Datasets
Viewing 1-10 of 54 datasets
StrategyQA
2,780 implicit multi-hop reasoning questionsAI2 Israel, Question Understanding, Aristo • 2021StrategyQA is a question-answering benchmark focusing on open-domain questions where the required reasoning steps are implicit in the question and should be inferred using a strategy. StrategyQA includes 2,780 examples, each consisting of a strategy question, its decomposition, and evidence paragraphs.ProofWriter
Updated RuleTaker datasets with 500k questions, answers and proofs over rulebases.Aristo • 2020These datasets accompany the paper "ProofWriter: Generating Implications, Proofs, and Abductive Statements over Natural Language". They contain updated RuleTaker-style datasets with 500k questions, answers and proofs over natural-language rulebases, used to show that Transformers can emulate reasoning over rules expressed in language, including proof generation. It includes variants using closed- and open-world semantics. Proofs include intermediate conclusions. Extra annotations provide data to train the iterative ProofWriter model as well as abductive reasoning to make uncertain statements certain.ARC Direct Answer Questions
A dataset of 2,985 grade-school level, direct-answer science questions derived from the ARC multiple-choice question set.2020A dataset of 2,985 grade-school level, direct-answer science questions derived from the ARC multiple-choice question set released as part of the AI2 Reasoning Challenge in 2018.RuleTaker: Transformers as Soft Reasoners over Language
Datasets used to teach transformers to reasonAristo • 2020Can transformers be trained to reason (or emulate reasoning) over rules expressed in language? In the associated paper and demo we provide evidence that they can. Our models, that we call RuleTakers, are trained on datasets of synthetic rule bases plus derived conclusions, provided here. The resulting models provide the first demonstration that this kind of soft reasoning over language is indeed learnable.ZEST: ZEroShot learning from Task descriptions
ZEST is a benchmark for zero-shot generalization to unseen NLP tasks, with 25K labeled instances across 1,251 different tasks.AI2 Irvine, Mosaic, AllenNLP • 2020ZEST tests whether NLP systems can perform unseen tasks in a zero-shot way, given a natural language description of the task. It is an instantiation of our proposed framework "learning from task descriptions". The tasks include classification, typed entity extraction and relationship extraction, and each task is paired with 20 different annotated (input, output) examples. ZEST's structure allows us to systematically test whether models can generalize in five different ways.Open PI
33K state changes over 4,050 sentences from 810 procedural, real-world paragraphsAristo, Mosaic • 2020Open PI is the first dataset for tracking state changes in procedural text from arbitrary domains by using an unrestricted (open) vocabulary. Our solution is a new task formulation in which just the text is provided, from which a set of state changes (entity, attribute, before, after) is generated for each step, where the entity, attribute, and values must all be predicted from an open vocabulary.Real Toxicity Prompts
A dataset of 100k sentence snippets from the web for researchers to further address the risk of neural toxic degeneration in models.Mosaic • 2020A dataset of 100k sentence snippets from the web for researchers to further address the risk of neural toxic degeneration in models.eQASC: Multihop Explanations for QASC
98k annotated explanations for the QASC datasetAristo • 2020This dataset contains 98k 2-hop explanations for questions in the QASC dataset, with annotations indicating if they are valid (~25k) or invalid (~73k) explanations.hasPart KB
A high-quality KB of hasPart relationsAristo • 2020A high-quality knowledge base of ~50k hasPart relationships, extracted from a large corpus of generic statements.SciDocs
Academic paper representation dataset accompanying the SPECTER paper/modelSemantic Scholar • 2020Representation learning is a critical ingredient for natural language processing systems. Recent Transformer language models like BERT learn powerful textual representations, but these models are targeted towards token- and sentence-level training objectives and do not leverage information on inter-document relatedness, which limits their document-level representation power. For applications on scientific documents, such as classification and recommendation, the embeddings power strong performance on end tasks. We propose SPECTER, a new method to generate document-level embedding of scientific documents based on pretraining a Transformer language model on a powerful signal of document-level relatedness: the citation graph. Unlike existing pretrained language models, SPECTER can be easily applied to downstream applications without task-specific fine-tuning. Additionally, to encourage further research on document-level models, we introduce SciDocs, a new evaluation benchmark consisting of seven document-level tasks ranging from citation prediction, to document classification and recommendation.