nassimjp/japanese-pashto-parallel-corpus

Japanese–Pashto Parallel Corpus

A Japanese–Pashto parallel text dataset designed to support machine translation, multilingual language models, Pashto NLP research, and cross-lingual language learning.

The dataset contains Japanese sentences paired with their Pashto translations. It is provided in JSONL format and can be used for translation research, supervised fine-tuning (SFT), and the development of Japanese–Pashto language technologies.

Dataset Statistics

Metric	Value
Language pair	Japanese–Pashto
Total records	12,395
Exact duplicate pairs removed	0
Invalid records detected	0
Empty records detected	0
Format	JSONL
Encoding	UTF-8

Dataset Format

Each line is a JSON object containing a Japanese text field (ja) and a Pashto translation field (ps).

{"ja": "堀之内駅は混雑していますか？", "ps": "د هورینوچي سټیشن ګڼه ګوڼه لري؟"}
{"ja": "補助艦隊の階級体系はどのようになっていますか？", "ps": "د مرستندویه بیړۍ د رتبې نظام څنګه دی؟"}
{"ja": "飲み物どう お二人さん", "ps": "څه څښاک غواړئ تاسو دواړه؟"}

Fields

• ja — Japanese source text.
• ps — Pashto translation.

Data Processing

The dataset was converted from a Japanese-to-Pashto translation cache into JSONL format.

The processing steps included:

• Conversion to JSONL.
• Exact duplicate-pair detection and removal.
• Validation for invalid and empty records.
• Random shuffling of records.

The final processed file contains 12,395 records. No exact duplicate pairs, invalid records, or empty records were detected in the reported validation run.

Note: Exact-pair deduplication does not guarantee that every Japanese sentence is unique or that every translation has been manually verified.

Intended Uses

This dataset may be useful for:

• Japanese–Pashto machine translation.
• Japanese–Pashto parallel corpus research.
• Supervised fine-tuning of multilingual language models.
• Pashto language technology development.
• Cross-lingual transfer learning.
• Translation quality evaluation and error analysis.
• Research on low-resource languages.

Limitations

This dataset has been processed programmatically, but the reported checks do not establish that every translation is linguistically accurate or contextually appropriate.

Users should independently evaluate translation quality and verify the licensing and permitted use of the underlying source material before using or redistributing the dataset.

Citation

If you use this dataset in your research, please cite it as follows:

@dataset{japanese_pashto_parallel_corpus_2026,
  title     = {Japanese-Pashto Parallel Corpus},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/nassimjp/japanese-pashto-parallel-corpus}
}

Project Goal

This dataset contributes to the development of open language resources for Pashto and supports efforts to improve multilingual AI systems for languages that have limited digital resources.

Dataset: nassimjp/japanese-pashto-parallel-corpus