{"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/why-self-attention-a-targeted-evaluation-of","title":"Why Self-Attention? A Targeted Evaluation of Neural Machine Translation Architectures","arxiv_id":"1808.08946","date":"2018-08-27","proceeding":"EMNLP 2018 10","authors":["Gongbo Tang","Mathias Müller","Annette Rios","Rico Sennrich"],"abstract":"Recently, non-recurrent architectures (convolutional, self-attentional) have\noutperformed RNNs in neural machine translation. CNNs and self-attentional\nnetworks can connect distant words via shorter network paths than RNNs, and it\nhas been speculated that this improves their ability to model long-range\ndependencies. However, this theoretical argument has not been tested\nempirically, nor have alternative explanations for their strong performance\nbeen explored in-depth. We hypothesize that the strong performance of CNNs and\nself-attentional networks could also be due to their ability to extract\nsemantic features from the source text, and we evaluate RNNs, CNNs and\nself-attention networks on two tasks: subject-verb agreement (where capturing\nlong-range dependencies is required) and word sense disambiguation (where\nsemantic feature extraction is required). Our experimental results show that:\n1) self-attentional networks and CNNs do not outperform RNNs in modeling\nsubject-verb agreement over long distances; 2) self-attentional networks\nperform distinctly better than RNNs and CNNs on word sense disambiguation.","url_abs":"http://arxiv.org/abs/1808.08946v3","url_pdf":"http://arxiv.org/pdf/1808.08946v3.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":"why-self-attention-a-targeted-evaluation-of","repo_url":"https://github.com/awslabs/sockeye","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"mxnet","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"translation","task_name":"Translation"},{"task_slug":"word-sense-disambiguation","task_name":"Word Sense Disambiguation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1808.08946","atlas_url":"https://app.syntology.ai/?focus=1808.08946","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}