{"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/improving-zero-shot-translation-of-low","title":"Improving Zero-Shot Translation of Low-Resource Languages","arxiv_id":"1811.01389","date":"2018-11-04","proceeding":"IWSLT 2017 12","authors":["Surafel M. Lakew","Quintino F. Lotito","Matteo Negri","Marco Turchi","Marcello Federico"],"abstract":"Recent work on multilingual neural machine translation reported competitive\nperformance with respect to bilingual models and surprisingly good performance\neven on (zeroshot) translation directions not observed at training time. We\ninvestigate here a zero-shot translation in a particularly lowresource\nmultilingual setting. We propose a simple iterative training procedure that\nleverages a duality of translations directly generated by the system for the\nzero-shot directions. The translations produced by the system (sub-optimal\nsince they contain mixed language from the shared vocabulary), are then used\ntogether with the original parallel data to feed and iteratively re-train the\nmultilingual network. Over time, this allows the system to learn from its own\ngenerated and increasingly better output. Our approach shows to be effective in\nimproving the two zero-shot directions of our multilingual model. In\nparticular, we observed gains of about 9 BLEU points over a baseline\nmultilingual model and up to 2.08 BLEU over a pivoting mechanism using two\nbilingual models. Further analysis shows that there is also a slight\nimprovement in the non-zero-shot language directions.","url_abs":"http://arxiv.org/abs/1811.01389v1","url_pdf":"http://arxiv.org/pdf/1811.01389v1.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":"improving-zero-shot-translation-of-low","repo_url":"https://github.com/surafelml/improving-zeroshot-nmt","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"translation","task_name":"Translation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1811.01389","atlas_url":"https://app.syntology.ai/?focus=1811.01389","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}