[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-160437-en":3,"doc-seo-160437-105":31,"detail-sidebar-cat-0-en-105":93},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":28,"seo_description":14,"update_tm":29,"read_time":30},160437,5909887254083,"\tWilliam","https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c",8,"Research & Report","ThinkStruct - RST-Aware Attention for Logical Reasoning in Machine Reading Comprehension - Research paper","Logical reasoning improves Machine Reading Comprehension by building logical structures in natural language, yet existing methods often misdivide logical units and produce inconsistent predictions for equivalent meanings. This paper proposes ThinkStruct, a transformer enhanced with Rhetorical Structure (RS) relations. RST splits text into Elementary Discourse Units (EDUs) and encodes relationships via an adjacency matrix, then integrates features for answer prediction. Contrastive learning further strengthens EDU relation understanding, outperforming state-of-the-art models on LogiQA and Reclor.","ThinkStruct: RST-Aware Attention for Logical Reasoning in Machine Reading Comprehension  \nNhi Thao Tran1,2 and Tien Le1,2 and Toan Pham1,2 and Quoc Hoang Vu1,2  \n1Faculty of Information Technology, University of Science, Ho Chi Minh City, Vietnam,  \n2Vietnam National University, Ho Chi Minh City, Vietnam, [tttnhi@fit.hcmus.edu.vn](tttnhi@fit.hcmus.edu.vn) , {lpttien21,[pktoan21}@clc.fitus.edu.vn](pktoan21}@clc.fitus.edu.vn) ,  \n[vqhoang@fit.hcmus.edu.vn](vqhoang@fit.hcmus.edu.vn)  \n[Correspondence:](Correspondence: vqhoang@fit.hcmus.edu.vn)[ vqhoang@fit.hcmus.edu.vn](Correspondence: vqhoang@fit.hcmus.edu.vn)  \nAbstract  \nLogical Reasoning is a novel approach to deal with challenging Machine Reading Comprehension tasks by utilizing the ability to construct logical structures in natural language. However, previous promising studies struggle with the accuracy of logical unit division and the consistency of model prediction on equivalent semantics. In this paper, we propose ThinkStruct, a new method that leverages a transformer network enhanced with the information of Rhetorical Structure (RS) relations for logical reasoning. Specifically, our method uses Rhetorical Structure Theory (RST) to split natural language text into Elementary Discourse Units (EDUs) and identify the relationship among these units. Node information is then fed into  \nthe fully connected transformer network, which is enhanced with logical relationships among the extracted units via adjacency matrix. Subsequently, the features of the transformer network are integrated before being passed into the answer prediction module. In addition, we employ a contrastive learning module for improving its understanding of the relationship between Elementary Discourse Units. Our experiments on the LogiQA and Reclor datasets demonstrate that our results outperform other state-of-the-art models.  \n1 Introduction  \nMachine Reading Comprehension (MRC), which facilitates machines in understanding natural language text, is a major focus (Huang et al., 2021 ; Liet al., 2022 ; Ouyang et al., 2021 ; Wang et al., 2021 ; Jiao et al., 2022) in Natural Language Processing (NLP) . However, traditional MRC models have unsatisfactory performance in many datasets that require a wider and deeper range of logical methods, such as LogiQA (Liu et al., 2020) and Reclor (Yu et al., 2020) due to the lack of robust logical reasoning ability. To address this limitation, logical reasoning in MRC emerged as a new approach.  \nFigure 1: An example of input to the RST parser and the subsequent graph construction process for Logical Reasoning. The green boxes represent EDUs segmented from the context, while the blue boxes represent EDUs from the option. The RS tree (left) is constructed from both context and option using an RST parser, where nodes with red numbers are nucleus and nodes with gray numbers are satellite. The right figure shows the RS-enhanced graph, in which additional edges are added for better information flow.  \nThis approach integrates comprehension ability by analyzing and evaluating the relationships among facts in natural language passage (Ouyang et al., 2023) . The LogiQA example shown in Fig 1, like previous MRC models, requires the context, question, and options to generate a predicted score for the option. A unique characteristic of these texts is the presence of both implicit and explicit logical structures across semantic meaning. In Fig 1 (a1 ), sentences are divided into smaller units (e.g., E1 − E12 ) by some explicit connective words or punctuation. Other information, which is presented in Fig 1 (b), includes logical relationships among Elementary Discourse Units (EDUs) . These relationships are encoded numerically, for instance: 12 represents Condition relation, 17 indicates CauseEffect relation.  \nWhile recent generative Large Language Models (LLMs) such as the GPT (Ouyang et al., 2022 ;  \n471  \nProceedings of the 30th Conference on Computational Natural Language Learning,","cbCaigItiaGDZh6r","https://ap.wps.com/l/cbCaigItiaGDZh6r","pdf",591873,5,1,10,"English","en",105,"# Abstract\n# 1 Introduction\n## Machine Reading Comprehension and logical reasoning motivation\n## RST-based EDU splitting and RS-enhanced graph construction\n## Limitations of prior models","[{\"question\":\"What problem does ThinkStruct target in logical reasoning for machine reading comprehension?\",\"answer\":\"It addresses inaccurate logical-unit division and inconsistent model predictions when semantic meanings are equivalent.\"},{\"question\":\"How does ThinkStruct represent logical structure within the model?\",\"answer\":\"It uses Rhetorical Structure Theory to split text into Elementary Discourse Units (EDUs) and encodes relations via an adjacency matrix that enhances transformer attention.\"},{\"question\":\"What evidence shows ThinkStruct improves performance?\",\"answer\":\"Experiments on LogiQA and Reclor demonstrate results that outperform other state-of-the-art models.\"}]","ThinkStruct - RST-Aware Attention for Logical Reasoning in Machine Reading Comprehension - Research paper | PDF",1788063398,25,{"code":4,"msg":32,"data":33},"ok",{"site_id":25,"language":24,"slug":34,"title":13,"keywords":35,"description":14,"schema_data":36,"social_meta":88,"head_meta":90,"extra_data":92,"updated_unix":29},"thinkstruct-rst-aware-attention-for-logical-reasoning-in-machine-reading-comprehension-research-paper","",{"@graph":37,"@context":87},[38,55,70],{"@type":39,"itemListElement":40},"BreadcrumbList",[41,45,49,52],{"item":42,"name":43,"@type":44,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":46,"name":47,"@type":44,"position":48},"https://docshare.wps.com/document/","Document",2,{"item":50,"name":12,"@type":44,"position":51},"https://docshare.wps.com/document/research-report/",3,{"item":53,"name":13,"@type":44,"position":54},"https://docshare.wps.com/document/thinkstruct-rst-aware-attention-for-logical-reasoning-in-machine-reading-comprehension-research-paper/160437/",4,{"url":53,"name":13,"@type":56,"author":57,"headline":13,"publisher":59,"fileFormat":62,"inLanguage":24,"description":14,"dateModified":63,"datePublished":64,"encodingFormat":62,"isAccessibleForFree":65,"interactionStatistic":66},"DigitalDocument",{"name":9,"@type":58},"Person",{"url":42,"name":60,"@type":61},"DocShare","Organization","application/pdf","2026-09-05","2026-08-30",true,{"@type":67,"interactionType":68,"userInteractionCount":20},"InteractionCounter",{"@type":69},"ViewAction",{"@type":71,"mainEntity":72},"FAQPage",[73,79,83],{"name":74,"@type":75,"acceptedAnswer":76},"What problem does ThinkStruct target in logical reasoning for machine reading comprehension?","Question",{"text":77,"@type":78},"It addresses inaccurate logical-unit division and inconsistent model predictions when semantic meanings are equivalent.","Answer",{"name":80,"@type":75,"acceptedAnswer":81},"How does ThinkStruct represent logical structure within the model?",{"text":82,"@type":78},"It uses Rhetorical Structure Theory to split text into Elementary Discourse Units (EDUs) and encodes relations via an adjacency matrix that enhances transformer attention.",{"name":84,"@type":75,"acceptedAnswer":85},"What evidence shows ThinkStruct improves performance?",{"text":86,"@type":78},"Experiments on LogiQA and Reclor demonstrate results that outperform other state-of-the-art models.","https://schema.org",{"og:url":53,"og:type":89,"og:title":13,"og:site_name":60,"og:description":14},"article",{"robots":91,"canonical":53},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":94},[95,99,103,107,111,116,121,124,129,132,135],{"id":21,"doc_module":4,"doc_module_name":47,"category_name":96,"show_sort_weight":97,"slug":98},"Story & Novel",90,"story-novel",{"id":48,"doc_module":4,"doc_module_name":47,"category_name":100,"show_sort_weight":101,"slug":102},"Literature",80,"literature",{"id":54,"doc_module":4,"doc_module_name":47,"category_name":104,"show_sort_weight":105,"slug":106},"Exam",70,"exam",{"id":20,"doc_module":4,"doc_module_name":47,"category_name":108,"show_sort_weight":109,"slug":110},"Comic",60,"comic",{"id":112,"doc_module":4,"doc_module_name":47,"category_name":113,"show_sort_weight":114,"slug":115},6,"Technology",50,"technology",{"id":117,"doc_module":4,"doc_module_name":47,"category_name":118,"show_sort_weight":119,"slug":120},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":47,"category_name":12,"show_sort_weight":122,"slug":123},30,"research-report",{"id":125,"doc_module":4,"doc_module_name":47,"category_name":126,"show_sort_weight":127,"slug":128},9,"Religion & Spirituality",20,"religion-spirituality",{"id":127,"doc_module":4,"doc_module_name":47,"category_name":130,"show_sort_weight":127,"slug":131},"World Cup","world-cup",{"id":22,"doc_module":4,"doc_module_name":47,"category_name":133,"show_sort_weight":22,"slug":134},"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":47,"category_name":137,"show_sort_weight":20,"slug":138},19,"General","general"]