{"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/on-adversarial-examples-for-character-level","title":"On Adversarial Examples for Character-Level Neural Machine Translation","arxiv_id":"1806.09030","date":"2018-06-23","proceeding":"COLING 2018 8","authors":["Javid Ebrahimi","Daniel Lowd","Dejing Dou"],"abstract":"Evaluating on adversarial examples has become a standard procedure to measure\nrobustness of deep learning models. Due to the difficulty of creating white-box\nadversarial examples for discrete text input, most analyses of the robustness\nof NLP models have been done through black-box adversarial examples. We\ninvestigate adversarial examples for character-level neural machine translation\n(NMT), and contrast black-box adversaries with a novel white-box adversary,\nwhich employs differentiable string-edit operations to rank adversarial\nchanges. We propose two novel types of attacks which aim to remove or change a\nword in a translation, rather than simply break the NMT. We demonstrate that\nwhite-box adversarial examples are significantly stronger than their black-box\ncounterparts in different attack scenarios, which show more serious\nvulnerabilities than previously known. In addition, after performing\nadversarial training, which takes only 3 times longer than regular training, we\ncan improve the model's robustness significantly.","url_abs":"http://arxiv.org/abs/1806.09030v1","url_pdf":"http://arxiv.org/pdf/1806.09030v1.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":"on-adversarial-examples-for-character-level","repo_url":"https://github.com/jebivid/adversarial-nmt","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"on-adversarial-examples-for-character-level","repo_url":"https://github.com/alankarj/robust_nlp","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"on-adversarial-examples-for-character-level","repo_url":"https://github.com/makcedward/nlpaug","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"nmt","task_name":"NMT"},{"task_slug":"translation","task_name":"Translation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1806.09030","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}