Five-way parallel sentences derived from
Helsinki-NLP/opus-100.
OPUS-100 is English-centric and ships ~1M sentence pairs per XX-en config.
For each of the four target languages we built a dictionary
english_sentence -> non_english from the train, validation, and test
splits of the relevant config (en-fr, de-en, ar-en, en-zh), then took
the intersection of English keys across all four. Each row therefore has… See the full description on the dataset page:
https://huggingface.co/datasets/nirmalendu01/opus100-en-fr-de-ar-zh.