{"url":"/dataset/cannot","name":"CANNOT","full_name":"Compilation of ANnotated, Negation-Oriented Text-pairs","description_markdown":"## Dataset Summary\r\n\r\n**CANNOT** is a dataset that focuses on negated textual pairs. It currently\r\ncontains **77,376 samples**, of which roughly of them are negated pairs of\r\nsentences, and the other half are not (they are paraphrased versions of each\r\nother).\r\n\r\nThe most frequent negation that appears in the dataset is verbal negation (e.g.,\r\nwill → won't), although it also contains pairs with antonyms (cold → hot).\r\n\r\n<br>\r\n\r\n## Languages\r\nCANNOT includes exclusively texts in **English**.\r\n\r\n<br>\r\n\r\n## Dataset Structure\r\n\r\nThe dataset is given as a\r\n[`.tsv`](https://en.wikipedia.org/wiki/Tab-separated_values) file with the\r\nfollowing structure:\r\n\r\n| premise     | hypothesis                                         | label |\r\n|:------------|:---------------------------------------------------|:-----:|\r\n| A sentence. | An equivalent, non-negated sentence (paraphrased). | 0     |\r\n| A sentence. | The sentence negated.                              | 1     |\r\n\r\n\r\nThe dataset can be easily loaded into a Pandas DataFrame by running:\r\n\r\n```Python\r\nimport pandas as pd\r\n\r\ndataset = pd.read_csv('negation_dataset_v1.0.tsv', sep='\\t')\r\n\r\n```\r\n\r\n<br>\r\n\r\n## Dataset Creation\r\n\r\nThe dataset has been created by cleaning up and merging the following datasets:\r\n\r\n1. _Not another Negation Benchmark: The NaN-NLI Test Suite for Sub-clausal\r\n    Negation_ (see\r\n[`datasets/nan-nli`](https://github.com/dmlls/cannot-dataset/tree/main/datasets/nan-nli)).\r\n\r\n2. _GLUE Diagnostic Dataset_ (see\r\n[`datasets/glue-diagnostic`](https://github.com/dmlls/cannot-dataset/tree/main/datasets/glue-diagnostic)).\r\n\r\n3. _Automated Fact-Checking of Claims from Wikipedia_ (see\r\n[`datasets/wikifactcheck-english`](https://github.com/dmlls/cannot-dataset/tree/main/datasets/wikifactcheck-english)).\r\n\r\n4. _From Group to Individual Labels Using Deep Features_ (see\r\n[`datasets/sentiment-labelled-sentences`](https://github.com/dmlls/cannot-dataset/tree/main/datasets/sentiment-labelled-sentences)).\r\nIn this case, the negated sentences were obtained by using the Python module\r\n[`negate`](https://github.com/dmlls/negate).\r\n\r\n5. _It Is Not Easy To Detect Paraphrases: Analysing Semantic Similarity With\r\nAntonyms and Negation Using the New SemAntoNeg Benchmark_ (see\r\n[`datasets/antonym-substitution`](https://github.com/dmlls/cannot-dataset/tree/main/datasets/antonym-substitution)).\r\n\r\n\r\nOnce processed, the number of remaining samples in each of the datasets above are:\r\n\r\n| Dataset                                                                   | Samples    |\r\n|:--------------------------------------------------------------------------|-----------:|\r\n| Not another Negation Benchmark                                            |      118   |\r\n| GLUE Diagnostic Dataset                                                   |      154   |\r\n| Automated Fact-Checking of Claims from Wikipedia                          |   14,970   |\r\n| From Group to Individual Labels Using Deep Features                       |    2,110   |\r\n| It Is Not Easy To Detect Paraphrases                                      |    8,597   |\r\n| <div align=\"right\"><b>Total</b></div>                                     | **25,949** |\r\n\r\n\r\nAdditionally, for each of the negated samples, another pair of non-negated\r\nsentences has been added by paraphrasing them with the pre-trained model\r\n[`🤗tuner007/pegasus_paraphrase`](https://huggingface.co/tuner007/pegasus_paraphrase).\r\n\r\nFinally, the swapped version of each pair (premise ⇋ hypothesis) has also been\r\nincluded, and any duplicates have been removed.\r\n\r\nWith this, the number of premises/hypothesis in the CANNOT dataset that appear\r\nin the original datasets are:\r\n\r\n| <div align=\"left\"><b>Dataset</b></div>                                                                   | <div align=\"center\"><b>Sentences</b></div>             |\r\n|:--------------------------------------------------------------------------|----------------------:|\r\n| Not another Negation Benchmark                                            |         552 &nbsp;&nbsp;&nbsp; (0.36 %) |\r\n| GLUE Diagnostic Dataset                                                   |         586 &nbsp;&nbsp;&nbsp; (0.38 %) |\r\n| Automated Fact-Checking of Claims from Wikipedia                          |      89,728 &nbsp; (59.98 %) |\r\n| From Group to Individual Labels Using Deep Features                       |      12,626 &nbsp;&nbsp;&nbsp; (8.16 %) |\r\n| It Is Not Easy To Detect Paraphrases                                      |      17,198 &nbsp; (11.11 %) |\r\n| <div align=\"right\"><b>Total</b></div>                                     | **120,690** &nbsp; (77.99 %) |\r\n\r\nThe percentages above are in relation to the total number of premises and\r\nhypothesis in the CANNOT dataset. The remaining 22.01 % (34,062 sentences) are\r\nthe novel premises/hypothesis added through paraphrase and rule-based negation.\r\n\r\n<br>\r\n\r\n## Additional Information\r\n\r\n<br>\r\n\r\n### Licensing Information\r\n\r\nThe CANNOT dataset is released under [CC BY-SA\r\n4.0](https://creativecommons.org/licenses/by-sa/4.0/).\r\n\r\n<a rel=\"license\" href=\"http://creativecommons.org/licenses/by-sa/4.0/\">\r\n    <img alt=\"Creative Commons License\" width=\"100px\" src=\"https://i.creativecommons.org/l/by-sa/4.0/88x31.png\"/>\r\n</a>\r\n\r\n<br>\r\n\r\n### Citation \r\nPlease cite our [INLG 2023 paper](https://arxiv.org/abs/2307.13989), if you use our dataset. \r\n**BibTeX:**\r\n```bibtex\r\n@misc{anschütz2023correct,\r\n      title={This is not correct! Negation-aware Evaluation of Language Generation Systems}, \r\n      author={Miriam Anschütz and Diego Miguel Lozano and Georg Groh},\r\n      year={2023},\r\n      eprint={2307.13989},\r\n      archivePrefix={arXiv},\r\n      primaryClass={cs.CL}\r\n}\r\n```\r\n\r\n<br>\r\n\r\n### Contributions\r\n\r\nContributions to the dataset can be submitted through the [project\r\nrepository](https://github.com/dmlls/cannot-dataset).","description_withheld":null,"homepage":"https://github.com/dmlls/cannot-dataset","introduced_date":"2023-07-18","introduced_date_note":null,"introduced_by":{"paper":"/paper/this-is-not-correct-negation-aware-evaluation","title":"This is not correct! Negation-aware Evaluation of Language Generation Systems","first_author":"Miriam Anschütz","url":null},"license":{"name":"CC BY-SA 4.0","url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode"},"modalities":[{"name":"Texts","url":"/datasets/modality/texts"}],"tasks":[{"name":"Text Generation","url":"/task/text-generation","datasets_with_task":"/datasets/task/text-generation"},{"name":"Embeddings Evaluation","url":"/task/embeddings-evaluation","datasets_with_task":"/datasets/task/embeddings-evaluation"}],"languages":[{"name":"English","url":"/datasets/language/english"}],"variants":["CANNOT"],"data_loaders":[],"num_papers_in_archive":3,"source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28"},"benchmarks":[],"papers_with_a_benchmark_row":[],"syntology_totals":{"read_at":"2026-09-24T18:15:14+00:00","papers_with_samples":0,"samples_harvested":0,"samples_ran":0,"samples_unverified":0,"pointer_only_for_licence":0,"papers_with_no_sample_that_ran":0,"note":"the per-paper counts above, summed; not a rate"},"papers_note":"The archive never published its papers-using-dataset list; these are papers with a leaderboard row on this dataset's benchmarks."}