YoruMed: A Yoruba Medical Terminology Dataset

Dataset Description

YoruMed is a structured Yoruba medical terminology dataset, comprising 1,000 English medical terms, their plain-language English definitions, and corresponding Yoruba translations.

Yoruba is spoken by over 50 million people across Nigeria, Benin, and Togo, yet remains severely underrepresented in biomedical NLP. YoruMed addresses this gap by providing a curated, linguistically annotated dataset that supports medical machine translation, terminology extraction, cross-lingual retrieval, and evaluation of multilingual language models in Yoruba.

• Curated by: Temitope Olagoke, Department of Linguistics, Obafemi Awolowo University, Ile-Ife, Nigeria
• Language(s): Yoruba (yo), English (en)
• License: CC-BY 4.0
• Originally created: 2019 (undergraduate linguistics thesis, OAU)
• Released as NLP dataset: 2026

---

Dataset Structure

Fields

Field	Type	Description
english_term	string	The English medical term
english_definition	string	Plain-language English definition of the term
yoruba_translation	string	Yoruba translation with full tonal diacritics
translation_strategy	categorical	Broad NLP category: MEANING-BASED or FORM-BASED
linguistic_strategy	categorical	Linguistic detail: MORPHOLOGICAL, EXISTING, or LOANWORD
medical_domain	categorical	Medical domain — see categories below

Translation Strategy

Translations were produced using a theoretically grounded framework based on Larson (1984) and Nida & Taber (1982):

Broad categories (NLP-friendly):

• MEANING-BASED (901 terms, 90.1%) — Yoruba-native translation, prioritising meaning equivalence over formal correspondence
• FORM-BASED (99 terms, 9.9%) — Phonological adaptation of English/Latin source term into Yoruba

Linguistic detail (3-way):

• MORPHOLOGICAL (809 terms, 80.9%) — New Yoruba terms coined using native morphological resources (compounding, derivation, descriptive circumlocution)
• EXISTING (92 terms, 9.2%) — Established Yoruba words applied to medical concepts
• LOANWORD (99 terms, 9.9%) — English/Latin terms adapted into Yoruba phonology and orthography

Medical Domains

Domain	Count
General Medicine	209
Anatomy	178
Haematology	162
Infectious Disease	128
Immunology	116
Pharmacology	62
Obstetrics & Gynaecology	50
Neurology	49
Cardiology	36
Surgery	10

---

Loading the Dataset

from datasets import load_dataset

dataset = load_dataset("temitopeolagoke/yorumed")
print(dataset)

Or load directly with pandas:

import pandas as pd

df = pd.read_csv("https://huggingface.co/datasets/temitopeolagoke/yorumed/resolve/main/YoruMed_final_v2.csv")
print(df.head())
print(f"Total terms: {len(df)}")

Example entries

# Filter by domain
infectious = df[df['medical_domain'] == 'Infectious Disease']

# Filter by translation strategy
morphological = df[df['linguistic_strategy'] == 'MORPHOLOGICAL']

# Filter by broad strategy
meaning_based = df[df['translation_strategy'] == 'MEANING-BASED']

---

Intended Uses

YoruMed supports the following NLP tasks:

1. Medical Machine Translation (English → Yoruba)

Evaluate whether multilingual models (NLLB-200, mBERT, Aya) correctly translate English medical terminology into Yoruba. Use as fine-tuning data to improve medical translation quality.

2. Biomedical Named Entity Recognition

Use as a reference lexicon for developing Yoruba biomedical NER systems to identify medical entities in Yoruba-language clinical text.

3. Cross-Lingual Information Retrieval

The parallel English definition and Yoruba translation structure supports retrieval tasks — assessing whether multilingual embedding models encode Yoruba medical concepts correctly.

4. LLM Evaluation in Yoruba

Benchmark large language models (GPT-4, Claude, Llama, Aya) on Yoruba medical terminology knowledge to quantify performance gaps relative to English.

5. Linguistic Analysis

The linguistic_strategy and medical_domain annotations support research into Yoruba lexical adaptation patterns, morphological productivity, and domain-specific vocabulary development.

---

Limitations

1. Single translator: All translations were produced by one linguist and validated by a faculty supervisor at OAU. Community-scale validation has not been conducted.
2. Dialect variation: Standard Yoruba orthography is used. Yoruba has significant dialect variation; some translations may not be equally natural across all dialect communities.
3. Domain concentration: Some concentration in infectious disease/HIV-AIDS terminology, reflecting the primary reference source (Yusuff et al., 2017).
4. Scale: At 1,000 entries, YoruMed is a terminology dataset, not a large-scale corpus. It is a benchmark and reference resource, not a language modelling training corpus.
5. Date of compilation: Originally compiled in 2019. Some drug names and clinical terminology may have evolved since.

---

Ethical Considerations

YoruMed is derived from publicly available English medical terminology and original Yoruba translations. No patient data or clinical records were used.

Important: YoruMed is an NLP research resource. It should not be used as a substitute for qualified medical interpreters or practitioners in clinical settings.

This dataset was created in the spirit of language equity — Yoruba speakers deserve access to health information and AI technology in their own language. Researchers using YoruMed are encouraged to engage with Yoruba-speaking medical communities for further validation.

---

Citation

If you use YoruMed in your research, please cite:

@dataset{olagoke2026yorumed,
  author    = {Olagoke, Temitope},
  title     = {YoruMed: A Yoruba Medical Terminology Dataset for Low-Resource African Language NLP},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/temitopeolagoke/yorumed},
  note      = {Originally developed as undergraduate linguistics thesis, Obafemi Awolowo University, 2019}
}

---

Related Work

• Yusuff, L. A., Adetunji, A. & Odoje, C. (2017). English-Yoruba Glossary of HIV, AIDS and Ebola-Related Terms. University Press Plc Ibadan.
• Adelani, D. et al. (2022). MAFAND-MT: A Benchmark for Low-Resource African Language Machine Translation. ACL 2022.
• Nekoto, W. et al. (2020). Participatory Research for Low-resourced Machine Translation: A Case Study in African Languages. EMNLP Findings 2020.
• Nida, E. A. & Taber, C. R. (1982). The Theory and Practice of Translation. E.J. Brill.
• Larson, L. M. (1984). Meaning-Based Translation. University Press of America.

---

Contact

Temitope Olagoke hello.temitopeolagoke@gmail.com HuggingFace: @temitopeolagoke