MAFAND-MT (en-swa)
Human-translated news-domain parallel data supporting English–Swahili machine translation.
- Swahili
- Parallel
- GitHub
The data, in context
Human-translated news-domain parallel data supporting English–Swahili machine translation.
CommonMind uses it to train and evaluate Swahili-gemma-1B-it, Swahili-nllb-1.3B, Swahili-nllb-600M.
DATASET INFORMATION
- Language
- Swahili
- Data type
- Parallel
- Domain
- News
- Format
- –
- Size
- –
- Samples used
- 32,414 pairs
- Version
- –
- License
- –
- Availability
- GitHub ↗
Explore what's in the dataset
Understand the language, content, and structure represented within the corpus.
DATA PREVIEW
| ID | Language | Type | Domain |
|---|---|---|---|
| 00001 | Swahili | Parallel | News |
| 00002 | Swahili | Parallel | News |
| 00003 | Swahili | Parallel | News |
| 00004 | Swahili | Parallel | News |
Sample rows will appear here once the preview is published.
What's represented
Understand the languages, data types, and contexts covered by this dataset.
- Language
- Swahili
- Data Type
- Parallel
- Source
- GitHub
- Version
- –
- News
How the dataset was built
Learn how the data was collected, processed, structured, reviewed, and prepared for civic language research.
- 01 — CollectionHow the source language data was gathered.
- 02 — ProcessingHow speech and related resources were prepared and structured.
- 03 — AnnotationHow language and civic information were organised within the dataset.
- 04 — ReviewHow the resulting data was assessed and prepared for use.
Use MAFAND-MT (en-swa)
Access the dataset for research, model development, evaluation, and civic language experimentation.
git clone https://github.com/masakhane-io/lafand-mtModels connected to this dataset
Explore models and language tools built with MAFAND-MT (en-swa).
- Swahili-gemma-1B-itA Gemma 3 1B model adapted using LoRA for English to Swahili translation.
- Swahili
- Translation
- Swahili-nllb-1.3BAn NLLB-200 1.3B model adapted using LoRA for English to Swahili translation.
- Swahili
- Translation
- Swahili-nllb-600MAn NLLB-200 600M model adapted using LoRA for English to Swahili translation.
- Swahili
- Translation
Understand the boundaries of the data
Review known limitations, representation gaps, intended uses, and areas where additional language data or research may be needed.
- Known Limitations
- Document known gaps and constraints within the dataset.
- Intended Use
- Describe the research and development contexts the dataset is designed to support.
- Out-of-Scope Use
- Clarify applications or interpretations the dataset was not designed to support.
- Representation
- Document relevant considerations around language and contextual coverage.
Dataset history
Track changes, additions, and improvements across releases.
| Version | Released | Status | Notes |
|---|---|---|---|
| – | – | In use | 32,414 pairs |