[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-461773-105":59,"doc-detail-461773-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","sommbench-assessing-sommelier-expertise-of-language-models","SommBench - Assessing Sommelier Expertise of Language Models","","SommBench systematically evaluates how well large language models emulate expert sommelier judgment across multilingual settings grounded in smell and taste. The benchmark tests whether textual descriptions provide sufficient grounding for sensory reasoning and cross-lingual consistency, separating language skill from domain expertise. It introduces three tasks—Wine Theory Question Answering, Wine Feature Completion, and Food-Wine Pairing—spanning eight languages and thousands of items curated with experts and native speakers. Results show strong performance on theory questions, while feature completion and pairing remain substantially more difficult.",{"@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/sommbench-assessing-sommelier-expertise-of-language-models/461773/",{"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/sommbench-assessing-sommelier-expertise-of-language-models/461773.png","ImageObject",300,407,{"name":92,"@type":93},"pop out","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-31","2026-09-30",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",12,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What is SommBench designed to measure in language models?","Question",{"text":113,"@type":114},"SommBench measures multilingual sommelier-domain expertise, testing whether models can use textual grounding to emulate expert-level judgments and maintain cross-lingual consistency.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"What are the three tasks in the SommBench benchmark?",{"text":118,"@type":114},"The benchmark comprises Wine Theory Question Answering (WTQA), Wine Feature Completion (WFC), and Food-Wine Pairing (FWP). Each task targets different aspects of wine knowledge and sensory or cultural reasoning.",{"name":120,"@type":111,"acceptedAnswer":121},"How do model performances differ across the SommBench tasks?",{"text":122,"@type":114},"The strongest models perform well on wine theory question answering, but feature completion and food-wine pairing are harder, showing lower peak accuracy and weaker performance metrics overall.","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},461773,1790844092,{"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},962090760408,"https://ap-avatar.wpscdn.com/davatar_085a072bc5b1113ac321206ff7593b45","SommBench: Assessing Sommelier Expertise of Language Models  \nWilliam Brach 1 ,2 , Tomas Bedej2 , Jacob Nielsen3 , Jacob Pichna2 ,  \nJuraj Bedej2 , Eemeli Saarensilta2 , Julie Dupouy2 , Gianluca Barmina3 , Andrea Blasi Núñez3 ,  \nPeter Schneider-Kamp3 , Kristian Košťál 1 , Michal Ries 1 , Lukas Galke Poech3  \n1 Slovak University of Technology, Bratislava, Slovakia  \n2 sommify, Helsinki, Finland  \n3 University of Southern Denmark, Odense, Denmark  \n[Correspondence : william.brach@stuba.sk](Correspondence : william.brach@stuba.sk)  \nAbstract  \nWith the rapid advances of large language models, it becomes increasingly important to systematically evaluate their multilingual and multicultural capabilities. Previous cultural evaluation benchmarks focus mainly on basic cultural knowledge that can be encoded in linguistic form. Here, we propose SommBench, a multilingual benchmark to assess sommelier expertise, a domain deeply grounded in the senses of smell and taste. While language models learn about sensory properties exclusively through textual descriptions, SommBench tests whether this textual grounding is sufficient to emulate expert-level sensory judgment. SommBench comprises three main tasks: Wine Theory Question Answering (WTQA), Wine Feature Completion (WFC), and Food-Wine Pairing (FWP) . SommBench is available in multiple languages: English, Slovak, Swedish, Finnish, German, Danish, Italian, and Spanish. This helps separate a language model’s wine expertise from its language skills. The benchmark datasets were developed in close collaboration with a professional sommelier and native speakers of the respective languages, resulting in 1,024 questions for wine theory question answering, 1,000 examples for wine feature completion, and 1,000 examples of food-wine pairing. We provide results for the most popular language models, including closed-weights models such as Gemini 2.5, and open-weights models, such as GPT-OSS and Qwen 3 . Our results show that the most capable models perform well on wine theory question answering (up to 97% correct with a closed-weights model), yet feature completion (peaking at 65%) and food-wine pairing show (MCC ranging between 0 and 0 .39) turn out to be more challenging. These results position SommBench as an interesting and challenging benchmark for evaluating the sommelier expertise of language models. The benchmark is publicly available at [https://github.com/sommify/sommbench](https://github.com/sommify/sommbench).  \nKeywords: Multilinguality, Question Answering, Language Modelling, Corpus (Creation, Annotation, etc.)  \n1. Introduction  \nLarge language models achieve increasingly strong performance on multilingual benchmarks (Pomerenke et al. , 2025 ; Hendrycks et al. , 2021), but a key open question concerns whether language models display consistent competencies in culturally-grounded, expert-level knowledge across languages, or whether they exhibit languagedependent behavior that reflects the cultural contexts embedded in their training data. This question matters as LLMs are deployed globally, where users interact in multiple languages and expect coherent, reliable answers (Ni et al. , 2025) .  \nWe investigate this question within the domain of sommelier expertise, a field rich with sensory, factual, and cultural knowledge (Chen , 2022 ; Eschevins et al. , 2019 ; Manske and Cordua , 2005) . Covering eight languages, our work examines whether a model’s understanding of wine production methods, regional classifications, food pairings, or varietal properties changes depending on the language used in the prompt, i.e., when a model is asked in German or in Spanish about the characteristics of a Grüner Veltliner, does it provide con-  \nsistent information, or does uneven training data across languages lead to contradictory responses?  \nWhile some tasks like food-wine pairing are inherently culturally situated, the foundational knowledge underlying them should remain consistent across languages (Li","cbCaiqa5MhA8pZVL","https://ap.wps.com/l/cbCaiqa5MhA8pZVL","pdf",877027,17,"English","# Introduction\n## Three SommBench Tasks\n### Wine Theory Question Answering (WTQA)\n### Wine Feature Completion (WFC)\n### Food-Wine Pairing (FWP)","[{\"question\":\"What is SommBench designed to measure in language models?\",\"answer\":\"SommBench measures multilingual sommelier-domain expertise, testing whether models can use textual grounding to emulate expert-level judgments and maintain cross-lingual consistency.\"},{\"question\":\"What are the three tasks in the SommBench benchmark?\",\"answer\":\"The benchmark comprises Wine Theory Question Answering (WTQA), Wine Feature Completion (WFC), and Food-Wine Pairing (FWP). Each task targets different aspects of wine knowledge and sensory or cultural reasoning.\"},{\"question\":\"How do model performances differ across the SommBench tasks?\",\"answer\":\"The strongest models perform well on wine theory question answering, but feature completion and food-wine pairing are harder, showing lower peak accuracy and weaker performance metrics overall.\"}]","SommBench - Assessing Sommelier Expertise of Language Models | PDF",1790762386,43,[148,152,156,160],{"page_no":8,"url":149,"width":150,"height":151},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/697073d6cd6d16d70c7dd5f8cc46f365.webp",983,1390,{"page_no":14,"url":153,"width":154,"height":155},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/31664ffcccfa0aa18d0b96b615ea4bc3.webp",982,1388,{"page_no":81,"url":157,"width":158,"height":159},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/767463c0b14cce607e6f844d4edcf9aa.webp",981,1387,{"page_no":19,"url":161,"width":162,"height":163},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/0d9c7c4bb4d377ca0a48fc76d0bb316f.webp",980,1385]