Note Filtered English sentences from C4 and FineWeb. ODC-By.

Note Harvard sentences as plain text. No licence tag on the repo.

Note About 11M Hebrew sentences from Wikipedia. CC BY-SA 3.0.

Note Short everyday sentences; use the en-he config. No licence tag on the repo.

Note Spoken-style talk sentences; use the en-he config. No licence tag on the repo.

Note News-style sentences; use the en-he config. No licence tag on the repo.

Note Short smart-home commands with intent labels. CC BY 4.0.

Note Browse text-only sentence datasets, mine and third party.

Note Standard multilingual set, source of FLEURS; use the eng_Latn and heb_Hebr configs. Gated. Recording it contaminates FLEURS evaluations. CC BY-SA 4.0.

Note Paragraph (median 87 words). English Wikipedia paragraphs, 47M rows, clean of markup. Encyclopedic register with some number-heavy rows. CC BY-SA 4.0.

Note Paragraph (median 145 words). Short GPT-written children's stories, 2.1M rows, simple vocabulary, split by paragraph for shorter reads. CDLA-Sharing 1.0.

Note Long (median 44 words). Five-sentence everyday stories, 98k rows, natural spoken register. CC BY 4.0 tag on this repackage; check the original corpus terms before redistribution.

Note Paragraph (median 148 words). Canonical Wikipedia paragraphs, in the context column. Dedupe first: many question rows share a paragraph. CC BY-SA 4.0.