This is a part of the Parallel Universal Dependencies (PUD) treebanks created for the CoNLL 2017 shared task on Multilingual Parsing from Raw Text to Universal Dependencies.

There are 1000 sentences in each language, always in the same order. (The sentence alignment is 1-1 but occasionally a sentence-level segment actually consists of two real sentences.) The sentences are taken from the news domain (sentence id starts in ‘n’) and from Wikipedia (sentence id starts with ‘w’). There are usually only a few sentences from each document, selected randomly, not necessarily adjacent. The digits on the second and third position in the sentence ids encode the original language of the sentence. The first 750 sentences are originally English (01). The remaining 250 sentences are originally German (02), French (03), Italian (04) or Spanish (05) and they were translated to other languages via English. Translation into German, French, Italian, Spanish, Arabic, Hindi, Chinese, Indonesian, Japanese, Korean, Portuguese, Russian, Thai and Turkish has been provided by DFKI and performed (except for German) by professional translators. Then the data has been annotated morphologically and syntactically by Google according to Google universal annotation guidelines; finally, it has been converted by members of the UD community to UD v2 guidelines.

Additional languages have been provided (both translation and native UD v2 annotation) by other teams: Czech by Charles University, Finnish by University of Turku and Swedish by Uppsala University.

The entire treebank is labeled as test set (and was used for testing in the shared task). If it is used for training in future research, the users should employ ten-fold cross-validation.

• 2024-11-15 v2.15
  • Nouns no longer distinguish Polarity. Negative nouns have negative lemmas.
  • Conditional auxiliary "by" does not have Person (besides 3, it could be also 2).
  • Short forms of adjectives now have Degree=Pos (instead of no Degree).
  • Disambiguated NumType=Mult,Sets.
• 2023-11-15 v2.13
  • Removed Style=Arch from all Czech UD treebanks.
  • Removed NumValue from all Czech UD treebanks.
• 2022-05-15 v2.10
  • Added VerbForm=Part|Voice=Pass to long forms of passive participles.
  • Added VerbForm=Vnoun to verbal nouns.
  • The word 'každý' is now DET instead of ADJ.
  • The verb 'být' is now AUX in all contexts.
  • Merged PRON/DET 'sám', 'samý'.
• 2020-05-15 v2.6
  • Genitive, dative and instrumental nominals are now considered oblique.
  • Added enhanced graphs (shared parents in coordination).
  • Added enhanced relations with case information.
  • Added enhanced relations around relative clauses.
  • Added enhanced external subjects in control verb constructions.
  • Added empty nodes to enhanced graphs (but orphans are just converted to dep).
• 2019-05-15 v2.4
  • Manually fixed errors, mostly "více/méně" as DET(quantifiers) instead of ADV
• 2018-11-15 v2.3
  • Added LDeriv for passive participles (the infinitive of the source verb).
• 2017-11-15 v2.1
  • First official release after it was used as a surprise dataset in the CoNLL 2017 shared task.