{"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/adversarial-propagation-and-zero-shot-cross","title":"Adversarial Propagation and Zero-Shot Cross-Lingual Transfer of Word Vector Specialization","arxiv_id":"1809.04163","date":"2018-09-11","proceeding":"EMNLP 2018 10","authors":["Edoardo Maria Ponti","Ivan Vulić","Goran Glavaš","Nikola Mrkšić","Anna Korhonen"],"abstract":"Semantic specialization is the process of fine-tuning pre-trained\ndistributional word vectors using external lexical knowledge (e.g., WordNet) to\naccentuate a particular semantic relation in the specialized vector space.\nWhile post-processing specialization methods are applicable to arbitrary\ndistributional vectors, they are limited to updating only the vectors of words\noccurring in external lexicons (i.e., seen words), leaving the vectors of all\nother words unchanged. We propose a novel approach to specializing the full\ndistributional vocabulary. Our adversarial post-specialization method\npropagates the external lexical knowledge to the full distributional space. We\nexploit words seen in the resources as training examples for learning a global\nspecialization function. This function is learned by combining a standard\nL2-distance loss with an adversarial loss: the adversarial component produces\nmore realistic output vectors. We show the effectiveness and robustness of the\nproposed method across three languages and on three tasks: word similarity,\ndialog state tracking, and lexical simplification. We report consistent\nimprovements over distributional word vectors and vectors specialized by other\nstate-of-the-art specialization frameworks. Finally, we also propose a\ncross-lingual transfer method for zero-shot specialization which successfully\nspecializes a full target distributional space without any lexical knowledge in\nthe target language and without any bilingual data.","url_abs":"http://arxiv.org/abs/1809.04163v1","url_pdf":"http://arxiv.org/pdf/1809.04163v1.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":"adversarial-propagation-and-zero-shot-cross","repo_url":"https://github.com/cambridgeltl/adversarial-postspec","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"cross-lingual-transfer","task_name":"Cross-Lingual Transfer"},{"task_slug":"lexical-simplification","task_name":"Lexical Simplification"},{"task_slug":"word-similarity","task_name":"Word Similarity"},{"task_slug":"zero-shot-cross-lingual-transfer","task_name":"Zero-Shot Cross-Lingual Transfer"},{"task_slug":null,"task_name":"dialog state tracking"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1809.04163","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}