{ "overview": { "where": { "has-leaderboard": "yes", "leaderboard-url": "[Leaderboard](https://github.com/Yale-LILY/dart#leaderboard)", "leaderboard-description": "Several state-of-the-art table-to-text models were evaluated on DART, such as BART ([Lewis et al., 2020](https://arxiv.org/pdf/1910.13461.pdf)), Seq2Seq-Att ([MELBOURNE](https://webnlg-challenge.loria.fr/files/melbourne_report.pdf)) and End-to-End Transformer ([Castro Ferreira et al., 2019](https://arxiv.org/pdf/1908.09022.pdf)).\nThe leaderboard reports BLEU, METEOR, TER, MoverScore, BERTScore and BLEURT scores.", "website": "n/a", "data-url": "[Github](https://github.com/Yale-LILY/dart)", "paper-url": "[ACL Anthology](https://aclanthology.org/2021.naacl-main.37/)", "contact-email": "{dragomir.radev, r.zhang}@yale.edu, {nazneen.rajani}@salesforce.com", "contact-name": "Dragomir Radev, Rui Zhang, Nazneen Rajani", "paper-bibtext": "```\n@inproceedings{nan-etal-2021-dart,\n title = \"{DART}: Open-Domain Structured Data Record to Text Generation\",\n author = \"Nan, Linyong and\n Radev, Dragomir and\n Zhang, Rui and\n Rau, Amrit and\n Sivaprasad, Abhinand and\n Hsieh, Chiachun and\n Tang, Xiangru and\n Vyas, Aadit and\n Verma, Neha and\n Krishna, Pranav and\n Liu, Yangxiaokang and\n Irwanto, Nadia and\n Pan, Jessica and\n Rahman, Faiaz and\n Zaidi, Ahmad and\n Mutuma, Mutethia and\n Tarabar, Yasin and\n Gupta, Ankit and\n Yu, Tao and\n Tan, Yi Chern and\n Lin, Xi Victoria and\n Xiong, Caiming and\n Socher, Richard and\n Rajani, Nazneen Fatema\",\n booktitle = \"Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies\",\n month = jun,\n year = \"2021\",\n address = \"Online\",\n publisher = \"Association for Computational Linguistics\",\n url = \"https://aclanthology.org/2021.naacl-main.37\",\n doi = \"10.18653/v1/2021.naacl-main.37\",\n pages = \"432--447\",\n abstract = \"We present DART, an open domain structured DAta Record to Text generation dataset with over 82k instances (DARTs). Data-to-text annotations can be a costly process, especially when dealing with tables which are the major source of structured data and contain nontrivial structures. To this end, we propose a procedure of extracting semantic triples from tables that encodes their structures by exploiting the semantic dependencies among table headers and the table title. Our dataset construction framework effectively merged heterogeneous sources from open domain semantic parsing and spoken dialogue systems by utilizing techniques including tree ontology annotation, question-answer pair to declarative sentence conversion, and predicate unification, all with minimum post-editing. We present systematic evaluation on DART as well as new state-of-the-art results on WebNLG 2017 to show that DART (1) poses new challenges to existing data-to-text datasets and (2) facilitates out-of-domain generalization. Our data and code can be found at https://github.com/Yale-LILY/dart.\",\n}\n```" }, "languages": { "is-multilingual": "no", "license": "mit: MIT License", "task-other": "N/A", "language-names": [ "English" ], "language-dialects": "It is an aggregated from multiple other datasets that use general US-American or British English without differentiation between dialects. ", "license-other": "N/A", "task": "Data-to-Text", "communicative": "The speaker is required to produce coherent sentences and construct a trees structured ontology of the column headers.\n\n", "language-speakers": "The dataset is aggregated from multiple others that were crowdsourced on different platforms. ", "intended-use": "The dataset is aimed to further research in natural language generation from semantic data." }, "credit": { "organization-type": [ "academic", "industry" ], "organization-names": "Yale University, Salesforce Research, Penn State University, The University of Hong Kong, MIT", "creators": "Linyong Nan, Dragomir Radev, Rui Zhang, Amrit Rau, Abhinand Sivaprasad, Chiachun Hsieh, Xiangru Tang, Aadit Vyas, Neha Verma, Pranav Krishna, Yangxiaokang Liu, Nadia Irwanto, Jessica Pan, Faiaz Rahman, Ahmad Zaidi, Mutethia Mutuma, Yasin Tarabar, Ankit Gupta, Tao Yu, Yi Chern Tan, Xi Victoria Lin, Caiming Xiong, Richard Socher, Nazneen Fatema Rajani", "funding": "n/a", "gem-added-by": "Miruna Clinciu contributed the original data card and Yacine Jernite wrote the initial data loader. Sebastian Gehrmann migrated the data card and the loader to the new format." }, "structure": { "data-fields": "-`tripleset`: a list of tuples, each tuple has 3 items\n-`subtree_was_extended`: a boolean variable (true or false)\n-`annotations`: a list of dict, each with source and text keys.\n-`source`: a string mentioning the name of the source table.\n-`text`: a sentence string.\n", "structure-example": "```\n {\n \"tripleset\": [\n [\n \"Ben Mauk\",\n \"High school\",\n \"Kenton\"\n ],\n [\n \"Ben Mauk\",\n \"College\",\n \"Wake Forest Cincinnati\"\n ]\n ],\n \"subtree_was_extended\": false,\n \"annotations\": [\n {\n \"source\": \"WikiTableQuestions_lily\",\n \"text\": \"Ben Mauk, who attended Kenton High School, attended Wake Forest Cincinnati for college.\"\n }\n ]\n }\n```", "structure-splits": "|Input Unit | Examples | Vocab Size | Words per SR | Sents per SR | Tables |\n| ------------- | ------------- || ------------- || ------------- || ------------- || ------------- |\n|Triple Set | 82,191 | 33.2K | 21.6 | 1.5 | 5,623 |\n\n| Train | Dev | Test|\n| ------------- | ------------- || ------------- |\n| 62,659 | 6,980 | 12,552|\n\n\nStatistics of DART decomposed by different collection methods. DART exhibits a great deal of topical variety in terms of the number of unique predicates, the number of unique triples, and the vocabulary size. These statistics are computed from DART v1.1.1; the number of unique predicates reported is post-unification (see Section 3.4). SR: Surface Realization.\n([details in Table 1 and 2](https://arxiv.org/pdf/2007.02871.pdf)).\n", "structure-description": "The structure is supposed to be able more complex structures beyond \"flat\" attribute-value pairs, instead encoding hierarchical relationships.", "structure-labels": "They are a combination of those from existing datasets and new annotations that take advantage of the hierarchical structure", "structure-splits-criteria": "For WebNLG 2017 and Cleaned E2E, DART use the original data splits. For the new annotation on WikiTableQuestions and WikiSQL, random splitting will make train, dev, and test splits contain similar tables and similar examples. They are thus split based on Jaccard similarity such that no training examples has a similarity with a test example of over 0.5", "structure-outlier": "n/a" }, "what": { "dataset": "DART is an English dataset aggregating multiple other data-to-text dataset in a common triple-based format. The new format is completely flat, thus not requiring a model to learn hierarchical structures, while still retaining the full information. " } }, "curation": { "original": { "is-aggregated": "yes", "aggregated-sources": "- human annotation on open-domain Wikipedia tables from WikiTableQuestions ([Pasupat and Liang,\n2015](https://www.aclweb.org/anthology/P15-1142.pdf)) and WikiSQL ([Zhong et al., 2017](https://arxiv.org/pdf/1709.00103.pdf))\n- automatic conversion of questions in WikiSQL to declarative sentences\n- incorporation of existing datasets including WebNLG 2017 (Gardent et al., 2017[a](https://www.aclweb.org/anthology/P17-1017.pdf),[b](https://www.aclweb.org/anthology/W17-3518.pdf); [Shimorina and Gardent, 2018](https://www.aclweb.org/anthology/W18-6543.pdf)) and Cleaned E2E ([Novikova et al., 2017b](https://arxiv.org/pdf/1706.09254.pdf); Du\u0161ek et al., [2018](https://arxiv.org/pdf/1810.01170.pdf), [2019](https://www.aclweb.org/anthology/W19-8652.pdf))\n", "rationale": "The dataset creators encourage through DART further research in natural language generation from semantic data. DART provides high-quality sentence annotations with each input being a set of entity-relation triples in a tree structure.\n\n", "communicative": "The speaker is required to produce coherent sentences and construct a trees structured ontology of the column headers.\n\n" }, "language": { "found": [ "Offline media collection" ], "crowdsourced": [], "created": "Creators proposed a two-stage annotation process for constructing triple set sentence pairs based on a tree-structured ontology of each table. First, internal skilled annotators denote the parent column for each column header. Then, a larger number of annotators provide a sentential description of an automatically-chosen subset of table cells in a row. To form a triple set sentence pair, the highlighted cells can be converted to a connected triple set automatically according to the column ontology for the given table.\n\n", "machine-generated": "N/A", "validated": "validated by crowdworker", "is-filtered": "not filtered", "filtered-criteria": "N/A", "obtained": [ "Found", "Created for the dataset" ], "producers-description": "No further information about the MTurk workers has been provided.\n\n", "topics": "The sub-datasets are from Wikipedia, DBPedia, and artificially created restaurant data. ", "pre-processed": "n/a" }, "annotations": { "origin": "none", "rater-number": "N/A", "rater-qualifications": "N/A", "rater-training-num": "N/A", "rater-test-num": "N/A", "rater-annotation-service-bool": "no", "rater-annotation-service": [], "values": "N/A", "quality-control": [], "quality-control-details": "N/A" }, "consent": { "has-consent": "no", "consent-policy": "N/A", "consent-other": "N/A", "no-consent-justification": "The new annotations are based on Wikipedia which is in the public domain and the other two datasets permit reuse (with attribution)" }, "pii": { "has-pii": "no PII", "no-pii-justification": "None of the datasets talk about individuals", "is-pii-identified": "N/A", "pii-identified-method": "N/A", "is-pii-replaced": "N/A", "pii-replaced-method": "N/A", "pii-categories": [] }, "maintenance": { "has-maintenance": "no", "description": "N/A", "contact": "N/A", "contestation-mechanism": "N/A", "contestation-link": "N/A", "contestation-description": "N/A" } }, "gem": { "rationale": { "sole-task-dataset": "yes", "sole-language-task-dataset": "no", "distinction-description": "The tree structure is unique among GEM datasets", "contribution": "DART is a large and open-domain structured DAta Record to Text generation corpus with high-quality sentence annotations with each input being a set of entity-relation triples following a tree-structured ontology. ", "model-ability": "Reasoning, surface realization" }, "curation": { "has-additional-curation": "no", "modification-types": [], "modification-description": "N/A", "has-additional-splits": "no", "additional-splits-description": "N/A", "additional-splits-capacicites": "N/A" }, "starting": { "research-pointers": "Experimental results on DART shows that BART model as the highest performance among three models with a BLEU score of 37.06. This is attributed to BART\u2019s generalization ability due to pretraining ([Table 4](https://arxiv.org/pdf/2007.02871.pdf)).\n", "technical-terms": "n/a" } }, "results": { "results": { "other-metrics-definitions": "N/A", "has-previous-results": "yes", "current-evaluation": "n/a ", "previous-results": "BART currently achieves the best performance according to the leaderboard.", "model-abilities": "Reasoning, surface realization", "metrics": [ "BLEU", "MoverScore", "BERT-Score", "BLEURT" ], "original-evaluation": "The leaderboard uses the combination of BLEU, METEOR, TER, MoverScore, BERTScore, PARENT and BLEURT to overcome the limitations of the n-gram overlap metrics. \nA small scale human annotation of 100 data points was conducted along the dimensions of (1) fluency - a sentence is natural and grammatical, and (2) semantic faithfulness - a sentence is supported by the input triples." } }, "considerations": { "pii": { "risks-description": "n/a" }, "licenses": { "dataset-restrictions-other": "N/A", "data-copyright-other": "N/A", "dataset-restrictions": [ "open license - commercial use allowed" ], "data-copyright": [ "open license - commercial use allowed" ] }, "limitations": { "data-technical-limitations": "The dataset may contain some social biases, as the input sentences are based on Wikipedia (WikiTableQuestions, WikiSQL, WebNLG). Studies have shown that the English Wikipedia contains gender biases([Dinan et al., 2020](https://www.aclweb.org/anthology/2020.emnlp-main.23.pdf)), racial biases([Papakyriakopoulos et al., 2020 (https://dl.acm.org/doi/pdf/10.1145/3351095.3372843)) and geographical bias([Livingstone et al., 2010](https://doi.org/10.5204/mcj.315)). [More info](https://en.wikipedia.org/wiki/Racial_bias_on_Wikipedia#cite_note-23).\n", "data-unsuited-applications": "The end-to-end transformer has the lowest performance since the transformer model needs intermediate pipeline planning steps to have higher performance. Similar findings can be found in [Castro Ferreira et al., 2019](https://arxiv.org/pdf/1908.09022.pdf).\n", "data-discouraged-use": "n/a" } }, "context": { "previous": { "is-deployed": "no", "described-risks": "N/A", "changes-from-observation": "N/A" }, "underserved": { "helps-underserved": "no", "underserved-description": "N/A" }, "biases": { "has-biases": "no", "bias-analyses": "N/A", "speaker-distibution": "No, the annotators are raters on crowdworking platforms and thus only represent their demographics." } } }