{"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/unifying-human-and-statistical-evaluation-for","title":"Unifying Human and Statistical Evaluation for Natural Language Generation","arxiv_id":"1904.02792","date":"2019-04-04","proceeding":"NAACL 2019 6","authors":["Tatsunori B. Hashimoto","Hugh Zhang","Percy Liang"],"abstract":"How can we measure whether a natural language generation system produces both\nhigh quality and diverse outputs? Human evaluation captures quality but not\ndiversity, as it does not catch models that simply plagiarize from the training\nset. On the other hand, statistical evaluation (i.e., perplexity) captures\ndiversity but not quality, as models that occasionally emit low quality samples\nwould be insufficiently penalized. In this paper, we propose a unified\nframework which evaluates both diversity and quality, based on the optimal\nerror rate of predicting whether a sentence is human- or machine-generated. We\ndemonstrate that this error rate can be efficiently estimated by combining\nhuman and statistical evaluation, using an evaluation metric which we call\nHUSE. On summarization and chit-chat dialogue, we show that (i) HUSE detects\ndiversity defects which fool pure human evaluation and that (ii) techniques\nsuch as annealing for improving quality actually decrease HUSE due to decreased\ndiversity.","url_abs":"http://arxiv.org/abs/1904.02792v1","url_pdf":"http://arxiv.org/pdf/1904.02792v1.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":"unifying-human-and-statistical-evaluation-for","repo_url":"https://worksheets.codalab.org/worksheets/0x88644b5ee189402eb19d39d721d1005c","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":null},{"paper_slug":"unifying-human-and-statistical-evaluation-for","repo_url":"https://github.com/hughbzhang/HUSE","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":{"status":"ok"}}],"tasks":[{"task_slug":"diversity","task_name":"Diversity"},{"task_slug":"sentence","task_name":"Sentence"},{"task_slug":"text-generation","task_name":"Text Generation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1904.02792","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}