{"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/post-specialisation-retrofitting-vectors-of","title":"Post-Specialisation: Retrofitting Vectors of Words Unseen in Lexical Resources","arxiv_id":"1805.03228","date":"2018-05-08","proceeding":"NAACL 2018 6","authors":["Ivan Vulić","Goran Glavaš","Nikola Mrkšić","Anna Korhonen"],"abstract":"Word vector specialisation (also known as retrofitting) is a portable,\nlight-weight approach to fine-tuning arbitrary distributional word vector\nspaces by injecting external knowledge from rich lexical resources such as\nWordNet. By design, these post-processing methods only update the vectors of\nwords occurring in external lexicons, leaving the representations of all unseen\nwords intact. In this paper, we show that constraint-driven vector space\nspecialisation can be extended to unseen words. We propose a novel\npost-specialisation method that: a) preserves the useful linguistic knowledge\nfor seen words; while b) propagating this external signal to unseen words in\norder to improve their vector representations as well. Our post-specialisation\napproach explicits a non-linear specialisation function in the form of a deep\nneural network by learning to predict specialised vectors from their original\ndistributional counterparts. The learned function is then used to specialise\nvectors of unseen words. This approach, applicable to any post-processing\nmodel, yields considerable gains over the initial specialisation models both in\nintrinsic word similarity tasks, and in two downstream tasks: dialogue state\ntracking and lexical text simplification. The positive effects persist across\nthree languages, demonstrating the importance of specialising the full\nvocabulary of distributional word vector spaces.","url_abs":"http://arxiv.org/abs/1805.03228v1","url_pdf":"http://arxiv.org/pdf/1805.03228v1.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":"post-specialisation-retrofitting-vectors-of","repo_url":"https://github.com/cambridgeltl/post-specialisation","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[{"task_slug":"dialogue-state-tracking","task_name":"Dialogue State Tracking"},{"task_slug":"text-simplification","task_name":"Text Simplification"},{"task_slug":"word-similarity","task_name":"Word Similarity"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1805.03228","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}