Komor-IA/French-ShiKomori-Parallel-Dataset

French-shiKomori Parallel Dataset v1.0

Dataset Description

The French-shiKomori Parallel Dataset v1.0 is a French–Shingazidja parallel corpus developed by Komor-IA as part of its efforts to build open language resources and artificial intelligence technologies for the Comorian languages.

This first version contains 1,424 aligned French–Shingazidja sentence pairs, with French sentences paired with their corresponding translations in Shingazidja.

The dataset is the first release of a broader French-shiKomori dataset series. Future versions may progressively incorporate additional varieties of Shikomori in order to provide broader linguistic coverage of the Comorian language continuum.

The resource is designed primarily for research and development in machine translation, computational linguistics, language education, speech technologies, and other natural language processing applications involving low-resource African languages.

Languages

• French (fr)
• Shingazidja

Note: Shingazidja is the variety represented in this first version. The broader French-shiKomori project is intended to evolve toward the inclusion of additional Shikomori varieties.

Dataset Contents

Version 1.0 contains:

• 1,424 parallel sentence pairs
• French source sentences
• Shingazidja translations
• 9 thematic categories

The thematic categories include:

1. Daily life
2. Press and news
3. Questions and dialogues
4. Education
5. Culture and proverbs
6. Health
7. Administration
8. Numbers, dates and proper names
9. Expressive and TTS-oriented sentences

Example

French	Shingazidja
Bonjour comment vas-tu aujourd'hui ?	ɓariza lewo, ngawe mnono ?
Où est la mosquée la plus proche ?	Omsihiri wakariɓu nguo nɗahu ?
Je bois du thé chaque matin.	Mi hunwa cai haina trasi.
Parle fort.	Rongoa hakuu.
Le médecin vaccine les nourrissons chaque semaine.	Etwaɓiɓu ye ucandja owana watiti haina mfumo.

Data Collection

The data were collected through the Komor-IA language data contribution campaign, which involves contributions from Comorian volunteers.

The objective of the campaign is to progressively build high-quality language resources for the development and evaluation of AI systems adapted to Comorian languages.

Contributors provide language data that can subsequently be reviewed and incorporated into Komor-IA datasets.

Data Validation and Quality Control

The dataset has been developed with attention to linguistic quality and alignment between the French and Shingazidja sentences.

Translations are reviewed with the involvement of native speakers and Komor-IA linguistic contributors.

Particular attention is given to the representation of sounds that are important for Shingazidja phonology and orthography, including implosive consonants such as ɓ and ɗ.

Despite these validation efforts, the dataset may contain errors, linguistic variation, orthographic variation, or sentences requiring further review.

Users are encouraged to report potential issues through the repository's discussion or contribution mechanisms.

Orthography and Linguistic Considerations

Shingazidja is a low-resource language for which standardized computational resources remain limited.

This dataset preserves linguistic forms used during the collection and validation process rather than attempting to impose a single computational normalization on all data.

The representation of phonological distinctions is particularly important for future speech and language technologies. The dataset therefore retains characters such as:

• ɓ
• ɗ

Users working with this resource should ensure that their processing pipelines correctly support Unicode and these characters.

Intended Uses

The dataset may be used for:

• French → Shingazidja machine translation research
• Shingazidja → French machine translation research
• Parallel corpus research
• Computational linguistics
• Low-resource language research
• African language NLP
• Language identification and linguistic analysis
• Educational language technologies
• Development of multilingual language models
• Preparation and development of speech datasets
• Text-to-speech research
• Automatic speech recognition research
• Evaluation of NLP systems involving Shingazidja

Researchers may also use the corpus as a starting point for the development of larger multilingual or dialect-aware datasets.

Out-of-Scope and Misuse

This dataset should not be considered a complete representation of Shingazidja or of the broader Shikomori language.

The dataset is relatively small and should not be assumed to provide comprehensive coverage of:

• all vocabulary,
• all grammatical constructions,
• all geographic varieties,
• all sociolinguistic registers,
• or all domains of everyday language.

The dataset should therefore be used with appropriate caution when developing systems intended for broad real-world deployment.

Limitations

Several limitations should be considered:

• The dataset contains only 1,424 sentence pairs in this first version.
• The current version focuses on Shingazidja.
• Additional Shikomori varieties are not yet represented in this release.
• The thematic distribution is not necessarily balanced.
• Some linguistic and orthographic variations may remain.
• The dataset should not be considered a definitive linguistic standard for Shingazidja.
• Machine translation systems trained exclusively on this dataset are likely to have limited generalization capabilities.

Future releases are expected to expand the number of sentence pairs, thematic coverage, linguistic validation, and dialectal coverage.

Data Splits

Version 1.0 is currently released as a single parallel corpus.

No official train/validation/test split is provided at this stage.

Users creating machine learning benchmarks are encouraged to define their own splits carefully and document their methodology to ensure reproducibility.

Data Format

The dataset is distributed as a CSV file containing aligned French and Shingazidja sentences.

The current file contains the following columns:

id
categorie
français
traduction
dialecte

The dialecte field identifies the Shikomori variety represented by the translation and is intended to support the progressive expansion of the dataset to additional Shikomori varieties in future releases.

Example:

001 | daily_life | Bonjour comment vas-tu aujourd'hui ? | ɓariza lewo, ngawe mnono ? | Shingazidja
002 | administration | Où est la mosquée la plus proche ? | Omsihiri wakariɓu nguo nɗahu ? | Shingazidja

Access

This dataset is publicly visible on Hugging Face but uses gated access.

Users must request access before downloading the underlying dataset files.

Access requests are reviewed manually by Komor-IA.

This mechanism allows the project team to monitor access to the resource while maintaining a public research-oriented dataset page.

License

This dataset is released under the:

Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.

Under this license, users may share and adapt the dataset subject to the applicable license conditions, including:

• appropriate attribution,
• indication of changes where applicable,
• and non-commercial use.

Users should consult the full license terms before redistributing or incorporating the dataset into a project.

Responsible Use

Users of this dataset are encouraged to:

• properly cite the dataset;
• preserve attribution to Komor-IA and the contributors where applicable;
• document modifications made to the data;
• report significant errors or problematic translations;
• avoid presenting the dataset as a complete representation of Shingazidja or Shikomori;
• respect the non-commercial conditions of the license.

Contributions and Feedback

Community feedback is welcome.

Users can report:

• translation errors,
• alignment problems,
• spelling or orthographic issues,
• incorrect linguistic interpretations,
• problematic examples,
• or suggestions for future releases.

Contributions should follow the repository's contribution mechanisms.

Dataset Development

The dataset is part of the broader Komor-IA language data initiative, whose objective is to create resources supporting the development of AI systems for Comorian languages.

The long-term objective is to progressively expand the resource beyond Shingazidja and include additional Shikomori varieties.

Future releases may therefore include:

• additional sentence pairs;
• additional domains;
• additional Shikomori varieties;
• improved linguistic validation;
• larger evaluation sets;
• and resources supporting speech and multilingual AI research.

Organization

Komor-IA

Komor-IA is a Comorian AI initiative working on artificial intelligence technologies adapted to Comorian realities, with a particular focus on language technologies and the development of resources for Shikomori.

Website: https://komor-ia.com

Acknowledgements

We thank the Comorian volunteers and contributors who participate in the Komor-IA data collection initiative.

Their contributions help build the language resources necessary for the development of future AI technologies for Comorian languages.

We also acknowledge the contribution of native speakers and linguistic reviewers involved in the validation and improvement of the data.

Citation

If you use this dataset in research, publications, benchmarks, or other public work, please cite the dataset as follows:

@dataset{komoria_french_shikomori_parallel_v1,
  author       = {Komor-IA},
  title        = {French-shiKomori Parallel Dataset v1.0},
  year         = {2026},
  publisher    = {Hugging Face},
  version      = {1.0},
  url          = {https://huggingface.co/datasets/Komor-IA/French-ShiKomori-Parallel-Dataset}
}

Version

Version: 1.0 Dataset: French-shiKomori Parallel Dataset Current variety: Shingazidja Parallel pairs: 1,424 License: CC BY-NC 4.0 Maintainer: Komor-IA