{"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/finding-better-subword-segmentation-for","title":"Finding Better Subword Segmentation for Neural Machine Translation","arxiv_id":"1807.09639","date":"2018-07-25","proceeding":null,"authors":["Yingting Wu","Hai Zhao"],"abstract":"For different language pairs, word-level neural machine translation (NMT)\nmodels with a fixed-size vocabulary suffer from the same problem of\nrepresenting out-of-vocabulary (OOV) words. The common practice usually\nreplaces all these rare or unknown words with a <UNK> token, which limits the\ntranslation performance to some extent. Most of recent work handled such a\nproblem by splitting words into characters or other specially extracted subword\nunits to enable open-vocabulary translation. Byte pair encoding (BPE) is one of\nthe successful attempts that has been shown extremely competitive by providing\neffective subword segmentation for NMT systems. In this paper, we extend the\nBPE style segmentation to a general unsupervised framework with three\nstatistical measures: frequency (FRQ), accessor variety (AV) and description\nlength gain (DLG). We test our approach on two translation tasks: German to\nEnglish and Chinese to English. The experimental results show that AV and DLG\nenhanced systems outperform the FRQ baseline in the frequency weighted schemes\nat different significant levels.","url_abs":"http://arxiv.org/abs/1807.09639v1","url_pdf":"http://arxiv.org/pdf/1807.09639v1.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":"finding-better-subword-segmentation-for","repo_url":"https://github.com/Lindsay125/gbpe","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"nmt","task_name":"NMT"},{"task_slug":"segmentation","task_name":"Segmentation"},{"task_slug":"translation","task_name":"Translation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}