Skip to content

Dataset · community-datasets

TaPaCo multilingual paraphrase corpus in 73 languages

community-datasets/tapaco

Between 1 and 10 million semantically equivalent sentences from Tatoeba for 73 languages.

What's inside

A free paraphrase corpus extracted from the Tatoeba crowdsourced example and translation database. The data is built on a network graph of sentences expressing the same idea across 73 languages (including Belarusian, Arabic, Bulgarian, and others). The dataset size is between 1 and 10 million text entries.

What you can do with it

  • Training text paraphrasing models and synonym generation
  • Training multilingual embeddings for semantic duplicate search
  • Text data augmentation for rare and low-resource languages
Tasks
Translation, Text classification
Languages
Afrikaans, Arabic, Azerbaijani, Belarusian, ber, Bulgarian, Bangla, Breton and 4 more
Size
1M–10M rows
Downloads this month
1K
Likes
48

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("community-datasets/tapaco")