The Universal Dependencies (UD) project seeks to develop cross-linguistically consistent treebank annotation of morphology and syntax for multiple languages. The first version of the dataset was released in 2015 and consisted of 10 treebanks over 10 languages. Version 2.7 released in 2020 consists of 183 treebanks over 104 languages. The annotation consists of UPOS (universal part-of-speech tags), XPOS (language-specific part-of-speech tags), Feats (universal morphological features), Lemmas, dependency heads and universal dependency labels.
505 PAPERS • 15 BENCHMARKS
The Cross-lingual Natural Language Inference (XNLI) corpus is the extension of the Multi-Genre NLI (MultiNLI) corpus to 15 languages. The dataset was created by manually translating the validation and test sets of MultiNLI into each of those 15 languages. The English training set was machine translated for all languages. The dataset is composed of 122k train, 2490 validation and 5010 test examples.
325 PAPERS • 11 BENCHMARKS
XQuAD (Cross-lingual Question Answering Dataset) is 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.
169 PAPERS • 2 BENCHMARKS
MLQA (MultiLingual Question Answering) is a benchmark dataset for evaluating cross-lingual question answering performance. MLQA consists of over 5K extractive QA instances (12K in English) in SQuAD format in seven languages - English, Arabic, German, Spanish, Hindi, Vietnamese and Simplified Chinese. MLQA is highly parallel, with QA instances parallel between 4 different languages on average.
151 PAPERS • 1 BENCHMARK
This corpus comprises of monolingual data for 100+ languages and also includes data for romanized languages. This was constructed using the urls and paragraph indices provided by the CC-Net repository by processing January-December 2018 Commoncrawl snapshots. Each file comprises of documents separated by double-newlines and paragraphs within the same document separated by a newline. The data is generated using the open source CC-Net repository.
96 PAPERS • NO BENCHMARKS YET
OSCAR or Open Super-large Crawled ALMAnaCH coRpus is a huge multilingual corpus obtained by language classification and filtering of the Common Crawl corpus using the goclassy architecture. The dataset used for training multilingual models such as BART incorporates 138 GB of text.
55 PAPERS • NO BENCHMARKS YET
WikiLingua includes ~770k article and summary pairs in 18 languages from WikiHow. Gold-standard article-summary alignments across languages are extracted by aligning the images that are used to describe each how-to step in an article.
50 PAPERS • 5 BENCHMARKS
Room-Across-Room (RxR) is a multilingual dataset for Vision-and-Language Navigation (VLN) for Matterport3D environments. In contrast to related datasets such as Room-to-Room (R2R), RxR is 10x larger, multilingual (English, Hindi and Telugu), with longer and more variable paths, and it includes and fine-grained visual groundings that relate each word to pixels/surfaces in the environment.
42 PAPERS • 1 BENCHMARK
XL-Sum is a comprehensive and diverse dataset for abstractive summarization comprising 1 million professionally annotated article-summary pairs from BBC, extracted using a set of carefully designed heuristics. The dataset covers 44 languages ranging from low to high-resource, for many of which no public dataset is currently available. XL-Sum is highly abstractive, concise, and of high quality, as indicated by human and intrinsic evaluation.
39 PAPERS • NO BENCHMARKS YET
Samanantar is the largest publicly available parallel corpora collection for Indic languages: Assamese, Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Oriya, Punjabi, Tamil, Telugu. The corpus has 49.6M sentence pairs between English to Indian Languages.
36 PAPERS • NO BENCHMARKS YET
Sentiment analysis of codemixed tweets.
27 PAPERS • NO BENCHMARKS YET
IndicCorp is a large monolingual corpora with around 9 billion tokens covering 12 of the major Indian languages. It has been developed by discovering and scraping thousands of web sources - primarily news, magazines and books, over a duration of several months.
24 PAPERS • NO BENCHMARKS YET
XGLUE is an evaluation benchmark XGLUE,which is composed of 11 tasks that span 19 languages. For each task, the training data is only available in English. This means that to succeed at XGLUE, a model must have a strong zero-shot cross-lingual transfer capability to learn from the English data of a specific task and transfer what it learned to other languages. Comparing to its concurrent work XTREME, XGLUE has two characteristics: First, it includes cross-lingual NLU and cross-lingual NLG tasks at the same time; Second, besides including 5 existing cross-lingual tasks (i.e. NER, POS, MLQA, PAWS-X and XNLI), XGLUE selects 6 new tasks from Bing scenarios as well, including News Classification (NC), Query-Ad Matching (QADSM), Web Page Ranking (WPR), QA Matching (QAM), Question Generation (QG) and News Title Generation (NTG). Such diversities of languages, tasks and task origin provide a comprehensive benchmark for quantifying the quality of a pre-trained model on cross-lingual natural lan
19 PAPERS • 3 BENCHMARKS
Belebele is a multiple-choice machine reading comprehension (MRC) dataset spanning 122 language variants. This dataset enables the evaluation of mono- and multi-lingual models in high-, medium-, and low-resource languages. Each question has four multiple-choice answers and is linked to a short passage from the FLORES-200 dataset. The human annotation procedure was carefully curated to create questions that discriminate between different levels of generalizable language comprehension and is reinforced by extensive quality checks. While all questions directly relate to the passage, the English dataset on its own proves difficult enough to challenge state-of-the-art language models. Being fully parallel, this dataset enables direct comparison of model performance across all languages. Belebele opens up new avenues for evaluating and analyzing the multilingual abilities of language models and NLP systems.
17 PAPERS • NO BENCHMARKS YET
X-FACT is a large publicly available multilingual dataset for factual verification of naturally existing real-world claims. The dataset contains short statements in 25 languages and is labeled for veracity by expert fact-checkers. The dataset includes a multilingual evaluation benchmark that measures both out-of-domain generalization, and zero-shot capabilities of the multilingual models.
13 PAPERS • 1 BENCHMARK
The Dakshina dataset is a collection of text in both Latin and native scripts for 12 South Asian languages. For each language, the dataset includes a large collection of native script Wikipedia text, a romanization lexicon which consists of words in the native script with attested romanizations, and some full sentence parallel data in both a native script of the language and the basic Latin alphabet.
11 PAPERS • NO BENCHMARKS YET
MINTAKA is a complex, natural, and multilingual dataset designed for experimenting with end-to-end question-answering models. It is composed of 20,000 question-answer pairs collected in English, annotated with Wikidata entities, and translated into Arabic, French, German, Hindi, Italian, Japanese, Portuguese, and Spanish for a total of 180,000 samples. Mintaka includes 8 types of complex questions, including superlative, intersection, and multi-hop questions, which were naturally elicited from crowd workers.
10 PAPERS • NO BENCHMARKS YET
XStoryCloze consists of the professionally translated version of the English StoryCloze dataset (Spring 2016 version) to 10 non-English languages. This dataset is intended to be used for evaluating the zero- and few-shot learning capabilities of multlingual language models. This dataset is released by Meta AI.
MM-COVID is a dataset for fake news detection related to COVID-19. This dataset provides the multilingual fake news and the relevant social context. It contains 3,981 pieces of fake news content and 7,192 trustworthy information from English, Spanish, Portuguese, Hindi, French and Italian, 6 different languages.
8 PAPERS • NO BENCHMARKS YET
Hindi Visual Genome is a multimodal dataset consisting of text and images suitable for English-Hindi multimodal machine translation task and multimodal research.
7 PAPERS • NO BENCHMARKS YET
Demetr is a diagnostic dataset with 31K English examples (translated from 10 source languages) for evaluating the sensitivity of MT evaluation metrics to 35 different linguistic perturbations spanning semantic, syntactic, and morphological error categories.
6 PAPERS • NO BENCHMARKS YET
MuMiN is a misinformation graph dataset containing rich social media data (tweets, replies, users, images, articles, hashtags), spanning 21 million tweets belonging to 26 thousand Twitter threads, each of which have been semantically linked to 13 thousand fact-checked claims across dozens of topics, events and domains, in 41 different languages, spanning more than a decade.
4 PAPERS • 3 BENCHMARKS
This dataset is an extension of MASAC, a multimodal, multi-party, Hindi-English code-mixed dialogue dataset compiled from the popular Indian TV show, ‘Sarabhai v/s Sarabhai’. WITS was created by augmenting MASAC with natural language explanations for each sarcastic dialogue. The dataset consists of the transcribed sarcastic dialogues from 55 episodes of the TV show, along with audio and video multimodal signals. It was designed to facilitate Sarcasm Explanation in Dialogue (SED), a novel task aimed at generating a natural language explanation for a given sarcastic dialogue, that spells out the intended irony. Each data instance in WITS is associated with a corresponding video, audio, and textual transcript where the last utterance is sarcastic in nature. All the final selected explanations contain the following attributes:
GeoCoV19 is a large-scale Twitter dataset containing more than 524 million multilingual tweets. The dataset contains around 378K geotagged tweets and 5.4 million tweets with Place information. The annotations include toponyms from the user location field and tweet content and resolve them to geolocations such as country, state, or city level. In this case, 297 million tweets are annotated with geolocation using the user location field and 452 million tweets using tweet content.
3 PAPERS • NO BENCHMARKS YET
The MCVQA dataset consists of 248, 349 training questions and 121, 512 validation questions for real images in Hindi and Code-mixed. For each Hindi question, we also provide its 10 corresponding answers in Hindi.
Naamapadam is a Named Entity Recognition (NER) dataset for the 11 major Indian languages from two language families. In each language, it contains more than 400k sentences annotated with a total of at least 100k entities from three standard entity categories (Person, Location and Organization) for 9 out of the 11 languages. The training dataset has been automatically created from the Samanantar parallel corpus by projecting automatically tagged entities from an English sentence to the corresponding Indian language sentence.
Stanceosaurus is a corpus of 28,033 tweets in English, Hindi, and Arabic annotated with stance towards 251 misinformation claims. The claims in Stanceosaurus originate from 15 fact-checking sources that cover diverse geographical regions and cultures. Unlike existing stance datasets, it introduces a more fine-grained 5-class labeling strategy with additional subcategories to distinguish implicit stance.
FixMyPose is a dataset for automated pose correction. It consists of descriptions to correct a "current" pose to look like a "target" pose, in English and Hindi. The collected descriptions have interesting linguistic properties such as egocentric relations to environment objects, analogous references, etc., requiring an understanding of spatial relations and commonsense knowledge about postures.
2 PAPERS • NO BENCHMARKS YET
MuCo-VQA consist of large-scale (3.7M) multilingual and code-mixed VQA datasets in multiple languages: Hindi (hi), Bengali (bn), Spanish (es), German (de), French (fr) and code-mixed language pairs: en-hi, en-bn, en-fr, en-de and en-es.
A Dataset for Politeness Classification in Nine Typologically Diverse Languages (TyDiP) is a dataset containing three-way politeness annotations for 500 examples in each language, totaling 4.5K examples.
It consists of an extensive collection of a high quality cross-lingual fact-to-text dataset in 11 languages: Assamese (as), Bengali (bn), Gujarati (gu), Hindi (hi), Kannada (kn), Malayalam (ml), Marathi (mr), Oriya (or), Punjabi (pa), Tamil (ta), Telugu (te), and monolingual dataset in English (en). This is the Wikipedia text <--> Wikidata KG aligned corpus used to train the data-to-text generation model. The Train & validation splits are created using distant supervision methods and Test data is generated through human annotations.
2 PAPERS • 1 BENCHMARK
The Archive Query Log (AQL) is a previously unused, comprehensive query log collected at the Internet Archive over the last 25 years. Its first version includes 356 million queries, 166 million search result pages, and 1.7 billion search results across 550 search providers. Although many query logs have been studied in the literature, the search providers that own them generally do not publish their logs to protect user privacy and vital business data. The AQL is the first publicly available query log that combines size, scope, and diversity, enabling research on new retrieval models and search engine analyses. Provided in a privacy-preserving manner, it promotes open research as well as more transparency and accountability in the search industry.
1 PAPER • NO BENCHMARKS YET
The dataset covers Hindi and Tamil, collected without the use of translation. It provides a realistic information-seeking task with questions written by native-speaking expert data annotators.
1 PAPER • 1 BENCHMARK
The GATITOS (Google's Additional Translations Into Tail-languages: Often Short) dataset is a high-quality, multi-way parallel dataset of tokens and short phrases, intended for training and improving machine translation models. This dataset consists in 4,000 English segments (4,500 tokens) that have been translated into each of 26 low-resource languages, as well as three higher-resource pivot languages (es, fr, hi). All translations were made directly from English, with the exception of Aymara, which was translated from the Spanish.
This dataset releases a significantly sized standard-abiding Hindi NER dataset containing 109,146 sentences and 2,220,856 tokens, annotated with 3 collapsed tags (PER, LOC, ORG).
This dataset releases a significantly sized standard-abiding Hindi NER dataset containing 109,146 sentences and 2,220,856 tokens, annotated with 11 tags.
Hinglish-TOP is a human annotated code-switched semantic parsing dataset containing 10k human annotations for Hindi-English (HINGLISH) code switched utterances, and over 170K CST5 generated code-switched utterances from the TOPv2 dataset.
IRLCov19 is a multilingual Twitter dataset related to Covid-19 collected in the period between February 2020 to July 2020 specifically for regional languages in India. It contains more than 13 million tweets.
The E-MASAC Dataset is a collection of code-mixed conversations sourced from an Indian TV series, focusing on Hindi-English interactions. It was derived from the MASAC dataset and specifically annotated for Emotion Recognition in Conversations (ERC) tasks. The dataset comprises 8,607 dialogues with 11,440 utterances, containing instances of sarcasm and humor. Emotions such as anger, fear, joy, sadness, surprise, contempt, and neutral are annotated for each utterance by three fluent English and Hindi-speaking linguists, ensuring a high inter-annotator agreement of 0.85.
MalayalamMixSentiment is a Sentiment Analysis Dataset for Code-Mixed Malayalam-English.
Mega-COV is a billion-scale dataset from Twitter for studying COVID-19. The dataset is diverse (covers 234 countries), longitudinal (goes as back as 2007), multilingual (comes in 65 languages), and has a significant number of location-tagged tweets (~32M tweets).
Mint is a new Multilingual intimacy analysis dataset covering 13,384 tweets in 10 languages including English, French, Spanish, Italian, Portuguese, Korean, Dutch, Chinese, Hindi, and Arabic. The dataset is released along with the SemEval 2023 Task 9: Multilingual Tweet Intimacy Analysis.
MultiTACRED is a multilingual version of the large-scale TAC Relation Extraction Dataset. It covers 12 typologically diverse languages from 9 language families, and was created by the Speech & Language Technology group of DFKI by machine-translating the instances of the original TACRED dataset and automatically projecting their entity annotations. For details of the original TACRED's data collection and annotation process, see the Stanford paper. Translations are syntactically validated by checking the correctness of the XML tag markup. Any translations with an invalid tag structure, e.g. missing or invalid head or tail tag pairs, are discarded (on average, 2.3% of the instances).
Test set of sentences in Hindi with simple gender-specific context used to measure gender bias in NMT systems for Hindi-English.
We release various types of word embeddings for multiple Indian languages. Please note that for a majority of our work, we had transliterated the corpora to the Devanagiri script and the script is changed. Word Embedding models using FastText, ElMo, and cross-lingual models based on an orthogonal alignment of monolingual models for all pairs of these languages.
India is a linguistic area with one of the longest histories of contact, influence, use, teaching and learning of English-in-diaspora in the world (Kachru and Nelson, 2006). Thus, a huge number of Indians active on the internet are able in English communication to some degree. India also enjoys huge diversity in language. Apart from Hindi, it has several regional languages that are the primary tongue of people native to the region. This is to the extent that social media including Facebook, WhatsApp, Twitter, etc. contain more than one language, and such phenomena are called code-mixing and code-switching. On the other side, the evolution of sentiments from such social media texts have also created many new opportunities for information access and language technology, but also many new challenges, making it one of the prime present-day research areas. Sentiment analysis in code-mixed data has several real-life applications in opinion mining from social media campaign to feedback analys
The ComMA Dataset v0.2 is a multilingual dataset annotated with a hierarchical, fine-grained tagset marking different types of aggression and the "context" in which they occur. The context, here, is defined by the conversational thread in which a specific comment occurs and also the "type" of discursive role that the comment is performing with respect to the previous comment. The initial dataset, being discussed here (and made available as part of the ComMA@ICON shared task), consists of a total 15,000 annotated comments in four languages - Meitei, Bangla, Hindi, and Indian English - collected from various social media platforms such as YouTube, Facebook, Twitter and Telegram. As is usual on social media websites, a large number of these comments are multilingual, mostly code-mixed with English.
WEATHub is a dataset containing 24 languages. It contains words organized into groups of (target1, target2, attribute1, attribute2) to measure the association target1:target2 :: attribute1:attribute2. For example target1 can be insects, target2 can be flowers. And we might be trying to measure whether we find insects or flowers pleasant or unpleasant. The measurement of word associations is quantified using the WEAT metric in our paper. It is a metric that calculates an effect size (Cohen's d) and also provides a p-value (to measure statistical significance of the results). In our paper, we use word embeddings from language models to perform these tests and understand biased associations in language models across different languages.