Datasets:

arbml

/

masader

like 7

ArCOV19-Rumors

https://huggingface.co/datasets/arbml/ArCOV19_claims

https://gitlab.com/bigirqu/ArCOV-19/-/tree/master/ArCOV19-Rumors

unknown

ar

mixed

social media

text

crawling and annotation(other)

The first Arabic dataset for rumors verification in Twitter

9,414

sentences

Medium

Qatar University

nan

ArCOV19-Rumors: Arabic COVID-19 Twitter Dataset for Misinformation Detection

https://aclanthology.org/2021.wanlp-1.8.pdf

Arab

No

GitLab

Free

nan

No

fact checking

WANLP

nan

workshop

Arabic Natural Language Processing Workshop

nan

nan

nan

Fatima Haouari

SenWave

nan

https://github.com/gitdevqiang/SenWave

unknown

multilingual

mixed

social media

text

crawling and annotation(other)

The largest fine-grained annotated Covid-19 tweets dataset

10,000

sentences

Medium

Multiple institutions

nan

SenWave: Monitoring the Global Sentiments under the COVID-19 Pandemic

https://arxiv.org/pdf/2006.10842.pdf

Arab

No

GitHub

Upon-Request

nan

No

emotion detection

arXiv

nan

preprint

nan

nan

nan

nan

Nora Alturayeif

OpenITI-proc

nan

https://zenodo.org/record/2535593#.YWh7FS8RozU

CC BY 4.0

ar

ar-CLS: (Arabic (Classic))

other

text

crawling and annotation(other)

A linguistically annotated version of the OpenITI corpus, with annotations for lemmas, POS tags, parse trees, and morphological segmentation

1,500,000,000

tokens

Low

Multiple institutions

OpenITI

Studying the History of the Arabic Language: Language Technology and a Large-Scale Historical Corpus

https://arxiv.org/pdf/1809.03891.pdf

Arab-Latn

Yes

zenodo

Free

nan

No

text generation, language modeling

LRE

nan

journal

Language Resources and Evaluation

Yonatan Belinkov, Alexander Magidow, Alberto Barrón-Cedeño, Avi Shmidman, Maxim Romanov

nan

Arabic is a widely-spoken language with a long and rich history, but existing corpora and language technology focus mostly on modern Arabic and its varieties. Therefore, studying the history of the language has so far been mostly limited to manual analyses on a small scale. In this work, we present a large-scale historical corpus of the written Arabic language, spanning 1400 years. We describe our efforts to clean and process this corpus using Arabic NLP tools, including the identification of reused text. We study the history of the Arabic language using a novel automatic periodization algorithm, as well as other techniques. Our findings confirm the established division of written Arabic into Modern Standard and Classical Arabic, and confirm other established periodizations, while suggesting that written Arabic may be divisible into still further periods of development.

Yonatan Belinkov

APGC v2.0: Arabic Parallel Gender Corpus v2.0

nan

https://camel.abudhabi.nyu.edu/arabic-parallel-gender-corpus/

custom

ar

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

manual curation

The Arabic Parallel Gender Corpus v2.0 (APGC v2.0) is designed to support research on gender bias and personalization in natural language processing applications working on Arabic. It expands on Habash et al. (2019)’s Arabic Parallel Gender Corpus (APGC v1.0) by adding 2nd person targets as well increasing the total number of sentences over 6.5 times, reaching over 590K words.

80,326

sentences

Medium

NYU Abu Dhabi

APGC v1.0

The Arabic Parallel Gender Corpus 2.0: Extensions and Analyses

https://arxiv.org/pdf/2110.09216.pdf

Arab

Yes

CAMeL Resources

Upon-Request

nan

Yes

gender identification, gender rewriting

arXiv

nan

preprint

nan

Bashar Alhafni and Nizar Habash and Houda Bouamor.

New York University Abu Dhabi, Carnegie Mellon University in Qatar

Gender bias in natural language processing (NLP) applications, particularly machine translation, has been receiving increasing attention. Much of the research on this issue has focused on mitigating gender bias in English NLP models and systems. Addressing the problem in poorly resourced, and/or morphologically rich languages has lagged behind, largely due to the lack of datasets and resources. In this paper, we introduce a new corpus for gender identification and rewriting in contexts involving one or two target users (I and/or You) -- first and second grammatical persons with independent grammatical gender preferences. We focus on Arabic, a gender-marking morphologically rich language. The corpus has multiple parallel components: four combinations of 1st and 2nd person in feminine and masculine grammatical genders, as well as English, and English to Arabic machine translation output. This corpus expands on Habash et al. (2019)'s Arabic Parallel Gender Corpus (APGC v1.0) by adding second person targets as well as increasing the total number of sentences over 6.5 times, reaching over 590K words. Our new dataset will aid the research and development of gender identification, controlled text generation, and post-editing rewrite systems that could be used to personalize NLP applications and provide users with the correct outputs based on their grammatical gender preferences. We make the Arabic Parallel Gender Corpus (APGC v2.0) publicly available.

Bashar Alhafni

The Nine Books Of Arabic Hadith

https://huggingface.co/datasets/arbml/Hadith

https://github.com/abdelrahmaan/Hadith-Data-Sets

unknown

ar

ar-CLS: (Arabic (Classic))

other

text

crawling and annotation(other)

There are two files of Hadith, the first one for all hadith With Tashkil and Without Tashkel from the Nine Books that are 62,169 Hadith. The second one it Hadith pre-processing data, which is applyed normalization and removeing stop words and lemmatization on it

62,169

documents

Low

nan

nan

nan

nan

Arab

No

GitHub

Free

nan

No

text classification, text Similarity

nan

nan

nan

nan

nan

nan

nan

Abdulrahman Kamar

Shamela et al Arabic Corpus

nan

https://github.com/tarekeldeeb/GloVe-Arabic/tree/master/arabic_corpus

CC BY 4.0

ar

ar-CLS: (Arabic (Classic))

other

text

crawling

The arabic corpus {1.9B word} consists of the following resources: ShamelaLibrary348.7z link {1.15B} UN arabic corpus mirror1 mirror2 {0.37B} AraCorpus.tar.gz link {0.14B} Arabic Wikipedia Latest Articles Dump link {0.11B} Tashkeela-arabic-diacritized-text-utf8-0.3.zip link {0.07B} Arabic Tweets link {0.03B} watan-2004.7z link {0.01B}

1,754,541,204

tokens

Low

nan

nan

nan

nan

Arab

Yes

GitHub

Free

nan

No

text generation, language modeling

nan

nan

nan

nan

nan

nan

nan

Tarek Eldeeb

Quran Speech: Imam + Users

https://huggingface.co/datasets/arbml/quran_uthmani

https://github.com/tarekeldeeb/DeepSpeech-Quran/tree/master/data/quran

CC BY 4.0

ar

ar-CLS: (Arabic (Classic))

transcribed audio

spoken

crawling and annotation(other)

7 full Quran recitations + 18K filtered user recitation

61,000

sentences

Low

quran.ksu.edu.sa + tarteel.io

nan

nan

nan

Arab

No

GitHub

Free

nan

No

speech recognition

nan

nan

nan

nan

nan

nan

nan

Tarek Eldeeb

OntoNotes Release 5

nan

https://catalog.ldc.upenn.edu/LDC2013T19

LDC User Agreement

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

news articles

text

other

The goal of the project was to annotate a large corpus comprising various genres of text (news, conversational telephone speech, weblogs, usenet newsgroups, broadcast, talk shows) in three languages (English, Chinese, and Arabic) with structural information (syntax and predicate argument structure) and shallow semantics (word sense linked to an ontology and coreference).

300,000

tokens

Medium

LDC

nan

CoNLL-2012 Shared Task: Modeling Multilingual Unrestricted Coreference in OntoNotes

https://aclanthology.org/W12-4501.pdf

Arab

No

LDC

Free

nan

No

coreference resolution, word sense disambiguation, named entity recognition

SIGDAT

nan

workshop

Special Interest Group on Linguistic data and corpus-based approaches to NLP

nan

nan

nan

Amr Keleg

ArSarcasm-v2

https://huggingface.co/datasets/arbml/ArSarcasm_v2

https://github.com/iabufarha/ArSarcasm-v2

MIT License

ar

mixed

social media

text

crawling and annotation(other)

ArSarcasm-v2 is an extension of the original ArSarcasm dataset published along with the paper From Arabic Sentiment Analysis to Sarcasm Detection: The ArSarcasm Dataset. ArSarcasm-v2 consists of ArSarcasm along with portions of DAICT corpus and some new tweets. Each tweet was annotated for sarcasm, sentiment and dialect. The final dataset consists of 15,548 tweets divided into 12,548 training tweets and 3,000 testing tweets. ArSarcasm-v2 was used and released as a part of the shared task on sarcasm detection and sentiment analysis in Arabic

15,548

sentences

Medium

Multiple institutions

ArSarcasm: https://github.com/iabufarha/ArSarcasm

Overview of the WANLP 2021 Shared Task on Sarcasm and Sentiment Detection in Arabic

https://aclanthology.org/2021.wanlp-1.36/

Arab

No

GitHub

Free

nan

Yes

sarcasm detection, sentiment analysis, dialect identification

WANLP

20.0

workshop

Arabic Natural Language Processing Workshop

Ibrahim Abu Farha, Wajdi Zaghouani, Walid Magdy

The University of Edinburgh, Hamad Bin Khalifa University

This paper provides an overview of the WANLP 2021 shared task on sarcasm and sentiment detection in Arabic. The shared task has two subtasks: sarcasm detection (subtask 1) and sentiment analysis (subtask 2). This shared task aims to promote and bring attention to Arabic sarcasm detection, which is crucial to improve the performance in other tasks such as sentiment analysis. The dataset used in this shared task, namely ArSarcasm-v2, consists of 15,548 tweets labelled for sarcasm, sentiment and dialect. We received 27 and 22 submissions for subtasks 1 and 2 respectively. Most of the approaches relied on using and fine-tuning pre-trained language models such as AraBERT and MARBERT. The top achieved results for the sarcasm detection and sentiment analysis tasks were 0.6225 F1-score and 0.748 F1-PN respectively.

Ibrahim Abu Farha

AraCovid19-SSD: Arabic COVID-19 Sentiment and Sarcasm Detection Dataset

nan

https://github.com/MohamedHadjAmeur/AraCovid19-SSD

CC BY-NC-SA 4.0

ar

mixed

social media

text

crawling and annotation(other)

AraCovid19-SSD is a manually annotated Arabic COVID-19 sarcasm and sentiment detection dataset containing 5,162 tweets.

5,162

sentences

High

Research Centre on Scientific and Technical Information (CERIST)

nan

ARACOVID19-SSD: ARABIC COVID-19 SENTIMENT AND SARCASM DETECTION DATASET

https://arxiv.org/pdf/2110.01948v1.pdf

Arab

No

other

Upon-Request

nan

No

sarcasm detection, sentiment detection

arXiv

nan

preprint

nan

Mohamed Seghir Hadj Ameur, Hassina Aliane

Research Centre on Scientific and Technical Information (CERIST)

Coronavirus disease (COVID-19) is an infectious respiratory disease that was first discovered in late December 2019, in Wuhan, China, and then spread worldwide causing a lot of panic and death. Users of social networking sites such as Facebook and Twitter have been focused on reading, publishing, and sharing novelties, tweets, and articles regarding the newly emerging pandemic. A lot of these users often employ sarcasm to convey their intended meaning in a humorous, funny, and indirect way making it hard for computer-based applications to automatically understand and identify their goal and the harm level that they can inflect. Motivated by the emerging need for annotated datasets that tackle these kinds of problems in the context of COVID-19, this paper builds and releases AraCOVID19-SSD1 a manually annotated Arabic COVID-19 sarcasm and sentiment detection dataset containing 5,162 tweets. To confirm the practical utility of the built dataset, it has been carefully analyzed and tested using several classification models.

Abdelrahman Kaseb

DiaLex

https://huggingface.co/datasets/arbml/dialex

https://github.com/UBC-NLP/dialex

unknown

ar

mixed

other

text

other

A Benchmark for Evaluating Multidialectal Arabic Word Embeddings

3,070

sentences

Low

Multiple institutions

nan

DiaLex: A Benchmark for Evaluating Multidialectal Arabic Word Embeddings

https://aclanthology.org/2021.wanlp-1.2

Arab

No

GitHub

Free

nan

No

benchmarking multidialectal word embeddings

WANLP

nan

workshop

Arabic Natural Language Processing Workshop

Muhammad Abdul-Mageed, Shady Elbassuoni, Jad Doughman, AbdelRahim Elmadany, El Moatez Billah Nagoudi, Yorgo Zoughby, Ahmad Shaher, Iskander Gaba, Ahmed Helal, and Mohammed El-Razzaz.

nan

Word embeddings are a core component of modern natural language processing systems, making the ability to thoroughly evaluate them a vital task. We describe DiaLex, a benchmark for intrinsic evaluation of dialectal Arabic word embeddings. DiaLex covers five important Arabic dialects: Algerian, Egyptian, Lebanese, Syrian, and Tunisian. Across these dialects, DiaLex provides a testbank for six syntactic and semantic relations, namely male to female, singular to dual, singular to plural, antonym, comparative, and genitive to past tense. DiaLex thus consists of a collection of word pairs representing each of the six relations in each of the five dialects. To demonstrate the utility of DiaLex, we use it to evaluate a set of existing and new Arabic word embeddings that we developed. Beyond evaluation of word embeddings, DiaLex supports efforts to integrate dialects into the Arabic language curriculum. It can be easily translated into Modern Standard Arabic and English, which can be useful for evaluating word translation. Our benchmark, evaluation code, and new word embedding models will be publicly available.

Iskander Gaba

COVID-19 Disinfo: COVID-19 Disinformation Twitter Dataset

https://huggingface.co/datasets/arbml/COVID_19_Disinformation_ar

https://github.com/firojalam/COVID-19-disinformation

CC BY-NC-SA 4.0

multilingual

mixed

social media

text

crawling and annotation(other)

With the emergence of the COVID-19 pandemic, the political and the medical aspects of disinformation merged as the problem got elevated to a whole new level to become the first global infodemic. Fighting this infodemic has been declared one of the most important focus areas of the World Health Organization, with dangers ranging from promoting fake cures, rumors, and conspiracy theories to spreading xenophobia and panic. Addressing the issue requires solving a number of challenging problems such as identifying messages containing claims, determining their check-worthiness and factuality, and their potential to do harm as well as the nature of that harm, to mention just a few. To address this gap, we release a large dataset of 16K manually annotated tweets for fine-grained disinformation analysis that focuses on COVID-19, combines the perspectives and the interests of journalists, fact-checkers, social media platforms, policy makers, and society, and covers Arabic, Bulgarian, Dutch, and English. Finally, we show strong evaluation results using pretrained Transformers, thus confirming the practical utility of the dataset in monolingual multilingual, and single task vs. multitask settings.

5,000

sentences

High

Multiple institutions

nan

Fighting the COVID-19 Infodemic: Modeling the Perspective of Journalists, Fact-Checkers, Social Media Platforms, Policy Makers, and the Society

https://pure.rug.nl/ws/portalfiles/portal/203339411/2021.findings_emnlp.56.pdf

Arab

No

GitHub

Free

nan

No

fact checking

Findings of EMNLP

nan

conference

findings of Conference on Empirical Methods in Natural Language Processing

Firoj Alam, Shaden Shaar, Fahim Dalvi, Hassan Sajjad, Alex Nikolov, Hamdy Mubarak, Giovanni Da San Martino, Ahmed Abdelali, Nadir Durrani, Kareem Darwish, Abdulaziz Al-Homaid, Wajdi Zaghouani, Tommaso Caselli, Gijs Danoe, Friso Stolk, Britt Bruntink, Preslav Nakov

nan

With the emergence of the COVID-19 pandemic, the political and the medical aspects of disinformation merged as the problem got elevated to a whole new level to become the first global infodemic. Fighting this infodemic has been declared one of the most important focus areas of the World Health Organization, with dangers ranging from promoting fake cures, rumors, and conspiracy theories to spreading xenophobia and panic. Addressing the issue requires solving a number of challenging problems such as identifying messages containing claims, determining their check-worthiness and factuality, and their potential to do harm as well as the nature of that harm, to mention just a few. To address this gap, we release a large dataset of 16K manually annotated tweets for fine-grained disinformation analysis that focuses on COVID-19, combines the perspectives and the interests of journalists, fact-checkers, social media platforms, policy makers, and society, and covers Arabic, Bulgarian, Dutch, and English. Finally, we show strong evaluation results using pretrained Transformers, thus confirming the practical utility of the dataset in monolingual multilingual, and single task vs. multitask settings.

Abdelrahman Kaseb

Senti lex

https://huggingface.co/datasets/senti_lex

https://www.kaggle.com/datasets/rtatman/sentiment-lexicons-for-81-languages

GPL-3.0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

wikipedia

text

crawling and annotation(other)

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

2,794

tokens

Low

Stony Brook University

nan

Building Sentiment Lexicons for All Major Languages

https://aclanthology.org/P14-2063.pdf

Arab

Yes

kaggle

Free

nan

No

sentiment analysis

ACL

186.0

conference

Assofications of computation linguisitcs

Yanqing Chen, S. Skiena

Stony Brook University

Sentiment analysis in a multilingual world remains a challenging problem, because developing language-specific sentiment lexicons is an extremely resourceintensive process. Such lexicons remain a scarce resource for most languages. In this paper, we address this lexicon gap by building high-quality sentiment lexicons for 136 major languages. We integrate a variety of linguistic resources to produce an immense knowledge graph. By appropriately propagating from seed words, we construct sentiment lexicons for each component language of our graph. Our lexicons have a polarity agreement of 95.7% with published lexicons, while achieving an overall coverage of 45.2%. We demonstrate the performance of our lexicons in an extrinsic analysis of 2,000 distinct historical figures’ Wikipedia articles on 30 languages. Despite cultural difference and the intended neutrality of Wikipedia articles, our lexicons show an average sentiment correlation of 0.28 across all language pairs.

Abdelrahman Kaseb

POLYGLOT-NER

nan

https://www3.cs.stonybrook.edu/~polyglot/ner2/

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

wikipedia

text

crawling and annotation(other)

Polyglot-NER A training dataset automatically generated from Wikipedia and Freebase the task of named entity recognition. The dataset contains the basic Wikipedia based training data for 40 languages we have (with coreference resolution) for the task of named entity recognition. The details of the procedure of generating them is outlined in Section 3 of the paper (https://arxiv.org/abs/1410.3791). Each config contains the data corresponding to a different language. For example, "es" includes only spanish examples.

10,000,144

tokens

Low

Stony Brook University

nan

POLYGLOT-NER: Massive Multilingual Named Entity Recognition

https://arxiv.org/pdf/1410.3791.pdf

Arab-Latn

Yes

other

Free

nan

No

named entity recognition

arXiv

161.0

preprint

nan

Rami Al-Rfou, Vivek Kulkarni, Bryan Perozzi, Steven Skiena

Stony Brook University

The increasing diversity of languages used on the web introduces a new level of complexity to Information Retrieval (IR) systems. We can no longer assume that textual content is written in one language or even the same language family. In this paper, we demonstrate how to build massive multilingual annotators with minimal human expertise and intervention. We describe a system that builds Named Entity Recognition (NER) annotators for 40 major languages using Wikipedia and Freebase. Our approach does not require NER human annotated datasets or language specific resources like treebanks, parallel corpora, and orthographic rules. The novelty of approach lies therein - using only language agnostic techniques, while achieving competitive performance. Our method learns distributed word representations (word embeddings) which encode semantic and syntactic features of words in each language. Then, we automatically generate datasets from Wikipedia link structure and Freebase attributes. Finally, we apply two preprocessing stages (oversampling and exact surface form matching) which do not require any linguistic expertise. Our evaluation is two fold: First, we demonstrate the system performance on human annotated datasets. Second, for languages where no gold-standard benchmarks are available, we propose a new method, distant evaluation, based on statistical machine translation.

Abdelrahman Kaseb

iSarcasmEval: SemEval-2022 Task 6

https://huggingface.co/datasets/arbml/iSarcasmEval_task_A

https://github.com/iabufarha/iSarcasmEval

unknown

multilingual

mixed

social media

text

crawling and annotation(other)

A Dataset of Intended Sarcasm

4,447

sentences

High

University of Edinburgh

nan

iSarcasm: A Dataset of Intended Sarcasm

https://arxiv.org/pdf/1911.03123.pdf

Arab-Latn

No

GitHub

Free

nan

Yes

sarcasm detection

SEMEVAL

17.0

workshop

International Workshop on Semantic Evaluation

Silviu Oprea, Walid Magdy

University of Edinburgh

We consider the distinction between intended and perceived sarcasm in the context of textual sarcasm detection. The former occurs when an utterance is sarcastic from the perspective of its author, while the latter occurs when the utterance is interpreted as sarcastic by the audience. We show the limitations of previous labelling methods in capturing intended sarcasm and introduce the iSarcasm dataset of tweets labeled for sarcasm directly by their authors. Examining the state-of-the-art sarcasm detection models on our dataset showed low performance compared to previously studied datasets, which indicates that these datasets might be biased or obvious and sarcasm could be a phenomenon under-studied computationally thus far. By providing the iSarcasm dataset, we aim to encourage future NLP research to develop methods for detecting sarcasm in text as intended by the authors of the text, not as labeled under assumptions that we demonstrate to be sub-optimal.

Abdelrahman Kaseb

Arabic Hate Speech 2022

nan

https://codalab.lisn.upsaclay.fr/competitions/2324

custom

ar

mixed

social media

text

crawling and annotation(other)

Fine-Grained Hate Speech Detection on Arabic Twitter

10,157

sentences

High

QCRI

nan

Emojis as Anchors to Detect Arabic Offensive Language and Hate Speech

https://arxiv.org/pdf/2201.06723.pdf

Arab

No

CodaLab

Free

nan

Yes

hate speech detection

OSACT

nan

workshop

Workshop on Open-Source Arabic Corpora and Processing Tools

Hamdy Mubarak, Sabit Hassan , and Shammur Absar Chowdhury

Qatar Computing Research Institute

We introduce a generic, language-independent method to collect a large percentage of offensive and hate tweets regardless of their topics or genres. We harness the extralinguistic information embedded in the emojis to collect a large number of offensive tweets. We apply the proposed method on Arabic tweets and compare it with English tweets -- analyzing some cultural differences. We observed a constant usage of these emojis to represent offensiveness in throughout different timelines in Twitter. We manually annotate and publicly release the largest Arabic dataset for offensive, fine-grained hate speech, vulgar and violence content. Furthermore, we benchmark the dataset for detecting offense and hate speech using different transformer architectures and performed in-depth linguistic analysis. We evaluate our models on external datasets -- a Twitter dataset collected using a completely different method, and a multi-platform dataset containing comments from Twitter, YouTube and Facebook, for assessing generalization capability. Competitive results on these datasets suggest that the data collected using our method captures universal characteristics of offensive language. Our findings also highlight the common words used in offensive communications; common targets for hate speech; specific patterns in violence tweets and pinpoints common classification errors due to the need to understand the context, consider culture and background and the presence of sarcasm among others.

Abdelrahman Kaseb

ArCorona: Analyzing Arabic Tweets in the Early Days of Coronavirus (COVID-19) Pandemic

https://huggingface.co/datasets/arbml/ArCorona

https://alt.qcri.org/resources/ArCorona.tsv

unknown

ar

mixed

social media

text

crawling and annotation(other)

Collected to prevent spreading of rumors and misinformation about the virus or bad cures

8,000

sentences

High

QCRI

nan

ArCorona: Analyzing Arabic Tweets in the Early Days of Coronavirus (COVID-19) Pandemic

https://arxiv.org/abs/2012.01462

Arab

No

other

Free

nan

No

COVID misinformation detection

arXiv

7.0

preprint

nan

Hamdy Mubarak, Sabit Hassan

Qatar Computing Research Institute

Over the past few months, there were huge numbers of circulating tweets and discussions about Coronavirus (COVID-19) in the Arab region. It is important for policy makers and many people to identify types of shared tweets to better understand public behavior, topics of interest, requests from governments, sources of tweets, etc. It is also crucial to prevent spreading of rumors and misinformation about the virus or bad cures. To this end, we present the largest manually annotated dataset of Arabic tweets related to COVID-19. We describe annotation guidelines, analyze our dataset and build effective machine learning and transformer based models for classification.

Abdelrahman Kaseb

Adult Content Detection on Arabic Twitter: Analysis and Experiments

nan

https://alt.qcri.org/resources/AdultContentDetection.zip

unknown

ar

mixed

social media

text

crawling and annotation(other)

Adult Content Detection on Arabic Twitter

50,000

sentences

High

QCRI

nan

Adult Content Detection on Arabic Twitter: Analysis and Experiments

https://aclanthology.org/2021.wanlp-1.14.pdf

Arab

No

other

Free

nan

Yes

adult language detection

arXiv

5.0

preprint

nan

Hamdy Mubarak, Sabit Hassan and Ahmed Abdelali

Qatar Computing Research Institute

With Twitter being one of the most popular social media platforms in the Arab region, it is not surprising to find accounts that post adult content in Arabic tweets; despite the fact that these platforms dissuade users from such content. In this paper, we present a dataset of Twitter accounts that post adult content. We perform an in-depth analysis of the nature of this data and contrast it with normal tweet content. Additionally, we present extensive experiments with traditional machine learning models, deep neural networks and contextual embeddings to identify such accounts. We show that from user information alone, we can identify such accounts with F1 score of 94.7% (macro average). With the addition of only one tweet as input, the F1 score rises to 96.8%.

Abdelrahman Kaseb

Understanding and Detecting Dangerous Speech in Social Media

https://huggingface.co/datasets/arbml/Dangerous_Dataset

https://github.com/UBC-NLP/Arabic-Dangerous-Dataset

unknown

ar

mixed

social media

text

crawling and annotation(other)

Dangerous speech detection

5,000

sentences

High

The University of British Columbia

nan

Understanding and Detecting Dangerous Speech in Social Media

https://arxiv.org/pdf/2005.06608.pdf

Arab

No

GitHub

Free

nan

No

dangerous speech detection

arXiv

8.0

preprint

nan

Ali Alshehri, El Moatez Billah Nagoudi, Muhammad Abdul-Mageed

The University of British Columbia

Social media communication has become a significant part of daily activity in modern societies. For this reason, ensuring safety in social media platforms is a necessity. Use of dangerous language such as physical threats in online environments is a somewhat rare, yet remains highly important. Although several works have been performed on the related issue of detecting offensive and hateful language, dangerous speech has not previously been treated in any significant way. Motivated by these observations, we report our efforts to build a labeled dataset for dangerous speech. We also exploit our dataset to develop highly effective models to detect dangerous content. Our best model performs at 59.60% macro F1, significantly outperforming a competitive baseline.

Abdelrahman Kaseb

APCD

https://huggingface.co/datasets/arbml/APCD

https://hci-lab.github.io/LearningMetersPoems/

unknown

ar

ar-CLS: (Arabic (Classic))

other

text

crawling

A dataset of Arabic poetry containing 1,831,770 along with there meters.

1,831,770

sentences

Low

Helwan University

nan

Learning meters of Arabic and English poems with Recurrent Neural Networks: a step forward for language understanding and synthesis

https://arxiv.org/pdf/1905.05700.pdf

Arab

No

GitHub

Free

nan

No

meter classification

arXiv

nan

preprint

nan

Waleed A. Yousefa,Omar M. Ibrahime,Taha M. Madboulya, Moustafa A. Mahmoud

Nile university, Nile university, Nile university

Recognizing a piece of writing as a poem or prose is usually easy for the majority of people; however, only specialists can determine which meter a poem belongs to. In this paper, we build Recurrent Neural Network (RNN) models that can classify poems according to their meters from plain text. The input text is encoded at the character level and directly fed to the models without feature handcrafting. This is a step forward for machine understanding and synthesis of languages in general, and Arabic language in particular. Among the 16 poem meters of Arabic and the 4 meters of English the networks were able to correctly classify poem with an overall accuracy of 96.38% and 82.31% respectively. The poem datasets used to conduct this research were massive, over 1.5 million of verses, and were crawled from different nontechnical sources, almost Arabic and English literature sites, and in different heterogeneous and unstructured formats. These datasets are now made publicly available in clean, structured, and documented format for other future research. To the best of the authors’ knowledge, this research is the first to address classifying poem meters in a machine learning approach, in general, and in RNN featureless based approach, in particular. In addition, the dataset is the first publicly available dataset ready for the purpose of future computational research. Index Terms—Poetry, Meters, Al-’arud, Arabic, English, Recurrent Neural Networks, RNN, Deep Learning, Deep Neural Networks, DNN, Classification, Text Mining.

Zaid Alyafeai

IDRISI-R

nan

https://github.com/rsuwaileh/IDRISI

custom

multilingual

mixed

social media

text

crawling and annotation(other)

IDRISI-R is the largest-scale publicly-available Twitter Location Mention Recognition (LMR) dataset, in both English and Arabic languages. It contains 41 disaster events of different types such as floods, fires, etc. In addition to tagging LMs in text, the LMs are labeled for location types such as countries, cities, streets, POIs, etc.

1,268,912

sentences

Medium

Qatar University, QCRI, Hamad Bin Khalifa University

Kawarith and humAID datasets

(Under review) IDRISI-R: Large-scale English and Arabic Location Mention Recognition Datasets for Disaster Response over Twitter

nan

Arab-Latn

Yes

GitHub

Free

nan

Yes

location mention recognition

nan

nan

journal

nan

Reem Suwaileh, Tamer Elsayed, Muhammad Imran

Computer Science and Engineering Department, College of Engineering, Qatar University, Doha, Qatar. Qatar Computing Research Institute (QCRI), Hamad Bin Khalifa University (HBKU), Doha, Qatar

While utilizing Twitter data for crisis management, a critical challenge that hinders authorities' response is the scarcity of geotagged messages. Although studies show the presence of toponyms in tweets and their effectiveness as alternative information to geotagged messages, limited focus has been given to location mention recognition in tweets. In fact, the community lacks a standard dataset to thrive research towards building robust models and solutions. To bridge this gap, we present two human-labeled datasets for the location mention recognition task in text messages, particularly tweets. The human annotation task labels toponym spans and assigns a location type (e.g., country, state, city) to them. The datasets contain tweets from 41 large-scale disaster events (e.g., floods, earthquakes) covering a wide geographical area of English and Arabic-speaking countries. Moreover, we benchmark the datasets using standard and deep learning models and present rigorous quantitative and qualitative analysis to highlight their superiority over past efforts. Last but not least, the trained models are used to process raw data comprising millions of tweets and offered as a silver dataset.

Reem Suwaileh

AFND

https://huggingface.co/datasets/arbml/AFND

https://data.mendeley.com/datasets/67mhx6hhzd/1

CC BY 4.0

ar

mixed

news articles

text

crawling

AFND consists of 606912 public news articles that were scraped from 134 public news websites of 19 different Arab countries over a 6-month period using Python scripts.

606,912

documents

Low

Multiple Institutions

nan

nan

AFND: Arabic fake news dataset for the detection and classification of articles credibility

Arab

No

Mendeley Data

Free

nan

No

news credibility detection

Data in Brief

nan

journal

Data in Brief

Ashwaq Khalila, Moath Jarrah, Monther Aldwairi, ManarJaradat

Department of Computer Engineering, Jordan University of Science and Technology, Department of Computer Engineering, Jordan University of Science and Technology,College of Technological Innovation, Zayed University, Department of Computer Engineering, The Hashemite University

The news credibility detection task has started to gain more attention recently due to the rapid increase of news on different social media platforms. This article provides a large, labeled, and diverse Arabic Fake News Dataset (AFND) that is collected from public Arabic news websites. This dataset enables the research community to use supervised and unsupervised machine learning algorithms to classify the credibility of Arabic news articles. AFND consists of 606912 public news articles that were scraped from 134 public news websites of 19 different Arab countries over a 6-month period using Python scripts. The Arabic fact-check platform, Misbar, is used manually to classify each public news source into credible, not credible, or undecided. Weak supervision is applied to label news articles with the same label as the public source. AFND is imbalanced in the number of articles in each class. Hence, it is useful for researchers who focus on finding solutions for imbalanced datasets. The dataset is available in JSON format and can be accessed from Mendeley Data repository.

Zaid Alyafeai

AT-ODTSA

https://huggingface.co/datasets/arbml/AT_ODSTA

https://github.com/sabudalfa/AT-ODTSA

unknown

ar

mixed

social media

text

crawling and annotation(other)

A dataset of Arabic Tweets for Open-Domain Targeted Sentiment Analysis, which includes Arabic tweets along with labels that specify targets (topics) and sentiments (opinions) expressed in the collected tweets.

3,000

sentences

Medium

Multiple Institutions

nan

AT-ODTSA: a Dataset of Arabic Tweets for Open Domain Targeted Sentiment Analysis

https://journal.uob.edu.bh/bitstream/handle/123456789/4607/IJCDS-1101105-1570749771.pdf

Arab

No

GitHub

Free

nan

No

open-domain targeted sentiment Analysis

IJCDS

nan

journal

International Journal of Computing and Digital Systems

Shaaban Sahmoud, Shadi Abudalfa, Wisam Elmasry

Department of Computer Engineering, Fatih Sultan Mehmet Vakif University, Information Technology Department, University College of Applied Sciences, 3Department of Computer Engineering, Istanbul Kultur University,

In the field of sentiment analysis, most of research has conducted experiments on datasets collected from Twitter for manipulating a specific language. Little number of datasets has been collected for detecting sentiments expressed in Arabic tweets. Moreover, very limited number of such datasets is suitable for conducting recent research directions such as target dependent sentiment analysis and open-domain targeted sentiment analysis. Thereby, there is a dire need for reliable datasets that are specifically acquired for open-domain targeted sentiment analysis with Arabic language. Therefore, in this paper, we introduce AT-ODTSA, a dataset of Arabic Tweets for Open-Domain Targeted Sentiment Analysis, which includes Arabic tweets along with labels that specify targets (topics) and sentiments (opinions) expressed in the collected tweets. To the best of our knowledge, our work presents the first dataset that manually annotated for applying Arabic open-domain targeted sentiment analysis. We also present a detailed statistical analysis of the dataset. The AT-ODTSA dataset is suitable for train numerous machine learning models such as a deep learning-based model.

Zaid Alyafeai

ArCovidVac

https://huggingface.co/datasets/arbml/ArCovidVac

https://alt.qcri.org/resources/ArCovidVac.zip

unknown

ar

mixed

social media

text

crawling and annotation(other)

the largest manually annotated Arabic tweet dataset, ArCovidVac, for the COVID-19 vaccination campaign, covering many countries in the Arab region

10,000

sentences

High

QCRI

nan

ArCovidVac: Analyzing Arabic Tweets About COVID-19 Vaccination

https://arxiv.org/pdf/2201.06496.pdf

Arab

No

QCRI Resources

Free

nan

Yes

informativeness, text classification, stance detection

LREC

nan

conference

Language Resources and Evaluation Conference

Hamdy Mubarak, Sabit Hassan, Shammur Absar Chowdhury, Firoj Alam

Qatar Computing Research Institute, HBKU; University of Pittsburgh;Qatar Computing Research Institute, HBKU

The emergence of the COVID-19 pandemic and the first global infodemic have changed our lives in many different ways. We relied on social media to get the latest information about COVID-19 pandemic and at the same time to disseminate information. The content in social media consisted not only health related advise, plans, and informative news from policymakers, but also contains conspiracies and rumors. It became important to identify such information as soon as they are posted to make an actionable decision (e.g., debunking rumors, or taking certain measures for traveling). To address this challenge, we develop and publicly release the first largest manually annotated Arabic tweet dataset, ArCovidVac, for the COVID-19 vaccination campaign, covering many countries in the Arab region. The dataset is enriched with different layers of annotation, including, (i) Informativeness (more vs. less importance of the tweets); (ii) fine-grained tweet content types (e.g., advice, rumors, restriction, authenticate news/information); and (iii) stance towards vaccination (pro-vaccination, neutral, anti-vaccination). Further, we performed in-depth analysis of the data, exploring the popularity of different vaccines, trending hashtags, topics and presence of offensiveness in the tweets. We studied the data for individual types of tweets and temporal changes in stance towards vaccine. We benchmarked the ArCovidVac dataset using transformer architectures for informativeness, content types, and stance detection.

Zaid Alyafeai

APCD2

https://huggingface.co/datasets/arbml/APCDv2

https://github.com/Gheith-Abandah/classify-arabic-poetry

unknown

ar

ar-CLS: (Arabic (Classic))

other

text

crawling

1657 k verses of poems and prose to develop neural networks to classify and diacritize Arabic poetry

1,831,770

sentences

Low

The University of Jordan

APCD

Classifying and diacritizing Arabic poems using deep recurrent neural networks

https://www.sciencedirect.com/science/article/pii/S1319157820305784/pdfft?md5=07be922e052bf43933bdb7bea5189718&pid=1-s2.0-S1319157820305784-main.pdf

Arab

No

GitHub

Free

nan

Yes

meter classification

nan

nan

journal

Journal of King Saud University - Computer and Information Sciences

Gheith A. Abandah, Mohammed Z. Khedher, Mohammad R. Abdel-Majeed, Hamdi M Mansour, Salma F Hulliel, Lara M Bisharata

School of Engineering, The University of Jordan,School of Engineering, The University of Jordan,School of Engineering, The University of Jordan,School of Arts, The University of Jordan, School of Engineering, The University of Jordan, School of Engineering, The University of Jordan

Poetry has a prominent history in Arabic literature. The classical Arabic poetry has 16 m that vary in rhythm and target purpose. Chanting a poem eloquently requires knowing the poem’s meter and obtaining a diacritized version of its verses (letters inscribed with their short vowels); diacritics are often not inscribed in Arabic texts. This work proposes solutions to classify input Arabic text into the 16 poetry meters and prose. It also investigates the automatic diacritization of Arabic poetry. We adopt machine learning approach using a large dataset of 1657 k verses of poems and prose to develop neural networks to classify and diacritize Arabic poetry. We propose deep and narrow recurrent neural networks with bidirectional long short-term memory cells for solving these problems. The proposed model classifies the input text with an average accuracy of 97.27%, which is significantly higher than previous work. We also propose a solution that achieves an accuracy that approaches 100% when multiple verses of the same poem are available through predicting the class from the aggregate probabilities of the multiple verses. Diacritizing poetry is much harder than diacritizing prose due to the poet’s meticulous selection of phrases and relaxation of some diacritization rules.

Zaid Alyafeai

Author Attribution Tweets

https://huggingface.co/datasets/arbml/Author_Attribution_Tweets

https://fada.birzeit.edu/handle/20.500.11889/6743

unknown

ar

ar-MSA: (Arabic (Modern Standard Arabic))

social media

text

crawling and annotation(other)

consists of 71,397 tweets for 45 authors for MSA collected from twitter.

71,397

sentences

Medium

Birzeit University

nan

Authorship Attribution of Modern Standard Arabic Short Texts

https://fada.birzeit.edu/bitstream/20.500.11889/6787/1/AA_PAPER___ACM.pdf

Arab

No

other

Free

nan

Yes

authorship attribution

nan

nan

preprint

nan

YARA ABUHAMMAD, YARA ADDABE, NATALY AYYAD, ADNAN YAHYA

Department of Electrical and Computer Engineering, Birzeit University, Palestine, Department of Electrical and Computer Engineering, Birzeit University, Palestine, Department of Electrical and Computer Engineering, Birzeit University, Palestine, Department of Electrical and Computer Engineering, Birzeit University, Palestine

Text data, including short texts, constitute a major share of web content. The availability of this data to billions of users triggers frequent plagiarism attacks. Authorship Attribution (AA) seeks to identify the most probable author of a given text based on similarity to the writing style of potential authors. In this paper, we approach AA as a writing style profile generation process, where we group text instances for each author into a single profile. We use Twitter as the source for our short Modern Standard Arabic (MSA) texts. Numerous experiments with various training approaches, tools and features allowed us to settle on a text representation method that relies on text concatenation of Arabic tweets to form chunks, which are then duplicated to reach a precalculated length. These chunks are used to train machine learning models for our 45 author profiles. This allowed us to achieve accuracies up to 99%, which compares favorably with the best results reported in the literature

Zaid Alyafeai

Sa`7r

https://huggingface.co/datasets/arbml/SaudiIrony

https://github.com/iwan-rg/Saudi-Dialect-Irony-Dataset

CC0

ar

ar-SA: (Arabic (Saudi Arabia))

social media

text

crawling and annotation(other)

The dataset was collected using Twitter API and it consists of 19,810 tweets, 8,089 of them are labeled as ironic tweets.

19,810

sentences

Medium

King Saud University

nan

Sa`7r: A Saudi Dialect Irony Dataset

nan

Arab

No

GitHub

Free

nan

No

irony detection

OSACT

nan

workshop

Open-Source Arabic Corpora and Processing Tools

Halah AlMazrua, Najla AlHazzani, Amaal AlDawod, Lama AlAwlaqi, Noura AlReshoudi, Hend Al-Khalifa and Luluh AlDhubayi

nan

In sentiment analysis, detecting irony is considered a major challenge. The key problem with detecting irony is the difficulty to recognize the implicit and indirect phrases which signifies the opposite meaning. In this paper, we present Sa`7r ساخرthe Saudi irony dataset, and describe our efforts in constructing it. The dataset was collected using Twitter API and it consists of 19,810 tweets, 8,089 of them are labeled as ironic tweets. We trained several models for irony detection task using machine learning models and deep learning models. The machine learning models include: K-Nearest Neighbor (KNN), Logistic Regression (LR), Support Vector Machine (SVM), and Naïve Bayes (NB). While the deep learning models include BiLSTM and AraBERT. The detection results show that among the tested machine learning models, the SVM outperformed other classifiers with an accuracy of 0.68. On the other hand, the deep learning models achieved an accuracy of 0.66 in the BiLSTM model and 0.71 in the AraBERT model. Thus, the AraBERT model achieved the most accurate result in detecting irony phrases in Saudi Dialect.

Zaid Alyafeai

Arabic Hate Speech 2022 Shared Task

https://huggingface.co/datasets/arbml/Arabic_Hate_Speech

https://sites.google.com/view/arabichate2022/home

custom

ar

mixed

social media

text

crawling and annotation(other)

largest Arabic dataset for offensive, fine-grained hate speech, vulgar and violence content

12,698

sentences

High

QCRI

nan

Emojis as Anchors to Detect Arabic Offensive Language and Hate Speech

https://arxiv.org/pdf/2201.06723.pdf

Arab

No

QCRI Resources

Free

nan

Yes

offensive language detection, hate speech detection

arXiv

nan

preprint

nan

nan

Hamdy Mubarak, Sabit Hassan , and Shammur Absar Chowdhury

We introduce a generic, language-independent method to collect a large percentage of offensive and hate tweets regardless of their topics or genres. We harness the extralinguistic information embedded in the emojis to collect a large number of offensive tweets. We apply the proposed method on Arabic tweets and compare it with English tweets – analysing key cultural differences. We observed a constant usage of these emojis to represent offensiveness throughout different timespans on Twitter. We manually annotate and publicly release the largest Arabic dataset for offensive, fine-grained hate speech, vulgar and violence content. Furthermore, we benchmark the dataset for detecting offensiveness and hate speech using different transformer architectures and perform in-depth linguistic analysis. We evaluate our models on external datasets – a Twitter dataset collected using a completely different method, and a multi-platform dataset containing comments from Twitter, YouTube and Facebook, for assessing generalization capability. Competitive results on these datasets suggest that the data collected using our method captures universal characteristics of offensive language. Our findings also highlight the common words used in offensive communications, common targets for hate speech, specific patterns in violence tweets; and pinpoint common classification errors that can be attributed to limitations of NLP models. We observe that even state-ofthe-art transformer models may fail to take into account culture, background and context or understand nuances present in real-world data such as sarcasm.

Zaid Alyafeai

xquad

https://huggingface.co/datasets/xquad

https://github.com/deepmind/xquad

CC BY-SA 4.0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

human translation

a benchmark dataset for evaluating cross-lingual question answering performance. The dataset consists of a subset of 240 paragraphs and 1190 question-answer pairs from the development set of SQuAD v1.1 (Rajpurkar et al., 2016) together with their professional translations into ten languages: Spanish, German, Greek, Russian, Turkish, Arabic, Vietnamese, Thai, Chinese, and Hindi. Consequently, the dataset is entirely parallel across 11 languages.

1,190

documents

Low

DeepMind

SQuAD

On the Cross-lingual Transferability of Monolingual Representations

https://aclanthology.org/2020.acl-main.421.pdf

Arab

No

GitHub

Free

nan

Yes

question answering

ACL

nan

conference

Association of Computation Linguistics

Mikel Artetxe†, Sebastian Ruder, Dani Yogatama

HiTZ Center, University of the Basque Country; DeepMind, DeepMind

State-of-the-art unsupervised multilingual models (e.g., multilingual BERT) have been shown to generalize in a zero-shot crosslingual setting. This generalization ability has been attributed to the use of a shared subword vocabulary and joint training across multiple languages giving rise to deep multilingual abstractions. We evaluate this hypothesis by designing an alternative approach that transfers a monolingual model to new languages at the lexical level. More concretely, we first train a transformer-based masked language model on one language, and transfer it to a new language by learning a new embedding matrix with the same masked language modeling objective—freezing parameters of all other layers. This approach does not rely on a shared vocabulary or joint training. However, we show that it is competitive with multilingual BERT on standard cross-lingual classification benchmarks and on a new Cross-lingual Question Answering Dataset (XQuAD). Our results contradict common beliefs of the basis of the generalization ability of multilingual models and suggest that deep monolingual models learn some abstractions that generalize across languages. We also release XQuAD as a more comprehensive cross-lingual benchmark, which comprises 240 paragraphs and 1190 question-answer pairs from SQuAD v1.1 translated into ten languages by professional translators.

Zaid Alyafeai

mC4

https://huggingface.co/datasets/mc4

https://www.tensorflow.org/datasets/catalog/c4#c4multilingual_nights_stay

CC BY 4.0

multilingual

mixed

other

text

crawling

A colossal, cleaned version of Common Crawl's web crawl corpus.

53,256,040

documents

Low

Google

C4

A colossal, cleaned version of Common Crawl's web crawl corpus.

https://arxiv.org/pdf/1910.10683.pdf

Arab

No

other

Free

nan

Yes

text generation, language modeling

JMLR

nan

journal

Journal of Machine Learning Research

Colin Raffel, Noam Shazeer, Adam Roberts, Ktherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, Peter J. Liu

Goolge; Goolge; Goolge; Goolge; Goolge; Goolge; Goolge;

Transfer learning, where a model is first pre-trained on a data-rich task before being finetuned on a downstream task, has emerged as a powerful technique in natural language processing (NLP). The effectiveness of transfer learning has given rise to a diversity of approaches, methodology, and practice. In this paper, we explore the landscape of transfer learning techniques for NLP by introducing a unified framework that converts all text-based language problems into a text-to-text format. Our systematic study compares pre-training objectives, architectures, unlabeled data sets, transfer approaches, and other factors on dozens of language understanding tasks. By combining the insights from our exploration with scale and our new “Colossal Clean Crawled Corpus”, we achieve state-of-the-art results on many benchmarks covering summarization, question answering, text classification, and more. To facilitate future work on transfer learning for NLP, we release our data set, pre-trained models, and code

Zaid Alyafeai

opus100

https://huggingface.co/datasets/opus100

https://data.statmt.org/opus-100-corpus/v1.0/

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

crawling

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.

1,040,000

sentences

Low

University of Edinburgh

nan

Improving Massively Multilingual Neural Machine Translation and Zero-Shot Translation

https://arxiv.org/pdf/2004.11867.pdf

Arab

No

other

Free

nan

Yes

machine translation

arXiv

nan

preprint

nan

Biao Zhang, Philip Williams, Ivan Titov, Rico Sennrich

School of Informatics, University of Edinburgh;School of Informatics, University of EdinburghSchool of Informatics, University of Edinburgh; Department of Computational Linguistics, University of Zurich

Massively multilingual models for neural machine translation (NMT) are theoretically attractive, but often underperform bilingual models and deliver poor zero-shot translations. In this paper, we explore ways to improve them. We argue that multilingual NMT requires stronger modeling capacity to support language pairs with varying typological characteristics, and overcome this bottleneck via language-specific components and deepening NMT architectures. We identify the off-target translation issue (i.e. translating into a wrong target language) as the major source of the inferior zero-shot performance, and propose random online backtranslation to enforce the translation of unseen training language pairs. Experiments on OPUS-100 (a novel multilingual dataset with 100 languages) show that our approach substantially narrows the performance gap with bilingual models in both oneto-many and many-to-many settings, and improves zero-shot performance by ∼10 BLEU, approaching conventional pivot-based methods

Zaid Alyafeai

CoVoST 2

https://huggingface.co/datasets/covost2

https://github.com/facebookresearch/covost

CC0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

transcribed audio

spoken

crawling and annotation(other)

a large-scale multilingual ST corpus based on Common Voice, to foster ST research with the largest ever open dataset. Its latest version covers translations from English into 15 languages---Arabic, Catalan, Welsh, German, Estonian, Persian, Indonesian, Japanese, Latvian, Mongolian, Slovenian, Swedish, Tamil, Turkish, Chinese

6

hours

Low

Facebook AI

Common Voice

CoVoST 2 and Massively Multilingual Speech-to-Text Translation

https://arxiv.org/pdf/2007.10310.pdf

Arab

No

GitHub

Free

nan

Yes

speech recognition

arXiv

nan

preprint

nan

Changhan Wang, Anne Wu, Juan Pino

Facebook AI;Facebook AI;Facebook AI

Speech-to-text translation (ST) has recently become an increasingly popular topic of research, partly due to the development of benchmark datasets. Nevertheless, current datasets cover a limited number of languages. With the aim to foster research in massive multilingual ST and ST for low resource language pairs, we release CoVoST 2, a large-scale multilingual ST corpus covering translations from 21 languages into English and from English into 15 languages. This represents the largest open dataset available to date from total volume and language coverage perspective. Data sanity checks provide evidence about the quality of the data, which is released under CC0 license. We also provide extensive speech recognition, bilingual and multilingual machine translation and ST baselines with open-source implementation

Zaid Alyafeai

News Commentary

https://huggingface.co/datasets/news_commentary

https://opus.nlpl.eu/News-Commentary.php

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

news articles

text

other

A parallel corpus of News Commentaries provided by WMT for training SMT

200,000

sentences

Low

OPUS

WMT 19

Parallel Data, Tools and Interfaces in OPUS

http://www.lrec-conf.org/proceedings/lrec2012/pdf/463_Paper.pdf

Arab

No

other

Free

nan

Yes

machine translation

LREC

nan

conference

Language Resources and Evaluation Conference

Jorg Tiedemann

Department of Linguistics and Philology Uppsala University, Uppsala/Sweden

This paper presents the current status of OPUS, a growing language resource of parallel corpora and related tools. The focus in OPUS is to provide freely available data sets in various formats together with basic annotation to be useful for applications in computational linguistics, translation studies and cross-linguistic corpus studies. In this paper, we report about new data sets and their features, additional annotation tools and models provided from the website and essential interfaces and on-line services included in the project.

Zaid Alyafeai

XGLUE

https://huggingface.co/datasets/xglue

https://github.com/microsoft/XGLUE

CC BY 4.0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

other

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 training data of each task is in English while the validation and test data is present in multiple different languages. The following table shows which languages are present as validation and test data for each config.

10,000

sentences

Low

Microsoft

Universal Dependencies, MLQA, XNLI

XGLUE: A New Benchmark Dataset for Cross-lingual Pre-training, Understanding and Generation

https://arxiv.org/pdf/2004.01401.pdf

Arab

No

GitHub

Free

nan

Yes

part of speech tagging, question answering, natural language inference

arXiv

nan

preprint

nan

Yaobo Liang, Nan Duan, Yeyun Gong, Ning Wu, Fenfei Guo, Weizhen Qi, Ming Gong, Linjun Shou,Daxin Jiang, Guihong Cao, Xiaodong Fan, Ruofei Zhang, Rahul Agrawal, Edward Cui, Sining Wei, Taroon Bharti,Ying Qiao, Jiun-Hung Chen, Winnie Wu, Shuguang Liu, Fan Yang, Daniel Campos, Rangan Majumder, Ming Zho

microsoft

In this paper, we introduce XGLUE, a new benchmark dataset that can be used to train large-scale cross-lingual pre-trained models using multilingual and bilingual corpora and evaluate their performance across a diverse set of cross-lingual tasks. Comparing to GLUE (Wang et al., 2019), which is labeled in English for natural language understanding tasks only, XGLUE has two main advantages: (1) it provides 11 diversified tasks that cover both natural language understanding and generation scenarios; (2) for each task, it provides labeled data in multiple languages. We extend a recent cross-lingual pre-trained model Unicoder (Huang et al., 2019) to cover both understanding and generation tasks, which is evaluated on XGLUE as a strong baseline. We also evaluate the base versions (12-layer) of Multilingual BERT, XLM and XLM-R for comparison

Zaid Alyafeai

TED TALKS IWSLT

https://huggingface.co/datasets/ted_talks_iwslt

https://drive.google.com/u/0/uc?id=1Cz1Un9p8Xn9IpEMMrg2kXSDt0dnjxc4z&export=download

CC BY-NC 4.0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

transcribed audio

spoken

crawling

The Web Inventory Talk is a collection of the original Ted talks and their translated version. The translations are available in more than 109+ languages, though the distribution is not uniform.

19,670,000

tokens

Low

Fondazione Bruno Kessler

TED

WIT3 : Web Inventory of Transcribed and Translated Talks

https://aclanthology.org/2012.eamt-1.60.pdf

Arab

No

Gdrive

Free

nan

Yes

speech recognition

EAMT

nan

conference

European Association for Machine Translation

Mauro Cettolo, Christian Girardi, Marcello Federico

FBK – Fondazione Bruno Kessler Trento, Italy

We describe here a Web inventory named WIT3 that offers access to a collection of transcribed and translated talks. 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.

Zaid Alyafeai

Universal Dependencies

https://huggingface.co/datasets/universal_dependencies

https://github.com/UniversalDependencies

unknown

ar

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

other

Universal Dependencies is a project that seeks to develop cross-linguistically consistent treebank annotation for many languages.

1,042,000

sentences

Low

Universal Dependencies(UD)

UDP (UDP-NYUAD), PADT, PUD

nan

nan

Arab

Yes

other

Free

nan

Yes

parts of speech tagging, morphological features, and syntactic dependencies

nan

nan

nan

nan

nan

nan

nan

Zaid Alyafeai

Wojood

nan

https://ontology.birzeit.edu/Wojood/

custom

ar

ar-MSA: (Arabic (Modern Standard Arabic))

news articles

text

crawling and annotation(other)

Wojood consists of about 550K tokens (MSA and dialect) that are manually annotated with 21 entity types (e.g., person, organization, location, event, date, etc). It covers multiple domains and was annotated with nested entities. The corpus contains about 75K entities and 22.5% of which are nested.

550,000

tokens

Low

Birzeit University

nan

Wojood: Nested Arabic Named Entity Corpus and Recognition using BERT

https://arxiv.org/pdf/2205.09651.pdf

Arab

No

other

Upon-Request

nan

Yes

named entity recognition

arXiv

nan

preprint

nan

Mustafa Jarrar, Mohammed Khalilia, Sana Ghanem

Birzeit University

This paper presents Wojood, a corpus for Arabic nested Named Entity Recognition (NER). Nested entities occur when one entity mention is embedded inside another entity mention. Wojood consists of about 550K Modern Standard Arabic (MSA) and dialect tokens that are manually annotated with 21 entity types including person, organization, location, event and date. More importantly, the corpus is annotated with nested entities instead of the more common flat annotations. The data contains about 75K entities and 22.5% of which are nested. The inter-annotator evaluation of the corpus demonstrated a strong agreement with Cohen’s Kappa of 0.979 and an F1-score of 0.976. To validate our data, we used the corpus to train a nested NER model based on multi-task learning using the pre-trained AraBERT (Arabic BERT). The model achieved an overall micro F1-score of 0.884. Our corpus, the annotation guidelines, the source code and the pre-trained model are publicly available.

Zaid Alyafeai

KDE4

https://huggingface.co/datasets/kde4

https://opus.nlpl.eu/KDE4.php

custom

ar

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

other

A parallel corpus of KDE4 localization files (v.2). 92 languages, 4,099 bitexts

700,000

sentences

Low

OPUS

nan

Parallel Data, Tools and Interfaces in OPUS

http://www.lrec-conf.org/proceedings/lrec2012/pdf/463_Paper.pdf

Arab

No

other

Free

nan

No

machine translation

LREC

nan

conference

Language Resources and Evaluation Conference

Jorg Tiedemann

Department of Linguistics and Philology Uppsala University, Uppsala/Sweden

nan

Zaid Alyafeai

Wikipedia

https://huggingface.co/datasets/wikipedia

https://dumps.wikimedia.org/

CC BY-SA 3.0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

wikipedia

text

crawling

Wikipedia dataset containing cleaned articles of all languages. The datasets are built from the Wikipedia dump (https://dumps.wikimedia.org/) with one split per language.

1,151,628

documents

Low

Wikimedia

nan

nan

nan

Arab

No

other

Free

nan

No

text generation, language modeling

nan

nan

nan

nan

nan

nan

nan

Zaid Alyafeai

WikiQAar

https://huggingface.co/datasets/wiki_qa_ar

https://github.com/qcri/WikiQAar

unknown

ar

ar-MSA: (Arabic (Modern Standard Arabic))

wikipedia

text

crawling

WIKIQAar is a bilingual English--Arabic Question Answering corpus built on top of WIKIQA

3,047

sentences

Low

QCRI

WikiQA

WIKIQA: A Challenge Dataset for Open-Domain Question Answering

https://aclanthology.org/D15-1237.pdf

Arab

No

GitHub

Free

nan

No

question answering

EMNLP

nan

conference

Empirical Methods in Natural Language Processing

Yi Yang, Wen-tau Yih Christopher Meek

Georgia Institute of Technology, Microsoft, Microsoft

We describe the WIKIQA dataset, a new publicly available set of question and sentence pairs, collected and annotated for research on open-domain question answering. Most previous work on answer sentence selection focuses on a dataset created using the TREC-QA data, which includes editor-generated questions and candidate answer sentences selected by matching content words in the question. WIKIQA is constructed using a more natural process and is more than an order of magnitude larger than the previous dataset. In addition, the WIKIQA dataset also includes questions for which there are no correct sentences, enabling researchers to work on answer triggering, a critical component in any QA system. We compare several systems on the task of answer sentence selection on both datasets and also describe the performance of a system on the problem of answer triggering using the WIKIQA dataset.

Zaid Alyafeai

Khaleej-2004

https://huggingface.co/datasets/arbml/khaleej_2004

https://sourceforge.net/projects/arabiccorpus/files/

unknown

ar

ar-MSA: (Arabic (Modern Standard Arabic))

news articles

text

crawling

Extracted from the daily Arabic news paper Akhbar al Khaleej, it includes 5120 news articles corresponding to 2,855,069 words covering four topics sport, local news, international news and economy

5,690

documents

Low

INRIA

nan

Comparison of Topic Identification methods for Arabic Language

https://hal.inria.fr/inria-00000448/document

Arab

No

sourceforge

Free

nan

No

topic classification

nan

nan

preprint

nan

M. Abbas and K. Smaili

INRIA-LORIA

In this paper we present two well-known methods for topic identification. The first one is a TFIDF classifier approach, and the second one is a based machine learning approach which is called Support Vector Machines (SVM). In our knowledge, we do not know several works on Arabic topic identification. So that we decide to investigate in this article. The corpus we used is extracted from the daily Arabic newspaper it Akhbar Al Khaleej, it includes 5120 news articles corresponding to 2.855.069 words covering four topics : sport, local news, international news and economy. According to our experiments, the results are encouraging both for SVM and TFIDF classifier, however we have noticed the superiority of the SVM classifier and its high capability to distinguish topics.

Zaid Alyafeai

CCAligned

https://huggingface.co/datasets/ccaligned_multilingual

https://opus.nlpl.eu/CCAligned.php

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

other

CCAligned consists of parallel or comparable web-document pairs in 137 languages aligned with English. These web-document pairs were constructed by performing language identification on raw web-documents, and ensuring corresponding language codes were corresponding in the URLs of web documents.

1,219,374

sentences

Low

Multiple Institutions

nan

CCAligned: A Massive Collection of Cross-lingual Web-Document Pairs

http://aclanthology.lst.uni-saarland.de/2020.emnlp-main.480.pdf

Arab

No

other

Free

nan

No

machine translation

EMNLP

nan

conference

Empirical Methods in Natural Language Processing

Ahmed El-Kishky, Vishrav Chaudhary, Francisco Guzmá, Philipp Koehn

Facebook AI; Facebook AI; Facebook AI; Johns Hopkins University

Cross-lingual document alignment aims to identify pairs of documents in two distinct languages that are of comparable content or translations of each other. In this paper, we exploit the signals embedded in URLs to label web documents at scale with an average precision of 94.5% across different language pairs. We mine sixty-eight snapshots of the Common Crawl corpus and identify web document pairs that are translations of each other. We release a new web dataset consisting of over 392 million URL pairs from Common Crawl covering documents in 8144 language pairs of which 137 pairs include English. In addition to curating this massive dataset, we introduce baseline methods that leverage crosslingual representations to identify aligned documents based on their textual content. Finally, we demonstrate the value of this parallel documents dataset through a downstream task of mining parallel sentences and measuring the quality of machine translations from models trained on this mined data. Our objective in releasing this dataset is to foster new research in cross-lingual NLP across a variety of low, medium, and high-resource languages

Zaid Alyafeai

Watan-2004

https://huggingface.co/datasets/arbml/watan_2004

https://sourceforge.net/projects/arabiccorpus/files/

unknown

ar

ar-MSA: (Arabic (Modern Standard Arabic))

news articles

text

crawling

Watan-2004 corpus contains about 20000 articles talking about the six following topics "categories": Culture, Religion, Economy, Local News, International News and sports. In this corpus, punctuation has been omitted intentionally in order to make it useful for Language Modeling.

20,000

sentences

Low

Multiple institutions

nan

Comparing TR-Classifier and KNN by using Reduced Sizes of Vocabularies

https://hal.archives-ouvertes.fr/hal-01586533/document

Arab

No

other

Free

nan

No

topic classification

CITALA

nan

conference

International Conference on Arabic Language Processing

M. Abbas, K. Smaili, and D. Berkani

CRSTDLA /Speech Processing Laboratory;NRIA-LORIA/Parole team, Villers les Nancy;NPS/ Signal and Communication laboratory

The aim of this study is topic identification by using two methods, in this case, a new one that we have proposed: TR-classifier which is based on computing triggers, and the well-known k Nearest Neighbors. Performances are acceptable, particularly for TR-classifier, though we have used reduced sizes of vocabularies. For the TR-Classifier, each topic is represented by a vocabulary which has been built using the corresponding training corpus. Whereas, the kNN method uses a general vocabulary, obtained by the concatenation of those used by the TR-Classifier. For the evaluation task, six topics have been selected to be identified: Culture, religion, economy, local news, international news and sports. An Arabic corpus has been used to achieve experiments.

Zaid Alyafeai

CCMatrix

https://huggingface.co/datasets/yhavinga/ccmatrix

https://github.com/facebookresearch/LASER/tree/main/tasks/CCMatrix

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

crawling

80 languages, we were able to mine 10.8 billion parallel sentences, out of which only 2.9 billion are aligned with English

196,000,000

sentences

Low

Facebook

nan

CCMatrix: Mining Billions of High-Quality Parallel Sentences on the WEB

https://arxiv.org/pdf/1911.04944.pdf

Arab

No

other

Free

nan

No

machine translation

arXiv

nan

preprint

nan

Holger Schwenk, Guillaume Wenzek, Sergey Edunov, Edouard Grave, Armand Joulin

Facebook AI

We show that margin-based bitext mining in a multilingual sentence space can be applied to monolingual corpora of billions of sentences. We are using ten snapshots of a curated common crawl corpus (Wenzek et al., 2019), totalling 32.7 billion unique sentences. Using one unified approach for 38 languages, we were able to mine 4.5 billions parallel sentences, out of which 661 million are aligned with English. 20 language pairs have more then 30 million parallel sentences, 112 more then 10 million, and most more than one million, including direct alignments between many European or Asian languages. To evaluate the quality of the mined bitexts, we train NMT systems for most of the language pairs and evaluate them on TED, WMT and WAT test sets. Using our mined bitexts only and no human translated parallel data, we achieve a new state-of-the-art for a single system on the WMT’19 test set for translation between English and German, Russian and Chinese, as well as German/French. In particular, our English/German system outperforms the best single one by close to 4 BLEU points and is almost on pair with best WMT’19 evaluation system which uses system combination and back-translation. We also achieve excellent results for distant languages pairs like Russian/Japanese, outperforming the best submission at the 2019 workshop on Asian Translation (WAT).

Zaid Alyafeai

CrossSum

https://huggingface.co/datasets/csebuetnlp/CrossSum

https://github.com/csebuetnlp/CrossSum

CC BY-NC-SA 4.0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

news articles

text

crawling

a large-scale dataset comprising 1.65 million cross-lingual article-summary samples in 1500+ language-pairs

72,795

documents

Low

Multiple Institutions

nan

CrossSum: Beyond English-Centric Cross-Lingual Abstractive Text Summarization for 1500+ Language Pairs

https://arxiv.org/pdf/2112.08804.pdf

Arab

No

GitHub

Free

nan

No

summarization

arXiv

nan

preprint

preprint

Tahmid Hasan, Abhik Bhattacharjee, Wasi Uddin Ahmad, Yuan-Fang Li, Yong-Bin Kang, Rifat Shahriyar

Bangladesh University of Engineering and Technology (BUET), University of California, Los Angeles, Monash University, Swinburne University of Technology

We present CrossSum, a large-scale dataset comprising 1.65 million cross-lingual articlesummary samples in 1500+ language-pairs constituting 45 languages. We use the multilingual XL-Sum dataset and align identical articles written in different languages via crosslingual retrieval using a language-agnostic representation model. We propose a multi-stage data sampling algorithm and fine-tune mT5, a multilingual pretrained model, with explicit cross-lingual supervision with CrossSum and introduce a new metric for evaluating crosslingual summarization. Results on established and our proposed metrics indicate that models fine-tuned on CrossSum outperforms summarization+translation baselines, even when the source and target language pairs are linguistically distant. To the best of our knowledge, CrossSum is the largest cross-lingual summarization dataset and also the first-ever that does not rely on English as the pivot language. We are releasing the dataset, alignment and training scripts, and the models to spur future research on cross-lingual abstractive summarization. The resources can be found at https: //github.com/csebuetnlp/CrossSum.

Zaid Alyafeai

Opus Wikipedia

https://huggingface.co/datasets/opus_wikipedia

https://data.statmt.org/cc-100/

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

wikipedia

text

crawling

This is a corpus of parallel sentences extracted from Wikipedia by Krzysztof Wołk and Krzysztof Marasek

1,000,000

sentences

Low

OPUS

nan

Building Subject-aligned Comparable Corpora and Mining it for Truly Parallel Sentence Pairs

https://www.sciencedirect.com/science/article/pii/S2212017314005453

Arab

No

other

Free

nan

No

machine translation

PT

nan

journal

Procedia Technology

Krzysztof Wołk, Krzysztof Marasek

Polish Japanese Institute of Information Technology, Warsaw, Poland

Parallel sentences are a relatively scarce but extremely useful resource for many applications including cross-lingual retrieval and statistical machine translation. This research explores our methodology for mining such data from previously obtained comparable corpora. The task is highly practical since non-parallel multilingual data exist in far greater quantities than parallel corpora, but parallel sentences are a much more useful resource. Here we propose a web crawling method for building subject-aligned comparable corpora from Wikipedia articles. We also introduce a method for extracting truly parallel sentences that are filtered out from noisy or just comparable sentence pairs. We describe our implementation of a specialized tool for this task as well as training and adaption of a machine translation system that supplies our filter with additional information about the similarity of comparable sentence pairs.

Zaid Alyafeai

SaudiNewsNet

https://huggingface.co/datasets/saudinewsnet

https://github.com/inparallel/SaudiNewsNet

CC BY-NC-SA 4.0

ar

ar-MSA: (Arabic (Modern Standard Arabic))

news articles

text

crawling

a set of 31,030 Arabic newspaper articles alongwith metadata, extracted from various online Saudi newspapers.

31,030

documents

Low

-

nan

nan

nan

Arab

No

GitHub

Free

nan

No

language modeling, text generation

nan

nan

nan

nan

nan

nan

nan

Zaid Alyafeai

OffensEval 2020

https://huggingface.co/datasets/strombergnlp/offenseval_2020

https://sites.google.com/site/offensevalsharedtask/results-and-paper-submission

CC BY 4.0

multilingual

mixed

social media

text

crawling and annotation(other)

The Arabic dataset consists of 10,000 tweets collected in April–May 2019 using the Twitter API with the language filter set to Arabic: lang:ar.

10,000

sentences

High

Multiple Institutions

Arabic OSACT4

SemEval-2020 Task 12: Multilingual Offensive Language Identification in Social Media (OffensEval 2020)

https://aclanthology.org/2020.semeval-1.188.pdf

Arab

No

other

Free

nan

Yes

offensive language detection

SemEval

nan

conference

Semantic Evaluation

Marcos Zampieri, Preslav Nakov, Sara Rosenthal, Pepa Atanasova, Georgi Karadzhov, Hamdy Mubarak, Leon Derczynski, Zeses Pitenis, Çağrı Çöltekin

nan

We present the results and the main findings of SemEval-2020 Task 12 on Multilingual Offensive Language Identification in Social Media (OffensEval-2020). The task included three subtasks corresponding to the hierarchical taxonomy of the OLID schema from OffensEval-2019, and it was offered in five languages: Arabic, Danish, English, Greek, and Turkish. OffensEval-2020 was one of the most popular tasks at SemEval-2020, attracting a large number of participants across all subtasks and languages: a total of 528 teams signed up to participate in the task, 145 teams submitted official runs on the test data, and 70 teams submitted system description papers.

Zaid Alyafeai

arwiki

https://huggingface.co/datasets/CALM/arwiki

https://huggingface.co/datasets/CALM/arwiki

unknown

ar

ar-MSA: (Arabic (Modern Standard Arabic))

wikipedia

text

crawling

This dataset is extracted using wikiextractor tool, from Wikipedia Arabic pages.

1,136,455

documents

Low

CALM

nan

nan

nan

Arab

No

other

Free

nan

No

text generation, language modeling

nan

nan

nan

nan

nan

nan

nan

Zaid Alyafeai

UniMorph

https://huggingface.co/datasets/universal_morphologies

https://github.com/unimorph/ara

CC BY-SA 3.0

ar

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

other

167 languages have been annotated according to the UniMorph schema.

140003

tokens

Low

Johns Hopkins University

nan

The Composition and Use of the Universal Morphological Feature Schema (UniMorph Schema)

https://unimorph.github.io/doc/unimorph-schema.pdf

Arab

No

GitHub

Free

nan

No

morphological analysis

nan

nan

preprint

nan

John Sylak-Glassman

Center for Language and Speech Processing Johns Hopkins University

nan

Zaid Alyafeai

CC-100

https://huggingface.co/datasets/cc100

https://data.statmt.org/cc-100/

unknown

ar

mixed

other

text

crawling

monolingual datasets from Common Crawl for a variety of languages

7,132,000

documents

Low

Facebook

Common Crawl

CCNet: Extracting High Quality Monolingual Datasets from Web Crawl Data

https://aclanthology.org/2020.lrec-1.494.pdf

Arab

No

other

Free

nan

No

text generation, language modeling

LREC

nan

conference

Language Resources and Evaluation Conference

Guillaume Wenzek, Marie-Anne Lachaux, Alexis Conneau, Vishrav Chaudhary, Francisco Guzmán, Armand Joulin, Edouard Grave

Facebook AI

Pre-training text representations have led to significant improvements in many areas of natural language processing. The quality of these models benefits greatly from the size of the pretraining corpora as long as its quality is preserved. In this paper, we describe an automatic pipeline to extract massive high-quality monolingual datasets from Common Crawl for a variety of languages. Our pipeline follows the data processing introduced in fastText (Mikolov et al., 2017; Grave et al., 2018), that deduplicates documents and identifies their language. We augment this pipeline with a filtering step to select documents that are close to high quality corpora like Wikipedia

Zaid Alyafeai

WiLI-2018

https://huggingface.co/datasets/wili_2018

https://zenodo.org/record/841984#.YpBRIahBxD8

ODbL-1.0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

wikipedia

text

crawling

WiLI-2018, the Wikipedia language identification benchmark dataset, contains 235000 paragraphs of 235 languages. The dataset is balanced and a train-test split is provided.

1,000

sentences

Low

-

nan

The WiLI benchmark dataset for written language identification

https://arxiv.org/pdf/1801.07779.pdf

Arab

No

zenodo

Free

nan

Yes

language identification

arXiv

nan

preprint

nan

Martin Thoma

nan

This paper describes the WiLI-2018 benchmark dataset for monolingual written natural language identification. WiLI-2018 is a publicly available,1 free of charge dataset of short text extracts from Wikipedia. It contains 1000 paragraphs of 235 languages, totaling in 235 000 paragraphs. WiLI is a classification dataset: Given an unknown paragraph written in one dominant language, it has to be decided which language it is.

Zaid Alyafeai

MMAC

nan

http://www.ashrafraouf.com/mmac

unknown

ar

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

crawling

The multi-modal Arabic corpus contains 6 million Arabic words selected from various sources covering old Arabic, religious texts, traditional language, modern language, different specialisations and very modern material from online “chat rooms.”

6,000,000

tokens

Low

Multiple Institutions

nan

Building a multi-modal Arabic corpus (MMAC)

https://link.springer.com/content/pdf/10.1007/s10032-010-0128-2.pdf

Arab

No

other

Free

nan

No

dictionary

IJDAR

nan

journal

International Journal on Document Analysis and Recognition

Ashraf AbdelRaouf, Colin A. Higgins, Tony Pridmore, Mahmoud Khalil

Misr International University, The University of Nottingham, The University of Nottingham, , Ain Shams University

Traditionally, a corpus is a large structured set of text, electronically stored and processed. Corpora have become very important in the study of languages. They have opened new areas of linguistic research, which were unknown until recently. Corpora are also key to the development of optical character recognition (OCR) applications. Access to a corpus of both language and images is essential during OCR development, particularly while training and testing a recognition application. Excellent corpora have been developed for Latin-based languages, but few relate to the Arabic language. This limits the penetration of both corpus linguistics and OCR in Arabic-speaking countries. This paper describes the construction and provides a comprehensive study and analysis of a multi-modal Arabic corpus (MMAC) that is suitable for use in both OCR development and linguistics. MMAC currently contains six million Arabic words and, unlike previous corpora, also includes connected segments or pieces of Arabic words (PAWs) as well as naked pieces of Arabic words (NPAWs) and naked words (NWords); PAWs and Words without diacritical marks. Multi-modal data is generated from both text, gathered from a wide variety of sources, and images of existing documents. Text-based data is complemented by a set of artificially generated images showing each of the Words, NWords, PAWs and NPAWs involved. Applications are provided to generate a natural-looking degradation to the generated images. A ground truth annotation is offered for each such image, while natural images showing small paragraphs and full pages are augmented with representations of the text they depict. A statistical analysis and verification of the dataset has been carried out and is presented. MMAC was also tested using commercial OCR software and is publicly and freely available.

Zaid Alyafeai

XOR-TyDi QA

https://huggingface.co/datasets/xor_tydi_qa

https://nlp.cs.washington.edu/xorqa/index.html

CC BY-SA 4.0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

other

XOR-TyDi QA brings together for the first time information-seeking questions, open-retrieval QA, and multilingual QA to create a multilingual open-retrieval QA dataset that enables cross-lingual answer retrieval. It consists of questions written by information-seeking native speakers in 7 typologically diverse languages and answer annotations that are retrieved from multilingual document collections.

5,235

sentences

Low

Multiple Institutions

TYDIQA

XOR QA: Cross-lingual Open-Retrieval Question Answering

https://arxiv.org/pdf/2010.11856.pdf

Arab

No

other

Free

nan

Yes

open-retrieval question answering

arXiv

nan

preprint

nan

Akari Asai, Jungo Kasai, Jonathan H. Clark,Kenton Lee, Eunsol Choi, Hannaneh Hajishirzi

University of Washington, University of Washington, Google Research, The University of Texas at Austin; Allen Institute for AI

Multilingual question answering tasks typically assume that answers exist in the same language as the question. Yet in practice, many languages face both information scarcity—where languages have few reference articles—and information asymmetry—where questions reference concepts from other cultures. This work extends open-retrieval question answering to a cross-lingual setting enabling questions from one language to be answered via answer content from another language. We construct a large-scale dataset built on 40K information-seeking questions across 7 diverse non-English languages that TYDI QA could not find same-language answers for. Based on this dataset, we introduce a task framework, called Cross-lingual OpenRetrieval Question Answering (XOR QA), that consists of three new tasks involving crosslingual document retrieval from multilingual and English resources. We establish baselines with state-of-the-art machine translation systems and cross-lingual pretrained models. Experimental results suggest that XOR QA is a challenging task that will facilitate the development of novel techniques for multilingual question answering. Our data and code are available at https://nlp.cs.washington. edu/xorqa/.

Zaid Alyafeai

Multilingual LAMA

https://huggingface.co/datasets/m_lama

https://github.com/norakassner/mlama

CC BY-NC 4.0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

machine translation

multilingual version of lama. The underlying idea of LAMA is to query knowledge from pretrained LMs using templates without any finetuning

19,354

sentences

Low

LMU Munich

nan

Multilingual LAMA: Investigating Knowledge in Multilingual Pretrained Language Models

https://arxiv.org/pdf/2102.00894.pdf

Arab

No

GitHub

Free

nan

Yes

lm probing

arXiv

nan

preprint

nan

Nora Kassner, Philipp Dufter, Hinrich Schutze

Center for Information and Language Processing (CIS), LMU Munich

Recently, it has been found that monolingual English language models can be used as knowledge bases. Instead of structural knowledge base queries, masked sentences such as “Paris is the capital of [MASK]” are used as probes. We translate the established benchmarks TREx and GoogleRE into 53 languages. Working with mBERT, we investigate three questions. (i) Can mBERT be used as a multilingual knowledge base? Most prior work only considers English. Extending research to multiple languages is important for diversity and accessibility. (ii) Is mBERT’s performance as knowledge base language-independent or does it vary from language to language? (iii) A multilingual model is trained on more text, e.g., mBERT is trained on 104 Wikipedias. Can mBERT leverage this for better performance? We find that using mBERT as a knowledge base yields varying performance across languages and pooling predictions across languages improves performance. Conversely, mBERT exhibits a language bias; e.g., when queried in Italian, it tends to predict Italy as the country of origin.

Zaid Alyafeai

infopankki v1

https://huggingface.co/datasets/opus_infopankki

https://opus.nlpl.eu/infopankki-v1.php

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

crawling

A parallel corpus of 12 languages, 66 bitexts.

63,000

sentences

Low

OPUS

nan

Parallel Data, Tools and Interfaces in OPUS

http://www.lrec-conf.org/proceedings/lrec2012/pdf/463_Paper.pdf

Arab

No

other

Free

nan

No

machine translation

LREC

nan

conference

Language Resources and Evaluation Conference

Jorg Tiedemann

Department of Linguistics and Philology Uppsala University

This paper presents the current status of OPUS, a growing language resource of parallel corpora and related tools. The focus in OPUS is to provide freely available data sets in various formats together with basic annotation to be useful for applications in computational linguistics, translation studies and cross-linguistic corpus studies. In this paper, we report about new data sets and their features, additional annotation tools and models provided from the website and essential interfaces and on-line services included in the project.

Zaid Alyafeai

United Nations General Assembly Resolutions

https://huggingface.co/datasets/un_ga

https://opus.nlpl.eu/UN.php

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

crawling

This is a collection of translated documents from the United Nations originally compiled into a translation memory by Alexandre Rafalovitch, Robert Dale

73,000

sentences

Low

OPUS

nan

United Nations General Assembly Resolutions: A Six-Language Parallel Corpus

https://aclanthology.org/2009.mtsummit-posters.15.pdf

Arab

No

other

Free

nan

No

machine translation

mtsummit

nan

conference

Machine Translation Summit XII

Alexandre Rafalovitch, Robert Dale

United Nations; Centre for Language Technology Macquarie University

In this paper we describe a six-ways parallel public-domain corpus consisting of 2100 United Nations General Assembly Resolutions with translations in the six official languages of the United Nations, with an average of around 3 million tokens per language. The corpus is available in a preprocessed, formatting-normalized TMX format with paragraphs aligned across multiple languages. We describe the background to the corpus and its content, the process of its construction, and some of its interesting properties.

Zaid Alyafeai

X-CSR

https://huggingface.co/datasets/xcsr

https://inklab.usc.edu//XCSR/xcsr_datasets

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

machine translation

automatically translate the original CSQA and CODAH datasets, which only have English versions, to 15 other languages, forming development and test sets for studying X-CSR

1,300

sentences

Low

University of Southern California

nan

Common Sense Beyond English: Evaluating and Improving Multilingual Language Models for Commonsense Reasoning

https://arxiv.org/pdf/2106.06937.pdf

Arab

No

other

Upon-Request

nan

Yes

commonsense reasoning

arXiv

nan

preprint

nan

Bill Yuchen Lin, Seyeon Lee, Xiaoyang Qiao, Xiang Ren

Department of Computer Science and Information Sciences Institute, University of Southern California

Commonsense reasoning research has so far been mainly limited to English. We aim to evaluate and improve popular multilingual language models (ML-LMs) to help advance commonsense reasoning (CSR) beyond English. We collect the Mickey corpus, consisting of 561k sentences in 11 different languages, which can be used for analyzing and improving ML-LMs. We propose Mickey Probe, a language-agnostic probing task for fairly evaluating the common sense of popular ML-LMs across different languages. Also, we create two new datasets, X-CSQA and XCODAH, by translating their English versions to 15 other languages, so that we can evaluate popular ML-LMs for cross-lingual commonsense reasoning. To improve the performance beyond English, we propose a simple yet effective method — multilingual contrastive pretraining (MCP). It significantly enhances sentence representations, yielding a large performance gain on both benchmarks (e.g., +2.7% accuracy for X-CSQA over XLM-RL).

Zaid Alyafeai

CommonLanguage

https://huggingface.co/datasets/anton-l/common_language

https://github.com/speechbrain/speechbrain/tree/develop/recipes/CommonLanguage

CC BY 4.0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

transcribed audio

spoken

other

This dataset is composed of speech recordings from languages that were carefully selected from the CommonVoice database. The total duration of audio recordings is 45.1 hours (i.e., 1 hour of material for each language). The dataset has been extracted from CommonVoice to train language-id systems.

1

hours

Low

SpeechBrain

CommonVoice

nan

nan

Arab

No

GitHub

Free

nan

No

speech recognition

nan

nan

nan

nan

nan

nan

nan

Zaid Alyafeai

Tatoeba Translation Challenge

https://huggingface.co/datasets/Helsinki-NLP/tatoeba_mt

https://github.com/Helsinki-NLP/Tatoeba-Challenge/

CC BY-NC-SA 4.0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

other

The Tatoeba Translation Challenge is a multilingual data set of machine translation benchmarks derived from user-contributed translations collected by Tatoeba.org and provided as parallel corpus from OPUS. This dataset includes test and development data sorted by language pair. It includes test sets for hundreds of language pairs and is continuously updated. Please, check the version number tag to refer to the release that your are using.

1,064,096,596

sentences

Low

Language Technology at the University of Helsinki

Tatoeba

nan

nan

Arab

No

GitHub

Free

nan

Yes

machine translation

nan

nan

nan

nan

nan

nan

nan

Zaid Alyafeai

MFQA

https://huggingface.co/datasets/clips/mqa

https://huggingface.co/datasets/clips/mqa

CC0

multilingual

mixed

web pages

text

crawling

MQA is a Multilingual corpus of Questions and Answers (MQA) parsed from the Common Crawl. Questions are divided in two types: Frequently Asked Questions (FAQ) and Community Question Answering (CQA).

3,017,456

sentences

Low

University of Antwerp

Common Crawl

MFAQ: a Multilingual FAQ Dataset

https://arxiv.org/pdf/2109.12870.pdf

Arab

No

HuggingFace

Free

nan

No

frequently asked questions, question answering

arXiv

nan

preprint

nan

Maxime De Bruyn, Ehsan Lotfi, Jeska Buhmann, Walter Daelemans

CLiPS Research Center University of Antwerp

In this paper, we present the first multilingual FAQ dataset publicly available. We collected around 6M FAQ pairs from the web, in 21 different languages. Although this is significantly larger than existing FAQ retrieval datasets, it comes with its own challenges: duplication of content and uneven distribution of topics. We adopt a similar setup as Dense Passage Retrieval (DPR) (Karpukhin et al., 2020) and test various bi-encoders on this dataset. Our experiments reveal that a multilingual model based on XLM-RoBERTa (Conneau et al., 2019) achieves the best results, except for English. Lower resources languages seem to learn from one another as a multilingual model achieves a higher MRR than language-specific ones. Our qualitative analysis reveals the brittleness of the model on simple word changes. We publicly release our dataset1 , model2 and training script

Zaid Alyafeai

OpenSubtitles

https://huggingface.co/datasets/open_subtitles

https://opus.nlpl.eu/OpenSubtitles.php

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

crawling

This is a new collection of translated movie subtitles from http://www.opensubtitles.org/.

83,600,000

sentences

Low

OPUS

nan

OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles

http://www.lrec-conf.org/proceedings/lrec2016/pdf/947_Paper.pdf

Arab

No

other

Free

nan

No

machine translation

LREC

nan

conference

Language Resources Evaluation Conference

Pierre Lison, Jorg Tiedemann

University of Oslo

We present a new major release of the OpenSubtitles collection of parallel corpora. The release is compiled from a large database of movie and TV subtitles and includes a total of 1689 bitexts spanning 2.6 billion sentences across 60 languages. The release also incorporates a number of enhancements in the preprocessing and alignment of the subtitles, such as the automatic correction of OCR errors and the use of meta-data to estimate the quality of each subtitle and score subtitle pairs.

Zaid Alyafeai

OSCAR Small

https://huggingface.co/datasets/nthngdy/oscar-small

https://huggingface.co/datasets/nthngdy/oscar-small

CC0

multilingual

mixed

web pages

text

other

a processed version of and smaller subset of OSCAR

408,438

documents

Low

-

OSCAR

nan

nan

Arab

No

HuggingFace

Free

nan

Yes

language modeling, text generation

nan

nan

nan

nan

nan

nan

nan

Zaid Alyafeai

GNOME

https://huggingface.co/datasets/opus_gnome

https://opus.nlpl.eu/GNOME.php

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

other

text

crawling

A parallel corpus of GNOME localization files

800,000

sentences

Low

OPUS

nan

Parallel Data, Tools and Interfaces in OPUS

http://www.lrec-conf.org/proceedings/lrec2012/pdf/463_Paper.pdf

Arab

No

other

Free

nan

No

machine translation

LREC

nan

conference

Language Resources Evaluation Conference

Jorg Tiedemann

Department of Linguistics and Philology Uppsala University

This paper presents the current status of OPUS, a growing language resource of parallel corpora and related tools. The focus in OPUS is to provide freely available data sets in various formats together with basic annotation to be useful for applications in computational linguistics, translation studies and cross-linguistic corpus studies. In this paper, we report about new data sets and their features, additional annotation tools and models provided from the website and essential interfaces and on-line services included in the project.

Zaid Alyafeai

OPUS Wikipedia

https://huggingface.co/datasets/opus_wikipedia

https://opus.nlpl.eu/Wikipedia.php

unknown

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

wikipedia

text

crawling

This is a corpus of parallel sentences extracted from Wikipedia

1,000,000

sentences

Low

OPUS

nan

Parallel Data, Tools and Interfaces in OPUS

http://www.lrec-conf.org/proceedings/lrec2012/pdf/463_Paper.pdf

Arab

No

other

Free

nan

No

machine translation

LREC

nan

conference

Language Resources Evaluation Conference

nan

Jorg Tiedemann ¨

This paper presents the current status of OPUS, a growing language resource of parallel corpora and related tools. The focus in OPUS is to provide freely available data sets in various formats together with basic annotation to be useful for applications in computational linguistics, translation studies and cross-linguistic corpus studies. In this paper, we report about new data sets and their features, additional annotation tools and models provided from the website and essential interfaces and on-line services included in the project.

Zaid Alyafeai

BnL Historical Newspapers

https://huggingface.co/datasets/bnl_newspapers

https://data.bnl.lu/data/historical-newspapers/

CC0

multilingual

ar-MSA: (Arabic (Modern Standard Arabic))

news articles

text

other

The BnL has digitised over 800.000 pages of Luxembourg newspapers

1

documents

Low

BnL

nan