{"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/linguistic-knowledge-as-memory-for-recurrent","title":"Linguistic Knowledge as Memory for Recurrent Neural Networks","arxiv_id":"1703.02620","date":"2017-03-07","proceeding":null,"authors":["Bhuwan Dhingra","Zhilin Yang","William W. Cohen","Ruslan Salakhutdinov"],"abstract":"Training recurrent neural networks to model long term dependencies is\ndifficult. Hence, we propose to use external linguistic knowledge as an\nexplicit signal to inform the model which memories it should utilize.\nSpecifically, external knowledge is used to augment a sequence with typed edges\nbetween arbitrarily distant elements, and the resulting graph is decomposed\ninto directed acyclic subgraphs. We introduce a model that encodes such graphs\nas explicit memory in recurrent neural networks, and use it to model\ncoreference relations in text. We apply our model to several text comprehension\ntasks and achieve new state-of-the-art results on all considered benchmarks,\nincluding CNN, bAbi, and LAMBADA. On the bAbi QA tasks, our model solves 15 out\nof the 20 tasks with only 1000 training examples per task. Analysis of the\nlearned representations further demonstrates the ability of our model to encode\nfine-grained entity information across a document.","url_abs":"http://arxiv.org/abs/1703.02620v1","url_pdf":"http://arxiv.org/pdf/1703.02620v1.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":[],"tasks":[{"task_slug":"lambada","task_name":"LAMBADA"},{"task_slug":"reading-comprehension","task_name":"Reading Comprehension"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/question-answering-on-cnn-daily-mail","task":"Question Answering","dataset":"CNN / Daily Mail","model":"GA+MAGE (32)","rank_in_archive_order":1,"of":16,"metrics":{"CNN":"78.6"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1703.02620","atlas_url":"https://app.syntology.ai/?focus=1703.02620","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}