{"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/strategies-for-training-large-vocabulary","title":"Strategies for Training Large Vocabulary Neural Language Models","arxiv_id":"1512.04906","date":"2015-12-15","proceeding":"ACL 2016 8","authors":["Welin Chen","David Grangier","Michael Auli"],"abstract":"Training neural network language models over large vocabularies is still\ncomputationally very costly compared to count-based models such as Kneser-Ney.\nAt the same time, neural language models are gaining popularity for many\napplications such as speech recognition and machine translation whose success\ndepends on scalability. We present a systematic comparison of strategies to\nrepresent and train large vocabularies, including softmax, hierarchical\nsoftmax, target sampling, noise contrastive estimation and self normalization.\nWe further extend self normalization to be a proper estimator of likelihood and\nintroduce an efficient variant of softmax. We evaluate each method on three\npopular benchmarks, examining performance on rare words, the speed/accuracy\ntrade-off and complementarity to Kneser-Ney.","url_abs":"http://arxiv.org/abs/1512.04906v1","url_pdf":"http://arxiv.org/pdf/1512.04906v1.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":"strategies-for-training-large-vocabulary","repo_url":"https://github.com/AshwinDeshpande96/Hierarchical-Softmax","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}},{"paper_slug":"strategies-for-training-large-vocabulary","repo_url":"https://github.com/jiali-ms/JLM","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"speech-recognition","task_name":"Speech Recognition"},{"task_slug":"translation","task_name":"Translation"},{"task_slug":"speech-recognition-1","task_name":"speech-recognition"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1512.04906","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}