What’s Out There
← Back to swiping
Language & textFreeAcademic

Multilingual parallel text corpora for East African languages

Makerere University NLP group

English text translated into five East African languages, for training machine translation.

A partial parallel corpus of English sentences translated into Acholi, Runyankore, Luganda, Lumasaba, and Swahili. Several of these languages have few other digital resources. It is meant for machine translation and NLP research for low-resource languages.

Access

Cost
FreeFree download from Harvard Dataverse; no account needed.
License
License not statedNot stated in the record
Sign-up
No sign-up
How you get it
Bulk downloadunknown

What's in it

One row is
One English sentence with its translations
Covers
Uganda and East Africa
Years
unknown (deposited 2023)
Updated
One-time release
Size
Partial; size not stated in the record
English sentenceAcholi translationRunyankore translationLuganda translationLumasaba translationSwahili translation

Quality

3 of 5

Valuable for under-resourced languages; one-off with unclear size.

  • Only a partial corpus
  • Translation quality is not independently reported

Connects to

  • Kiswahili Monolingual Corpus: Monolingual text for the same project
  • Sunbird AI Salt dataset: Related Ugandan language translation set
  • FLORES-200: Benchmark covering Swahili and Luganda

Questions it could answer

  • How well do translation models handle Acholi?
  • How much do the translations overlap in vocabulary?

Similar data sets

  • OPUS Parallel Corpora

    Over 100 billion aligned sentence pairs across more than 1,000 languages.

    Language & textUniversity of Helsinki (Language Technology Research Group)Free

  • MMMLU (Multilingual MMLU)

    The MMLU test set translated into 14 languages by professional human translators.

    EducationOpenAIFree

  • SmolLM-Corpus

    The mix of synthetic textbooks, educational web text, and Python code that trained small SmolLM models.

    Language & textHugging Face (SmolLM team)Free

  • MADLAD-400

    A Common Crawl corpus in 419 languages whose lower-resource portions were manually audited.

    Language & textGoogle Research and Allen Institute for AIFree

  • Dolma

    The three-trillion-token corpus behind OLMo, with the recipe and tooling published alongside it.

    Language & textAllen Institute for AIFree

  • MMLU-Pro

    A harder MMLU with ten answer options and 12,000 reasoning-heavy questions.

    EducationTIGER-Lab (University of Waterloo and collaborators)Free