{"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/learning-to-compose-task-specific-tree","title":"Learning to Compose Task-Specific Tree Structures","arxiv_id":"1707.02786","date":"2017-07-10","proceeding":null,"authors":["Jihun Choi","Kang Min Yoo","Sang-goo Lee"],"abstract":"For years, recursive neural networks (RvNNs) have been shown to be suitable\nfor representing text into fixed-length vectors and achieved good performance\non several natural language processing tasks. However, the main drawback of\nRvNNs is that they require structured input, which makes data preparation and\nmodel implementation hard. In this paper, we propose Gumbel Tree-LSTM, a novel\ntree-structured long short-term memory architecture that learns how to compose\ntask-specific tree structures only from plain text data efficiently. Our model\nuses Straight-Through Gumbel-Softmax estimator to decide the parent node among\ncandidates dynamically and to calculate gradients of the discrete decision. We\nevaluate the proposed model on natural language inference and sentiment\nanalysis, and show that our model outperforms or is at least comparable to\nprevious models. We also find that our model converges significantly faster\nthan other models.","url_abs":"http://arxiv.org/abs/1707.02786v4","url_pdf":"http://arxiv.org/pdf/1707.02786v4.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":"learning-to-compose-task-specific-tree","repo_url":"https://github.com/jihunchoi/unsupervised-treelstm","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"natural-language-inference","task_name":"Natural Language Inference"},{"task_slug":"sentiment-analysis","task_name":"Sentiment Analysis"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/natural-language-inference-on-snli","task":"Natural Language Inference","dataset":"SNLI","model":"600D Gumbel TreeLSTM encoders","rank_in_archive_order":62,"of":98,"metrics":{"% Test Accuracy":"86.0","% Train Accuracy":"93.1","Parameters":"10m"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-snli","task":"Natural Language Inference","dataset":"SNLI","model":"300D Gumbel TreeLSTM encoders","rank_in_archive_order":69,"of":98,"metrics":{"% Test Accuracy":"85.6","% Train Accuracy":"91.2","Parameters":"2.9m"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1707.02786","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}