ML-ready metadata · Croissant 1.1 · CC BY 4.0

Bopomofo ML Dataset - All 37 Zhuyin Symbols in Croissant 1.1

A small, deterministic reference dataset for AI agents, data catalogs and language-tool developers working with Taiwan Mandarin phonetics.

37 records12 documented fieldsCSV + JSONLCroissant 1.1 validated

Download the dataset

CSV is the canonical table loaded by the Croissant RecordSet. JSONL contains the same records, while the metadata file declares fields, checksums, stable identifiers and provenance.

CSV tableUTF-8 CSV · 8,350 bytes · SHA-256 6d9235a013dd…JSONL recordsJSON Lines · 14,734 bytes · SHA-256 68729f91871c…Croissant metadataMLCommons Croissant 1.1 · 13,777 bytes · SHA-256 d96d4b5a8d50…

Dataset schema

Each row represents one symbol. symbol_id is the unique key; concept_uri and category_uri connect the table to the companion SKOS vocabulary.

Field	Type	Meaning
`order`	sc:Integer	Stable educational display order from 1 to 37.
`symbol_id`	sc:Text	ASCII identifier derived from the uppercase Unicode code point.
`concept_uri`	sc:URL	Stable IRI for the symbol in the companion SKOS vocabulary.
`symbol`	sc:Text	The Bopomofo character.
`unicode`	sc:Text	Unicode notation in U+XXXX form.
`pinyin`	sc:Text	Hanyu Pinyin correspondence.
`ipa`	sc:Text	Broad IPA transcription without brackets.
`category`	sc:Text	One of initial, medial or final.
`category_uri`	sc:URL	IRI for the broader SKOS category concept.
`example_character`	sc:Text	Traditional Chinese example character.
`example_pinyin`	sc:Text	Tone-marked Pinyin for the example.
`example_meaning_en`	sc:Text	Short English meaning for the example.

Load with mlcroissant

The official Python reader can validate the metadata, fetch the CSV and yield each row from the symbols RecordSet.

import mlcroissant as mlc

dataset = mlc.Dataset(jsonld="https://alice51849.github.io/ios-app-guide/data/zhuyin-bopomofo-ml-dataset.croissant.jsonld")
records = dataset.records(record_set="symbols")
first = next(iter(records))

Intended uses

Bopomofo-aware text processing and Unicode normalization tests
Educational search, glossary and knowledge-graph prototypes
Data-catalog demonstrations and small interoperability fixtures
Joining tabular records to stable SKOS concept IRIs

Limitations and non-uses

Not an audio, speech or acoustic-training corpus
Not sufficient to train or benchmark Mandarin pronunciation models
Not a learner assessment or diagnostic dataset
Example mappings are compact reference aids, not full phonological rules

Preview all 37 records

Order	Symbol	Category	Pinyin	IPA	Example
1	ㄅ	Initial	b	[p]	爸 (bà)
2	ㄆ	Initial	p	[pʰ]	怕 (pà)
3	ㄇ	Initial	m	[m]	媽 (mā)
4	ㄈ	Initial	f	[f]	飛 (fēi)
5	ㄉ	Initial	d	[t]	大 (dà)
6	ㄊ	Initial	t	[tʰ]	天 (tiān)
7	ㄋ	Initial	n	[n]	你 (nǐ)
8	ㄌ	Initial	l	[l]	來 (lái)
9	ㄍ	Initial	g	[k]	狗 (gǒu)
10	ㄎ	Initial	k	[kʰ]	看 (kàn)
11	ㄏ	Initial	h	[x]	好 (hǎo)
12	ㄐ	Initial	j	[tɕ]	家 (jiā)
13	ㄑ	Initial	q	[tɕʰ]	去 (qù)
14	ㄒ	Initial	x	[ɕ]	小 (xiǎo)
15	ㄓ	Initial	zh	[ʈʂ]	中 (zhōng)
16	ㄔ	Initial	ch	[ʈʂʰ]	吃 (chī)
17	ㄕ	Initial	sh	[ʂ]	是 (shì)
18	ㄖ	Initial	r	[ʐ]	人 (rén)
19	ㄗ	Initial	z	[ts]	早 (zǎo)
20	ㄘ	Initial	c	[tsʰ]	菜 (cài)
21	ㄙ	Initial	s	[s]	三 (sān)
22	ㄧ	Medial	i / yi	[i]	一 (yī)
23	ㄨ	Medial	u / wu	[u]	五 (wǔ)
24	ㄩ	Medial	ü / yu	[y]	魚 (yú)
25	ㄚ	Final	a	[a]	大 (dà)
26	ㄛ	Final	o	[ɔ]	波 (bō)
27	ㄜ	Final	e	[ɤ]	喝 (hē)
28	ㄝ	Final	ê	[ɛ]	誒 (ề)
29	ㄞ	Final	ai	[ai]	愛 (ài)
30	ㄟ	Final	ei	[ei]	給 (gěi)
31	ㄠ	Final	ao	[au]	好 (hǎo)
32	ㄡ	Final	ou	[ou]	有 (yǒu)
33	ㄢ	Final	an	[an]	安 (ān)
34	ㄣ	Final	en	[ən]	恩 (ēn)
35	ㄤ	Final	ang	[aŋ]	忙 (máng)
36	ㄥ	Final	eng	[əŋ]	冷 (lěng)
37	ㄦ	Final	er	[ɚ]	二 (èr)

Provenance and standards

The table is generated deterministically from the site's open 37-symbol reference. Croissant structure follows MLCommons 1.1; code points follow Unicode, and the source page cites Taiwan's Ministry of Education.

View the source dataset Open the SKOS vocabulary Open the no-key symbol API Open the Data Package 2.0 edition MLCommons Croissant 1.1 specification

License and privacy

The dataset and original metadata are reusable under CC BY 4.0 with attribution to Lumi Apps - iOS App Guide.

CC BY 4.0

No people, learner records, personal data, cookies, executable JavaScript or runtime API are included in the downloadable artifacts.

Questions

What is Croissant?

Croissant is an MLCommons JSON-LD format that describes dataset files, record structure, fields, provenance and usage conditions.

Do the CSV and JSONL files contain the same records?

Yes. Automated checks compare all 37 records and field values.

Does this dataset contain every Bopomofo symbol?

Yes. It contains 21 initials, 3 medials and 13 finals: 37 symbols.

Can this train a speech model?

No. It is a compact symbolic reference without audio, recordings or enough observations for speech-model training.

Optional game-based iPhone practice

Lumi Bopomofo offers a separate on-device way to practise Zhuyin through short activities. The open dataset remains free and independent.

View Lumi Bopomofo on the App Store