and produce figures and summary tables.- Included: `scripts/` (generation + analysis + visualization), `data/raw/` (sample JSON logs), `data/stimuli/` (example image), `results/` (CSV and plots), `README.md` and `README_jp.md` (how to run), `CITATION.cff`, `LICENSES/`.- Reproducibility: Python 3.9+. Install with `python3 -m pip install -r requirements.txt`. Run analysis with `python3 scripts/analyze_5tails_allturns.py --in ./data/raw --out ./results` and `python3 scripts/visualize_5tails_results.py`. Optionally generate logs with `python3 scripts/run_gpt5_5tails.py --image ./data/stimuli/stimulus.jpeg --model gpt-5 --replicates 10 --out .`.- Outputs: Condition×turn summary (`results/summary_by_condition_turn.csv`), final‑turn summary (`results/final_turn_summary_by_condition.csv`), per‑turn metrics (`results/metrics_per_turn.csv`), and figures in `results/plots/` (length trends, category densities, element coverage, etc.).- Methods: Lightweight dictionary‑based counts (no morphological analysis) for rapid reporting; cumulative statistics script available for 5‑turn aggregates.- License: Code under MIT (`LICENSES/LICENSE_CODE_MIT`); data/text/figures under CC BY 4.0 (`LICENSES/LICENSE_CONTENT_CC-BY-4.0.txt`). Please cite “Five‑Tails Multi‑Turn Visual Reasoning (5T‑MVR)” (see `CITATION.cff`).- Authors: Keisuke Sato (National Institute of Technology, Ibaraki College; `[e-mail]`).- Keywords: visual reasoning, multi‑turn dialogue, evaluation, analysis, image stimulus, prompting.