|
"citation": "@inproceedings{huang-etal-2019-cosmos,\n title = \"Cosmos {QA}: Machine Reading Comprehension with Contextual Commonsense Reasoning\",\n author = \"Huang, Lifu and\n Le Bras, Ronan and\n Bhagavatula, Chandra and\n Choi, Yejin\",\n booktitle = \"Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)\",\n month = nov,\n year = \"2019\",\n address = \"Hong Kong, China\",\n publisher = \"Association for Computational Linguistics\",\n url = \"https://www.aclweb.org/anthology/D19-1243\",\n doi = \"10.18653/v1/D19-1243\",\n pages = \"2391--2401\",\n}\n\n\n\n@InProceedings{paws2019naacl,\n title = {{PAWS: Paraphrase Adversaries from Word Scrambling}},\n author = {Zhang, Yuan and Baldridge, Jason and He, Luheng},\n booktitle = {Proc. of NAACL},\n year = {2019}\n}\n\n@inproceedings{DBLP:conf/aaai/RogersKDR20,\n author = {Anna Rogers and\n Olga Kovaleva and\n Matthew Downey and\n Anna Rumshisky},\n title = {Getting Closer to {AI} Complete Question Answering: {A} Set of Prerequisite\n Real Tasks},\n booktitle = {The Thirty-Fourth {AAAI} Conference on Artificial Intelligence, {AAAI}\n 2020, The Thirty-Second Innovative Applications of Artificial Intelligence\n Conference, {IAAI} 2020, The Tenth {AAAI} Symposium on Educational\n Advances in Artificial Intelligence, {EAAI} 2020, New York, NY, USA,\n February 7-12, 2020},\n pages = {8722--8731},\n publisher = {{AAAI} Press},\n year = {2020},\n url = {https://aaai.org/ojs/index.php/AAAI/article/view/6398},\n timestamp = {Thu, 04 Jun 2020 13:18:48 +0200},\n biburl = {https://dblp.org/rec/conf/aaai/RogersKDR20.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n@article{2016arXiv160605250R,\n author = {{Rajpurkar}, Pranav and {Zhang}, Jian and {Lopyrev},\n Konstantin and {Liang}, Percy},\n title = \"{SQuAD: 100,000+ Questions for Machine Comprehension of Text}\",\n journal = {arXiv e-prints},\n year = 2016,\n eid = {arXiv:1606.05250},\n pages = {arXiv:1606.05250},\narchivePrefix = {arXiv},\n eprint = {1606.05250},\n}\n\n@article{sundream2018,\n title={{DREAM}: A Challenge Dataset and Models for Dialogue-Based Reading Comprehension},\n author={Sun, Kai and Yu, Dian and Chen, Jianshu and Yu, Dong and Choi, Yejin and Cardie, Claire},\n journal={Transactions of the Association for Computational Linguistics},\n year={2019},\n url={https://arxiv.org/abs/1902.00164v1}\n}\n\n@article{allenai:qasc,\n author = {Tushar Khot and Peter Clark and Michal Guerquin and Peter Jansen and Ashish Sabharwal},\n title = {QASC: A Dataset for Question Answering via Sentence Composition},\n journal = {arXiv:1910.11473v2},\n year = {2020},\n}\n\n@misc{welbl2018constructing,\n title={Constructing Datasets for Multi-hop Reading Comprehension Across Documents},\n author={Johannes Welbl and Pontus Stenetorp and Sebastian Riedel},\n year={2018},\n eprint={1710.06481},\n archivePrefix={arXiv},\n primaryClass={cs.CL}\n}\n\n@article{lai2017large,\n title={RACE: Large-scale ReAding Comprehension Dataset From Examinations},\n author={Lai, Guokun and Xie, Qizhe and Liu, Hanxiao and Yang, Yiming and Hovy, Eduard},\n journal={arXiv preprint arXiv:1704.04683},\n year={2017}\n}\n\n@inproceedings{zellers2019hellaswag,\n title={HellaSwag: Can a Machine Really Finish Your Sentence?},\n author={Zellers, Rowan and Holtzman, Ari and Bisk, Yonatan and Farhadi, Ali and Choi, Yejin},\n booktitle ={Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics},\n year={2019}\n}\n\n@inproceedings{clark2019boolq,\n title={BoolQ: Exploring the Surprising Difficulty of Natural Yes/No Questions},\n author={Clark, Christopher and Lee, Kenton and Chang, Ming-Wei, and Kwiatkowski, Tom and Collins, Michael, and Toutanova, Kristina},\n booktitle={NAACL},\n year={2019}\n}\n@article{wang2019superglue,\n title={SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems},\n author={Wang, Alex and Pruksachatkun, Yada and Nangia, Nikita and Singh, Amanpreet and Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R},\n journal={arXiv preprint arXiv:1905.00537},\n year={2019}\n}\n\nNote that each SuperGLUE dataset has its own citation. Please see the source to\nget the correct citation for each contained dataset.\n\n@article{bartolo2020beat,\n author = {Bartolo, Max and Roberts, Alastair and Welbl, Johannes and Riedel, Sebastian and Stenetorp, Pontus},\n title = {Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension},\n journal = {Transactions of the Association for Computational Linguistics},\n volume = {8},\n number = {},\n pages = {662-678},\n year = {2020},\n doi = {10.1162/tacl_a_00338},\n URL = { https://doi.org/10.1162/tacl_a_00338 },\n eprint = { https://doi.org/10.1162/tacl_a_00338 },\n abstract = { Innovations in annotation methodology have been a catalyst for Reading Comprehension (RC) datasets and models. One recent trend to challenge current RC models is to involve a model in the annotation process: Humans create questions adversarially, such that the model fails to answer them correctly. In this work we investigate this annotation methodology and apply it in three different settings, collecting a total of 36,000 samples with progressively stronger models in the annotation loop. This allows us to explore questions such as the reproducibility of the adversarial effect, transfer from data collected with varying model-in-the-loop strengths, and generalization to data collected without a model. We find that training on adversarially collected samples leads to strong generalization to non-adversarially collected datasets, yet with progressive performance deterioration with increasingly stronger models-in-the-loop. Furthermore, we find that stronger models can still learn from datasets collected with substantially weaker models-in-the-loop. When trained on data collected with a BiDAF model in the loop, RoBERTa achieves 39.9F1 on questions that it cannot answer when trained on SQuAD\u2014only marginally lower than when trained on data collected using RoBERTa itself (41.0F1). }\n}\n\n@article{allenai:quoref,\n author = {Pradeep Dasigi and Nelson F. Liu and Ana Marasovic and Noah A. Smith and Matt Gardner},\n title = {Quoref: A Reading Comprehension Dataset with Questions Requiring Coreferential Reasoning},\n journal = {arXiv:1908.05803v2 },\n year = {2019},\n}\n\n@inproceedings{DuoRC,\nauthor = { Amrita Saha and Rahul Aralikatte and Mitesh M. Khapra and Karthik Sankaranarayanan},title = {{DuoRC: Towards Complex Language Understanding with Paraphrased Reading Comprehension}},\nbooktitle = {Meeting of the Association for Computational Linguistics (ACL)},\nyear = {2018}\n}\n\n@inproceedings{Lin2019ReasoningOP,\n title={Reasoning Over Paragraph Effects in Situations},\n author={Kevin Lin and Oyvind Tafjord and Peter Clark and Matt Gardner},\n booktitle={MRQA@EMNLP},\n year={2019}\n}\n\n@article{zhang2018record,\n title={Record: Bridging the gap between human and machine commonsense reading comprehension},\n author={Zhang, Sheng and Liu, Xiaodong and Liu, Jingjing and Gao, Jianfeng and Duh, Kevin and Van Durme, Benjamin},\n journal={arXiv preprint arXiv:1810.12885},\n year={2018}\n}\n@article{wang2019superglue,\n title={SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems},\n author={Wang, Alex and Pruksachatkun, Yada and Nangia, Nikita and Singh, Amanpreet and Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R},\n journal={arXiv preprint arXiv:1905.00537},\n year={2019}\n}\n\nNote that each SuperGLUE dataset has its own citation. Please see the source to\nget the correct citation for each contained dataset.\n\n@inproceedings{mcauley2013hidden,\n title={Hidden factors and hidden topics: understanding rating dimensions with review text},\n author={McAuley, Julian and Leskovec, Jure},\n booktitle={Proceedings of the 7th ACM conference on Recommender systems},\n pages={165--172},\n year={2013}\n}\n\n@InProceedings{Zurich Open Repository and\nArchive:dataset,\ntitle = {Software Applications User Reviews},\nauthors={Grano, Giovanni; Di Sorbo, Andrea; Mercaldo, Francesco; Visaggio, Corrado A; Canfora, Gerardo;\nPanichella, Sebastiano},\nyear={2017}\n}\n\n@InProceedings{maas-EtAl:2011:ACL-HLT2011,\n author = {Maas, Andrew L. and Daly, Raymond E. and Pham, Peter T. and Huang, Dan and Ng, Andrew Y. and Potts, Christopher},\n title = {Learning Word Vectors for Sentiment Analysis},\n booktitle = {Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies},\n month = {June},\n year = {2011},\n address = {Portland, Oregon, USA},\n publisher = {Association for Computational Linguistics},\n pages = {142--150},\n url = {http://www.aclweb.org/anthology/P11-1015}\n}\n\n@InProceedings{Pang+Lee:05a,\n author = {Bo Pang and Lillian Lee},\n title = {Seeing stars: Exploiting class relationships for sentiment\n categorization with respect to rating scales},\n booktitle = {Proceedings of the ACL},\n year = 2005\n}\n\n@inproceedings{zhang2015character,\n title={Character-level convolutional networks for text classification},\n author={Zhang, Xiang and Zhao, Junbo and LeCun, Yann},\n booktitle={Advances in neural information processing systems},\n pages={649--657},\n year={2015}\n}\n\n@article{DBLP:journals/corr/SeeLM17,\n author = {Abigail See and\n Peter J. Liu and\n Christopher D. Manning},\n title = {Get To The Point: Summarization with Pointer-Generator Networks},\n journal = {CoRR},\n volume = {abs/1704.04368},\n year = {2017},\n url = {http://arxiv.org/abs/1704.04368},\n archivePrefix = {arXiv},\n eprint = {1704.04368},\n timestamp = {Mon, 13 Aug 2018 16:46:08 +0200},\n biburl = {https://dblp.org/rec/bib/journals/corr/SeeLM17},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n@inproceedings{hermann2015teaching,\n title={Teaching machines to read and comprehend},\n author={Hermann, Karl Moritz and Kocisky, Tomas and Grefenstette, Edward and Espeholt, Lasse and Kay, Will and Suleyman, Mustafa and Blunsom, Phil},\n booktitle={Advances in neural information processing systems},\n pages={1693--1701},\n year={2015}\n}\n\n@article{graff2003english,\n title={English gigaword},\n author={Graff, David and Kong, Junbo and Chen, Ke and Maeda, Kazuaki},\n journal={Linguistic Data Consortium, Philadelphia},\n volume={4},\n number={1},\n pages={34},\n year={2003}\n}\n\n@article{Rush_2015,\n title={A Neural Attention Model for Abstractive Sentence Summarization},\n url={http://dx.doi.org/10.18653/v1/D15-1044},\n DOI={10.18653/v1/d15-1044},\n journal={Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing},\n publisher={Association for Computational Linguistics},\n author={Rush, Alexander M. and Chopra, Sumit and Weston, Jason},\n year={2015}\n}\n\n@article{gliwa2019samsum,\n title={SAMSum Corpus: A Human-annotated Dialogue Dataset for Abstractive Summarization},\n author={Gliwa, Bogdan and Mochol, Iwona and Biesek, Maciej and Wawer, Aleksander},\n journal={arXiv preprint arXiv:1911.12237},\n year={2019}\n}\n\n@article{Narayan2018DontGM,\n title={Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization},\n author={Shashi Narayan and Shay B. Cohen and Mirella Lapata},\n journal={ArXiv},\n year={2018},\n volume={abs/1808.08745}\n}\n\n@inproceedings{Zhang2015CharacterlevelCN,\n title={Character-level Convolutional Networks for Text Classification},\n author={Xiang Zhang and Junbo Jake Zhao and Yann LeCun},\n booktitle={NIPS},\n year={2015}\n}\n\n@article{lehmann2015dbpedia,\n title={DBpedia--a large-scale, multilingual knowledge base extracted from Wikipedia},\n author={Lehmann, Jens and Isele, Robert and Jakob, Max and Jentzsch, Anja and Kontokostas,\n Dimitris and Mendes, Pablo N and Hellmann, Sebastian and Morsey, Mohamed and Van Kleef,\n Patrick and Auer, S{\"o}ren and others},\n journal={Semantic web},\n volume={6},\n number={2},\n pages={167--195},\n year={2015},\n publisher={IOS Press}\n}\n\n@inproceedings{dolan2005automatically,\n title={Automatically constructing a corpus of sentential paraphrases},\n author={Dolan, William B and Brockett, Chris},\n booktitle={Proceedings of the Third International Workshop on Paraphrasing (IWP2005)},\n year={2005}\n}\n@inproceedings{wang2019glue,\n title={{GLUE}: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding},\n author={Wang, Alex and Singh, Amanpreet and Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R.},\n note={In the Proceedings of ICLR.},\n year={2019}\n}\n\n@inproceedings{levesque2012winograd,\n title={The winograd schema challenge},\n author={Levesque, Hector and Davis, Ernest and Morgenstern, Leora},\n booktitle={Thirteenth International Conference on the Principles of Knowledge Representation and Reasoning},\n year={2012}\n}\n@article{wang2019superglue,\n title={SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems},\n author={Wang, Alex and Pruksachatkun, Yada and Nangia, Nikita and Singh, Amanpreet and Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R},\n journal={arXiv preprint arXiv:1905.00537},\n year={2019}\n}\n\nNote that each SuperGLUE dataset has its own citation. Please see the source to\nget the correct citation for each contained dataset.\n\n@article{DBLP:journals/corr/abs-1808-09121,\n author={Mohammad Taher Pilehvar and os{'{e}} Camacho{-}Collados},\n title={WiC: 10, 000 Example Pairs for Evaluating Context-Sensitive Representations},\n journal={CoRR},\n volume={abs/1808.09121},\n year={2018},\n url={http://arxiv.org/abs/1808.09121},\n archivePrefix={arXiv},\n eprint={1808.09121},\n timestamp={Mon, 03 Sep 2018 13:36:40 +0200},\n biburl={https://dblp.org/rec/bib/journals/corr/abs-1808-09121},\n bibsource={dblp computer science bibliography, https://dblp.org}\n}\n@article{wang2019superglue,\n title={SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems},\n author={Wang, Alex and Pruksachatkun, Yada and Nangia, Nikita and Singh, Amanpreet and Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R},\n journal={arXiv preprint arXiv:1905.00537},\n year={2019}\n}\n\nNote that each SuperGLUE dataset has its own citation. Please see the source to\nget the correct citation for each contained dataset.\n\n@inproceedings{roemmele2011choice,\n title={Choice of plausible alternatives: An evaluation of commonsense causal reasoning},\n author={Roemmele, Melissa and Bejan, Cosmin Adrian and Gordon, Andrew S},\n booktitle={2011 AAAI Spring Symposium Series},\n year={2011}\n}\n@article{wang2019superglue,\n title={SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems},\n author={Wang, Alex and Pruksachatkun, Yada and Nangia, Nikita and Singh, Amanpreet and Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R},\n journal={arXiv preprint arXiv:1905.00537},\n year={2019}\n}\n\nNote that each SuperGLUE dataset has its own citation. Please see the source to\nget the correct citation for each contained dataset.\n\n@inproceedings{dagan2005pascal,\n title={The PASCAL recognising textual entailment challenge},\n author={Dagan, Ido and Glickman, Oren and Magnini, Bernardo},\n booktitle={Machine Learning Challenges Workshop},\n pages={177--190},\n year={2005},\n organization={Springer}\n}\n@inproceedings{bar2006second,\n title={The second pascal recognising textual entailment challenge},\n author={Bar-Haim, Roy and Dagan, Ido and Dolan, Bill and Ferro, Lisa and Giampiccolo, Danilo and Magnini, Bernardo and Szpektor, Idan},\n booktitle={Proceedings of the second PASCAL challenges workshop on recognising textual entailment},\n volume={6},\n number={1},\n pages={6--4},\n year={2006},\n organization={Venice}\n}\n@inproceedings{giampiccolo2007third,\n title={The third pascal recognizing textual entailment challenge},\n author={Giampiccolo, Danilo and Magnini, Bernardo and Dagan, Ido and Dolan, Bill},\n booktitle={Proceedings of the ACL-PASCAL workshop on textual entailment and paraphrasing},\n pages={1--9},\n year={2007},\n organization={Association for Computational Linguistics}\n}\n@inproceedings{bentivogli2009fifth,\n title={The Fifth PASCAL Recognizing Textual Entailment Challenge.},\n author={Bentivogli, Luisa and Clark, Peter and Dagan, Ido and Giampiccolo, Danilo},\n booktitle={TAC},\n year={2009}\n}\n@article{wang2019superglue,\n title={SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems},\n author={Wang, Alex and Pruksachatkun, Yada and Nangia, Nikita and Singh, Amanpreet and Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R},\n journal={arXiv preprint arXiv:1905.00537},\n year={2019}\n}\n\nNote that each SuperGLUE dataset has its own citation. Please see the source to\nget the correct citation for each contained dataset.\n\n@article{de marneff_simons_tonhauser_2019,\n title={The CommitmentBank: Investigating projection in naturally occurring discourse},\n journal={proceedings of Sinn und Bedeutung 23},\n author={De Marneff, Marie-Catherine and Simons, Mandy and Tonhauser, Judith},\n year={2019}\n}\n@article{wang2019superglue,\n title={SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems},\n author={Wang, Alex and Pruksachatkun, Yada and Nangia, Nikita and Singh, Amanpreet and Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R},\n journal={arXiv preprint arXiv:1905.00537},\n year={2019}\n}\n\nNote that each SuperGLUE dataset has its own citation. Please see the source to\nget the correct citation for each contained dataset.\n\n@InProceedings{nie2019adversarial,\n title={Adversarial NLI: A New Benchmark for Natural Language Understanding},\n author={Nie, Yixin\n and Williams, Adina\n and Dinan, Emily\n and Bansal, Mohit\n and Weston, Jason\n and Kiela, Douwe},\n booktitle = \"Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics\",\n year = \"2020\",\n publisher = \"Association for Computational Linguistics\",\n}\n\n@InProceedings{quartz,\n author = {Oyvind Tafjord and Matt Gardner and Kevin Lin and Peter Clark},\n title = {\"QUARTZ: An Open-Domain Dataset of Qualitative Relationship\nQuestions\"},\n year = {\"2019\"},\n}", |
|
"description": "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\n\nWe 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\u2019s 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\u2019 abilities to reason about the social implications of everyday events and situations. (Less)\n\nPAWS: Paraphrase Adversaries from Word Scrambling\n\nThis dataset contains 108,463 human-labeled and 656k noisily labeled pairs that feature\nthe importance of modeling structure, context, and word order information for the problem\nof paraphrase identification. The dataset has two subsets, one based on Wikipedia and the\nother one based on the Quora Question Pairs (QQP) dataset.\n\nFor further details, see the accompanying paper: PAWS: Paraphrase Adversaries from Word Scrambling\n(https://arxiv.org/abs/1904.01130)\n\nPAWS-QQP is not available due to license of QQP. It must be reconstructed by downloading the original\ndata and then running our scripts to produce the data and attach the labels.\n\nNOTE: There might be some missing or wrong labels in the dataset and we have replaced them with -1.\n\nQuAIL is a reading comprehension dataset. QuAIL contains 15K multi-choice questions in texts 300-350 tokens long 4 domains (news, user stories, fiction, blogs).QuAIL is balanced and annotated for question types.\n\nStanford 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.\n\nDREAM is a multiple-choice Dialogue-based REAding comprehension exaMination dataset. In contrast to existing reading comprehension datasets, DREAM is the first to focus on in-depth multi-turn multi-party dialogue understanding.\n\nQASC is a question-answering dataset with a focus on sentence composition. It consists of 9,980 8-way multiple-choice\nquestions about grade school science (8,134 train, 926 dev, 920 test), and comes with a corpus of 17M sentences.\n\nWikiHop is open-domain and based on Wikipedia articles; the goal is to recover Wikidata information by hopping through documents. The goal is to answer text understanding queries by combining multiple facts that are spread across different documents.\n\nRace is a large-scale reading comprehension dataset with more than 28,000 passages and nearly 100,000 questions. The\n dataset is collected from English examinations in China, which are designed for middle school and high school students.\nThe dataset can be served as the training and test sets for machine comprehension.\n\nHellaSwag: Can a Machine Really Finish Your Sentence? is a new dataset for commonsense NLI. A paper was published at ACL2019.\n\nSuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after\nGLUE with a new set of more difficult language understanding tasks, improved\nresources, and a new public leaderboard.\n\nBoolQ (Boolean Questions, Clark et al., 2019a) is a QA task where each example consists of a short\npassage and a yes/no question about the passage. The questions are provided anonymously and\nunsolicited by users of the Google search engine, and afterwards paired with a paragraph from a\nWikipedia article containing the answer. Following the original work, we evaluate with accuracy.\n\nAdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.\nWe use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.\nThe adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.\n\nQuoref is a QA dataset which tests the coreferential reasoning capability of reading comprehension systems. In this\nspan-selection benchmark containing 24K questions over 4.7K paragraphs from Wikipedia, a system must resolve hard\ncoreferences before selecting the appropriate span(s) in the paragraphs for answering questions.\n\nDuoRC contains 186,089 unique question-answer pairs created from a collection of 7680 pairs of movie plots where each pair in the collection reflects two versions of the same movie.\n\nROPES (Reasoning Over Paragraph Effects in Situations) is a QA dataset\nwhich tests a system's ability to apply knowledge from a passage\nof text to a new situation. A system is presented a background\npassage containing a causal or qualitative relation(s) (e.g.,\n\"animal pollinators increase efficiency of fertilization in flowers\"),\na novel situation that uses this background, and questions that require\nreasoning about effects of the relationships in the background\npassage in the background of the situation.\n\nSuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after\nGLUE with a new set of more difficult language understanding tasks, improved\nresources, and a new public leaderboard.\n\n(Reading Comprehension with Commonsense Reasoning Dataset, Zhang et al., 2018) is a\nmultiple-choice QA task. Each example consists of a news article and a Cloze-style question about\nthe article in which one entity is masked out. The system must predict the masked out entity from a\ngiven list of possible entities in the provided passage, where the same entity may be expressed using\nmultiple different surface forms, all of which are considered correct. Articles are drawn from CNN\nand Daily Mail. Following the original work, we evaluate with max (over all mentions) token-level\nF1 and exact match (EM).\n\nThe Amazon reviews dataset consists of reviews from amazon.\nThe data span a period of 18 years, including ~35 million reviews up to March 2013.\nReviews include product and user information, ratings, and a plaintext review.\n\nIt is a large dataset of Android applications belonging to 23 differentapps categories, which provides an overview of the types of feedback users report on the apps and documents the evolution of the related code metrics. The dataset contains about 395 applications of the F-Droid repository, including around 600 versions, 280,000 user reviews (extracted with specific text mining approaches)\n\nLarge Movie Review Dataset.\nThis 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.\n\nMovie Review Dataset.\nThis is a dataset of containing 5,331 positive and 5,331 negative processed\nsentences from Rotten Tomatoes movie reviews. This data was first used in Bo\nPang and Lillian Lee, ``Seeing stars: Exploiting class relationships for\nsentiment categorization with respect to rating scales.'', Proceedings of the\nACL, 2005.\n\nThe Yelp reviews dataset consists of reviews from Yelp. It is extracted from the Yelp Dataset Challenge 2015 data.\nThe Yelp reviews full star dataset is constructed by Xiang Zhang ([email protected]) from the above dataset.\nIt is first used as a text classification benchmark in the following paper: Xiang Zhang, Junbo Zhao, Yann LeCun.\nCharacter-level Convolutional Networks for Text Classification. Advances in Neural Information Processing Systems 28 (NIPS 2015).\n\nCNN/DailyMail non-anonymized summarization dataset.\n\nThere are two features:\n - article: text of news article, used as the document to be summarized\n - highlights: joined text of highlights with <s> and </s> around each\n highlight, which is the target summary\n\nHeadline-generation on a corpus of article pairs from Gigaword consisting of\naround 4 million articles. Use the 'org_data' provided by\nhttps://github.com/microsoft/unilm/ which is identical to\nhttps://github.com/harvardnlp/sent-summary but with better format.\n\nThere are two features:\n - document: article.\n - summary: headline.\n\nSAMSum Corpus contains over 16k chat dialogues with manually annotated\nsummaries.\nThere are two features:\n - dialogue: text of dialogue.\n - summary: human written summary of the dialogue.\n - id: id of a example.\n\nExtreme Summarization (XSum) Dataset.\n\nThere are three features:\n - document: Input news article.\n - summary: One sentence summary of the article.\n - id: BBC ID of the article.\n\nAG is a collection of more than 1 million news articles. News articles have been\ngathered from more than 2000 news sources by ComeToMyHead in more than 1 year of\nactivity. ComeToMyHead is an academic news search engine which has been running\nsince July, 2004. The dataset is provided by the academic comunity for research\npurposes in data mining (clustering, classification, etc), information retrieval\n(ranking, search, etc), xml, data compression, data streaming, and any other\nnon-commercial activity. For more information, please refer to the link\nhttp://www.di.unipi.it/~gulli/AG_corpus_of_news_articles.html .\n\nThe AG's news topic classification dataset is constructed by Xiang Zhang\n([email protected]) from the dataset above. It is used as a text\nclassification benchmark in the following paper: Xiang Zhang, Junbo Zhao, Yann\nLeCun. Character-level Convolutional Networks for Text Classification. Advances\nin Neural Information Processing Systems 28 (NIPS 2015).\n\nThe DBpedia ontology classification dataset is constructed by picking 14 non-overlapping classes\nfrom DBpedia 2014. They are listed in classes.txt. From each of thse 14 ontology classes, we\nrandomly choose 40,000 training samples and 5,000 testing samples. Therefore, the total size\nof the training dataset is 560,000 and testing dataset 70,000.\nThere are 3 columns in the dataset (same for train and test splits), corresponding to class index\n(1 to 14), title and content. The title and content are escaped using double quotes (\"), and any\ninternal double quote is escaped by 2 double quotes (\"\"). There are no new lines in title or content.\n\nGLUE, the General Language Understanding Evaluation benchmark\n(https://gluebenchmark.com/) is a collection of resources for training,\nevaluating, and analyzing natural language understanding systems.\n\nSuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after\nGLUE with a new set of more difficult language understanding tasks, improved\nresources, and a new public leaderboard.\n\nThe Winograd Schema Challenge (WSC, Levesque et al., 2012) is a reading comprehension\ntask in which a system must read a sentence with a pronoun and select the referent of that pronoun\nfrom a list of choices. Given the difficulty of this task and the headroom still left, we have included\nWSC in SuperGLUE and recast the dataset into its coreference form. The task is cast as a binary\nclassification problem, as opposed to N-multiple choice, in order to isolate the model's ability to\nunderstand the coreference links within a sentence as opposed to various other strategies that may\ncome into play in multiple choice conditions. With that in mind, we create a split with 65% negative\nmajority class in the validation set, reflecting the distribution of the hidden test set, and 52% negative\nclass in the training set. The training and validation examples are drawn from the original Winograd\nSchema dataset (Levesque et al., 2012), as well as those distributed by the affiliated organization\nCommonsense Reasoning. The test examples are derived from fiction books and have been shared\nwith us by the authors of the original dataset. Previously, a version of WSC recast as NLI as included\nin GLUE, known as WNLI. No substantial progress was made on WNLI, with many submissions\nopting to submit only majority class predictions. WNLI was made especially difficult due to an\nadversarial train/dev split: Premise sentences that appeared in the training set sometimes appeared\nin the development set with a different hypothesis and a flipped label. If a system memorized the\ntraining set without meaningfully generalizing, which was easy due to the small size of the training\nset, it could perform far below chance on the development set. We remove this adversarial design\nin the SuperGLUE version of WSC by ensuring that no sentences are shared between the training,\nvalidation, and test sets.\n\nHowever, the validation and test sets come from different domains, with the validation set consisting\nof ambiguous examples such that changing one non-noun phrase word will change the coreference\ndependencies in the sentence. The test set consists only of more straightforward examples, with a\nhigh number of noun phrases (and thus more choices for the model), but low to no ambiguity.\n\nThis version fixes issues where the spans are not actually substrings of the text.\n\nSuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after\nGLUE with a new set of more difficult language understanding tasks, improved\nresources, and a new public leaderboard.\n\nThe Word-in-Context (WiC, Pilehvar and Camacho-Collados, 2019) dataset supports a word\nsense disambiguation task cast as binary classification over sentence pairs. Given two sentences and a\npolysemous (sense-ambiguous) word that appears in both sentences, the task is to determine whether\nthe word is used with the same sense in both sentences. Sentences are drawn from WordNet (Miller,\n1995), VerbNet (Schuler, 2005), and Wiktionary. We follow the original work and evaluate using\naccuracy.\n\nSuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after\nGLUE with a new set of more difficult language understanding tasks, improved\nresources, and a new public leaderboard.\n\nThe Choice Of Plausible Alternatives (COPA, Roemmele et al., 2011) dataset is a causal\nreasoning task in which a system is given a premise sentence and two possible alternatives. The\nsystem must choose the alternative which has the more plausible causal relationship with the premise.\nThe method used for the construction of the alternatives ensures that the task requires causal reasoning\nto solve. Examples either deal with alternative possible causes or alternative possible effects of the\npremise sentence, accompanied by a simple question disambiguating between the two instance\ntypes for the model. All examples are handcrafted and focus on topics from online blogs and a\nphotography-related encyclopedia. Following the recommendation of the authors, we evaluate using\naccuracy.\n\nSuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after\nGLUE with a new set of more difficult language understanding tasks, improved\nresources, and a new public leaderboard.\n\nThe Recognizing Textual Entailment (RTE) datasets come from a series of annual competitions\non textual entailment, the problem of predicting whether a given premise sentence entails a given\nhypothesis sentence (also known as natural language inference, NLI). RTE was previously included\nin GLUE, and we use the same data and format as before: We merge data from RTE1 (Dagan\net al., 2006), RTE2 (Bar Haim et al., 2006), RTE3 (Giampiccolo et al., 2007), and RTE5 (Bentivogli\net al., 2009). All datasets are combined and converted to two-class classification: entailment and\nnot_entailment. Of all the GLUE tasks, RTE was among those that benefited from transfer learning\nthe most, jumping from near random-chance performance (~56%) at the time of GLUE's launch to\n85% accuracy (Liu et al., 2019c) at the time of writing. Given the eight point gap with respect to\nhuman performance, however, the task is not yet solved by machines, and we expect the remaining\ngap to be difficult to close.\n\nSuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after\nGLUE with a new set of more difficult language understanding tasks, improved\nresources, and a new public leaderboard.\n\nThe CommitmentBank (De Marneffe et al., 2019) is a corpus of short texts in which at least\none sentence contains an embedded clause. Each of these embedded clauses is annotated with the\ndegree to which we expect that the person who wrote the text is committed to the truth of the clause.\nThe resulting task framed as three-class textual entailment on examples that are drawn from the Wall\nStreet Journal, fiction from the British National Corpus, and Switchboard. Each example consists\nof a premise containing an embedded clause and the corresponding hypothesis is the extraction of\nthat clause. We use a subset of the data that had inter-annotator agreement above 0.85. The data is\nimbalanced (relatively fewer neutral examples), so we evaluate using accuracy and F1, where for\nmulti-class F1 we compute the unweighted average of the F1 per class.\n\nThe Adversarial Natural Language Inference (ANLI) is a new large-scale NLI benchmark dataset,\nThe dataset is collected via an iterative, adversarial human-and-model-in-the-loop procedure.\nANLI is much more difficult than its predecessors including SNLI and MNLI.\nIt contains three rounds. Each round has train/dev/test splits.\n\nQuaRTz is a crowdsourced dataset of 3864 multiple-choice questions about open domain qualitative relationships. Each\nquestion is paired with one of 405 different background sentences (sometimes short paragraphs).\nThe QuaRTz dataset V1 contains 3864 questions about open domain qualitative relationships. Each question is paired with\none of 405 different background sentences (sometimes short paragraphs).\nThe dataset is split into train (2696), dev (384) and test (784). A background sentence will only appear in a single split.", |
