[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-346097-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-346097-en":130},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","ocr-error-post-correction-with-llms-in-historical-documents-no-free-lunches","OCR Error Post-Correction with LLMs in Historical Documents - No Free Lunches","","Optical Character Recognition (OCR) systems often introduce transcription errors in historical documents, creating noise that limits downstream digital research. This study evaluates open-weight large language models for OCR post-correction on historical English and Finnish datasets. It tests parameter optimization, quantization, segment-length effects, and text-continuation strategies. Results show LLMs reduce character error rates in English, while practical gains for Finnish remain unmet. The work clarifies both potential and limits for scaling post-correction to large corpora.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/ocr-error-post-correction-with-llms-in-historical-documents-no-free-lunches/346097/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/ocr-error-post-correction-with-llms-in-historical-documents-no-free-lunches/346097.png","ImageObject",300,407,{"name":42,"@type":43},"Bulrr","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-11-06","2026-09-22",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",10,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"Why is OCR post-correction needed for historical documents?","Question",{"text":63,"@type":64},"OCR transcription introduces noise whose severity depends on source quality and OCR technology. Post-correction improves text quality without repeating full digitization and transcription workflows.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"Which strategies are evaluated for LLM-based OCR error correction?",{"text":68,"@type":64},"The study tests hyperparameter optimization, quantization, input segment length effects, text continuation methods, and additional output post-processing and correction strategies.",{"name":70,"@type":61,"acceptedAnswer":71},"What are the main findings for English versus Finnish datasets?",{"text":72,"@type":64},"Modern LLMs show promise for reducing character error rates in English. For Finnish, the study reports that practically useful performance was not reached.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},346097,1791162761,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,115,120,123,126],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":107,"doc_module":4,"doc_module_name":25,"category_name":108,"show_sort_weight":109,"slug":110},7,"Healthcare",40,"healthcare",{"id":112,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":113,"slug":114},8,30,"research-report",{"id":116,"doc_module":4,"doc_module_name":25,"category_name":117,"show_sort_weight":118,"slug":119},9,"Religion & Spirituality",20,"religion-spirituality",{"id":118,"doc_module":4,"doc_module_name":25,"category_name":121,"show_sort_weight":118,"slug":122},"World Cup","world-cup",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":55,"slug":125},"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":97,"slug":129},19,"General","general",{"code":4,"msg":82,"data":131},{"doc_id":79,"user_id":132,"nickname":42,"user_avatar":133,"doc_module":4,"category_id":112,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":55,"language":139,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":140,"faqs":141,"seo_title":142,"seo_description":12,"update_tm":143,"read_time":144,"preview_page_count":33,"preview_urls":145},5909892115043,"https://ap-avatar.wpscdn.com/davatar_6f874abed73319feea01a86fa6f0fab8","OCR Error Post-Correction with LLMs in Historical Documents: No Free  \nLunches  \nJenna Kanerva, Cassandra Ledins, Siiri Käpyaho, and Filip Ginter  \nTurkuNLP, Department of Computing  \nUniversity of Turku, Finland  \n{jmnybl, caledi, siakap, figint}@utu.fi  \nAbstract  \nOptical Character Recognition (OCR) systems often introduce errors when transcribing historical documents, leaving room for post-correction to improve text quality. This study evaluates the use of openweight LLMs for OCR error correction in historical English and Finnish datasets. We explore various strategies, including parameter optimization, quantization, segment length effects, and text continuation methods. Our results demonstrate that while modern LLMs show promise in reducing character error rates (CER) in English, a practically useful performance for Finnish was not reached. Our findings highlight the potential and limitations of LLMs in scaling OCR post-correction for large historical corpora.  \n1 Introduction  \nDigitizing and transcribing historical documents and literature is vital for preserving our cultural heritage and making it accessible for modern digital research methods. The transcription process relies on OCR, which naturally introduces noise into the output. The noise varies in severity depending on the quality of the source material and the OCR technology used, impacting the research usage of the data (Chiron et al., 2017b) . The OCR output at two noise levels is illustrated in Figure 1 . Although modern OCR systems are becoming increasingly accurate (Li et al., 2023), reprocessing large collections of historical literature remains a significant challenge, as the resources available to the institutions maintaining these collections are often insufficient for such an undertaking. Consequently, OCR error post-correction has been suggested as means of improving the historical collections without the need to repeat the whole transcription process (Nguyen et al., 2021) .  \nMild noise (0.04 CER):  \nA work of art, (be it what it may, house, pi&ure, book, or garden,) however beautiful in it ' s underparts, loses half it ' s value, if the gneralfcope  \nof it be not obvio ' ,s to conception.  \nSevere noise (0.19 CER):  \nbke up at Sx in the Mo . r aig. ll the eauing Withr he went from Cbaud to Cbhh every Suday, «d from Play . bote~PIOoaBcu evi Niuht m the Week, but vd  \nFigure 1: Example extracts of texts at two different OCR noise levels from the ECCO dataset of 18th century literature.  \nRecent studies (Boros et al., 2024; Bourne, 2024) have proposed the application of LLMs to the task, with varying degrees of success. Currently, there is no clear consensus as to how LLMs can be applied to the task and how to deal with the various methodological challenges it poses. Our objective is to address some of these challenges as well as to assess several open LLMs to correct OCR-generated text when prompted to. We study hyperparameter optimization, quantization levels, input lengths, output post-processing and several novel correction methods so as to benchmark and improve the LLM performance on this task.  \nAs our long-term goal is to post-correct two large historical datasets, one in English and the other in Finnish, we focus on these two languages as well as open-weight LLMs, since post-correction of large datasets with commercial models is infeasible costwise.  \n2 Related work  \nDespite decades of active research, post-correction of historical documents remains a challenge. The ICDAR 2017 and 2019 shared tasks (Chiron et al.,  \n38  \nProceedings of the Third Workshop on Resources and Representations for Under-Resourced Languages and Domains (RESOURCEFUL 2025) , pages 38–47  \nMarch 2, 2025 ©2025 Association for Computational Linguistics  \n2017a; Rigaud et al., 2019) addressed the lack of adequate benchmarks for evaluating OCR performance across several languages, introducing two tracks: detecting OCR errors, and correcting previously detected errors. This s","cbCaiea2syQeVZSS","https://ap.wps.com/l/cbCaiea2syQeVZSS","pdf",608244,"English","# 1 Introduction\n# 2 Related work","[{\"question\":\"Why is OCR post-correction needed for historical documents?\",\"answer\":\"OCR transcription introduces noise whose severity depends on source quality and OCR technology. Post-correction improves text quality without repeating full digitization and transcription workflows.\"},{\"question\":\"Which strategies are evaluated for LLM-based OCR error correction?\",\"answer\":\"The study tests hyperparameter optimization, quantization, input segment length effects, text continuation methods, and additional output post-processing and correction strategies.\"},{\"question\":\"What are the main findings for English versus Finnish datasets?\",\"answer\":\"Modern LLMs show promise for reducing character error rates in English. For Finnish, the study reports that practically useful performance was not reached.\"}]","OCR Error Post-Correction with LLMs in Historical Documents - No Free Lunches | PDF",1790060075,25,[146,150,154,158],{"page_no":22,"url":147,"width":148,"height":149},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/672f3ed779d153691ce9376cc7a6e4eb.webp",983,1390,{"page_no":26,"url":151,"width":152,"height":153},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/b1fb17815b16fc2f35d7fb6ddccce0a7.webp",982,1388,{"page_no":30,"url":155,"width":156,"height":157},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/a5b6790060b72a1fd2518327790d32fe.webp",981,1387,{"page_no":33,"url":159,"width":160,"height":161},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/170e073c700e9cd402e55ddbf076eba7.webp",980,1385]