A collection of email messages of employees in the Enron Corporation. There are two features: - email_body: email body text. - subject_line: email subject text.

AG is a collection of more than 1 million news articles. News articles have been gathered from more than 2000 news sources by ComeToMyHead in more than 1 year of activity. ComeToMyHead is an academic news search engine which has been running since July, 2004. The dataset is provided by the academic comunity for research purposes in data mining (clustering, classification, etc), information retrieval (ranking, search, etc), xml, data compression, data streaming, and any other non-commercial activity. For more information, please refer to the link http://www.di.unipi.it/~gulli/AG_corpus_of_news_articles.html . The AG's news topic classification dataset is constructed by Xiang Zhang (xiang.zhang@nyu.edu) from the dataset above. It is used as a text classification benchmark in the following paper: Xiang Zhang, Junbo Zhao, Yann LeCun. Character-level Convolutional Networks for Text Classification. Advances in Neural Information Processing Systems 28 (NIPS 2015).

A new dataset of 7,787 genuine grade-school level, multiple-choice science questions, assembled to encourage research in advanced question-answering. The dataset is partitioned into a Challenge Set and an Easy Set, where the former contains only questions answered incorrectly by both a retrieval-based algorithm and a word co-occurrence algorithm. We are also including a corpus of over 14 million science sentences relevant to the task, and an implementation of three neural baseline models for this dataset. We pose ARC as a challenge to the community.

Allocine Dataset: A Large-Scale French Movie Reviews Dataset. This is a dataset for binary sentiment classification, made of user reviews scraped from Allocine.fr. It contains 100k positive and 100k negative reviews divided into 3 balanced splits: train (160k reviews), val (20k) and test (20k).

The Adversarial Natural Language Inference (ANLI) is a new large-scale NLI benchmark dataset, The dataset is collected via an iterative, adversarial human-and-model-in-the-loop procedure. ANLI is much more difficult than its predecessors including SNLI and MNLI. It contains three rounds. Each round has train/dev/test splits.

Arabic Reading Comprehension Dataset (ARCD) composed of 1,395 questions posed by crowdworkers on Wikipedia articles.

the Abductive Natural Language Inference Dataset from AI2

BillSum, summarization of US Congressional and California state bills. There are several features: - text: bill text. - summary: summary of the bills. - title: title of the bills. features for us bills. ca bills does not have. - text_len: number of chars in text. - sum_len: number of chars in summary.

We introduce BIOMRC, a large-scale cloze-style biomedical MRC dataset. Care was taken to reduce noise, compared to the previous BIOREAD dataset of Pappas et al. (2018). Experiments show that simple heuristics do not perform well on the new dataset and that two neural MRC models that had been tested on BIOREAD perform much better on BIOMRC, indicating that the new dataset is indeed less noisy or at least that its task is more feasible. Non-expert human performance is also higher on the new dataset compared to BIOREAD, and biomedical experts perform even better. We also introduce a new BERT-based MRC model, the best version of which substantially outperforms all other methods tested, reaching or surpassing the accuracy of biomedical experts in some experiments. We make the new dataset available in three different sizes, also releasing our code, and providing a leaderboard.

A dataset of 7k conversations explicitly designed to exhibit multiple conversation modes: displaying personality, having empathy, and demonstrating knowledge.

BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars.

The Blog Authorship Corpus consists of the collected posts of 19,320 bloggers gathered from blogger.com in August 2004. The corpus incorporates a total of 681,288 posts and over 140 million words - or approximately 35 posts and 7250 words per person. Each blog is presented as a separate file, the name of which indicates a blogger id# and the blogger’s self-provided gender, age, industry and astrological sign. (All are labeled for gender and age but for many, industry and/or sign is marked as unknown.) All bloggers included in the corpus fall into one of three age groups: · 8240 "10s" blogs (ages 13-17), · 8086 "20s" blogs(ages 23-27) · 2994 "30s" blogs (ages 33-47). For each age group there are an equal number of male and female bloggers. Each blog in the corpus includes at least 200 occurrences of common English words. All formatting has been stripped with two exceptions. Individual posts within a single blogger are separated by the date of the following post and links within a post are denoted by the label urllink. The corpus may be freely used for non-commercial research purposes

Books are a rich source of both fine-grained information, how a character, an object or a scene looks like, as well as high-level semantics, what someone is thinking, feeling and how these states evolve through a story.This work aims to align books to their movie releases in order to providerich descriptive explanations for visual content that go semantically farbeyond the captions available in current datasets. \

BoolQ is a question answering dataset for yes/no questions containing 15942 examples. These questions are naturally occurring ---they are generated in unprompted and unconstrained settings. Each example is a triplet of (question, passage, answer), with the title of the page as optional additional context. The text-pair classification setup is similar to existing natural language inference tasks.

Break is a human annotated dataset of natural language questions and their Question Decomposition Meaning Representations (QDMRs). Break consists of 83,978 examples sampled from 10 question answering datasets over text, images and databases. This repository contains the Break dataset along with information on the exact data format.

A colossal, cleaned version of Common Crawl's web crawl corpus. Based on Common Crawl dataset: "https://commoncrawl.org" Due to the overhead of cleaning the dataset, it is recommend you prepare it with a distributed service like Cloud Dataflow. More info at https://www.tensorflow.org/datasets/beam_datasets.

The CFQ dataset (and it's splits) for measuring compositional generalization. See https://arxiv.org/abs/1912.09713.pdf for background. Example usage: data = datasets.load_dataset('cfq/mcd1')

The comments in this dataset come from an archive of the Civil Comments platform, a commenting plugin for independent news sites. These public comments were created from 2015 - 2017 and appeared on approximately 50 English-language news sites across the world. When Civil Comments shut down in 2017, they chose to make the public comments available in a lasting open archive to enable future research. The original data, published on figshare, includes the public comment text, some associated metadata such as article IDs, timestamps and commenter-generated "civility" labels, but does not include user ids. Jigsaw extended this dataset by adding additional labels for toxicity and identity mentions. This data set is an exact replica of the data released for the Jigsaw Unintended Bias in Toxicity Classification Kaggle challenge. This dataset is released under CC0, as is the underlying comment text.

CLUE, A Chinese Language Understanding Evaluation Benchmark (https://www.cluebenchmarks.com/) is a collection of resources for training, evaluating, and analyzing Chinese language understanding systems.

A Span-Extraction dataset for Chinese machine reading comprehension to add language diversities in this area. The dataset is composed by near 20,000 real questions annotated on Wikipedia paragraphs by human experts. We also annotated a challenge set which contains the questions that need comprehensive understanding and multi-sentence inference throughout the context.

CNN/DailyMail non-anonymized summarization dataset. There are two features: - article: text of news article, used as the document to be summarized - highlights: joined text of highlights with <s> and </s> around each highlight, which is the target summary

dataset contains discourse annotation and relation on threads from reddit during 2016

ComQA is a dataset of 11,214 questions, which were collected from WikiAnswers, a community question answering website. By collecting questions from such a site we ensure that the information needs are ones of interest to actual users. Moreover, questions posed there are often cannot be answered by commercial search engines or QA technology, making them more interesting for driving future research compared to those collected from an engine's query log. The dataset contains questions with various challenging phenomena such as the need for temporal reasoning, comparison (e.g., comparatives, superlatives, ordinals), compositionality (multiple, possibly nested, subquestions with multiple entities), and unanswerable questions (e.g., Who was the first human being on Mars?). Through a large crowdsourcing effort, questions in ComQA are grouped into 4,834 paraphrase clusters that express the same information need. Each cluster is annotated with its answer(s). ComQA answers come in the form of Wikipedia entities wherever possible. Wherever the answers are temporal or measurable quantities, TIMEX3 and the International System of Units (SI) are used for normalization.

CommonGen is a constrained text generation task, associated with a benchmark dataset, to explicitly test machines for the ability of generative commonsense reasoning. Given a set of common concepts; the task is to generate a coherent sentence describing an everyday scenario using these concepts. CommonGen is challenging because it inherently requires 1) relational reasoning using background commonsense knowledge, and 2) compositional generalization ability to work on unseen concept combinations. Our dataset, constructed through a combination of crowd-sourcing from AMT and existing caption corpora, consists of 30k concept-sets and 50k sentences in total.

CommonsenseQA is a new multiple-choice question answering dataset that requires different types of commonsense knowledge to predict the correct answers . It contains 12,102 questions with one correct answer and four distractor answers. The dataset is provided in two major training/validation/testing set splits: "Random split" which is the main evaluation split, and "Question token split", see paper for details.

CompGuessWhat?! is an instance of a multi-task framework for evaluating the quality of learned neural representations, in particular concerning attribute grounding. Use this dataset if you want to use the set of games whose reference scene is an image in VisualGenome. Visit the website for more details: https://compguesswhat.github.io

Text chunking consists of dividing a text in syntactically correlated parts of words. For example, the sentence He reckons the current account deficit will narrow to only # 1.8 billion in September . can be divided as follows: [NP He ] [VP reckons ] [NP the current account deficit ] [VP will narrow ] [PP to ] [NP only # 1.8 billion ] [PP in ] [NP September ] . Text chunking is an intermediate step towards full parsing. It was the shared task for CoNLL-2000. Training and test data for this task is available. This data consists of the same partitions of the Wall Street Journal corpus (WSJ) as the widely used data for noun phrase chunking: sections 15-18 as training data (211727 tokens) and section 20 as test data (47377 tokens). The annotation of the data has been derived from the WSJ corpus by a program written by Sabine Buchholz from Tilburg University, The Netherlands.

The shared task of CoNLL-2003 concerns language-independent named entity recognition. We will concentrate on four types of named entities: persons, locations, organizations and names of miscellaneous entities that do not belong to the previous three groups. The CoNLL-2003 shared task data files contain four columns separated by a single space. Each word has been put on a separate line and there is an empty line after each sentence. The first item on each line is a word, the second a part-of-speech (POS) tag, the third a syntactic chunk tag and the fourth the named entity tag. The chunk tags and the named entity tags have the format I-TYPE which means that the word is inside a phrase of type TYPE. Only if two phrases of the same type immediately follow each other, the first word of the second phrase will have tag B-TYPE to show that it starts a new phrase. A word with tag O is not part of a phrase. Note the dataset uses IOB2 tagging scheme, whereas the original dataset uses IOB1. For more details see https://www.clips.uantwerpen.be/conll2003/ner/ and https://www.aclweb.org/anthology/W03-0419

CoQA: A Conversational Question Answering Challenge

This corpus contains a large metadata-rich collection of fictional conversations extracted from raw movie scripts: - 220,579 conversational exchanges between 10,292 pairs of movie characters - involves 9,035 characters from 617 movies - in total 304,713 utterances - movie metadata included: - genres - release year - IMDB rating - number of IMDB votes - IMDB rating - character metadata included: - gender (for 3,774 characters) - position on movie credits (3,321 characters)

Common Sense Explanations (CoS-E) allows for training language models to automatically generate explanations that can be used during training and inference in a novel Commonsense Auto-Generated Explanation (CAGE) framework.

Cosmos QA is a large-scale dataset of 35.6K problems that require commonsense-based reading comprehension, formulated as multiple-choice questions. It focuses on reading between the lines over a diverse collection of people's everyday narratives, asking questions concerning on the likely causes or effects of events that require reasoning beyond the exact text spans in the context

Storytelling with Dialogue: A Critical Role Dungeons and Dragons Dataset. Critical Role is an unscripted, live-streamed show where a fixed group of people play Dungeons and Dragons, an open-ended role-playing game. The dataset is collected from 159 Critical Role episodes transcribed to text dialogues, consisting of 398,682 turns. It also includes corresponding abstractive summaries collected from the Fandom wiki. The dataset is linguistically unique in that the narratives are generated entirely through player collaboration and spoken interaction. For each dialogue, there are a large number of turns, multiple abstractive summaries with varying levels of detail, and semantic ties to the previous dialogues.

\

We develop a high-quality multi-turn dialog dataset, DailyDialog, which is intriguing in several aspects. The language is human-written and less noisy. The dialogues in the dataset reflect our daily communication way and cover various topics about our daily life. We also manually label the developed dataset with communication intention and emotion information. Then, we evaluate existing approaches on DailyDialog dataset and hope it benefit the research field of dialog systems.

Composed by 30 students from one of the author's undergraduate classes. These sentence pairs cover topics ranging from real events (e.g., Iran's plan to attack the Saudi ambassador to the U.S.) to events/characters in movies (e.g., Batman) and purely imaginary situations, largely reflecting the pop culture as perceived by the American kids born in the early 90s. Each annotated example spans four lines: the first line contains the sentence, the second line contains the target pronoun, the third line contains the two candidate antecedents, and the fourth line contains the correct antecedent. If the target pronoun appears more than once in the sentence, its first occurrence is the one to be resolved.

DISCOFUSE is a large scale dataset for discourse-based sentence fusion.

Multiple entities in a document generally exhibit complex inter-sentence relations, and cannot be well handled by existing relation extraction (RE) methods that typically focus on extracting intra-sentence relations for single entity pairs. In order to accelerate the research on document-level RE, we introduce DocRED, a new dataset constructed from Wikipedia and Wikidata with three features: - DocRED annotates both named entities and relations, and is the largest human-annotated dataset for document-level RE from plain text. - DocRED requires reading multiple sentences in a document to extract entities and infer their relations by synthesizing all information of the document. - Along with the human-annotated data, we also offer large-scale distantly supervised data, which enables DocRED to be adopted for both supervised and weakly supervised scenarios.

DoQA is a dataset for accessing Domain Specific FAQs via conversational QA that contains 2,437 information-seeking question/answer dialogues (10,917 questions in total) on three different domains: cooking, travel and movies. Note that we include in the generic concept of FAQs also Community Question Answering sites, as well as corporate information in intranets which is maintained in textual form similar to FAQs, often referred to as internal “knowledge bases”. These dialogues are created by crowd workers that play the following two roles: the user who asks questions about a given topic posted in Stack Exchange (https://stackexchange.com/), and the domain expert who replies to the questions by selecting a short span of text from the long textual reply in the original post. The expert can rephrase the selected span, in order to make it look more natural. The dataset covers unanswerable questions and some relevant dialogue acts. DoQA enables the development and evaluation of conversational QA systems that help users access the knowledge buried in domain specific FAQs.

DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs. . DROP is a crowdsourced, adversarially-created, 96k-question benchmark, in which a system must resolve references in a question, perhaps to multiple input positions, and perform discrete operations over them (such as addition, counting, or sorting). These operations require a much more comprehensive understanding of the content of paragraphs than what was necessary for prior datasets.

Explain Like I'm 5 long form QA dataset

In this dataset, given a textual dialogue i.e. an utterance along with two previous turns of context, the goal was to infer the underlying emotion of the utterance by choosing from four emotion classes - Happy, Sad, Angry and Others.

Emotion is a dataset of English Twitter messages with eight basic emotions: anger, anticipation, disgust, fear, joy, sadness, surprise, and trust. For more detailed information please refer to the paper.

PyTorch original implementation of Towards Empathetic Open-domain Conversation Models: a New Benchmark and Dataset

Eraser Multi RC is a dataset for queries over multi-line passages, along with answers and a rationalte. Each example in this dataset has the following 5 parts 1. A Mutli-line Passage 2. A Query about the passage 3. An Answer to the query 4. A Classification as to whether the answer is right or wrong 5. An Explanation justifying the classification

The e-SNLI dataset extends the Stanford Natural Language Inference Dataset to include human-annotated natural language explanations of the entailment relations.

In Event2Mind, we explore the task of understanding stereotypical intents and reactions to events. Through crowdsourcing, we create a large corpus with 25,000 events and free-form descriptions of their intents and reactions, both of the event's subject and (potentially implied) other participants.

With billions of individual pages on the web providing information on almost every conceivable topic, we should have the ability to collect facts that answer almost every conceivable question. However, only a small fraction of this information is contained in structured sources (Wikidata, Freebase, etc.) – we are therefore limited by our ability to transform free-form text to structured knowledge. There is, however, another problem that has become the focus of a lot of recent research and media coverage: false information coming from unreliable sources. [1] [2] The FEVER workshops are a venue for work in verifiable knowledge extraction and to stimulate progress in this direction.

Evaluation datasets for low-resource machine translation: Nepali-English and Sinhala-English.

FQuAD: French Question Answering Dataset We introduce FQuAD, a native French Question Answering Dataset. FQuAD contains 25,000+ question and answer pairs. Finetuning CamemBERT on FQuAD yields a F1 score of 88% and an exact match of 77.9%.

GAP is a gender-balanced dataset containing 8,908 coreference-labeled pairs of (ambiguous pronoun, antecedent name), sampled from Wikipedia and released by Google AI Language for the evaluation of coreference resolution in practical applications.

The GermEval 2014 NER Shared Task builds on a new dataset with German Named Entity annotation with the following properties: - The data was sampled from German Wikipedia and News Corpora as a collection of citations. - The dataset covers over 31,000 sentences corresponding to over 590,000 tokens. - The NER annotation uses the NoSta-D guidelines, which extend the Tübingen Treebank guidelines, using four main NER categories with sub-structure, and annotating embeddings among NEs such as [ORG FC Kickers [LOC Darmstadt]].

Headline-generation on a corpus of article pairs from Gigaword consisting of around 4 million articles. Use the 'org_data' provided by https://github.com/microsoft/unilm/ which is identical to https://github.com/harvardnlp/sent-summary but with better format. There are two features: - document: article. - summary: headline.

GLUE, the General Language Understanding Evaluation benchmark (https://gluebenchmark.com/) is a collection of resources for training, evaluating, and analyzing natural language understanding systems.

A dataset cross-topic authorship attribution. The dataset is provided by Stamatatos 2013. 1- The cross-topic scenarios are based on Table-4 in Stamatatos 2017 (Ex. cross_topic_1 => row 1:P S U&W ). 2- The cross-genre scenarios are based on Table-5 in the same paper. (Ex. cross_genre_1 => row 1:B P S&U&W). 3- The same-topic/genre scenario is created by grouping all the datasts as follows. For ex., to use same_topic and split the data 60-40 use: train_ds = load_dataset('guardian_authorship', name="cross_topic_<<#>>", split='train[:60%]+validation[:60%]+test[:60%]') tests_ds = load_dataset('guardian_authorship', name="cross_topic_<<#>>", split='train[-40%:]+validation[-40%:]+test[-40%:]') IMPORTANT: train+validation+test[:60%] will generate the wrong splits becasue the data is imbalanced * See https://huggingface.co/docs/datasets/splits.html for detailed/more examples

The HANS dataset is an NLI evaluation set that tests specific hypotheses about invalid heuristics that NLI models are likely to learn.

This release contains 1.3 million pairs of aligned text chunks (sentences or smaller fragments) from the official records (Hansards) of the 36th Canadian Parliament. The complete Hansards of the debates in the House and Senate of the 36th Canadian Parliament, as far as available, were aligned. The corpus was then split into 5 sets of sentence pairs: training (80% of the sentence pairs), two sets of sentence pairs for testing (5% each), and two sets of sentence pairs for final evaluation (5% each). The current release consists of the training and testing sets. The evaluation sets are reserved for future MT evaluation purposes and currently not available. Caveats 1. This release contains only sentence pairs. Even though the order of the sentences is the same as in the original, there may be gaps resulting from many-to-one, many-to-many, or one-to-many alignments that were filtered out. Therefore, this release may not be suitable for discourse-related research. 2. Neither the sentence splitting nor the alignments are perfect. In particular, watch out for pairs that differ considerably in length. You may want to filter these out before you do any statistical training. The alignment of the Hansards was performed as part of the ReWrite project under funding from the DARPA TIDES program.

Hyperpartisan News Detection was a dataset created for PAN @ SemEval 2019 Task 4. Given a news article text, decide whether it follows a hyperpartisan argumentation, i.e., whether it exhibits blind, prejudiced, or unreasoning allegiance to one party, faction, cause, or person. There are 2 parts: - byarticle: Labeled through crowdsourcing on an article basis. The data contains only articles for which a consensus among the crowdsourcing workers existed. - bypublisher: Labeled by the overall bias of the publisher as provided by BuzzFeed journalists or MediaBiasFactCheck.com.

Large Movie Review Dataset. This is a dataset for binary sentiment classification containing substantially more data than previous benchmark datasets. We provide a set of 25,000 highly polar movie reviews for training, and 25,000 for testing. There is additional unlabeled data for use as well.\

The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five: German, English, Italian, Dutch, Romanian. For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.

Dataset containing 216,930 Jeopardy questions, answers and other data. The json file is an unordered list of questions where each question has 'category' : the question category, e.g. "HISTORY" 'value' : integer $ value of the question as string, e.g. "200" Note: This is "None" for Final Jeopardy! and Tiebreaker questions 'question' : text of question Note: This sometimes contains hyperlinks and other things messy text such as when there's a picture or video question 'answer' : text of answer 'round' : one of "Jeopardy!","Double Jeopardy!","Final Jeopardy!" or "Tiebreaker" Note: Tiebreaker questions do happen but they're very rare (like once every 20 years) 'show_number' : int of show number, e.g '4680' 'air_date' : string of the show air date in format YYYY-MM-DD

KILT tasks training and evaluation data. - [FEVER](https://fever.ai) | Fact Checking | fever - [AIDA CoNLL-YAGO](https://www.mpi-inf.mpg.de/departments/databases-and-information-systems/research/ambiverse-nlu/aida/downloads) | Entity Linking | aidayago2 - [WNED-WIKI](https://github.com/U-Alberta/wned) | Entity Linking | wned - [WNED-CWEB](https://github.com/U-Alberta/wned) | Entity Linking | cweb - [T-REx](https://hadyelsahar.github.io/t-rex) | Slot Filling | trex - [Zero-Shot RE](http://nlp.cs.washington.edu/zeroshot) | Slot Filling | structured_zeroshot - [Natural Questions](https://ai.google.com/research/NaturalQuestions) | Open Domain QA | nq - [HotpotQA](https://hotpotqa.github.io) | Open Domain QA | hotpotqa - [TriviaQA](http://nlp.cs.washington.edu/triviaqa) | Open Domain QA | triviaqa - [ELI5](https://facebookresearch.github.io/ELI5/explore.html) | Open Domain QA | eli5 - [Wizard of Wikipedia](https://parl.ai/projects/wizard_of_wikipedia) | Dialogue | wow To finish linking TriviaQA questions to the IDs provided, follow the instructions [here](http://github.com/huggingface/datasets/datasets/kilt_tasks/README.md).

KILT-Wikipedia: Wikipedia pre-processed for KILT.

Korean Natural Language Inference datasets

LC-QuAD 2.0 is a Large Question Answering dataset with 30,000 pairs of question and its corresponding SPARQL query. The target knowledge base is Wikidata and DBpedia, specifically the 2018 version. Please see our paper for details about the dataset creation process and framework.

Language modeling resources to be used in conjunction with the LibriSpeech ASR corpus.

LinCE is a centralized Linguistic Code-switching Evaluation benchmark (https://ritual.uh.edu/lince/) that contains data for training and evaluating NLP systems on code-switching tasks.

A benchmark corpus to be used for measuring progress in statistical language modeling. This has almost one billion words in the training data.

Mathematics database. This dataset code generates mathematical question and answer pairs, from a range of question types at roughly school-level difficulty. This is designed to test the mathematical learning and algebraic reasoning skills of learning models. Original paper: Analysing Mathematical Reasoning Abilities of Neural Models (Saxton, Grefenstette, Hill, Kohli). Example usage: train_examples, val_examples = datasets.load_dataset( 'math_dataset/arithmetic__mul', split=['train', 'test'], as_supervised=True)

Our dataset is gathered by using a new representation language to annotate over the AQuA-RAT dataset. AQuA-RAT has provided the questions, options, rationale, and the correct options.

MATINF is the first jointly labeled large-scale dataset for classification, question answering and summarization. MATINF contains 1.07 million question-answer pairs with human-labeled categories and user-generated question descriptions. Based on such rich information, MATINF is applicable for three major NLP tasks, including classification, question answering, and summarization. We benchmark existing methods and a novel multi-task baseline over MATINF to inspire further research. Our comprehensive comparison and experiments over MATINF and other datasets demonstrate the merits held by MATINF.

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.

We present MLSUM, the first large-scale MultiLingual SUMmarization dataset. Obtained from online newspapers, it contains 1.5M+ article/summary pairs in five different languages -- namely, French, German, Spanish, Russian, Turkish. Together with English newspapers from the popular CNN/Daily mail dataset, the collected data form a large scale multilingual dataset which can enable new research directions for the text summarization community. We report cross-lingual comparative analyses based on state-of-the-art systems. These highlight existing biases which motivate the use of a multi-lingual dataset.

The movie rationale dataset contains human annotated rationales for movie reviews.

Starting with a paper released at NIPS 2016, MS MARCO is a collection of datasets focused on deep learning in search. The first dataset was a question answering dataset featuring 100,000 real Bing questions and a human generated answer. Since then we released a 1,000,000 question dataset, a natural langauge generation dataset, a passage ranking dataset, keyphrase extraction dataset, crawling dataset, and a conversational search. There have been 277 submissions. 20 KeyPhrase Extraction submissions, 87 passage ranking submissions, 0 document ranking submissions, 73 QnA V2 submissions, 82 NLGEN submisions, and 15 QnA V1 submissions This data comes in three tasks/forms: Original QnA dataset(v1.1), Question Answering(v2.1), Natural Language Generation(v2.1). The original question answering datset featured 100,000 examples and was released in 2016. Leaderboard is now closed but data is availible below. The current competitive tasks are Question Answering and Natural Language Generation. Question Answering features over 1,000,000 queries and is much like the original QnA dataset but bigger and with higher quality. The Natural Language Generation dataset features 180,000 examples and builds upon the QnA dataset to deliver answers that could be spoken by a smart speaker.

Multi-News, consists of news articles and human-written summaries of these articles from the site newser.com. Each summary is professionally written by editors and includes links to the original articles cited. There are two features: - document: text of news articles seperated by special token "|||||". - summary: news summary.

The Multi-Genre Natural Language Inference (MultiNLI) corpus is a crowd-sourced collection of 433k sentence pairs annotated with textual entailment information. The corpus is modeled on the SNLI corpus, but differs in that covers a range of genres of spoken and written text, and supports a distinctive cross-genre generalization evaluation. The corpus served as the basis for the shared task of the RepEval 2017 Workshop at EMNLP in Copenhagen.

The Multi-Genre Natural Language Inference (MultiNLI) corpus is a crowd-sourced collection of 433k sentence pairs annotated with textual entailment information. The corpus is modeled on the SNLI corpus, but differs in that covers a range of genres of spoken and written text, and supports a distinctive cross-genre generalization evaluation. The corpus served as the basis for the shared task of the RepEval 2017 Workshop at EMNLP in Copenhagen.

Examples taken from the Winograd Schema Challenge modified to ensure that answers are a single word from the context. This modified Winograd Schema Challenge (MWSC) ensures that scores are neither inflated nor deflated by oddities in phrasing.

The NQ corpus contains questions from real users, and it requires QA systems to read and comprehend an entire Wikipedia article that may or may not contain the answer to the question. The inclusion of real user questions, and the requirement that solutions should read an entire page to find the answer, cause NQ to be a more realistic and challenging task than prior QA datasets.

The 20 Newsgroups data set is a collection of approximately 20,000 newsgroup documents, partitioned (nearly) evenly across 20 different newsgroups. The 20 newsgroups collection has become a popular data set for experiments in text applications of machine learning techniques, such as text classification and text clustering.

NEWSROOM is a large dataset for training and evaluating summarization systems. It contains 1.3 million articles and summaries written by authors and editors in the newsrooms of 38 major publications. Dataset features includes: - text: Input news text. - summary: Summary for the news. And additional features: - title: news title. - url: url of the news. - date: date of the article. - density: extractive density. - coverage: extractive coverage. - compression: compression ratio. - density_bin: low, medium, high. - coverage_bin: extractive, abstractive. - compression_bin: low, medium, high. This dataset can be downloaded upon requests. Unzip all the contents "train.jsonl, dev.josnl, test.jsonl" to the tfds folder.

An open-source replication of the WebText dataset from OpenAI.

The Opinosis Opinion Dataset consists of sentences extracted from reviews for 51 topics. Topics and opinions are obtained from Tripadvisor, Edmunds.com and Amazon.com.

This repository contains the PG-19 language modeling benchmark. It includes a set of books extracted from the Project Gutenberg books library, that were published before 1919. It also contains metadata of book titles and publication dates. PG-19 is over double the size of the Billion Word benchmark and contains documents that are 20X longer, on average, than the WikiText long-range language modelling benchmark. Books are partitioned into a train, validation, and test set. Book metadata is stored in metadata.csv which contains (book_id, short_book_title, publication_date). Unlike prior benchmarks, we do not constrain the vocabulary size --- i.e. mapping rare words to an UNK token --- but instead release the data as an open-vocabulary benchmark. The only processing of the text that has been applied is the removal of boilerplate license text, and the mapping of offensive discriminatory words as specified by Ofcom to placeholder tokens. Users are free to model the data at the character-level, subword-level, or via any mechanism that can model an arbitrary string of text. To compare models we propose to continue measuring the word-level perplexity, by calculating the total likelihood of the dataset (via any chosen subword vocabulary or character-based scheme) divided by the number of tokens --- specified below in the dataset statistics table. One could use this dataset for benchmarking long-range language models, or use it to pre-train for other natural language processing tasks which require long-range reasoning, such as LAMBADA or NarrativeQA. We would not recommend using this dataset to train a general-purpose language model, e.g. for applications to a production-system dialogue agent, due to the dated linguistic style of old texts and the inherent biases present in historical writing.

Piaf is a reading comprehension dataset. This version, published in February 2020, contains 3835 questions on French Wikipedia.

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.

QA4MRE dataset was created for the CLEF 2011/2012/2013 shared tasks to promote research in question answering and reading comprehension. The dataset contains a supporting passage and a set of questions corresponding to the passage. Multiple options for answers are provided for each question, of which only one is correct. The training and test datasets are available for the main track. Additional gold standard documents are available for two pilot studies: one on alzheimers data, and the other on entrance exams data.

A dataset reducing relation extraction to simple reading comprehension questions

We have created two new Reading Comprehension datasets focussing on multi-hop (alias multi-step) inference. Several pieces of information often jointly imply another fact. In multi-hop inference, a new fact is derived by combining facts via a chain of multiple steps. Our aim is to build Reading Comprehension methods that perform multi-hop inference on text, where individual facts are spread out across different documents. The two QAngaroo datasets provide a training and evaluation resource for such methods.

The Qanta dataset is a question answering dataset based on the academic trivia game Quizbowl.

QASC is a question-answering dataset with a focus on sentence composition. It consists of 9,980 8-way multiple-choice questions about grade school science (8,134 train, 926 dev, 920 test), and comes with a corpus of 17M sentences.

QuaRel is a crowdsourced dataset of 2771 multiple-choice story questions, including their logical forms.

QuaRTz is a crowdsourced dataset of 3864 multiple-choice questions about open domain qualitative relationships. Each question is paired with one of 405 different background sentences (sometimes short paragraphs). The QuaRTz dataset V1 contains 3864 questions about open domain qualitative relationships. Each question is paired with one of 405 different background sentences (sometimes short paragraphs). The dataset is split into train (2696), dev (384) and test (784). A background sentence will only appear in a single split.

Quoref is a QA dataset which tests the coreferential reasoning capability of reading comprehension systems. In this span-selection benchmark containing 24K questions over 4.7K paragraphs from Wikipedia, a system must resolve hard coreferences before selecting the appropriate span(s) in the paragraphs for answering questions.

Race is a large-scale reading comprehension dataset with more than 28,000 passages and nearly 100,000 questions. The dataset is collected from English examinations in China, which are designed for middle school and high school students. The dataset can be served as the training and test sets for machine comprehension.

Logical reasoning is an important ability to examine, analyze, and critically evaluate arguments as they occur in ordinary language as the definition from LSAC. ReClor is a dataset extracted from logical reasoning questions of standardized graduate admission examinations. Empirical results show that the state-of-the-art models struggle on ReClor with poor performance indicating more research is needed to essentially enhance the logical reasoning ability of current models. We hope this dataset could help push Machine Reading Comprehension (MRC) towards more complicated reasonin

This corpus contains preprocessed posts from the Reddit dataset. The dataset consists of 3,848,330 posts with an average length of 270 words for content, and 28 words for the summary. Features includes strings: author, body, normalizedBody, content, summary, subreddit, subreddit_id. Content is used as document and summary is used as summary.

Reddit dataset, where TIFU denotes the name of subbreddit /r/tifu. As defined in the publication, styel "short" uses title as summary and "long" uses tldr as summary. Features includes: - document: post text without tldr. - tldr: tldr line. - title: trimmed title without tldr. - ups: upvotes. - score: score. - num_comments: number of comments. - upvote_ratio: upvote ratio.

The Reuters-21578 dataset is one of the most widely used data collections for text categorization research. It is collected from the Reuters financial newswire service in 1987.

Movie Review Dataset. This is a dataset of containing 5,331 positive and 5,331 negative processed sentences from Rotten Tomatoes movie reviews. This data was first used in Bo Pang and Lillian Lee, ``Seeing stars: Exploiting class relationships for sentiment categorization with respect to rating scales.'', Proceedings of the ACL, 2005.

SCAN tasks with various splits. SCAN is a set of simple language-driven navigation tasks for studying compositional learning and zero-shot generalization. See https://github.com/brendenlake/SCAN for a description of the splits. Example usage: data = datasets.load_dataset('scan/length')

This is a dataset for classifying citation intents in academic papers. The main citation intent label for each Json object is specified with the label key while the citation context is specified in with a context key. Example: { 'string': 'In chacma baboons, male-infant relationships can be linked to both formation of friendships and paternity success [30,31].' 'sectionName': 'Introduction', 'label': 'background', 'citingPaperId': '7a6b2d4b405439', 'citedPaperId': '9d1abadc55b5e0', ... } You may obtain the full information about the paper using the provided paper ids with the Semantic Scholar API (https://api.semanticscholar.org/). The labels are: Method, Background, Result

Scientific papers datasets contains two sets of long and structured documents. The datasets are obtained from ArXiv and PubMed OpenAccess repositories. Both "arxiv" and "pubmed" have two features: - article: the body of the document, pagragraphs seperated by "/n". - abstract: the abstract of the document, pagragraphs seperated by "/n". - section_names: titles of sections, seperated by "/n".

SciFact, a dataset of 1.4K expert-written scientific claims paired with evidence-containing abstracts, and annotated with labels and rationales

The SciQ dataset contains 13,679 crowdsourced science exam questions about Physics, Chemistry and Biology, among others. The questions are in multiple-choice format with 4 answer options each. For the majority of the questions, an additional paragraph with supporting evidence for the correct answer is provided.

The SciTail dataset is an entailment dataset created from multiple-choice science exams and web sentences. Each question and the correct answer choice are converted into an assertive statement to form the hypothesis. We use information retrieval to obtain relevant text from a large text corpus of web sentences, and use these sentences as a premise P. We crowdsource the annotation of such premise-hypothesis pair as supports (entails) or not (neutral), in order to create the SciTail dataset. The dataset contains 27,026 examples with 10,101 examples with entails label and 16,925 examples with neutral label

We publicly release a new large-scale dataset, called SearchQA, for machine comprehension, or question-answering. Unlike recently released datasets, such as DeepMind CNN/DailyMail and SQuAD, the proposed SearchQA was constructed to reflect a full pipeline of general question-answering. That is, we start not from an existing article and generate a question-answer pair, but start from an existing question-answer pair, crawled from J! Archive, and augment it with text snippets retrieved by Google. Following this approach, we built SearchQA, which consists of more than 140k question-answer pairs with each pair having 49.6 snippets on average. Each question-answer-context tuple of the SearchQA comes with additional meta-data such as the snippet's URL, which we believe will be valuable resources for future research. We conduct human evaluation as well as test two baseline methods, one simple word selection and the other deep learning based, on the SearchQA. We show that there is a meaningful gap between the human and machine performances. This suggests that the proposed dataset could well serve as a benchmark for question-answering.

Sentiment140 consists of Twitter messages with emoticons, which are used as noisy labels for sentiment classification. For more detailed information please refer to the paper.

The SNLI corpus (version 1.0) is a collection of 570k human-written English sentence pairs manually labeled for balanced classification with the labels entailment, contradiction, and neutral, supporting the task of natural language inference (NLI), also known as recognizing textual entailment (RTE).

Social Bias Frames is a new way of representing the biases and offensiveness that are implied in language. For example, these frames are meant to distill the implication that "women (candidates) are less qualified" behind the statement "we shouldn’t lower our standards to hire more women."

We introduce Social IQa: Social Interaction QA, a new question-answering benchmark for testing social commonsense intelligence. Contrary to many prior benchmarks that focus on physical or taxonomic knowledge, Social IQa focuses on reasoning about people’s actions and their social implications. For example, given an action like "Jesse saw a concert" and a question like "Why did Jesse do this?", humans can easily infer that Jesse wanted "to see their favorite performer" or "to enjoy the music", and not "to see what's happening inside" or "to see if it works". The actions in Social IQa span a wide variety of social situations, and answer candidates contain both human-curated answers and adversarially-filtered machine-generated candidates. Social IQa contains over 37,000 QA pairs for evaluating models’ abilities to reason about the social implications of everyday events and situations. (Less)

The Sogou News dataset is a mixture of 2,909,551 news articles from the SogouCA and SogouCS news corpora, in 5 categories. The number of training samples selected for each class is 90,000 and testing 12,000. Note that the Chinese characters have been converted to Pinyin. classification labels of the news are determined by their domain names in the URL. For example, the news with URL http://sports.sohu.com is categorized as a sport class.

Stanford Question Answering Dataset (SQuAD) is a reading comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles, where the answer to every question is a segment of text, or span, from the corresponding reading passage, or the question might be unanswerable.

automatic translation of the Stanford Question Answering Dataset (SQuAD) v2 into Spanish

SQuAD-it is derived from the SQuAD dataset and it is obtained through semi-automatic translation of the SQuAD dataset into Italian. It represents a large-scale dataset for open question answering processes on factoid questions in Italian. The dataset contains more than 60,000 question/answer pairs derived from the original English dataset. The dataset is split into training and test sets to support the replicability of the benchmarking of QA systems:

Portuguese translation of the SQuAD dataset. The translation was performed automatically using the Google Cloud API.

combines the 100,000 questions in SQuAD1.1 with over 50,000 unanswerable questions written adversarially by crowdworkers to look similar to answerable ones. To do well on SQuAD2.0, systems must not only answer questions when possible, but also determine when no answer is supported by the paragraph and abstain from answering.

The goal of the style change detection task is to identify text positions within a given multi-author document at which the author switches. Detecting these positions is a crucial part of the authorship identification process, and for multi-author document analysis in general. Access to the dataset needs to be requested from zenodo.

SuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after GLUE with a new set of more difficult language understanding tasks, improved resources, and a new public leaderboard.

Data sets derived from TED talk transcripts for comparing similar language pairs where one is high resource and the other is low resource.

Massively multilingual (60 language) data set derived from TED Talk transcripts. Each record consists of parallel arrays of language and text. Missing and incomplete translations will be filtered out.

40,000 lines of Shakespeare from a variety of Shakespeare's plays. Featured in Andrej Karpathy's blog post 'The Unreasonable Effectiveness of Recurrent Neural Networks': http://karpathy.github.io/2015/05/21/rnn-effectiveness/. To use for e.g. character modelling: ``` d = datasets.load_dataset(name='tiny_shakespeare')['train'] d = d.map(lambda x: datasets.Value('strings').unicode_split(x['text'], 'UTF-8')) # train split includes vocabulary for other splits vocabulary = sorted(set(next(iter(d)).numpy())) d = d.map(lambda x: {'cur_char': x[:-1], 'next_char': x[1:]}) d = d.unbatch() seq_len = 100 batch_size = 2 d = d.batch(seq_len) d = d.batch(batch_size) ```

The Text REtrieval Conference (TREC) Question Classification dataset contains 5500 labeled questions in training set and another 500 for test set. The dataset has 6 labels, 47 level-2 labels. Average length of each sentence is 10, vocabulary size of 8700. Data are collected from four sources: 4,500 English questions published by USC (Hovy et al., 2001), about 500 manually constructed questions for a few rare classes, 894 TREC 8 and TREC 9 questions, and also 500 questions from TREC 10 which serves as the test set.

TriviaqQA is a reading comprehension dataset containing over 650K question-answer-evidence triples. TriviaqQA includes 95K question-answer pairs authored by trivia enthusiasts and independently gathered evidence documents, six per question on average, that provide high quality distant supervision for answering the questions.

TyDi QA is a question answering dataset covering 11 typologically diverse languages with 204K question-answer pairs. The languages of TyDi QA are diverse with regard to their typology -- the set of linguistic features that each language expresses -- such that we expect models performing well on this set to generalize across a large number of the languages in the world. It contains language phenomena that would not be found in English-only corpora. To provide a realistic information-seeking task and avoid priming effects, questions are written by people who want to know the answer, but don’t know the answer yet, (unlike SQuAD and its descendents) and the data is collected directly in each language without the use of translation (unlike MLQA and XQuAD).

Ubuntu Dialogue Corpus, a dataset containing almost 1 million multi-turn dialogues, with a total of over 7 million utterances and 100 million words. This provides a unique resource for research into building dialogue managers based on neural language models that can make use of large amounts of unlabeled data. The dataset has both the multi-turn property of conversations in the Dialog State Tracking Challenge datasets, and the unstructured nature of interactions from microblog services such as Twitter.

The Web Of Science (WOS) dataset is a collection of data of published papers available from the Web of Science. WOS has been released in three versions: WOS-46985, WOS-11967 and WOS-5736. WOS-46985 is the full dataset. WOS-11967 and WOS-5736 are two subsets of WOS-46985.

This dataset consists of 6,642 question/answer pairs. The questions are supposed to be answerable by Freebase, a large knowledge graph. The questions are mostly centered around a single named entity. The questions are popular ones asked on the web (at least in 2013).

Clean-up text for 40+ Wikipedia languages editions of pages correspond to entities. The datasets have train/dev/test splits per language. The dataset is cleaned up by page filtering to remove disambiguation pages, redirect pages, deleted pages, and non-entity pages. Each example contains the wikidata id of the entity, and the full Wikipedia article after page processing that removes non-content sections and structured objects.

This is the wikipedia split used to evaluate the Dense Passage Retrieval (DPR) model. It contains 21M passages from wikipedia along with their DPR embeddings. The wikipedia articles were split into multiple, disjoint text blocks of 100 words as passages.

Wiki Question Answering corpus from Microsoft

Wikipedia version split into plain text snippets for dense semantic indexing.

One million English sentences, each split into two sentences that together preserve the original meaning, extracted from Wikipedia Google's WikiSplit dataset was constructed automatically from the publicly available Wikipedia revision history. Although the dataset contains some inherent noise, it can serve as valuable training data for models that split or merge sentences.

WikiHow is a new large-scale dataset using the online WikiHow (http://www.wikihow.com/) knowledge base. There are two features: - text: wikihow answers texts. - headline: bold lines as summary. There are two separate versions: - all: consisting of the concatenation of all paragraphs as the articles and the bold lines as the reference summaries. - sep: consisting of each paragraph and its summary. Download "wikihowAll.csv" and "wikihowSep.csv" from https://github.com/mahnazkoupaee/WikiHow-Dataset and place them in manual folder https://www.tensorflow.org/datasets/api_docs/python/tfds/download/DownloadConfig. Train/validation/test splits are provided by the authors. Preprocessing is applied to remove short articles (abstract length < 0.75 article length) and clean up extra commas.

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. Each example contains the content of one full Wikipedia article with cleaning to strip markdown and unwanted sections (references, etc.).

A large crowd-sourced dataset for developing natural language interfaces for relational databases

The WikiText language modeling dataset is a collection of over 100 million tokens extracted from the set of verified Good and Featured articles on Wikipedia. The dataset is available under the Creative Commons Attribution-ShareAlike License.

WinoGrande is a new collection of 44k problems, inspired by Winograd Schema Challenge (Levesque, Davis, and Morgenstern 2011), but adjusted to improve the scale and robustness against the dataset-specific bias. Formulated as a fill-in-a-blank task with binary options, the goal is to choose the right option for a given sentence which requires commonsense reasoning.

The WIQA dataset V1 has 39705 questions containing a perturbation and a possible effect in the context of a paragraph. The dataset is split into 29808 train questions, 6894 dev questions and 3003 test questions.

WNUT 17: Emerging and Rare entity recognition This shared task focuses on identifying unusual, previously-unseen entities in the context of emerging discussions. Named entities form the basis of many modern approaches to other tasks (like event clustering and summarisation), but recall on them is a real problem in noisy text - even among annotators. This drop tends to be due to novel entities and surface forms. Take for example the tweet “so.. kktny in 30 mins?” - even human experts find entity kktny hard to detect and resolve. This task will evaluate the ability to detect and classify novel, emerging, singleton named entities in noisy text. The goal of this task is to provide a definition of emerging and of rare entities, and based on that, also datasets for detecting these entities.

The x-stance dataset contains more than 150 political questions, and 67k comments written by candidates on those questions. It can be used to train and evaluate stance detection systems.

XCOPA: A Multilingual Dataset for Causal Commonsense Reasoning The Cross-lingual Choice of Plausible Alternatives dataset is a benchmark to evaluate the ability of machine learning models to transfer commonsense reasoning across languages. The dataset is the translation and reannotation of the English COPA (Roemmele et al. 2011) and covers 11 languages from 11 families and several areas around the globe. The dataset is challenging as it requires both the command of world knowledge and the ability to generalise to new languages. All the details about the creation of XCOPA and the implementation of the baselines are available in the paper.\n

XNLI is a subset of a few thousand examples from MNLI which has been translated into a 14 different languages (some low-ish resource). As with MNLI, the goal is to predict textual entailment (does sentence A imply/contradict/neither sentence B) and is a classification task (given two sentences, predict one of three labels).

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.

Extreme Summarization (XSum) Dataset. There are two features: - document: Input news article. - summary: One sentence summary of the article.

The Cross-lingual TRansfer Evaluation of Multilingual Encoders (XTREME) benchmark is a benchmark for the evaluation of the cross-lingual generalization ability of pre-trained multilingual models. It covers 40 typologically diverse languages (spanning 12 language families) and includes nine tasks that collectively require reasoning about different levels of syntax and semantics. The languages in XTREME are selected to maximize language diversity, coverage in existing tasks, and availability of training data. Among these are many under-studied languages, such as the Dravidian languages Tamil (spoken in southern India, Sri Lanka, and Singapore), Telugu and Malayalam (spoken mainly in southern India), and the Niger-Congo languages Swahili and Yoruba, spoken in Africa.

Large Yelp Review Dataset. This is a dataset for binary sentiment classification. We provide a set of 560,000 highly polar yelp reviews for training, and 38,000 for testing. ORIGIN The Yelp reviews dataset consists of reviews from Yelp. It is extracted from the Yelp Dataset Challenge 2015 data. For more information, please refer to http://www.yelp.com/dataset_challenge The Yelp reviews polarity dataset is constructed by Xiang Zhang (xiang.zhang@nyu.edu) from the above dataset. It is first used as a text classification benchmark in the following paper: Xiang Zhang, Junbo Zhao, Yann LeCun. Character-level Convolutional Networks for Text Classification. Advances in Neural Information Processing Systems 28 (NIPS 2015). DESCRIPTION The Yelp reviews polarity dataset is constructed by considering stars 1 and 2 negative, and 3 and 4 positive. For each polarity 280,000 training samples and 19,000 testing samples are take randomly. In total there are 560,000 trainig samples and 38,000 testing samples. Negative polarity is class 1, and positive class 2. The files train.csv and test.csv contain all the training samples as comma-sparated values. There are 2 columns in them, corresponding to class index (1 and 2) and review text. The review texts are escaped using double quotes ("), and any internal double quote is escaped by 2 double quotes (""). New lines are escaped by a backslash followed with an "n" character, that is "\n".