{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/reaching-human-level-performance-in-automatic","title":"Reaching Human-level Performance in Automatic Grammatical Error Correction: An Empirical Study","arxiv_id":"1807.01270","date":"2018-07-03","proceeding":null,"authors":["Tao Ge","Furu Wei","Ming Zhou"],"abstract":"Neural sequence-to-sequence (seq2seq) approaches have proven to be successful\nin grammatical error correction (GEC). Based on the seq2seq framework, we\npropose a novel fluency boost learning and inference mechanism. Fluency\nboosting learning generates diverse error-corrected sentence pairs during\ntraining, enabling the error correction model to learn how to improve a\nsentence's fluency from more instances, while fluency boosting inference allows\nthe model to correct a sentence incrementally with multiple inference steps.\nCombining fluency boost learning and inference with convolutional seq2seq\nmodels, our approach achieves the state-of-the-art performance: 75.72 (F_{0.5})\non CoNLL-2014 10 annotation dataset and 62.42 (GLEU) on JFLEG test set\nrespectively, becoming the first GEC system that reaches human-level\nperformance (72.58 for CoNLL and 62.37 for JFLEG) on both of the benchmarks.","url_abs":"http://arxiv.org/abs/1807.01270v5","url_pdf":"http://arxiv.org/pdf/1807.01270v5.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"reaching-human-level-performance-in-automatic","repo_url":"https://github.com/getao/human-performance-gec","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"grammatical-error-correction","task_name":"Grammatical Error Correction"},{"task_slug":"sentence","task_name":"Sentence"}],"methods":[{"method_slug":"lstm","method_name":"LSTM"},{"method_slug":"seq2seq","method_name":"Seq2Seq"},{"method_slug":"sigmoid-activation","method_name":"Sigmoid Activation"},{"method_slug":"tanh-activation","method_name":"Tanh Activation"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/grammatical-error-correction-on-unrestricted","task":"Grammatical Error Correction","dataset":"Unrestricted","model":"CNN Seq2Seq + Fluency Boost","rank_in_archive_order":1,"of":3,"metrics":{"F0.5":"61.34"},"uses_additional_data":false},{"leaderboard":"/sota/grammatical-error-correction-on-unrestricted","task":"Grammatical Error Correction","dataset":"Unrestricted","model":"CNN Seq2Seq + Fluency Boost and inference","rank_in_archive_order":3,"of":3,"metrics":{"GLEU":"62.37"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1807.01270","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}