{"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/distilling-an-ensemble-of-greedy-dependency","title":"Distilling an Ensemble of Greedy Dependency Parsers into One MST Parser","arxiv_id":"1609.07561","date":"2016-09-24","proceeding":"EMNLP 2016 11","authors":["Adhiguna Kuncoro","Miguel Ballesteros","Lingpeng Kong","Chris Dyer","Noah A. Smith"],"abstract":"We introduce two first-order graph-based dependency parsers achieving a new\nstate of the art. The first is a consensus parser built from an ensemble of\nindependently trained greedy LSTM transition-based parsers with different\nrandom initializations. We cast this approach as minimum Bayes risk decoding\n(under the Hamming cost) and argue that weaker consensus within the ensemble is\na useful signal of difficulty or ambiguity. The second parser is a\n\"distillation\" of the ensemble into a single model. We train the distillation\nparser using a structured hinge loss objective with a novel cost that\nincorporates ensemble uncertainty estimates for each possible attachment,\nthereby avoiding the intractable cross-entropy computations required by\napplying standard distillation objectives to problems with structured outputs.\nThe first-order distillation parser matches or surpasses the state of the art\non English, Chinese, and German.","url_abs":"http://arxiv.org/abs/1609.07561v1","url_pdf":"http://arxiv.org/pdf/1609.07561v1.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":"distilling-an-ensemble-of-greedy-dependency","repo_url":"https://github.com/adhigunasurya/distillation_parser","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}}],"tasks":[{"task_slug":"dependency-parsing","task_name":"Dependency Parsing"}],"methods":[{"method_slug":"lstm","method_name":"LSTM"},{"method_slug":"sigmoid-activation","method_name":"Sigmoid Activation"},{"method_slug":"tanh-activation","method_name":"Tanh Activation"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/dependency-parsing-on-penn-treebank","task":"Dependency Parsing","dataset":"Penn Treebank","model":"Distilled neural FOG","rank_in_archive_order":18,"of":22,"metrics":{"LAS":"92.06","POS":"97.44","UAS":"94.26"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1609.07561","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}