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.
- 2024-05-15 v2.14
- Added sentence transliteration.
- Added lemmas predicted by UDPipe trained on UD Hindi HDTB 2.12.
- The light verbs कर and हो are no longer analyzed as copulas.
- Some other pseudo-auxiliaries reanalyzed using xcomp.
- The Animacy feature is not used in Hindi.
- Verbal lemma is infinitive instead of stem.
- 2021-05-15 v2.8
- Relation det:predet changed to just det. The subtype is not needed and is not used in the other Hindi treebanks.
- Relation compound:conjv changed to compound:lvc (light-verb construction), which is used in other UD treebanks for this construction.
- 2019-05-15 v2.4
- Normalized Unicode.
- Fixed punctuation.
- Fixed some other errors found by the new validator.
- 2018-07-01 v2.2
- Fixed right-headed relations.
- 2017-11-15 v2.1
- First official release after it was used as a surprise dataset in the CoNLL 2017 shared task.
A description of how the treebanks were generated can be found in:
Universal Dependency Annotation for Multilingual Parsing Ryan McDonald, Joakim Nivre, Yvonne Quirmbach-Brundage, Yoav Goldberg, Dipanjan Das, Kuzman Ganchev, Keith Hall, Slav Petrov, Hao Zhang, Oscar Tackstrom, Claudia Bedini, Nuria Bertomeu Castello and Jungmee Lee Proceedings of ACL 2013
A more detailed description of each relation type in our harmonized scheme is included in the file universal-guidelines.pdf.
Each file is formatted according to the CoNLL 2006/2007 guidelines:
http://ilk.uvt.nl/conll/#dataformat
The treebank annotations use basic Stanford Style dependencies, modified minimally to be sufficient for each language and be maximally consistent across languages. The original English Stanford guidelines can be found here:
http://nlp.stanford.edu/software/dependencies_manual.pdf
In the CoNLL file format there is a coarse part-of-speech tag field (4) and a fine-grained part-of-speech tag field (5). In this data release, we use the coarse field to store the normalized universal part-of-speech tags that are consistent across languages. The fine-grained field contains potentially richer part-of-speech information depending on the language, e.g., a richer tag representation for clitics.
We will distinguish between two portions of the data:
-
The underlying text for sentences and corresponding translations. This data Google asserts no ownership over and no copyright over. The source of the texts is randomly selected Wikipedia (www.wikipedia.org) sentences. Some or all of these sentences may be copyrighted in some jurisdictions. Where copyrighted, Google collected these sentences under exceptions to copyright or implied license rights. GOOGLE MAKES THEM AVAILABLE TO YOU under CC-BY-SA 3.0, WITHOUT ANY WARRANTY OF ANY KIND, WHETHER EXPRESS OR IMPLIED.See attached LICENSE file for the text of CC BY-SA 3.0.
-
The annotations -- part-of-speech tags and dependency annotations. GOOGLE MAKES THEM AVAILABLE TO YOU 'AS IS', WITHOUT ANY WARRANTY OF ANY KIND, WHETHER EXPRESS OR IMPLIED.
We are greatful to the many people who made this dataset possible: Fernando Pereira, Hans Uszkoreit, Aljoscha Burchardt, Vivien Macketanz, Ali Elkahky, Abhijit Barde, Tolga Kayadelen, ...