kaihanzi/hanzi-sino-vietnamese

HSK × Sino-Vietnamese (Hán-Việt) character dataset

768 HSK characters joined with their Sino-Vietnamese (Hán-Việt) readings, radical breakdowns and hand-written memory hooks in Vietnamese.

Open HSK wordlists are plentiful. The Sino-Vietnamese layer is what is missing from all of them — and it is the layer that matters most for the ~1 million Vietnamese speakers studying Chinese, because roughly 60% of Vietnamese vocabulary descends from Chinese. A learner meeting 学 (xué) already knows the morpheme through học — học sinh, đại học, học phí. Attach the reading and thousands of characters stop being arbitrary shapes.

Every sinoViet reading, mnemonic, radical gloss and stroke hint is hand-authored in Vietnamese, not model-generated and not scraped.

Load

from datasets import load_dataset

chars = load_dataset("<your-org>/hanzi-sino-vietnamese", "characters", split="train")
rads  = load_dataset("<your-org>/hanzi-sino-vietnamese", "radicals",   split="train")

chars[0]
# {'hanzi': '一', 'pinyin': 'yī', 'sino_viet': 'Nhất', 'meaning_vi': 'một, số một',
#  'stroke_count': 1, 'hsk_level': 1, 'radicals': '一', 'monolithic': True, ...}

JSON equivalents (data/characters.json, data/radicals.json) and per-level splits (data/by-level/hsk1.json … hsk6.json) ship in the same repo.

Schema

characters

radicals

Level distribution

HSK 4–6 coverage is deliberately partial: those levels only include characters that were prioritised for the source site, so do not treat this as a full HSK 4–6 character list.

Uses

• —Vietnamese-facing Chinese-learning apps, Anki decks, flashcard generators
• —Research on Sino-Xenic readings — Sino-Vietnamese ↔ Sino-Japanese on'yomi ↔ Sino-Korean hanja
• —Any task needing radical decomposition with a plain-language gloss instead of a bare Kangxi index

Limitations

• —Not a wordlist. Characters only, no multi-character vocabulary. For HSK 3.0 vocabulary use complete-hsk-vocabulary or hsk30.
• —HSK 2.0 banding. Levels follow HSK 2.0, not the HSK 3.0 nine-band standard.
• —No stroke data. Stroke paths and medians are derived from hanzi-writer-data and are excluded here for licensing clarity — take them from upstream directly.
• —Vietnamese-language fields. meaning_vi, mnemonic, name_vi and stroke_hint are Vietnamese; there is no English gloss column.
• —Some characters carry more than one Sino-Vietnamese reading in the literature; the field records the reading used in teaching materials for that character's common sense.

Citation

@misc{hanzi_sino_vietnamese_2026,
  title  = {HSK × Sino-Vietnamese (Hán-Việt) character dataset},
  year   = {2026},
  note   = {768 HSK characters with Sino-Vietnamese readings, radical breakdowns and mnemonics},
  url    = {https://github.com/binhbuithithanh/hanzi-sino-vietnamese}
}

CC BY 4.0 — reuse freely, including commercially, with attribution to the upstream repository.