Darayut/bilingual-en-km

Bilingual English–Khmer Parallel Corpus

A combined, deduplicated English↔Khmer parallel corpus for training sequence-to-sequence machine-translation models. It merges four publicly available English–Khmer datasets into a single bidirectional training set, formatted with NLLB-style direction tokens so a single model can be trained to translate in both directions.

Dataset summary

Split	Pairs	Files
train	4,214,606	data/train-*.parquet
validation	42,572	data/validation-*.parquet

Each example is bidirectional: the corpus contains both EN→KM and KM→EN directions, indicated by a leading direction token on the source field.

Format

The data is stored as Parquet with two string columns, source and target. Each example looks like:

{"source": "<2km> A year later, the scientists were able to detect these waves.", "target": "មួយឆ្នាំក្រោយមក អ្នកវិទ្យាសាស្ត្រអាចរកឃើញរលកទាំងនេះ។"}
{"source": "<2en> សីតុណ្ហភាពប្រតិបត្តិការ -20 ~ 55 ℃", "target": "Operating Temperature -20~55℃"}

• source — the input text, prefixed with a direction token:
  • <2km> → translate the following English text into Khmer
  • <2en> → translate the following Khmer text into English
• target — the reference translation in the target language.

Usage

from datasets import load_dataset

ds = load_dataset("Darayut/bilingual-en-km")
print(ds["train"][0])
# {'source': '<2km> ...', 'target': '...'}

Source datasets

This corpus is a combination of the following four datasets. Please cite the original sources if you use this data, and consult each source for its own licensing terms.

• lyfeyvutha/nllb-en-km-316K
• Darayut/khmer-english-pairs-raw
• KrorngAI/ParaCrawl-English-Khmer-v2
• SeyhaLite/Translate-English-Khmer-All

Licensing

This combined dataset card is released under CC-BY-4.0. The underlying text originates from the four source datasets listed above, each of which carries its own license and terms. Users are responsible for reviewing and complying with the license of each source dataset before redistribution or commercial use.

Citation

If you use this dataset, please cite this repository along with the four source datasets:

@misc{bilingual_en_km,
  title        = {Bilingual English--Khmer Parallel Corpus},
  author       = {Darayut},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/datasets/Darayut/bilingual-en-km}},
  note         = {Combined from lyfeyvutha/nllb-en-km-316K, Darayut/khmer-english-pairs-raw,
                  KrorngAI/ParaCrawl-English-Khmer-v2, and SeyhaLite/Translate-English-Khmer-All}
}