The ASSIN (Avaliação de Similaridade Semântica e INferência textual) corpus is a corpus annotated with pairs of sentences written in
Portuguese that is suitable for the exploration of textual entailment and paraphrasing classifiers. The corpus contains pairs of sentences
extracted from news articles written in European Portuguese (EP) and Brazilian Portuguese (BP), obtained from Google News Portugal
and Brazil, respectively. To create the corpus, the authors started by collecting a set of news articles describing the
same event (one news article from Google News Portugal and another from Google News Brazil) from Google News.
Then, they employed Latent Dirichlet Allocation (LDA) models to retrieve pairs of similar sentences between sets of news
articles that were grouped together around the same topic. For that, two LDA models were trained (for EP and for BP)
on external and large-scale collections of unannotated news articles from Portuguese and Brazilian news providers, respectively.
Then, the authors defined a lower and upper threshold for the sentence similarity score of the retrieved pairs of sentences,
taking into account that high similarity scores correspond to sentences that contain almost the same content (paraphrase candidates),
and low similarity scores correspond to sentences that are very different in content from each other (no-relation candidates).
From the collection of pairs of sentences obtained at this stage, the authors performed some manual grammatical corrections
and discarded some of the pairs wrongly retrieved. Furthermore, from a preliminary analysis made to the retrieved sentence pairs
the authors noticed that the number of contradictions retrieved during the previous stage was very low. Additionally, they also
noticed that event though paraphrases are not very frequent, they occur with some frequency in news articles. Consequently,
in contrast with the majority of the currently available corpora for other languages, which consider as labels “neutral”, “entailment”
and “contradiction” for the task of RTE, the authors of the ASSIN corpus decided to use as labels “none”, “entailment” and “paraphrase”.
Finally, the manual annotation of pairs of sentences was performed by human annotators. At least four annotators were randomly
selected to annotate each pair of sentences, which is done in two steps: (i) assigning a semantic similarity label (a score between 1 and 5,
from unrelated to very similar); and (ii) providing an entailment label (one sentence entails the other, sentences are paraphrases,
or no relation). Sentence pairs where at least three annotators do not agree on the entailment label were considered controversial
and thus discarded from the gold standard annotations. The full dataset has 10,000 sentence pairs, half of which in Brazilian Portuguese
and half in European Portuguese. Either language variant has 2,500 pairs for training, 500 for validation and 2,000 for testing.
The ASSIN 2 corpus is composed of rather simple sentences. Following the procedures of SemEval 2014 Task 1.
The training and validation data are composed, respectively, of 6,500 and 500 sentence pairs in Brazilian Portuguese,
annotated for entailment and semantic similarity. Semantic similarity values range from 1 to 5, and text entailment
classes are either entailment or none. The test data are composed of approximately 3,000 sentence pairs with the same
annotation. All data were manually annotated.
This is a multilingual parallel corpus created from translations of the Bible compiled by Christos Christodoulopoulos and Mark Steedman.
102 languages, 5,148 bitexts
total number of files: 107
total number of tokens: 56.43M
total number of sentence fragments: 2.84M
The BrWaC (Brazilian Portuguese Web as Corpus) is a large corpus constructed following the Wacky framework,
which was made public for research purposes. The current corpus version, released in January 2017, is composed by
3.53 million documents, 2.68 billion tokens and 5.79 million types. Please note that this resource is available
solely for academic research purposes, and you agreed not to use it for any commercial applications.
Manually download at https://www.inf.ufrgs.br/pln/wiki/index.php?title=BrWaC
A parallel corpus of theses and dissertations abstracts in English and Portuguese were collected from the CAPES website (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior) - Brazil. The corpus is sentence aligned for all language pairs. Approximately 240,000 documents were collected and aligned using the Hunalign algorithm.
This corpus is an attempt to recreate the dataset used for training XLM-R. This corpus comprises of monolingual data for 100+ languages and also includes data for romanized languages (indicated by *_rom). This was constructed using the urls and paragraph indices provided by the CC-Net repository by processing January-December 2018 Commoncrawl snapshots. Each file comprises of documents separated by double-newlines and paragraphs within the same document separated by a newline. The data is generated using the open source CC-Net repository. No claims of intellectual property are made on the work of preparation of the corpus.
This dataset is designed to provide training data
for common sense relationships pulls together from various sources.
The dataset is multi-lingual. See langauge codes and language info
This dataset provides an interface for the conceptnet5 csv file, and
some (but not all) of the raw text data used to build conceptnet5:
omcsnet_sentences_free.txt, and omcsnet_sentences_more.txt.
One use of this dataset would be to learn to extract the conceptnet
relationship from the omcsnet sentences.
Conceptnet5 has 34,074,917 relationships. Of those relationships,
there are 2,176,099 surface text sentences related to those 2M
omcsnet_sentences_free has 898,161 lines. omcsnet_sentences_more has
Original downloads are available here
https://github.com/commonsense/conceptnet5/wiki/Downloads. For more
information, see: https://github.com/commonsense/conceptnet5/wiki
The omcsnet data comes with the following warning from the authors of
the above site: Remember: this data comes from various forms of
crowdsourcing. Sentences in these files are not necessarily true,
useful, or appropriate.
Original source: Website and documentatuion from the European Central Bank, compiled and made available by Alberto Simoes (thank you very much!)
19 languages, 170 bitexts
total number of files: 340
total number of tokens: 757.37M
total number of sentence fragments: 30.55M
This is a parallel corpus made out of PDF documents from the European Medicines Agency. All files are automatically converted from PDF to plain text using pdftotext with the command line arguments -layout -nopgbrk -eol unix. There are some known problems with tables and multi-column layouts - some of them are fixed in the current version.
22 languages, 231 bitexts
total number of files: 41,957
total number of tokens: 311.65M
total number of sentence fragments: 26.51M
In October 2012, the European Union's (EU) Directorate General for Education and Culture ( DG EAC) released a translation memory (TM), i.e. a collection of sentences and their professionally produced translations, in twenty-six languages. This resource bears the name EAC Translation Memory, short EAC-TM.
EAC-TM covers up to 26 languages: 22 official languages of the EU (all except Irish) plus Icelandic, Croatian, Norwegian and Turkish. EAC-TM thus contains translations from English into the following 25 languages: Bulgarian, Czech, Danish, Dutch, Estonian, German, Greek, Finnish, French, Croatian, Hungarian, Icelandic, Italian, Latvian, Lithuanian, Maltese, Norwegian, Polish, Portuguese, Romanian, Slovak, Slovenian, Spanish, Swedish and Turkish.
All documents and sentences were originally written in English (source language is English) and then translated into the other languages. The texts were translated by staff of the National Agencies of the Lifelong Learning and Youth in Action programmes. They are typically professionals in the field of education/youth and EU programmes. They are thus not professional translators, but they are normally native speakers of the target language.
In October 2012, the European Union (EU) agency 'European Centre for Disease Prevention and Control' (ECDC) released a translation memory (TM), i.e. a collection of sentences and their professionally produced translations, in twenty-five languages. This resource bears the name EAC Translation Memory, short EAC-TM.
ECDC-TM covers 25 languages: the 23 official languages of the EU plus Norwegian (Norsk) and Icelandic. ECDC-TM was created by translating from English into the following 24 languages: Bulgarian, Czech, Danish, Dutch, English, Estonian, Gaelige (Irish), German, Greek, Finnish, French, Hungarian, Icelandic, Italian, Latvian, Lithuanian, Maltese, Norwegian (NOrsk), Polish, Portuguese, Romanian, Slovak, Slovenian, Spanish and Swedish.
All documents and sentences were thus originally written in English. They were then translated into the other languages by professional translators from the Translation Centre CdT in Luxembourg.
EXAMS is a benchmark dataset for multilingual and cross-lingual question answering from high school examinations.
It consists of more than 24,000 high-quality high school exam questions in 16 languages,
covering 8 language families and 24 school subjects from Natural Sciences and Social Sciences, among others.
A dataset to study Fake News in Portuguese, presenting a supposedly false News along with their respective fact check and classification.
The data is collected from the ClaimReview, a structured data schema used by fact check agencies to share their results in search engines, enabling data collect in real time.
The FACTCK.BR dataset contains 1309 claims with its corresponding label.
The HAREM is a Portuguese language corpus commonly used for Named Entity Recognition tasks. It includes about 93k words, from 129 different texts,
from several genres, and language varieties. The split of this dataset version follows the division made by , where 7% HAREM
documents are the validation set and the miniHAREM corpus (with about 65k words) is the test set. There are two versions of the dataset set,
a version that has a total of 10 different named entity classes (Person, Organization, Location, Value, Date, Title, Thing, Event,
Abstraction, and Other) and a "selective" version with only 5 classes (Person, Organization, Location, Value, and Date).
It's important to note that the original version of the HAREM dataset has 2 levels of NER details, namely "Category" and "Sub-type".
The dataset version processed here ONLY USE the "Category" level of the original dataset.
 Souza, Fábio, Rodrigo Nogueira, and Roberto Lotufo. "BERTimbau: Pretrained BERT Models for Brazilian Portuguese." Brazilian Conference on Intelligent Systems. Springer, Cham, 2020.
Portuguese dataset for hate speech detection composed of 5,668 tweets with binary annotations (i.e. 'hate' vs. 'no-hate').
A parallel corpus of KDE4 localization files (v.2).
92 languages, 4,099 bitexts
total number of files: 75,535
total number of tokens: 60.75M
total number of sentence fragments: 8.89M
LeNER-Br is a Portuguese language dataset for named entity recognition
applied to legal documents. LeNER-Br consists entirely of manually annotated
legislation and legal cases texts and contains tags for persons, locations,
time entities, organizations, legislation and legal cases.
To compose the dataset, 66 legal documents from several Brazilian Courts were
collected. Courts of superior and state levels were considered, such as Supremo
Tribunal Federal, Superior Tribunal de Justiça, Tribunal de Justiça de Minas
Gerais and Tribunal de Contas da União. In addition, four legislation documents
were collected, such as "Lei Maria da Penha", giving a total of 70 documents
Mac-Morpho is a corpus of Brazilian Portuguese texts annotated with part-of-speech tags.
Its first version was released in 2003 , and since then, two revisions have been made in order
to improve the quality of the resource [2, 3].
The corpus is available for download split into train, development and test sections.
These are 76%, 4% and 20% of the corpus total, respectively (the reason for the unusual numbers
is that the corpus was first split into 80%/20% train/test, and then 5% of the train section was
set aside for development). This split was used in , and new POS tagging research with Mac-Morpho
is encouraged to follow it in order to make consistent comparisons possible.
 Aluísio, S., Pelizzoni, J., Marchi, A.R., de Oliveira, L., Manenti, R., Marquiafável, V. 2003.
An account of the challenge of tagging a reference corpus for brazilian portuguese.
In: Proceedings of the 6th International Conference on Computational Processing of the Portuguese Language. PROPOR 2003
 Fonseca, E.R., Rosa, J.L.G. 2013. Mac-morpho revisited: Towards robust part-of-speech.
In: Proceedings of the 9th Brazilian Symposium in Information and Human Language Technology – STIL
 Fonseca, E.R., Aluísio, Sandra Maria, Rosa, J.L.G. 2015.
Evaluating word embeddings and a revised corpus for part-of-speech tagging in Portuguese.
Journal of the Brazilian Computer Society.
We introduce MKQA, an open-domain question answering evaluation set comprising 10k question-answer pairs sampled from the Google Natural Questions dataset, aligned across 26 typologically diverse languages (260k question-answer pairs in total). For each query we collected new passage-independent answers. These queries and answers were then human translated into 25 Non-English languages.
The Microsoft Terminology Collection can be used to develop localized versions of applications that integrate with Microsoft products.
It can also be used to integrate Microsoft terminology into other terminology collections or serve as a base IT glossary
for language development in the nearly 100 languages available. Terminology is provided in .tbx format, an industry standard for terminology exchange.
Parallel corpora from Web Crawls collected in the ParaCrawl project and further processed for making it a multi-parallel corpus by pivoting via English. Here we only provide the additional language pairs that came out of pivoting. The bitexts for English are available from the ParaCrawl release.
40 languages, 669 bitexts
total number of files: 40
total number of tokens: 10.14G
total number of sentence fragments: 505.48M
Please, acknowledge the ParaCrawl project at http://paracrawl.eu. This version is derived from the original release at their website adjusted for redistribution via the OPUS corpus collection. Please, acknowledge OPUS as well for this service.
A parallel corpus of News Commentaries provided by WMT for training SMT. The source is taken from CASMACAT: http://www.casmacat.eu/corpus/news-commentary.html
12 languages, 63 bitexts
total number of files: 61,928
total number of tokens: 49.66M
total number of sentence fragments: 1.93M
OffComBR: an annotated dataset containing for hate speech detection in Portuguese composed of news comments on the Brazilian Web.
This is a new collection of translated movie subtitles from http://www.opensubtitles.org/.
IMPORTANT: If you use the OpenSubtitle corpus: Please, add a link to http://www.opensubtitles.org/ to your website and to your reports and publications produced with the data!
This is a slightly cleaner version of the subtitle collection using improved sentence alignment and better language checking.
62 languages, 1,782 bitexts
total number of files: 3,735,070
total number of tokens: 22.10G
total number of sentence fragments: 3.35G
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
This is a collection of copyright free books aligned by Andras Farkas, which are available from http://www.farkastranslations.com/bilingual_books.php
Note that the texts are rather dated due to copyright issues and that some of them are manually reviewed (check the meta-data at the top of the corpus files in XML). The source is multilingually aligned, which is available from http://www.farkastranslations.com/bilingual_books.php. In OPUS, the alignment is formally bilingual but the multilingual alignment can be recovered from the XCES sentence alignment files. Note also that the alignment units from the original source may include multi-sentence paragraphs, which are split and sentence-aligned in OPUS.
All texts are freely available for personal, educational and research use. Commercial use (e.g. reselling as parallel books) and mass redistribution without explicit permission are not granted. Please acknowledge the source when using the data!
16 languages, 64 bitexts
total number of files: 158
total number of tokens: 19.50M
total number of sentence fragments: 0.91M
A collection of translation memories provided by the JRC. Source: https://ec.europa.eu/jrc/en/language-technologies/dgt-translation-memory
25 languages, 299 bitexts
total number of files: 817,410
total number of tokens: 2.13G
total number of sentence fragments: 113.52M
A parallel corpus collected from the European Constitution for 21 language.
A parallel corpus of GNOME localization files. Source: https://l10n.gnome.org
187 languages, 12,822 bitexts
total number of files: 113,344
total number of tokens: 267.27M
total number of sentence fragments: 58.12M
Parallel corpora from Web Crawls collected in the ParaCrawl project
40 languages, 41 bitexts
total number of files: 20,995
total number of tokens: 21.40G
total number of sentence fragments: 1.12G
A parallel corpus of Ubuntu localization files. Source: https://translations.launchpad.net
244 languages, 23,988 bitexts
total number of files: 30,959
total number of tokens: 29.84M
total number of sentence fragments: 7.73M
This is a corpus of parallel sentences extracted from Wikipedia by Krzysztof Wołk and Krzysztof Marasek. Please cite the following publication if you use the data: Krzysztof Wołk and Krzysztof Marasek: Building Subject-aligned Comparable Corpora and Mining it for Truly Parallel Sentence Pairs., Procedia Technology, 18, Elsevier, p.126-132, 2014
20 languages, 36 bitexts
total number of files: 114
total number of tokens: 610.13M
total number of sentence fragments: 25.90M
ParaPat: The Multi-Million Sentences Parallel Corpus of Patents Abstracts
This dataset contains the developed parallel corpus from the open access Google
Patents dataset in 74 language pairs, comprising more than 68 million sentences
and 800 million tokens. Sentences were automatically aligned using the Hunalign algorithm
for the largest 22 language pairs, while the others were abstract (i.e. paragraph) aligned.
The QCRI Educational Domain Corpus (formerly QCRI AMARA Corpus) is an open multilingual collection of subtitles for educational videos and lectures collaboratively transcribed and translated over the AMARA web-based platform.
Developed by: Qatar Computing Research Institute, Arabic Language Technologies Group
The QED Corpus is made public for RESEARCH purpose only.
The corpus is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. Copyright Qatar Computing Research Institute. All rights reserved.
225 languages, 9,291 bitexts
total number of files: 271,558
total number of tokens: 371.76M
total number of sentence fragments: 30.93M
A parallel corpus of full-text scientific articles collected from Scielo database in the following languages: English, Portuguese and Spanish. The corpus is sentence aligned for all language pairs, as well as trilingual aligned for a small subset of sentences. Alignment was carried out using the Hunalign algorithm.
This dataset add sentiment lexicons for 81 languages generated via graph propagation based on a knowledge graph--a graphical representation of real-world entities and the links between them.
This is a collection of Quran translations compiled by the Tanzil project
The translations provided at this page are for non-commercial purposes only. If used otherwise, you need to obtain necessary permission from the translator or the publisher.
If you are using more than three of the following translations in a website or application, we require you to put a link back to this page to make sure that subsequent users have access to the latest updates.
42 languages, 878 bitexts
total number of files: 105
total number of tokens: 22.33M
total number of sentence fragments: 1.01M
A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a crowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular linguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links between sentences “meaning the same thing”. This graph is then traversed to extract sets of paraphrases. Several language-independent filters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows that between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or near-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 – 250 000 sentences per language. It covers a range of languages for which, to our knowledge,no other paraphrase dataset exists.
This is a collection of translated sentences from Tatoeba
359 languages, 3,403 bitexts
total number of files: 750
total number of tokens: 65.54M
total number of sentence fragments: 8.96M
A parallel corpus of TED talk subtitles provided by CASMACAT: http://www.casmacat.eu/corpus/ted2013.html. The files are originally provided by https://wit3.fbk.eu.
15 languages, 14 bitexts
total number of files: 28
total number of tokens: 67.67M
total number of sentence fragments: 3.81M
The core of WIT3 is the TED Talks corpus, that basically redistributes the original content published by the TED Conference website (http://www.ted.com). Since 2007,
the TED Conference, based in California, has been posting all video recordings of its talks together with subtitles in English
and their translations in more than 80 languages. Aside from its cultural and social relevance, this content, which is published under the Creative Commons BYNC-ND license, also represents a precious
language resource for the machine translation research community, thanks to its size, variety of topics, and covered languages.
This effort repurposes the original content in a way which is more convenient for machine translation researchers.
This is the Tilde MODEL Corpus – Multilingual Open Data for European Languages.
The data has been collected from sites allowing free use and reuse of its content, as well as from Public Sector web sites. The activities have been undertaken as part of the ODINE Open Data Incubator for Europe, which aims to support the next generation of digital businesses and fast-track the development of new products and services. The corpus includes the following parts:
Tilde MODEL - EESC is a multilingual corpus compiled from document texts of European Economic and Social Committee document portal. Source: http://dm.eesc.europa.eu/
Tilde MODEL - RAPID multilingual parallel corpus is compiled from all press releases of Press Release Database of European Commission released between 1975 and end of 2016 as available from http://europa.eu/rapid/
Tilde MODEL - ECB multilingual parallel corpus is compiled from the multilingual pages of European Central Bank web site http://ebc.europa.eu/
Tilde MODEL - EMA is a corpus compiled from texts of European Medicines Agency document portal as available in http://www.ema.europa.eu/ at the end of 2016
Tilde MODEL - World Bank is a corpus compiled from texts of World Bank as available in http://www.worldbank.org/ in 2017
Tilde MODEL - AirBaltic.com Travel Destinations is a multilingual parallel corpus compiled from description texts of AirBaltic.com travel destinations as available in https://www.airbaltic.com/en/destinations/ in 2017
Tilde MODEL - LiveRiga.com is a multilingual parallel corpus compiled from Riga tourist attractions description texts of http://liveriga.com/ web site in 2017
Tilde MODEL - Lithuanian National Philharmonic Society is a parallel corpus compiled from texts of Lithuanian National Philharmonic Society web site http://www.filharmonija.lt/ in 2017
Tilde MODEL - mupa.hu is a parallel corpus from texts of Müpa Budapest - web site of Hungarian national culture house and concert venue https://www.mupa.hu/en/ compiled in spring of 2017
Tilde MODEL - fold.lv is a parallel corpus from texts of fold.lv portal http://www.fold.lv/en/ of the best of Latvian and foreign creative industries as compiled in spring of 2017
Tilde MODEL - czechtourism.com is a multilingual parallel corpus from texts of http://czechtourism.com/ portal compiled in spring of 2017
30 languages, 274 bitexts
total number of files: 125
total number of tokens: 1.43G
total number of sentence fragments: 62.44M
The Universal Declaration of Human Rights (UDHR) is a milestone document in the history of human rights. Drafted by
representatives with different legal and cultural backgrounds from all regions of the world, it set out, for the
first time, fundamental human rights to be universally protected. The Declaration was adopted by the UN General
Assembly in Paris on 10 December 1948 during its 183rd plenary meeting. The dataset includes translations of the
document in 464 languages and dialects.
© 1996 – 2009 The Office of the High Commissioner for Human Rights
This plain text version prepared by the “UDHR in Unicode” project, https://www.unicode.org/udhr.
Universal Dependencies is a project that seeks to develop cross-linguistically consistent treebank annotation for many languages, with the goal of facilitating multilingual parser development, cross-lingual learning, and parsing research from a language typology perspective. The annotation scheme is based on (universal) Stanford dependencies (de Marneffe et al., 2006, 2008, 2014), Google universal part-of-speech tags (Petrov et al., 2012), and the Interset interlingua for morphosyntactic tagsets (Zeman, 2008).
WikiLingua is a large-scale multilingual dataset for the evaluation of
crosslingual abstractive summarization systems. The dataset includes ~770k
article and summary pairs in 18 languages from WikiHow. The gold-standard
article-summary alignments across languages was done by aligning the images
that are used to describe each how-to step in an article.
WikiANN (sometimes called PAN-X) is a multilingual named entity recognition dataset consisting of Wikipedia articles annotated with LOC (location), PER (person), and ORG (organisation) tags in the IOB2 format. This version corresponds to the balanced train, dev, and test splits of Rahimi et al. (2019), which supports 176 of the 282 languages from the original WikiANN corpus.
It is a benchmark dataset for language identification and contains 235000 paragraphs of 235 languages
XGLUE is a new benchmark dataset to evaluate the performance of cross-lingual pre-trained
models with respect to cross-lingual natural language understanding and generation.
The benchmark is composed of the following 11 tasks:
- POS Tagging (POS)
- News Classification (NC)
- Query-Ad Matching (QADSM)
- Web Page Ranking (WPR)
- QA Matching (QAM)
- Question Generation (QG)
- News Title Generation (NTG)
For more information, please take a look at https://microsoft.github.io/XGLUE/.