Datasets:
License:
Commit History
Yesterday
-
Missing text: issue column, editor notes, English stand-ins, absent articles
glot-buildermissing is now true in three cases, given in the new issue column: placeholder (now also units that are only an editor's note in square brackets, e.g. [Missing 15.2], 138 units), english (the English leaflet in the Ao Naga notes, 6 units; GlotLID over every unit found no others) and absent (whole articles not in the file: one empty row per English unit, 425 units in 11 translations, so every translation has all 30 articles). README: Missing text section.76f21f1 -
Corrections: an outside source for each relabel
glot-builderrelabel and the README Corrections table now give, besides the evidence in the text, an outside source for each change: OHCHR's page for nbl (source UNIC Zimbabwe, recording lng=nde), English Wikipedia's sample of Olonets Karelian (the krl text), OHCHR's own Picard label of the pcd text with its Walloon spelling and DTW Walloon dictionary entries, and OHCHR's source for tzm (Algeria) with its article 1 matching the Kabyle translation.dfedfd2 -
Remove the Latin-script Central Kurdish translation: byte-identical to Northern Kurdish
glot-builderudhr_ckb.xml (ckb_Latn) has the same text as udhr_kmr.xml, byte for byte (the only identical pair in the collection). It is left out (build_udhr.py: DROP); kmr is now the subset kmr_Latn. 532 translations, 47,555 units. README updated.e3c4df3 -
UDHR: every translation, parallel with English, one subset per translation
glot-builderFrom the UDHR XML collection formerly at unicode.org/udhr (version of 2026-07-28), built with build_udhr.py: 533 translations (462 language_script codes, 446 ISO 639-3 languages, 43 scripts), 47,645 units with structural ids (title, preamble.recital.1-7, preamble.proclamation, article.K.title, article.K.pM, article.K.iM, note.N.pM), the English text of each unit and an aligned flag. Nothing is removed. lang has the corrected code and legacy_lang the collection's (nbl -> nde, krl -> olo, pcd -> wln, tzm -> kab, ckb_Latn -> kmr, with the evidence in relabel). Subsets: one per translation (data/<lang>[_<variant>].parquet), "all" (default) and "translations".7c7cb58