Datasets › Universal Dependencies

Universal Dependencies

Introduced by Joakim Nivre et al. in Universal Dependencies v1: A Multilingual Treebank Collection1 Jan 2016 archive 2025-07-28

The Universal Dependencies (UD) project seeks to develop cross-linguistically consistent treebank annotation of morphology and syntax for multiple languages. The first version of the dataset was released in 2015 and consisted of 10 treebanks over 10 languages. Version 2.7 released in 2020 consists of 183 treebanks over 104 languages. The annotation consists of UPOS (universal part-of-speech tags), XPOS (language-specific part-of-speech tags), Feats (universal morphological features), Lemmas, dependency heads and universal dependency labels.

Source: Evaluating Contextualized Embeddings on 54 Languagesin POS Tagging, Lemmatization and Dependency Parsing Image Source: https://universaldependencies.org/introduction.html

Homepage Benchmarks Papers JSON

Benchmarks archive 2025-07-28

All 5 leaderboards whose dataset resolves to this page shown (sort by any header). "First row" is the archive's own first row at snapshot, in the archive's row order; nothing here re-ranks and metric direction is not asserted.

		First row (archive order)	Paper	Code
Dependency Parsing	Universal Dependencies	UDPipe 2.0 + mBERT + FLAIR LAS 84.60	Evaluating Contextualized Embeddings on 54 Languages in...	—	6	Compare
Part-Of-Speech Tagging	UD	BiLSTM-LAN Avg accuracy 96.88	Hierarchically-Refined Label Attention Network for...	Nealcly/LAN +1	5	Compare
Cross-lingual zero-shot dependency parsing	Universal Dependency Treebank	Cross-Lingual ELMo LAS 77.3	Cross-Lingual Alignment of Contextual Word Embeddings,...	TalSchuster/CrossLingualContextualEmb +1	3	Compare
Dependency Parsing	Universal Dependency Treebank	Stanza Macro F1 81.13	Stanza: A Python Natural Language Processing Toolkit for...	stanfordnlp/stanza +4	1	Compare
Language Identification	Universal Dependencies	Apple bi-LSTM Accuracy 86.93	A reproduction of Apple's bi-directional LSTM models for...	AU-DIS/LSTM_langid	1	Compare

Papers archive 2025-07-28

16 shown of 16 papers with a leaderboard row on this dataset's benchmarks, newest first. The archive's own "papers using this dataset" list was never published, so this is the benchmark-backed subset; the archive's count for this dataset is 520. The Syntology column is from Syntology's graph (read 2026-09-24), stated per sample; it is not part of any archive number.

			Date	Samples run Syntology
A reproduction of Apple's bi-directional LSTM models for language identification in short strings	1	1	11 Feb 2021	not harvested
Stanza: A Python Natural Language Processing Toolkit for Many Human Languages	5	1	16 Mar 2020	ran 16 of 30 samples (14 unverified; 28 pointer-only for licence)
Hierarchically-Refined Label Attention Network for Sequence Labeling	2	1	23 Aug 2019	not harvested
Evaluating Contextualized Embeddings on 54 Languages in POS Tagging, Lemmatization and Dependency Parsing	0	1	20 Aug 2019	not harvested
Sequence Tagging with Contextual and Non-Contextual Subword Representations: A Multilingual Evaluation	1	1	4 Jun 2019	not harvested
75 Languages, 1 Model: Parsing Universal Dependencies Universally	3	1	3 Apr 2019	ran 0 of 12 samples (12 unverified)
Cross-Lingual Alignment of Contextual Word Embeddings, with Applications to Zero-shot Dependency Parsing	2	1	25 Feb 2019	ran 2 of 3 samples (1 unverified)
Universal Dependency Parsing from Scratch	1	1	29 Jan 2019	not harvested
UDPipe 2.0 Prototype at CoNLL 2018 UD Shared Task	0	1	1 Oct 2018	not harvested
Turku Neural Parser Pipeline: An End-to-End System for the CoNLL 2018 Shared Task	0	1	1 Oct 2018	not harvested
Towards Better UD Parsing: Deep Contextualized Word Embeddings, Ensemble, and Treebank Concatenation	1	1	9 Jul 2018	not harvested
Robust Multilingual Part-of-Speech Tagging via Adversarial Training	1	1	14 Nov 2017	not harvested
A Novel Neural Network Model for Joint POS Tagging and Graph-based Dependency Parsing	1	1	16 May 2017	not harvested
Multilingual Part-of-Speech Tagging with Bidirectional Long Short-Term Memory Models and Auxiliary Loss	3	1	19 Apr 2016	ran 1 of 4 samples (3 unverified; 1 pointer-only for licence)
A Representation Learning Framework for Multi-Source Transfer Parsing	0	1	5 Mar 2016	not harvested
Many Languages, One Parser	1	1	4 Feb 2016	not harvested

Dataset loaders archive 2025-07-28

huggingface.co/datasets/universal_dependencies huggingface/datasetstfpytorchjax

huggingface.co/datasets/albertvillanova/universal_dependencies huggingface/datasetstfpytorchjax

huggingface.co/datasets/BramVanroy/ud_dutch_lassysmall huggingface/datasetstfpytorchjax

huggingface.co/datasets/BramVanroy/ud_dutch_alpino huggingface/datasetstfpytorchjax

huggingface.co/datasets/universal-dependencies/universal_dependencies huggingface/datasetstfpytorchjax

docs.allennlp.org/models/main/models/structured_prediction/dataset_readers/universal_dependencies/allenai/allennlp-modelspytorch

6 loaders as listed in the archive; links are outbound and not re-checked here.

Tasks archive 2025-07-28

Dependency Parsing Part-Of-Speech Tagging Language Identification POS Cross-lingual zero-shot dependency parsing Cross-Lingual POS Tagging TAG SENTS MORPH UNLABELED_DEPENDENCIES LABELED_DEPENDENCIES LEMMA

License archive 2025-07-28

Various (see link)

Modalities archive 2025-07-28

Texts Audio

Languages archive 2025-07-28

English French Spanish German Italian Chinese Japanese Russian Portuguese Afrikaans Akkadian Akuntsu Albanian Amharic Ancient Greek Apurinã Arabic Armenian Assyrian Neo-Aramaic Bambara Basque Belarusian Bhojpuri Breton Bulgarian Catalan Chukot Church Slavic Coptic Croatian Czech Danish Dutch Erzya Estonian Faroese Finnish Galician Gothic Hebrew Hindi Hungarian Icelandic Indonesian Irish Karelian Kazakh Khunsari Komi-Permyak Komi-Zyrian Korean Latin Latvian Literary Chinese Lithuanian Livvi Maltese Manx Marathi Mbyá Guaraní Modern Greek Moksha Mundurukú Nayini Nigerian Pidgin Northern Kurdish Northern Sami Norwegian Old French Old Russian Old Turkish Persian Polish Romanian Russia Buriat Sanskrit Scottish Gaelic Serbian Skolt Sami Slovak Slovenian Soi South Levantine Arabic Swedish Swedish Sign Language Swiss German Tagalog Tamil Telugu Thai Tupinambá Turkish Uighur Ukrainian Upper Sorbian Urdu Vietnamese Warlpiri Welsh Wolof Yoruba Yue Chinese

Variants archive 2025-07-28

UD
Universal Dependencies
Universal Dependency Treebank
Universal Dependencies v2.8
universal_dependencies sk_snk
Maltese Universal Dependencies Treebank (MUDT)
UD Danish DDT

7 variant names, as the archive lists them.

At a glance archive 2025-07-28

Introduced: 1 Jan 2016
Papers: 520
Benchmarks: 5
With a result: 16
Licence: Various (see link)
Modalities: Texts, Audio
Languages: English, French, Spanish, German, Italian, Chinese, Japanese, Russian, Portuguese, Afrikaans, Akkadian, Akuntsu, Albanian, Amharic, Ancient Greek, Apurinã, Arabic, Armenian, Assyrian Neo-Aramaic, Bambara, Basque, Belarusian, Bhojpuri, Breton, Bulgarian, Catalan, Chukot, Church Slavic, Coptic, Croatian, Czech, Danish, Dutch, Erzya, Estonian, Faroese, Finnish, Galician, Gothic, Hebrew, Hindi, Hungarian, Icelandic, Indonesian, Irish, Karelian, Kazakh, Khunsari, Komi-Permyak, Komi-Zyrian, Korean, Latin, Latvian, Literary Chinese, Lithuanian, Livvi, Maltese, Manx, Marathi, Mbyá Guaraní, Modern Greek, Moksha, Mundurukú, Nayini, Nigerian Pidgin, Northern Kurdish, Northern Sami, Norwegian, Old French, Old Russian, Old Turkish, Persian, Polish, Romanian, Russia Buriat, Sanskrit, Scottish Gaelic, Serbian, Skolt Sami, Slovak, Slovenian, Soi, South Levantine Arabic, Swedish, Swedish Sign Language, Swiss German, Tagalog, Tamil, Telugu, Thai, Tupinambá, Turkish, Uighur, Ukrainian, Upper Sorbian, Urdu, Vietnamese, Warlpiri, Welsh, Wolof, Yoruba, Yue Chinese
Loaders: 6
Homepage: universaldependencies.org/

Samples run Syntology

Of the 16 papers with a result here, 4 have harvested code samples in Syntology's graph. Summed over those: Syntology ran 19 of 49 samples; 30 unverified; 29 pointer-only for licence. 1 of the 4 has no sample that ran. Each row in the Papers table states its own "ran N of M samples" with the unverified count; a sample that ran executed on a synthesized fixture, which is not a correctness claim about any result. Read from the graph 2026-09-24.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections