Dataset

A New English–Vietnamese Dictionary Dataset for Vietnamese Mathematics

September 13, 2026 · Harry Tran

For mathematical literature to be fully usable in modern computing, it must be searchable and interpretable by both scholars and software.

Today, VietAlpha Lab is releasing the VietAlpha English–Vietnamese Mathematics Dictionary, a machine-readable bilingual lexicon digitized from Cung Kim Tiến's Từ Điển Toán Học Anh – Việt, Việt – Anh. Spanning 26,205 entries across English-to-Vietnamese and Vietnamese-to-English sections, each record is published with stable identifiers, separated senses, and direct references back to the original printed page.

By structuring this material, we aim to provide a dependable terminology layer for Vietnamese scientific computing. In practice, this gives researchers an evidentiary baseline to trace historical translation choices, equips retrieval engines with domain-accurate vocabulary, and helps translation systems preserve mathematical consistency across technical literature. All of these capabilities stem from the same foundational step: converting an authoritative printed reference into clean, structured data.

From printed reference to structured data

The source volume spans 709 pages divided into two complementary sections: an English-to-Vietnamese section containing 14,888 entries (pages 1–408), followed by a Vietnamese-to-English section with 11,317 entries (pages 410–709). Across both parts, the dictionary documents pure and applied mathematics alongside foundational terminology from computing, physics, and astronomy.

At its simplest, an extracted pair appears as a direct equivalence:

Abelian group → nhóm Aben, nhóm giao hoán

In the compiled dataset, however, this single pairing is augmented with the contextual metadata required for programmatic retrieval and analysis: an immutable identifier, lookup direction, sequence number, separated senses, and the source page where it appears.

{
  "id": "envi-00011",
  "direction": "en-vi",
  "entry_no": 11,
  "headword": "Abelian group",
  "translation": "nhóm Aben, nhóm giao hoán",
  "senses": ["nhóm Aben, nhóm giao hoán"],
  "en": "Abelian group",
  "vi": "nhóm Aben, nhóm giao hoán",
  "page": 1
}

Retaining that page reference is essential. While optical character recognition automates the conversion of page images to text, building a dependable research resource requires an unbroken audit trail back to the primary source.

Terminology as research infrastructure

At VietAlpha, our work focuses on Vietnam's mathematical record: digitizing historical texts, studying how mathematical concepts were formalized and taught, and applying those linguistic insights to modern software.

A narrow aisle in a secondhand bookshop, with used books stacked floor to ceiling on both sides and a blue plastic stool in the foreground. — A secondhand bookshop, stacked floor to ceiling. Much of the printed record still lives on shelves like these.

Terminology forms the connective tissue across each of these stages. Rather than functioning merely as literal translations, mathematical terms document institutional choices: how an imported concept was assimilated into Vietnamese, which transliteration gained currency, and which synonyms were favored across publishing eras. Historical forms such as Aben (Abelian), afin (affine), and alep (aleph) anchor this volume in the specific conventions of mid-twentieth-century scientific publishing.

Beyond historical inquiry, this standardized vocabulary addresses immediate bottlenecks in modern language technology. For instance, OCR post-correction pipelines perform far more reliably when guided by expected domain vocabulary. Similarly, cross-lingual search engines can map English and Vietnamese technical queries to the same underlying concepts, while machine translation models can retrieve recorded mathematical glosses instead of improvising ad-hoc translations.

Together, these applications form a continuous feedback loop: digitized historical records supply the structured domain vocabulary needed by modern systems, and practical use of those systems helps surface corrections that feed back into the archive.

Intended use cases

We structured the dataset to support several immediate workflows:

Bilingual terminology lookup: Direct programmatic querying of mathematical terms across both language directions.
Constrained machine translation: Providing glossary guidance to translation models to enforce consistent terminology across technical papers and textbooks.
Retrieval-Augmented Generation (RAG): Enhancing technical retrieval pipelines with canonical English–Vietnamese term pairs.
Vocabulary audits and benchmarking: Evaluating lexical coverage for OCR post-processors, tokenizers, and domain-adapted language models.
Archival indexing: Enabling semantic search and cataloging across digitized textbooks, historical syllabi, and academic papers.
Linguistic and historical research: Analyzing the historical development, transliteration choices, and standardization of Vietnamese mathematical discourse.

Crucially, this release is scoped specifically as a lexical resource: it provides atomic, term-level mappings rather than parallel sentences, worked problem sets, or mathematical reasoning traces. Maintaining this focused boundary ensures high fidelity at the term level, where ambiguity is lowest and reusability across diverse pipelines is highest.

Limitations

Because this dataset is derived from optical character recognition rather than manual re-keying, systematic line-by-line proofreading remains an ongoing effort. Users should anticipate typical OCR artifacts in this initial release: diacritics may occasionally shift, certain words may misalign across the headword–translation boundary, and densely packed rows can occasionally merge.

This approach reflects our philosophy for public research data: provide immediate utility, maintain full transparency regarding known limitations, and offer the tooling necessary for ongoing community refinement.

Licensing and terms of use

To respect the provenance of the source text while encouraging open technical reuse, the release operates under a two-part licensing structure:

Tooling, schema, and metadata: VietAlpha's transcription pipeline, data schema, build scripts, and documentation are licensed under the Apache License 2.0.
Underlying lexical content: The original dictionary compiled by Cung Kim Tiến is cataloged on the Internet Archive under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).

Accordingly, downstream use of the dictionary entries must preserve proper attribution, remain non-commercial, and adhere to the restrictions regarding the distribution of modified dictionary content. For full legal boundaries and citation guidelines, please refer to the repository's NOTICE file and dataset card.

Availability and access

The complete dataset and accompanying build tools are available on Hugging Face under VietAlpha Lab. The repository bundles both directional splits (en-vi and vi-en), a consolidated JSON package, extraction statistics, intermediate OCR artifacts, and the Python build script used to generate the release.

Hugging Face: huggingface.co/datasets/VietAlphaLabs/vi-en-mathematics-dictionary
Source scan: archive.org/details/tu-dien-toan-hoc-anh-viet-viet-anh-cung-kim-tien

As community feedback and proofreading corrections arrive, we will continue updating the structured source files and publishing versioned iterations, ensuring every revision maintains its explicit link to the printed record.

Preserving and revitalizing Vietnam's scientific heritage requires moving beyond static document scans toward open, verifiable data infrastructure. We hope this dictionary provides a practical, shared foundation for researchers, educators, and engineers working to advance Vietnamese language technology.