{"paper":{"arxiv_id":"1910.13461","title":"BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension","abstract":"We present BART, a denoising autoencoder for pretraining sequence-to-sequence models. BART is trained by (1) corrupting text with an arbitrary noising function, and (2) learning a model to reconstruct the original text. It uses a standard Tranformer-based neural machine translation architecture which, despite its simplicity, can be seen as generalizing BERT (due to the bidirectional encoder), GPT (with the left-to-right decoder), and many other more recent pretraining schemes. We evaluate a number of noising approaches, finding the best performance by both randomly shuffling the order of the original sentences and using a novel in-filling scheme, where spans of text are replaced with a single mask token. BART is particularly effective when fine tuned for text generation but also works well for comprehension tasks. It matches the performance of RoBERTa with comparable training resources on GLUE and SQuAD, achieves new state-of-the-art results on a range of abstractive dialogue, question answering, and summarization tasks, with gains of up to 6 ROUGE.","primary_category":"cs.CL","venue":"ACL 2020","published_at":null,"latest_version":1,"withdrawn":false},"latest_version":{"id":"11a66ad3-3596-4e72-82eb-528a6dd125f7","version":1,"source_url":"https://arxiv.org/abs/1910.13461","rendered_html_url":null,"rendering_engine":null},"verdict":{"id":"da7cc30d-0751-4d9e-ac06-bd24c3eaef9b","kind":"POST","status":"partial","score":0.42137434129818424,"confidence":0.6,"agent_version":"v0.1.0-bart-cnndm-200slice","computed_at":"2026-05-14T23:22:17.517Z","is_current":true,"claim_citation":{"paper_arxiv_id":"1910.13461","section":"Table 3","row":"BART","column":"R-L","reported_value":40.9,"reported_metric":"rougeLsum","quoted_text":"BART 44.16 21.28 40.90","pdf_page":7,"notes":"Table 3 (CNN/DailyMail summarization, abstractive systems). The paper's ROUGE-L is summary-level LCS per Lin 2004 — Google's `rouge_score` library exposes this as `rougeLsum`. An earlier reproduction measured `rougeL` (sentence-level) and recorded a false-positive WRONG verdict; PR #104 corrected the metric and this PR adds the citation contract."},"protocol_match":"exact"},"verdicts":{"post":{"id":"da7cc30d-0751-4d9e-ac06-bd24c3eaef9b","kind":"POST","status":"partial","score":0.42137434129818424,"confidence":0.6,"agent_version":"v0.1.0-bart-cnndm-200slice","computed_at":"2026-05-14T23:22:17.517Z","is_current":true,"claim_citation":{"paper_arxiv_id":"1910.13461","section":"Table 3","row":"BART","column":"R-L","reported_value":40.9,"reported_metric":"rougeLsum","quoted_text":"BART 44.16 21.28 40.90","pdf_page":7,"notes":"Table 3 (CNN/DailyMail summarization, abstractive systems). The paper's ROUGE-L is summary-level LCS per Lin 2004 — Google's `rouge_score` library exposes this as `rougeLsum`. An earlier reproduction measured `rougeL` (sentence-level) and recorded a false-positive WRONG verdict; PR #104 corrected the metric and this PR adds the citation contract."},"protocol_match":"exact"},"pre":null}}