{"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/generating-descriptions-from-structured-data","title":"Generating Descriptions from Structured Data Using a Bifocal Attention Mechanism and Gated Orthogonalization","arxiv_id":"1804.07789","date":"2018-04-20","proceeding":"NAACL 2018 6","authors":["Preksha Nema","Shreyas Shetty","Parag Jain","Anirban Laha","Karthik Sankaranarayanan","Mitesh M. Khapra"],"abstract":"In this work, we focus on the task of generating natural language\ndescriptions from a structured table of facts containing fields (such as\nnationality, occupation, etc) and values (such as Indian, actor, director,\netc). One simple choice is to treat the table as a sequence of fields and\nvalues and then use a standard seq2seq model for this task. However, such a\nmodel is too generic and does not exploit task-specific characteristics. For\nexample, while generating descriptions from a table, a human would attend to\ninformation at two levels: (i) the fields (macro level) and (ii) the values\nwithin the field (micro level). Further, a human would continue attending to a\nfield for a few timesteps till all the information from that field has been\nrendered and then never return back to this field (because there is nothing\nleft to say about it). To capture this behavior we use (i) a fused bifocal\nattention mechanism which exploits and combines this micro and macro level\ninformation and (ii) a gated orthogonalization mechanism which tries to ensure\nthat a field is remembered for a few time steps and then forgotten. We\nexperiment with a recently released dataset which contains fact tables about\npeople and their corresponding one line biographical descriptions in English.\nIn addition, we also introduce two similar datasets for French and German. Our\nexperiments show that the proposed model gives 21% relative improvement over a\nrecently proposed state of the art method and 10% relative improvement over\nbasic seq2seq models. The code and the datasets developed as a part of this\nwork are publicly available.","url_abs":"http://arxiv.org/abs/1804.07789v1","url_pdf":"http://arxiv.org/pdf/1804.07789v1.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":"generating-descriptions-from-structured-data","repo_url":"https://github.com/PrekshaNema25/StructuredData_To_Descriptions","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null},{"paper_slug":"generating-descriptions-from-structured-data","repo_url":"https://github.com/parajain/StructuredData_To_Descriptions","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[],"methods":[{"method_slug":"lstm","method_name":"LSTM"},{"method_slug":"seq2seq","method_name":"Seq2Seq"},{"method_slug":"sigmoid-activation","method_name":"Sigmoid Activation"},{"method_slug":"tanh-activation","method_name":"Tanh Activation"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1804.07789","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}