{"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/recursive-subtree-composition-in-lstm-based","title":"Recursive Subtree Composition in LSTM-Based Dependency Parsing","arxiv_id":"1902.09781","date":"2019-02-26","proceeding":"NAACL 2019 6","authors":["Miryam de Lhoneux","Miguel Ballesteros","Joakim Nivre"],"abstract":"The need for tree structure modelling on top of sequence modelling is an open\nissue in neural dependency parsing. We investigate the impact of adding a tree\nlayer on top of a sequential model by recursively composing subtree\nrepresentations (composition) in a transition-based parser that uses features\nextracted by a BiLSTM. Composition seems superfluous with such a model,\nsuggesting that BiLSTMs capture information about subtrees. We perform model\nablations to tease out the conditions under which composition helps. When\nablating the backward LSTM, performance drops and composition does not recover\nmuch of the gap. When ablating the forward LSTM, performance drops less\ndramatically and composition recovers a substantial part of the gap, indicating\nthat a forward LSTM and composition capture similar information. We take the\nbackward LSTM to be related to lookahead features and the forward LSTM to the\nrich history-based features both crucial for transition-based parsers. To\ncapture history-based information, composition is better than a forward LSTM on\nits own, but it is even better to have a forward LSTM as part of a BiLSTM. We\ncorrelate results with language properties, showing that the improved lookahead\nof a backward LSTM is especially important for head-final languages.","url_abs":"http://arxiv.org/abs/1902.09781v1","url_pdf":"http://arxiv.org/pdf/1902.09781v1.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":"recursive-subtree-composition-in-lstm-based","repo_url":"https://github.com/mdelhoneux/uuparser-composition","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"dependency-parsing","task_name":"Dependency Parsing"}],"methods":[{"method_slug":"bilstm","method_name":"BiLSTM"},{"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":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1902.09781","atlas_url":"https://app.syntology.ai/?focus=1902.09781","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}