Luganda Sentences Corpus

A cleaned corpus of Luganda sentences extracted from various publicly available text dumps and corpora.

Motivation

Publicly available text dumps can contain text from languages other than the language they are intended to represent. This can introduce unwanted language contamination into downstream NLP models, even when the original dumps have already undergone cleaning.

This corpus focuses on reducing that contamination by filtering the collected text for Luganda. Sentences identified as belonging to another language were removed using lingua-language-detector.

During the cleaning process, paragraphs were also split into individual sentences using a Luganda Punkt tokenizer. Therefore, this dataset consists of sentences rather than paragraphs or full documents.

Data Sources

The corpus was constructed from multiple publicly available text dumps and corpora, including sources such as:

• Wikipedia - LICENSE
• cis-lmu/GlotCC-V1
• cis-lmu/Glot500
• HuggingFaceFW/fineweb-2 - LICENSE

The sentences originate from these source materials and were not generated synthetically.

No synthetic data is included in this dataset.

Processing

The cleaning pipeline included:

1. Collecting Luganda text from publicly available dumps and corpora.
2. Performing initial text cleaning and normalization.
3. Splitting paragraphs into individual sentences using a Luganda Punkt tokenizer.
4. Detecting the language of each sentence using lingua-language-detector.
5. Removing sentences detected as belonging to languages other than Luganda.
6. Retaining the resulting Luganda sentence-level corpus.

The language filtering step is intended to reduce multilingual contamination that can otherwise propagate into models trained on low-resource language corpora.

Intended Uses

The corpus can be used for:

• Training Luganda embedding models
• Semantic similarity and retrieval datasets
• Constructing grammar-correction datasets
• Training or adapting Luganda language models
• Language identification research
• Corpus and linguistic analysis
• Creating downstream Luganda NLP datasets
• Other sentence-level NLP tasks

Because the data is provided at sentence level, it can also serve as a base corpus for constructing specialized datasets for downstream tasks.

Data Characteristics

• Language: Luganda (lg)
• Unit: Sentence
• Data type: Naturally occurring text
• Synthetic data: None
• Language filtering: lingua-language-detector
• Sentence segmentation: Luganda Punkt tokenizer
• Sources: Multiple publicly available text dumps and corpora

Maintenance

This dataset is released and maintained by SAGE POND as part of ongoing work to improve the availability and quality of resources for Luganda NLP.

The corpus may be updated as additional source material is processed and further cleaning and auditing are performed.

Citation

If you use this dataset in research or applications, please cite SAGE POND:

@dataset{sagepond_luganda_sentences,
  author       = {{SAGE POND}},
  title        = {Luganda Sentences Corpus},
  year         = {2026},
  publisher    = {Hugging Face},
  note         = {Cleaned Luganda sentence corpus extracted from publicly available text dumps and filtered for language contamination}
}

License

The licensing and redistribution conditions of the resulting corpus depend on the licenses and terms of the original source materials from which the sentences were collected.

Users should review the licensing requirements of the respective source datasets and attribute the original sources where required before redistributing derived material.