{"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/deep-architectures-for-neural-machine","title":"Deep Architectures for Neural Machine Translation","arxiv_id":"1707.07631","date":"2017-07-24","proceeding":"WS 2017 9","authors":["Antonio Valerio Miceli Barone","Jindřich Helcl","Rico Sennrich","Barry Haddow","Alexandra Birch"],"abstract":"It has been shown that increasing model depth improves the quality of neural\nmachine translation. However, different architectural variants to increase\nmodel depth have been proposed, and so far, there has been no thorough\ncomparative study.\n  In this work, we describe and evaluate several existing approaches to\nintroduce depth in neural machine translation. Additionally, we explore novel\narchitectural variants, including deep transition RNNs, and we vary how\nattention is used in the deep decoder. We introduce a novel \"BiDeep\" RNN\narchitecture that combines deep transition RNNs and stacked RNNs.\n  Our evaluation is carried out on the English to German WMT news translation\ndataset, using a single-GPU machine for both training and inference. We find\nthat several of our proposed architectures improve upon existing approaches in\nterms of speed and translation quality. We obtain best improvements with a\nBiDeep RNN of combined depth 8, obtaining an average improvement of 1.5 BLEU\nover a strong shallow baseline.\n  We release our code for ease of adoption.","url_abs":"http://arxiv.org/abs/1707.07631v1","url_pdf":"http://arxiv.org/pdf/1707.07631v1.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":"deep-architectures-for-neural-machine","repo_url":"https://github.com/EdinburghNLP/nematus","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"deep-architectures-for-neural-machine","repo_url":"https://github.com/Avmb/deep-nmt-architectures","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":{"status":"unanswered"}},{"paper_slug":"deep-architectures-for-neural-machine","repo_url":"https://github.com/bhaddow/dev-nematus","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"decoder","task_name":"Decoder"},{"task_slug":null,"task_name":"GPU"},{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"translation","task_name":"Translation"}],"methods":[{"method_slug":"speed","method_name":"SPEED"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1707.07631","atlas_url":"https://app.syntology.ai/?focus=1707.07631","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}