{"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/subword-regularization-improving-neural","title":"Subword Regularization: Improving Neural Network Translation Models with Multiple Subword Candidates","arxiv_id":"1804.10959","date":"2018-04-29","proceeding":"ACL 2018 7","authors":["Taku Kudo"],"abstract":"Subword units are an effective way to alleviate the open vocabulary problems\nin neural machine translation (NMT). While sentences are usually converted into\nunique subword sequences, subword segmentation is potentially ambiguous and\nmultiple segmentations are possible even with the same vocabulary. The question\naddressed in this paper is whether it is possible to harness the segmentation\nambiguity as a noise to improve the robustness of NMT. We present a simple\nregularization method, subword regularization, which trains the model with\nmultiple subword segmentations probabilistically sampled during training. In\naddition, for better subword sampling, we propose a new subword segmentation\nalgorithm based on a unigram language model. We experiment with multiple\ncorpora and report consistent improvements especially on low resource and\nout-of-domain settings.","url_abs":"http://arxiv.org/abs/1804.10959v1","url_pdf":"http://arxiv.org/pdf/1804.10959v1.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":"subword-regularization-improving-neural","repo_url":"https://github.com/google/sentencepiece","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"subword-regularization-improving-neural","repo_url":"https://github.com/gretelai/gretel-synthetics","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"NOASSERTION"}}],"tasks":[{"task_slug":"language-modeling","task_name":"Language Modeling"},{"task_slug":"language-modelling","task_name":"Language Modelling"},{"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":[{"slug":"unigram-segmentation","name":"Unigram Segmentation","full_name":"Unigram Segmentation"}],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1804.10959","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}