{"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/dynamic-multi-level-multi-task-learning-for","title":"Dynamic Multi-Level Multi-Task Learning for Sentence Simplification","arxiv_id":"1806.07304","date":"2018-06-19","proceeding":"COLING 2018 8","authors":["Han Guo","Ramakanth Pasunuru","Mohit Bansal"],"abstract":"Sentence simplification aims to improve readability and understandability,\nbased on several operations such as splitting, deletion, and paraphrasing.\nHowever, a valid simplified sentence should also be logically entailed by its\ninput sentence. In this work, we first present a strong pointer-copy mechanism\nbased sequence-to-sequence sentence simplification model, and then improve its\nentailment and paraphrasing capabilities via multi-task learning with related\nauxiliary tasks of entailment and paraphrase generation. Moreover, we propose a\nnovel 'multi-level' layered soft sharing approach where each auxiliary task\nshares different (higher versus lower) level layers of the sentence\nsimplification model, depending on the task's semantic versus lexico-syntactic\nnature. We also introduce a novel multi-armed bandit based training approach\nthat dynamically learns how to effectively switch across tasks during\nmulti-task learning. Experiments on multiple popular datasets demonstrate that\nour model outperforms competitive simplification systems in SARI and FKGL\nautomatic metrics, and human evaluation. Further, we present several ablation\nanalyses on alternative layer sharing methods, soft versus hard sharing,\ndynamic multi-armed bandit sampling approaches, and our model's learned\nentailment and paraphrasing skills.","url_abs":"http://arxiv.org/abs/1806.07304v1","url_pdf":"http://arxiv.org/pdf/1806.07304v1.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":"multi-task-learning","task_name":"Multi-Task Learning"},{"task_slug":"paraphrase-generation","task_name":"Paraphrase Generation"},{"task_slug":"sentence","task_name":"Sentence"},{"task_slug":"text-simplification","task_name":"Text Simplification"},{"task_slug":null,"task_name":"valid"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/text-simplification-on-newsela","task":"Text Simplification","dataset":"Newsela","model":"Pointer + Multi-task Entailment and Paraphrase Generation","rank_in_archive_order":2,"of":13,"metrics":{"BLEU":"11.14","SARI":"33.22"},"uses_additional_data":false},{"leaderboard":"/sota/text-simplification-on-pwkp-wikismall","task":"Text Simplification","dataset":"PWKP / WikiSmall","model":"Pointer + Multi-task Entailment and Paraphrase Generation","rank_in_archive_order":6,"of":11,"metrics":{"BLEU":"27.23","SARI":"29.58"},"uses_additional_data":false},{"leaderboard":"/sota/text-simplification-on-turkcorpus","task":"Text Simplification","dataset":"TurkCorpus","model":"Pointer + Multi-task Entailment and Paraphrase Generation","rank_in_archive_order":12,"of":25,"metrics":{"BLEU":"81.49","SARI (EASSE>=0.2.1)":"37.45"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1806.07304","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}