{"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/multi-range-reasoning-for-machine","title":"Multi-range Reasoning for Machine Comprehension","arxiv_id":"1803.09074","date":"2018-03-24","proceeding":null,"authors":["Yi Tay","Luu Anh Tuan","Siu Cheung Hui"],"abstract":"We propose MRU (Multi-Range Reasoning Units), a new fast compositional\nencoder for machine comprehension (MC). Our proposed MRU encoders are\ncharacterized by multi-ranged gating, executing a series of parameterized\ncontract-and-expand layers for learning gating vectors that benefit from long\nand short-term dependencies. The aims of our approach are as follows: (1)\nlearning representations that are concurrently aware of long and short-term\ncontext, (2) modeling relationships between intra-document blocks and (3) fast\nand efficient sequence encoding. We show that our proposed encoder demonstrates\npromising results both as a standalone encoder and as well as a complementary\nbuilding block. We conduct extensive experiments on three challenging MC\ndatasets, namely RACE, SearchQA and NarrativeQA, achieving highly competitive\nperformance on all. On the RACE benchmark, our model outperforms DFN (Dynamic\nFusion Networks) by 1.5%-6% without using any recurrent or convolution layers.\nSimilarly, we achieve competitive performance relative to AMANDA on the\nSearchQA benchmark and BiDAF on the NarrativeQA benchmark without using any\nLSTM/GRU layers. Finally, incorporating MRU encoders with standard BiLSTM\narchitectures further improves performance, achieving state-of-the-art results.","url_abs":"http://arxiv.org/abs/1803.09074v1","url_pdf":"http://arxiv.org/pdf/1803.09074v1.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":"reading-comprehension","task_name":"Reading Comprehension"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/question-answering-on-race","task":"Question Answering","dataset":"RACE","model":"BiAttention MRU","rank_in_archive_order":5,"of":7,"metrics":{"RACE":"53.3","RACE-h":"50.3","RACE-m":"60.2"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}