{"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/studying-the-inductive-biases-of-rnns-with","title":"Studying the Inductive Biases of RNNs with Synthetic Variations of Natural Languages","arxiv_id":"1903.06400","date":"2019-03-15","proceeding":"NAACL 2019 6","authors":["Shauli Ravfogel","Yoav Goldberg","Tal Linzen"],"abstract":"How do typological properties such as word order and morphological case\nmarking affect the ability of neural sequence models to acquire the syntax of a\nlanguage? Cross-linguistic comparisons of RNNs' syntactic performance (e.g., on\nsubject-verb agreement prediction) are complicated by the fact that any two\nlanguages differ in multiple typological properties, as well as by differences\nin training corpus. We propose a paradigm that addresses these issues: we\ncreate synthetic versions of English, which differ from English in one or more\ntypological parameters, and generate corpora for those languages based on a\nparsed English corpus. We report a series of experiments in which RNNs were\ntrained to predict agreement features for verbs in each of those synthetic\nlanguages. Among other findings, (1) performance was higher in\nsubject-verb-object order (as in English) than in subject-object-verb order (as\nin Japanese), suggesting that RNNs have a recency bias; (2) predicting\nagreement with both subject and object (polypersonal agreement) improves over\npredicting each separately, suggesting that underlying syntactic knowledge\ntransfers across the two tasks; and (3) overt morphological case makes\nagreement prediction significantly easier, regardless of word order.","url_abs":"http://arxiv.org/abs/1903.06400v2","url_pdf":"http://arxiv.org/pdf/1903.06400v2.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":"studying-the-inductive-biases-of-rnns-with","repo_url":"https://github.com/Shaul1321/rnn_typology","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"studying-the-inductive-biases-of-rnns-with","repo_url":"https://github.com/573phn/rnn_typology","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}}],"tasks":[{"task_slug":"object","task_name":"Object"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1903.06400","atlas_url":"https://app.syntology.ai/?focus=1903.06400","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}