{"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/mass-masked-sequence-to-sequence-pre-training","title":"MASS: Masked Sequence to Sequence Pre-training for Language Generation","arxiv_id":"1905.02450","date":"2019-05-07","proceeding":null,"authors":["Kaitao Song","Xu Tan","Tao Qin","Jianfeng Lu","Tie-Yan Liu"],"abstract":"Pre-training and fine-tuning, e.g., BERT, have achieved great success in language understanding by transferring knowledge from rich-resource pre-training task to the low/zero-resource downstream tasks. Inspired by the success of BERT, we propose MAsked Sequence to Sequence pre-training (MASS) for the encoder-decoder based language generation tasks. MASS adopts the encoder-decoder framework to reconstruct a sentence fragment given the remaining part of the sentence: its encoder takes a sentence with randomly masked fragment (several consecutive tokens) as input, and its decoder tries to predict this masked fragment. In this way, MASS can jointly train the encoder and decoder to develop the capability of representation extraction and language modeling. By further fine-tuning on a variety of zero/low-resource language generation tasks, including neural machine translation, text summarization and conversational response generation (3 tasks and totally 8 datasets), MASS achieves significant improvements over the baselines without pre-training or with other pre-training methods. Specially, we achieve the state-of-the-art accuracy (37.5 in terms of BLEU score) on the unsupervised English-French translation, even beating the early attention-based supervised model.","url_abs":"https://arxiv.org/abs/1905.02450v5","url_pdf":"https://arxiv.org/pdf/1905.02450v5.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":"mass-masked-sequence-to-sequence-pre-training","repo_url":"https://github.com/microsoft/MASS","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"NOASSERTION"}},{"paper_slug":"mass-masked-sequence-to-sequence-pre-training","repo_url":"https://github.com/jiaruncao/BioCopyMechanism","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"mass-masked-sequence-to-sequence-pre-training","repo_url":"https://github.com/michael-wzhu/mpnet_zh","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"mass-masked-sequence-to-sequence-pre-training","repo_url":"https://github.com/microsoft/MPNet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"mass-masked-sequence-to-sequence-pre-training","repo_url":"https://github.com/MindSpore-paper-code-3/code9/tree/main/mass","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null},{"paper_slug":"mass-masked-sequence-to-sequence-pre-training","repo_url":"https://github.com/cui0523/Code6/tree/main/mass","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null},{"paper_slug":"mass-masked-sequence-to-sequence-pre-training","repo_url":"https://github.com/mindspore-ai/models/tree/master/official/nlp/mass","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null}],"tasks":[{"task_slug":"conversational-response-generation","task_name":"Conversational Response Generation"},{"task_slug":"decoder","task_name":"Decoder"},{"task_slug":"language-modeling","task_name":"Language Modeling"},{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"response-generation","task_name":"Response Generation"},{"task_slug":"sentence","task_name":"Sentence"},{"task_slug":"text-generation","task_name":"Text Generation"},{"task_slug":"text-summarization","task_name":"Text Summarization"},{"task_slug":"translation","task_name":"Translation"},{"task_slug":"unsupervised-machine-translation","task_name":"Unsupervised Machine Translation"}],"methods":[{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"attention-dropout","method_name":"Attention Dropout"},{"method_slug":"bert","method_name":"BERT"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"linear-warmup-with-linear-decay","method_name":"Linear Warmup With Linear Decay"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"weight-decay","method_name":"Weight Decay"},{"method_slug":"wordpiece","method_name":"WordPiece"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/text-summarization-on-gigaword","task":"Text Summarization","dataset":"GigaWord","model":"MASS","rank_in_archive_order":20,"of":41,"metrics":{"ROUGE-1":"38.73","ROUGE-2":"19.71","ROUGE-L":"35.96"},"uses_additional_data":true},{"leaderboard":"/sota/unsupervised-machine-translation-on-wmt2014-2","task":"Unsupervised Machine Translation","dataset":"WMT2014 English-French","model":"MASS (6-layer Transformer)","rank_in_archive_order":2,"of":7,"metrics":{"BLEU":"37.5"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-machine-translation-on-wmt2014-1","task":"Unsupervised Machine Translation","dataset":"WMT2014 French-English","model":"MASS (6-layer Transformer)","rank_in_archive_order":2,"of":7,"metrics":{"BLEU":"34.9"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-machine-translation-on-wmt2016","task":"Unsupervised Machine Translation","dataset":"WMT2016 English-German","model":"MASS (6-layer Transformer)","rank_in_archive_order":2,"of":7,"metrics":{"BLEU":"28.3"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-machine-translation-on-wmt2016-2","task":"Unsupervised Machine Translation","dataset":"WMT2016 English-Romanian","model":"MASS (6-layer Transformer)","rank_in_archive_order":3,"of":3,"metrics":{"BLEU":"35.2"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-machine-translation-on-wmt2016-1","task":"Unsupervised Machine Translation","dataset":"WMT2016 German-English","model":"MASS (6-layer Transformer)","rank_in_archive_order":2,"of":7,"metrics":{"BLEU":"35.2"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-machine-translation-on-wmt2016-3","task":"Unsupervised Machine Translation","dataset":"WMT2016 Romanian-English","model":"MASS (6-layer Transformer)","rank_in_archive_order":2,"of":3,"metrics":{"BLEU":"33.1"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1905.02450","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}