CoNLL 2017 and 2018 shared tasks:
Multilingual Parsing from Raw Text to Universal Dependencies
This package contains the test data in the form in which they ware presented
to the participating systems: raw text files and files preprocessed by UDPipe.
The metadata.json files contain lists of files to process and to output;
README files in the respective folders describe the syntax of metadata.json.
For full training, development and gold standard test data, see
Universal Dependencies 2.0 (CoNLL 2017)
Universal Dependencies 2.2 (CoNLL 2018)
See the download links at http://universaldependencies.org/.
For more information on the shared tasks, see
http://universaldependencies.org/conll17/
http://universaldependencies.org/conll18/
Contents:
conll17-ud-test-2017-05-09 ... CoNLL 2017 test data
conll18-ud-test-2018-05-06 ... CoNLL 2018 test data
conll18-ud-test-2018-05-06-for-conll17 ... CoNLL 2018 test data with metadata
and filenames modified so that it is digestible by the 2017 systems.
We present DaMuEL, a large Multilingual Dataset for Entity Linking containing data in 53 languages. DaMuEL consists of two components: a knowledge base that contains language-agnostic information about entities, including their claims from Wikidata and named entity types (PER, ORG, LOC, EVENT, BRAND, WORK_OF_ART, MANUFACTURED); and Wikipedia texts with entity mentions linked to the knowledge base, along with language-specific text from Wikidata such as labels, aliases, and descriptions, stored separately for each language. The Wikidata QID is used as a persistent, language-agnostic identifier, enabling the combination of the knowledge base with language-specific texts and information for each entity. Wikipedia documents deliberately annotate only a single mention for every entity present; we further automatically detect all mentions of named entities linked from each document. The dataset contains 27.9M named entities in the knowledge base and 12.3G tokens from Wikipedia texts. The dataset is published under the CC BY-SA licence.
Deep Universal Dependencies is a collection of treebanks derived semi-automatically from Universal Dependencies (http://hdl.handle.net/11234/1-2988). It contains additional deep-syntactic and semantic annotations. Version of Deep UD corresponds to the version of UD it is based on. Note however that some UD treebanks have been omitted from Deep UD.
Deep Universal Dependencies is a collection of treebanks derived semi-automatically from Universal Dependencies (http://hdl.handle.net/11234/1-3105). It contains additional deep-syntactic and semantic annotations. Version of Deep UD corresponds to the version of UD it is based on. Note however that some UD treebanks have been omitted from Deep UD.
Deep Universal Dependencies is a collection of treebanks derived semi-automatically from Universal Dependencies (http://hdl.handle.net/11234/1-3226). It contains additional deep-syntactic and semantic annotations. Version of Deep UD corresponds to the version of UD it is based on. Note however that some UD treebanks have been omitted from Deep UD.
Deep Universal Dependencies is a collection of treebanks derived semi-automatically from Universal Dependencies (http://hdl.handle.net/11234/1-3424). It contains additional deep-syntactic and semantic annotations. Version of Deep UD corresponds to the version of UD it is based on. Note however that some UD treebanks have been omitted from Deep UD.
Deep Universal Dependencies is a collection of treebanks derived semi-automatically from Universal Dependencies (http://hdl.handle.net/11234/1-3687). It contains additional deep-syntactic and semantic annotations. Version of Deep UD corresponds to the version of UD it is based on. Note however that some UD treebanks have been omitted from Deep UD.
Texts in 107 languages from the W2C corpus (http://hdl.handle.net/11858/00-097C-0000-0022-6133-9), first 1,000,000 tokens per language, tagged by the delexicalized tagger described in Yu et al. (2016, LREC, Portorož, Slovenia).
Texts in 107 languages from the W2C corpus (http://hdl.handle.net/11858/00-097C-0000-0022-6133-9), first 1,000,000 tokens per language, tagged by the delexicalized tagger described in Yu et al. (2016, LREC, Portorož, Slovenia).
Changes in version 1.1:
1. Universal Dependencies tagset instead of the older and smaller Google Universal POS tagset.
2. SVM classifier trained on Universal Dependencies 1.2 instead of HamleDT 2.0.
3. Balto-Slavic languages, Germanic languages and Romance languages were tagged by classifier trained only on the respective group of languages. Other languages were tagged by a classifier trained on all available languages. The "c7" combination from version 1.0 is no longer used.
HamleDT 2.0 is a collection of 30 existing treebanks harmonized into a common annotation style, the Prague Dependencies, and further transformed into Stanford Dependencies, a treebank annotation style that became popular recently. We use the newest basic Universal Stanford Dependencies, without added language-specific subtypes.
Wikipedia plain text data obtained from Wikipedia dumps with WikiExtractor in February 2018.
The data come from all Wikipedias for which dumps could be downloaded at [https://dumps.wikimedia.org/]. This amounts to 297 Wikipedias, usually corresponding to individual languages and identified by their ISO codes. Several special Wikipedias are included, most notably "simple" (Simple English Wikipedia) and "incubator" (tiny hatching Wikipedias in various languages).
For a list of all the Wikipedias, see [https://meta.wikimedia.org/wiki/List_of_Wikipedias].
The script which can be used to get new version of the data is included, but note that Wikipedia limits the download speed for downloading a lot of the dumps, so it takes a few days to download all of them (but one or a few can be downloaded fast).
Also, the format of the dumps changes time to time, so the script will probably eventually stop working one day.
The WikiExtractor tool [http://medialab.di.unipi.it/wiki/Wikipedia_Extractor] used to extract text from the Wikipedia dumps is not mine, I only modified it slightly to produce plaintext outputs [https://github.com/ptakopysk/wikiextractor].
Tools and scripts used to create the cross-lingual parsing models submitted to VarDial 2017 shared task (https://bitbucket.org/hy-crossNLP/vardial2017), as described in the linked paper. The trained UDPipe models themselves are published in a separate submission (https://lindat.mff.cuni.cz/repository/xmlui/handle/11234/1-1971).
For each source (SS, e.g. sl) and target (TT, e.g. hr) language,
you need to add the following into this directory:
- treebanks (Universal Dependencies v1.4):
SS-ud-train.conllu
TT-ud-predPoS-dev.conllu
- parallel data (OpenSubtitles from Opus):
OpenSubtitles2016.SS-TT.SS
OpenSubtitles2016.SS-TT.TT
!!! If they are originally called ...TT-SS... instead of ...SS-TT...,
you need to symlink them (or move, or copy) !!!
- target tagging model
TT.tagger.udpipe
All of these can be obtained from https://bitbucket.org/hy-crossNLP/vardial2017
You also need to have:
- Bash
- Perl 5
- Python 3
- word2vec (https://code.google.com/archive/p/word2vec/); we used rev 41 from 15th Sep 2014
- udpipe (https://github.com/ufal/udpipe); we used commit 3e65d69 from 3rd Jan 2017
- Treex (https://github.com/ufal/treex); we used commit d27ee8a from 21st Dec 2016
The most basic setup is the sl-hr one (train_sl-hr.sh):
- normalization of deprels
- 1:1 word-alignment of parallel data with Monolingual Greedy Aligner
- simple word-by-word translation of source treebank
- pre-training of target word embeddings
- simplification of morpho feats (use only Case)
- and finally, training and evaluating the parser
Both da+sv-no (train_ds-no.sh) and cs-sk (train_cs-sk.sh) add some cross-tagging, which seems to be useful only in
specific cases (see paper for details).
Moreover, cs-sk also adds more morpho features, selecting those that
seem to be very often shared in parallel data.
The whole pipeline takes tens of hours to run, and uses several GB of RAM, so make sure to use a powerful computer.
Pretrained model weights for the UDify model, and extracted BERT weights in pytorch-transformers format. Note that these weights slightly differ from those used in the paper.
Tokenizer, POS Tagger, Lemmatizer and Parser models for 123 treebanks of 69 languages of Universal Depenencies 2.10 Treebanks, created solely using UD 2.10 data (https://hdl.handle.net/11234/1-4758). The model documentation including performance can be found at https://ufal.mff.cuni.cz/udpipe/2/models#universal_dependencies_210_models .
To use these models, you need UDPipe version 2.0, which you can download from https://ufal.mff.cuni.cz/udpipe/2 .
Tokenizer, POS Tagger, Lemmatizer and Parser models for 131 treebanks of 72 languages of Universal Depenencies 2.12 Treebanks, created solely using UD 2.12 data (https://hdl.handle.net/11234/1-5150). The model documentation including performance can be found at https://ufal.mff.cuni.cz/udpipe/2/models#universal_dependencies_212_models .
To use these models, you need UDPipe version 2.0, which you can download from https://ufal.mff.cuni.cz/udpipe/2 .
Tokenizer, POS Tagger, Lemmatizer and Parser models for 90 treebanks of 60 languages of Universal Depenencies 2.4 Treebanks, created solely using UD 2.4 data (http://hdl.handle.net/11234/1-2988). The model documentation including performance can be found at http://ufal.mff.cuni.cz/udpipe/models#universal_dependencies_24_models .
To use these models, you need UDPipe binary version at least 1.2, which you can download from http://ufal.mff.cuni.cz/udpipe .
In addition to models itself, all additional data and value of hyperparameters used for training are available in the second archive, allowing reproducible training.
Tokenizer, POS Tagger, Lemmatizer and Parser models for 94 treebanks of 61 languages of Universal Depenencies 2.5 Treebanks, created solely using UD 2.5 data (http://hdl.handle.net/11234/1-3105). The model documentation including performance can be found at http://ufal.mff.cuni.cz/udpipe/models#universal_dependencies_25_models .
To use these models, you need UDPipe binary version at least 1.2, which you can download from http://ufal.mff.cuni.cz/udpipe .
In addition to models itself, all additional data and value of hyperparameters used for training are available in the second archive, allowing reproducible training.
Tokenizer, POS Tagger, Lemmatizer and Parser models for 99 treebanks of 63 languages of Universal Depenencies 2.6 Treebanks, created solely using UD 2.6 data (https://hdl.handle.net/11234/1-3226). The model documentation including performance can be found at https://ufal.mff.cuni.cz/udpipe/2/models#universal_dependencies_26_models .
To use these models, you need UDPipe version 2.0, which you can download from https://ufal.mff.cuni.cz/udpipe/2 .
A set of corpora for 120 languages automatically collected from wikipedia and the web.
Collected using the W2C toolset: http://hdl.handle.net/11858/00-097C-0000-0022-60D6-1