PangeanicYueJa: 55,000 Cantonese-Japanese Parallel Sentences for Machine Translation

PangeanicYueJa is a parallel corpus containing 55,000 Cantonese-Japanese sentence pairs sampled from a larger collection of approximately 3.08 million pairs. It was created by Pangeanic and released on Hugging Face, with a last recorded update in June 2026. The corpus is designed for training and evaluating machine translation and multilingual language models.

Use Cases

• Train machine translation models based on the parallel sentence pairs.
• Develop multilingual large language models (LLMs) based on the bilingual text data.
• Conduct cross-lingual NLP research based on the aligned sentences.
• Build retrieval-augmented generation (RAG) systems based on the bilingual corpus.
• Create bilingual embeddings or perform instruction tuning based on the parallel text.

Strengths

• Contains 55,000 parallel sentence pairs, providing a substantial foundation for model training.
• Sampled from a larger corpus of approximately 3.08 million sentence pairs, suggesting a potential source for scaling.
• Explicitly designed for multiple advanced NLP tasks like machine translation, LLM training, and RAG.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full 3.08 million pair source is unknown, which may limit suitability assessment for large-scale projects.
• Description metadata is limited; actual data quality and sampling methodology require manual inspection after download.

Provenance

Source

Pangeanic

Freshness

Last updated 2026-06-18 08:48:19; freshness should be verified.