{"url":"/sota/text-generation-on-coco-captions","task":{"name":"Text Generation","url":"/task/text-generation","note":null},"dataset":{"name":"COCO Captions","url":"/dataset/coco-captions"},"category":"Natural Language Processing","categories":["Adversarial","Computer Code","Natural Language Processing","Speech"],"category_note":null,"description":"**Text Generation** is the task of generating text with the goal of appearing indistinguishable to human-written text. This task is more formally known as \"natural language generation\" in the literature.\r\n\r\nText generation can be addressed with Markov processes or deep generative models like LSTMs. Recently, some of the most advanced methods for text generation include [BART](/method/bart), [GPT](/method/gpt) and other [GAN-based approaches](/method/gan). Text generation systems are evaluated either through human ratings or automatic evaluation metrics like METEOR, ROUGE, and BLEU. \r\n\r\nFurther readings:\r\n\r\n- [The survey: Text generation models in deep learning](https://www.sciencedirect.com/science/article/pii/S1319157820303360)\r\n- [Modern Methods for Text Generation](https://arxiv.org/abs/2009.04968)\r\n\r\n<span style=\"color:grey; opacity: 0.6\">( Image credit: [Adversarial Ranking for Language Generation](https://arxiv.org/abs/1705.11001) )</span>","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["BLEU-2","BLEU-3","BLEU-4","BLEU-5"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"BLEU-2":"higher","BLEU-3":"higher","BLEU-4":"higher","BLEU-5":"higher"}},"counts":{"rows":5,"rows_with_code":5,"rows_with_paper_page":5,"rows_dated":5,"rows_using_additional_data":0},"rows":[{"rank_in_archive_order":1,"model":"LeakGAN","metrics":{"BLEU-2":"0.950","BLEU-3":"0.880","BLEU-4":"0.778","BLEU-5":"0.686"},"uses_additional_data":false,"paper_date":"2017-09-24","paper":"/paper/long-text-generation-via-adversarial-training","paper_url":"http://arxiv.org/abs/1709.08624v2","paper_title":"Long Text Generation via Adversarial Training with Leaked Information","code":"https://github.com/CR-Gjx/LeakGAN","n_code_links":6,"syntology":{"n_ran":2,"n_unverified":0,"n_samples":2,"n_pointer_only_licence":2}},{"rank_in_archive_order":2,"model":"partGAN","metrics":{"BLEU-2":"0.910","BLEU-3":"0.713","BLEU-4":"O.753","BLEU-5":"0.590"},"uses_additional_data":false,"paper_date":"2017-09-24","paper":"/paper/long-text-generation-via-adversarial-training","paper_url":"http://arxiv.org/abs/1709.08624v2","paper_title":"Long Text Generation via Adversarial Training with Leaked Information","code":"https://github.com/CR-Gjx/LeakGAN","n_code_links":6,"syntology":{"n_ran":2,"n_unverified":0,"n_samples":2,"n_pointer_only_licence":2}},{"rank_in_archive_order":3,"model":"RankGAN","metrics":{"BLEU-2":"0.850","BLEU-3":"0.672","BLEU-4":"0.557","BLEU-5":"0.544"},"uses_additional_data":false,"paper_date":"2017-05-31","paper":"/paper/adversarial-ranking-for-language-generation","paper_url":"http://arxiv.org/abs/1705.11001v3","paper_title":"Adversarial Ranking for Language Generation","code":"https://github.com/desire2020/RankGAN","n_code_links":1,"syntology":null},{"rank_in_archive_order":4,"model":"RelGAN (100)","metrics":{"BLEU-2":"0.849","BLEU-3":"0.687","BLEU-4":"0.502"},"uses_additional_data":false,"paper_date":"2019-05-01","paper":"/paper/relgan-relational-generative-adversarial","paper_url":"https://openreview.net/forum?id=rJedV3R5tm","paper_title":"RelGAN: Relational Generative Adversarial Networks for Text Generation","code":"https://github.com/weilinie/RelGAN","n_code_links":1,"syntology":null},{"rank_in_archive_order":5,"model":"SeqGAN","metrics":{"BLEU-2":"0.831","BLEU-3":"0.642","BLEU-4":"0.521","BLEU-5":"0.427"},"uses_additional_data":false,"paper_date":"2016-09-18","paper":"/paper/seqgan-sequence-generative-adversarial-nets","paper_url":"http://arxiv.org/abs/1609.05473v6","paper_title":"SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient","code":"https://github.com/LantaoYu/SeqGAN","n_code_links":23,"syntology":{"n_ran":10,"n_unverified":11,"n_samples":21,"n_pointer_only_licence":13}}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 6,264 of the 9,581 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9581,"papers_checked":6264,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":3316},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":[],"entries":[]},"syntology":{"read_at":"2026-09-24T18:15:14+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":3,"rows_with_any_sample_ran":3,"distinct_papers_with_graph_line":2,"distinct_papers_with_any_sample_ran":2,"samples_over_distinct_papers":{"n_ran":12,"n_unverified":11,"n_samples":23,"n_pointer_only_licence":15,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":14,"n_unverified":11,"n_samples":25,"n_pointer_only_licence":17,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}