Universal Dependencies Treebanks
Universal Dependencies consortium
Sentences in 150+ languages, each word tagged and linked to its grammatical head the same way.
A shared annotation scheme and 200+ treebanks in which sentences are split into words, tagged for part of speech and morphology, and parsed into dependency trees. Because every language uses the same labels, researchers can compare grammar across languages and train parsers that work across them. Releases come out about every six months.
Access
- Cost
- FreeFree download of each release.
- License
- Open licenseVaries by treebank (CC BY, CC BY-SA, CC BY-NC-SA, GPL, and others)
- Sign-up
- No sign-up
- How you get it
- Bulk downloadCoNLL-U
What's in it
- One row is
- One row per word (token) in an annotated sentence
- Covers
- Over 150 languages
- Years
- 2014-present (release 2.19, November 2026)
- Updated
- Irregular
- Size
- Over 200 treebanks from 600+ contributors
Quality
The field standard for cross-lingual syntax, with versioned releases, validation tools, and documentation for each language.
- Some treebanks are tiny (a few hundred sentences)
- Some treebanks carry non-commercial licenses
- Genres vary: some languages have only Bible or news text
Connects to
- WALS (World Atlas of Language Structures): Compare word-order patterns measured in treebanks with WALS typology codes
- OPUS Parallel Corpora: Parallel text in the same languages for cross-lingual parser transfer
Questions it could answer
- How does the average distance between a word and its head differ across languages?
- Which languages put the verb last most consistently in real sentences?