[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-515143-105":59,"doc-detail-515143-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","protrix-building-models-for-planning-and-reasoning-over-tables-with-sentence-context","PROTRIX: Building Models for Planning and Reasoning - over Tables with Sentence Context","","Tables serve as a central way to present information across domains, yet table-based question answering is often slow for humans due to large content volumes. PROTRIX proposes a Plan-then-Reason framework that plans reasoning paths using both tabular and sentence context, then assigns each step to program-based or textual reasoning. Experiments with GPT-3.5-Turbo and a finetuned PROTRIX family show improved table reasoning, strong generalization to unseen tasks with 6k instances, and faithful explanations for complex free-form queries.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/protrix-building-models-for-planning-and-reasoning-over-tables-with-sentence-context/515143/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/protrix-building-models-for-planning-and-reasoning-over-tables-with-sentence-context/515143.png","ImageObject",300,407,{"name":92,"@type":93},"Aran","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-15","2026-10-06",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",15,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What is the core idea of PROTRIX?","Question",{"text":113,"@type":114},"PROTRIX introduces a Plan-then-Reason framework that first plans reasoning paths over both tables and sentence context, then chooses program-based or textual reasoning per step.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"How does PROTRIX combine tabular and sentence context?",{"text":118,"@type":114},"The model extracts needed information from the table and uses sentence context as evidence to support conclusions, including cases requiring commonsense knowledge.",{"name":120,"@type":111,"acceptedAnswer":121},"What evidence does the document provide about effectiveness?",{"text":122,"@type":114},"Experiments report that GPT-3.5-Turbo with the Plan-then-Reason framework outperforms prompting baselines without self-consistency, while finetuned PROTRIX models generalize to unseen tabular tasks with only 6k training instances and generate accurate explanations.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},515143,1791392048,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":140,"language":141,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":67,"update_tm":145,"read_time":146,"preview_page_count":19,"preview_urls":147},137455076865,"https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0","PROTRIX: Building Models for Planning and Reasoning  \nover Tables with Sentence Context  \nZirui Wu, Yansong Feng􀀌  \nWangxuan Institute of Computer Technology, Peking University, China {ziruiwu, [fengyansong}@pku.edu.cn](fengyansong}@pku.edu.cn)  \nAbstract  \nTables play a crucial role in conveying information of various domains. We propose a novel Plan-then-Reason framework to answer different types of user queries over tables with sentence context. Our framework first plans the reasoning paths over both tabular and sentence context, then assigns each step to programbased or textual reasoning to reach the final answer. This framework enhances the table reasoning abilities for both in-context learning and finetuning methods. GPT-3.5-Turbo following our Plan-then-Reason framework surpasses other prompting baselines that do not employ self-consistency while using fewer API calls and in-context demonstrations. We also construct an instruction tuning set TrixInstruct to evaluate the effectiveness of finetuning with this framework. We present a PROTRIX model family by finetuning models on TrixInstruct. Our experiments show that the PROTRIX family can generalize to diverse unseen tabular tasks with only 6k training instances. We further demonstrate that PROTRIX can generate accurate and faithful explanations to answer complex free-form questions. Our work underscores the importance of the planning and reasoning abilities towards a model over tabular tasks with generalizability and interpretability. We open-source our dataset and models at [https://github.com/WilliamZR/ProTrix](https://github.com/WilliamZR/ProTrix).  \n1 Introduction  \nTables, serve as a fundamental tool for organizing and presenting information across various domains. Whether in business reports, or scientific publications, tables are commonly employed to represent complex data effectively. Despite their widespread utility, the process of human beings answering questions involving tables appears tobe time-consuming, given the often substantial amount of content involved. Recognizing this challenge, there arises a need to leverage Large Lan-  \nFigure 1: Demonstrations of user queries to a table in Wikipedia. Some of the sentences with hyperlinks to the table are presented as sentence context.  \nguage Models (LLMs) to understand and respond to user queries automatically.  \nFigure 1 demonstrates three kinds of user queries for a table from Wikipedia. In the first example, the user query is how many medals were won by Kenya. This question is annotated as a programunsolvable question by SQL experts (Shi et al., 2020) attributed to the absence of an explicit column for medals in the table. To resolve this, the model must fill the gap between the query and the  \n4378  \nFindings of the Association for Computational Linguistics: EMNLP 2024 , pages 4378–4406 November 12-16, 2024 ©2024 Association for Computational Linguistics  \nFigure 2: Demonstration of disadvantages of program-based and textual reasoning on tabular tasks. Program-based reasoning fails to answer the query since it tries to compare general concepts with a math operator. The textual reasoning fails on a program-solvable query that needs to count distinct countries in the table.  \ntable by recalling the commonsense knowledge that only the top three players can win medals.  \nThe second query delves into a multi-hop scenario asking whether the silver medalist at the 2001 Goodwill Games has ever won an Olympic gold medal. Addressing such queries raises two challenges (1) Decompose the query into sub-tasks. Such as the model plans to derive the silver medalists first and then verify their record of Olympic medals. (2) Combining structured and unstructured context. The model must extract the athletic name from the table and derive the information from the sentence context that Kosgei has won an Olympic gold medal since he is the youngest ever winner of an Olympic gold medal.  \nThe last query also requires the model to ","cbCaiaItWmH5LO1R","https://ap.wps.com/l/cbCaiaItWmH5LO1R","pdf",846058,29,"English","# Introduction\n## Tables and table-based question answering challenges\n## Textual vs program-based reasoning\n## Dynamic planning vs global strategy planning","[{\"question\":\"What is the core idea of PROTRIX?\",\"answer\":\"PROTRIX introduces a Plan-then-Reason framework that first plans reasoning paths over both tables and sentence context, then chooses program-based or textual reasoning per step.\"},{\"question\":\"How does PROTRIX combine tabular and sentence context?\",\"answer\":\"The model extracts needed information from the table and uses sentence context as evidence to support conclusions, including cases requiring commonsense knowledge.\"},{\"question\":\"What evidence does the document provide about effectiveness?\",\"answer\":\"Experiments report that GPT-3.5-Turbo with the Plan-then-Reason framework outperforms prompting baselines without self-consistency, while finetuned PROTRIX models generalize to unseen tabular tasks with only 6k training instances and generate accurate explanations.\"}]","PROTRIX: Building Models for Planning and Reasoning - over Tables with Sentence Context | PDF",1791308146,73,[148,152,156,160],{"page_no":8,"url":149,"width":150,"height":151},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/ac487064852a47e9fc28976579640ba6.webp",983,1390,{"page_no":14,"url":153,"width":154,"height":155},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/5c4f8924d8f4a817cd1dc44942673ddf.webp",982,1388,{"page_no":81,"url":157,"width":158,"height":159},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/837ad217d43b3d8bc8f767c2bb6801d3.webp",981,1387,{"page_no":19,"url":161,"width":162,"height":163},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/cf7f4f94d431f3b97eba98fa34e64f26.webp",980,1385]