Tatoeba

Tatoeba is a free, community-built collection of short example sentences and their translations, started in 2006 by Trang Ho and maintained since by volunteer contributors. Its name comes from the Japanese word for "for example". The corpus holds more than 13 million sentences across over 400 languages, linked into a translation graph so that a given sentence connects to its equivalents in other languages; the download page reports roughly 8.5 million unique translation pairs. Around 69 languages have 10,000 or more sentences, and over a million sentences carry audio recordings contributed by native speakers.

You access it by bulk download from the downloads page, which publishes compressed CSV and tar.bz2 files for sentences, translation links, tags, lists and audio metadata; the files are refreshed every Saturday. The volume is modest text, so ingestion is undemanding. Several third-party mirrors also republish it, including Hugging Face datasets.

For RAG, Tatoeba is a clean, sentence-aligned parallel corpus useful for multilingual retrieval, translation-aware indexing and building glossaries. For an AI tutor it is more directly a bank of practice items: short, natural sentences, paired across languages for translation drills, reading comprehension and vocabulary work, with audio for pronunciation. A tutor can sequence sentences by length or by how common a language is, and generate cloze or translation exercises from aligned pairs.

The licence needs care. Sentences default to CC BY 2.0 FR, which permits commercial use provided you attribute Tatoeba, and a subset is released under CC0 1.0; per-sentence licences can differ, so check individual records. Unlike the subject-matter sources we list, such as OpenStax, Khan Academy and CK-12, which supply curriculum content, Tatoeba supplies the raw language material a tutor drills with, complementing rather than overlapping them.