[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-513656-105":59,"doc-detail-513656-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","teaching-vision-language-models-to-ask-resolving-ambiguity-in-visual-questions","Teaching Vision-Language Models to Ask - Resolving Ambiguity in Visual Questions","","Teaching vision-language models to ask clarification questions addresses ambiguity in visual question answering where users phrase queries with varying habits. Existing work mainly rephrases questions and ignores interactive clarification via user feedback. Two gaps remain: lack of benchmarks to evaluate interactive ambiguity resolution, and training objectives that favor answering over asking. ClearVQA benchmark targets referential, intent underspecification, and spelling ambiguities, with an automated pipeline generating ambiguity-clarification question pairs. Training on generated data improves the model’s clarification quality and answer specificity using feedback.",{"@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/teaching-vision-language-models-to-ask-resolving-ambiguity-in-visual-questions/513656/",{"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/teaching-vision-language-models-to-ask-resolving-ambiguity-in-visual-questions/513656.png","ImageObject",300,407,{"name":92,"@type":93},"jay","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-16","2026-10-06",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",11,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"Why do ambiguous visual questions cause problems for visual-language models?","Question",{"text":113,"@type":114},"Ambiguous questions arise from differing expression habits and language proficiency, leading models to make undesired or overly general guesses when asked to produce a single answer.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"What are the main challenges for interactive clarification research in VQA?",{"text":118,"@type":114},"There are two major issues: missing benchmarks to evaluate interactive ambiguity resolution capacity and metrics, and the tendency of most VLMs to be trained for answering rather than asking clarification questions.",{"name":120,"@type":111,"acceptedAnswer":121},"How does ClearVQA help models resolve ambiguity interactively?",{"text":122,"@type":114},"ClearVQA provides a benchmark covering three ambiguity categories and multiple VQA scenarios, and it supports an automated pipeline that generates ambiguity-clarification question pairs so models can be trained to ask reasonable clarification questions and then generate more accurate, specific answers from user feedback.","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},513656,1791361541,{"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":46,"language":140,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":67,"update_tm":144,"read_time":31,"preview_page_count":19,"preview_urls":145},3985747858866,"https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd","Teaching Vision-Language Models to Ask: Resolving Ambiguity in Visual Questions  \nPu Jian1,2 , Donglei Yu1,2 , Wen Yang1,2 , Shuo Ren1 , Jiajun Zhang1,2,3 *  \n1Institute of Automation, Chinese Academy of Sciences  \n2 School of Artificial Intelligence, University of Chinese Academy of Sciences  \n3Wuhan AI Research  \n{jianpu2023, yudonglei2021, yangwen2023, [shuo.ren}@ia.ac.cn](shuo.ren}@ia.ac.cn)  \n[jjzhang@nlpr.ia.ac.cn](jjzhang@nlpr.ia.ac.cn)  \nAbstract  \nIn visual question answering (VQA) context, users often pose ambiguous questions to visual language models (VLMs) due to varying expression habits. Existing research addresses such ambiguities primarily by rephrasing questions. These approaches neglect the inherently interactive nature of user interactions with VLMs, where ambiguities can be clarified through user feedback. However, research on interactive clarification faces two major challenges: (1) Benchmarks are absent to assess VLMs’ capacity for resolving ambiguities through interaction; (2) VLMs are trained toprefer answering rather than asking, preventing them from seeking clarification. To overcome these challenges, we introduce ClearVQA benchmark1 , which targets three common categories of ambiguity in VQA context, and encompasses various VQA scenarios. Furthermore, we propose an automated pipeline to generate ambiguity-clarification question pairs.  \nExperimental results demonstrate that training based on the automated generated data enables VLMs to ask reasonable clarification questions, thereby generating more accurate and specific answers based on user feedback.  \n1 Introduction  \nThe visual question answering (VQA) task aims to provide a natural language answer to a question based on a given image (Antol et al., 2015 ; Gao et al., 2022) . Recent studies in VQA field (StengelEskin et al., 2023 ; Prasad et al., 2023) highlight that users often pose ambiguous visual questions due to differences in language proficiency and expression habits. For example, the question in Figure 1 could be asking about the type of \"vehicle\" or\"business\" near the crowd, or it might refer to the \"pizza\" in the foreground. Since the prevailing assumption of VQA task is to return a single  \n* Corresponding Author  \n1The proposed ClearVQA dataset and related codes can be found at [https://github.com/jian0805/ClearVQA](https://github.com/jian0805/ClearVQA)  \nFigure 1: In the traditional VQA context, ambiguous questions may confuse VLMs and lead to undesired answers. In such cases, we emphasize that VLMs should pose clarification questions and generate desired answers based on user feedback.  \nanswer, when directly addressing ambiguous questions, even state-of-the-art large visual-language models (VLMs) (Liu et al., 2024a) that excel in visual understanding and cross-modal reasoning may generate undesired responses.  \nSeveral studies have investigated how VLMs address ambiguous instructions or questions. Pezzelle (2023) emphasize that the ambiguity of instructions is a significant source of errors in visual-language tasks. One feasible solution to such ambiguity is additional pre-training to align ambiguous text with images better, enabling VLMs to interpret such questions in a more human-like manner (Sun et al.,  \n3619  \nProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages 3619–3638 July 27-August 1, 2025 ©2025 Association for Computational Linguistics  \n2024 ; Lu et al., 2023) . However, this approach leads to significant computational costs. StengelEskin et al. (2023) introduce a question generation model that rephrases ambiguous visual questions based on images to reduce ambiguity. Similarly, Prasad et al. (2023) leverage well-trained VLMs to extract visual details omitted from the questions and employ a large language model (LLM) torephrase ambiguous questions in a zero-shot manner, thus improving VQA accuracy.  \nDespite their contributions, these studies focus solely ","cbCaiufC4lP2tGcR","https://ap.wps.com/l/cbCaiufC4lP2tGcR","pdf",4618773,"English","# Abstract\n# Introduction\n# ClearVQA Benchmark\n## Task Formulation","[{\"question\":\"Why do ambiguous visual questions cause problems for visual-language models?\",\"answer\":\"Ambiguous questions arise from differing expression habits and language proficiency, leading models to make undesired or overly general guesses when asked to produce a single answer.\"},{\"question\":\"What are the main challenges for interactive clarification research in VQA?\",\"answer\":\"There are two major issues: missing benchmarks to evaluate interactive ambiguity resolution capacity and metrics, and the tendency of most VLMs to be trained for answering rather than asking clarification questions.\"},{\"question\":\"How does ClearVQA help models resolve ambiguity interactively?\",\"answer\":\"ClearVQA provides a benchmark covering three ambiguity categories and multiple VQA scenarios, and it supports an automated pipeline that generates ambiguity-clarification question pairs so models can be trained to ask reasonable clarification questions and then generate more accurate, specific answers from user feedback.\"}]","Teaching Vision-Language Models to Ask - Resolving Ambiguity in Visual Questions | PDF",1791288554,[146,150,154,158],{"page_no":8,"url":147,"width":148,"height":149},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/9ab76d0114c874a7245d1165478f8469.webp",983,1390,{"page_no":14,"url":151,"width":152,"height":153},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/9483b08cc3e47f30d6ea2eedb4a64543.webp",982,1388,{"page_no":81,"url":155,"width":156,"height":157},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/fc429a361740c2ddbb57d56dea79d3c9.webp",981,1387,{"page_no":19,"url":159,"width":160,"height":161},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/356c7b64169df5e48232885e2be20547.webp",980,1385]