Dataset
September 17, 2026 · Harry Tran
VietAlpha Lab is releasing the VietAlpha French–Vietnamese Mathematics Dictionary, a machine-readable edition of Danh-từ Toán-học Pháp-Việt. The source was compiled in Saigon in 1964 by the Tiểu-ban Toán-học, the mathematics subcommittee of the national committee charged with compiling specialized dictionaries, and published by Bộ Văn-hóa Giáo-dục. Our edition contains 4,095 dictionary entries and a 1,369-item Vietnamese index, with stable identifiers and links back to the printed record.
The volume documents an institutional effort to set down a Vietnamese technical vocabulary for modern mathematics. It organizes terms for algebra, geometry, analysis, probability, and related fields, with alternate senses, compound expressions, and a Vietnamese lookup index. Digitizing that system gives researchers direct access to the choices recorded on its pages.
A technical dictionary's entries show which Vietnamese term was paired with a French concept, which synonyms stood together, how compound expressions were organized, and how readers were expected to find a term from either side of the book. The 1964 dictionary gives this history a precise setting: it belongs to a state-backed programme to make scientific knowledge legible in Vietnamese, two decades after Hoàng Xuân Hãn's Danh từ khoa học (1942) first attempted the same task for the sciences as a whole.
The entries themselves are small records of mathematical language taking form. One gives Algèbre two Vietnamese senses: Đại-số học and Đại-số hệ. Another gives Angle as Góc and Giác. Under Calcul, a subentry records calcul des probabilités as xác-suất. Hyphenated compounds such as Đại-số học are the source's own orthography, and we have preserved them.
The Vietnamese index extends the picture. It maps terms such as Đại-số-học and Góc back to numbered French entries, preserving the route a reader of the printed volume would have followed. The dictionary therefore records both the terminology and the architecture used to retrieve it.
We reconstructed the volume as two linked datasets. The dictionary file contains 1,769 headword entries and 2,326 subentries, for a total of 4,095 records across source pages 16 through 188, covering 1,716 unique headwords. The index contains 1,369 Vietnamese items carrying 1,879 references across pages 190 through 228.
Each dictionary record carries a stable entry ID, alphabetical section, entry number, source page, record type, headword, French text, Vietnamese text, and structured senses where the source numbers them. Headwords and subentries remain distinct records, which preserves the relationship between a main term and the expressions listed beneath it while keeping every row addressable.
{
"entry_id": "A0116",
"section": "A",
"entry_number": 116,
"printed_entry_number": 116,
"page": 20,
"printed_page": 25,
"kind": "headword",
"headword": "Algèbre",
"french": "Algèbre, f",
"french_lemma": "Algèbre",
"part_of_speech": "f",
"vietnamese": "1) Đại-số học",
"senses": [
{"n": 1, "vietnamese": "Đại-số học"},
{"n": 2, "vietnamese": "Đại-số hệ"}
]
}An index record links a Vietnamese term to one or more dictionary entries by ID. Every one of the 1,879 references resolves to a dictionary record, and each index record also keeps its original line of text. That combination supports automated lookup and direct inspection of the source transcription.
{
"term": "Đại-số-học",
"refs": [
{"section": "A", "number": 116, "entry_id": "A0116"}
],
"text": "Đại-số-học, A 116",
"page": 199,
"printed_page": null,
"column": "right"
}Retaining the page reference on every record is the point. Optical character recognition automates the conversion of page images to text; a dependable research resource also needs an unbroken path back to the primary source.
The dataset is designed for concrete research and engineering work:
The release is a lexical resource. Its scope covers terms, senses, subentries, and index links. Sentence pairs, worked problems, and mathematical reasoning traces sit outside that scope, and keeping the boundary at the term level is what keeps the data reusable across pipelines.
The dataset comes from OCR and structured reconstruction. Historical spelling, punctuation, and hyphenation remain visible because they are part of the source. Every dictionary record includes a source page, and printed page numbers are included where they were captured. Researchers quoting an entry or tracing its historical form should return to the cited page.
Index references that could not be linked to a reconstructed dictionary entry, and index lines whose Vietnamese term did not survive OCR, were removed from this release rather than published with unconfirmed links. Every remaining reference resolves.
This is how we approach historical data: preserve the source form, expose the structure, publish the measured limits, and keep the path back to the page.
VietAlpha's OCR transcription, entry reconstruction, normalization, schema, and documentation are released under the Apache License 2.0. The repository's NOTICE file identifies the 1964 compilation by the Tiểu-ban Toán-học as the underlying historical work, and redistributions must retain that notice and its attribution.
The release includes JSON Lines and CSV versions of the dictionary, a separate JSON Lines index, full statistics, and licensing documentation, published on Hugging Face under VietAlpha Lab. It joins our English–Vietnamese mathematics dictionary released earlier this month.
This release makes one more part of Vietnam's mathematical record searchable, citable, and ready for independent study.