# Attribution and third-party sources

The Chinese Name Corpus is released under **CC BY 4.0** (see `LICENSE`). That licence covers the
selection, structure and writing in this package. The underlying facts and the third-party data
they were derived from carry their own terms, listed here.

If you redistribute this package, keep this file with it.

---

## 1. Real-name frequency counts — `data/name_character_counts.jsonl`

- **Source:** [`wainshine/Chinese-Names-Corpus`](https://github.com/wainshine/Chinese-Names-Corpus),
  file `Chinese_Names_Corpus_Gender` (1,144,226 records).
- **Licence:** Apache License 2.0.
- **What we did (statement of changes, required by Apache-2.0 §4b):** we did **not** redistribute
  the corpus. We counted, for each character, how often it appears as the first character of a
  two-character given name, as the second character, as a single-character name, and in names
  labelled male / female. Only those aggregate integers appear here.
- **Attribution:** the counts derive from work by the Chinese-Names-Corpus maintainers. Any use of
  `name_character_counts.jsonl` should credit them alongside this corpus.

## 2. Classical source texts — `sources/`

The works themselves (Analects, Mencius, the Great Learning, the Doctrine of the Mean, the Book of
Songs, the Songs of Chu, and Tang poems) are **public domain**. The digitised transcriptions we
ship are derived from:

| Work | Transcription source | Licence |
|---|---|---|
| 论语 · 孟子 · 大学 · 中庸 · 诗经 · 楚辞 · 全唐诗 | [`chinese-poetry/chinese-poetry`](https://github.com/chinese-poetry/chinese-poetry) | MIT |

MIT requires the copyright notice be retained; see `sources/MANIFEST.json` for the exact files and
their SHA-256 hashes. We keep the original character forms (`orig`) alongside the simplified
reading (`text`) so the transcription can be checked against a print edition.

## 3. Character metadata — `data/characters.jsonl`

| Field | Source | Licence / nature |
|---|---|---|
| `strokes` | Unicode **Unihan** `kTotalStrokes` | Unicode License v3 |
| `radical_rs` | Unicode **Unihan** `kRSUnicode` | Unicode License v3 |
| `table_level` | 《通用规范汉字表》 (2013) | Public government standard; factual |
| `pinyin_tone3` | [`pypinyin`](https://github.com/mozillazg/python-pinyin) | MIT |
| `zipf` | [`wordfreq`](https://github.com/rspeer/wordfreq) | Apache-2.0 |
| `rarity` / `rarity_label` | derived from `zipf` by our thresholds | our work, CC BY 4.0 |

## 4. Surname frequency tiers — `data/surnames.jsonl`

`population_tier` follows the ordering published in 公安部《二〇一九年全国姓名报告》. We publish
**tiers, not ranks**: within a tier the order changes depending on the statistics consulted, so a
rank number would be false precision.

## 5. Written by us

`data/characters_deep.jsonl`, `data/quotations.jsonl` (selections and English translations, except
where `translation_attrib` says otherwise), `data/homophones.jsonl`, `data/combos.jsonl`,
`sources/MANIFEST.json`, `tools/verify_quotes.py`, and all prose in this package.

Quotation translations marked `our translation` are ours and are CC BY 4.0.

---

## Not included, and why

- **The character glosses from our legacy library** (`char_meaning.json`, 2,028 entries). Their own
  metadata records `provenance = legacy, verified: false` — the source bibliographies were never
  checked one by one. Publishing them under a licence that implies verification would be a false
  claim. They are being re-researched character by character; the 143 verified ones are in
  `characters_deep.jsonl`.
- **The raw 1.14 million name corpus.** Apache-2.0 permits redistribution with notices, but the
  counts answer the question people actually ask ("is this character used in names, and how often")
  without shipping a file full of real people's names.

## Corrections

A misattributed source is a defect of the same severity as a wrong quotation. Write to
support@uion.top and it will be corrected in the next version.