The dataset
Every secret in 20Q comes from one public, open dataset: concrete, commonly-known things a normal person can hold in mind and answer twenty questions about. No trick answers. It doubles as the benchmark sample, the play whitelist, and fine-tuning teacher data.
–
secrets and counting
Categories
animalfoodhouseholdtoolvehicleclothingplacenaturebodyinstrumentsporttechabstract-lite
Format
{"version": 1, "license": "CC-BY-4.0", "items": [
{"id": "lasagna", "name": "lasagna",
"category": "food", "difficulty": "medium",
"aliases": ["lasagne"], "adult_ok": true}
]}
Licensed CC-BY-4.0 — use it, remix it, cite it. Difficulty labels: easy / medium / hard. Aliases count as correct guesses.