[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-273560-105":59,"doc-detail-273560-en":130},{"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":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","temporal-multi-signal-fusion-for-token-level-hallucination-detection","Temporal Multi-Signal Fusion for Token-Level Hallucination Detection","","Token-level hallucination detectors score each token independently from a single signal, and fail when a language model is confidently wrong. This work models hallucination as a temporally extended span and performs sequence labeling with a 33-dimensional feature stream that fuses text statistics, Natural Language Inference entailment, and language-model surprisal, without any access to model internals. A BiGRU labeler reaches 0.840 AUC on RAGTruth (10 seeds), with most gains explained by temporal order and span-level evidence propagation; temporal fusion also transfers across architectures and closed-source models.",{"@graph":69,"@context":122},[70,84,105],{"@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/temporal-multi-signal-fusion-for-token-level-hallucination-detection/273560/",{"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/temporal-multi-signal-fusion-for-token-level-hallucination-detection/273560.png","ImageObject",300,407,{"name":92,"@type":93},"Noah","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-16","2026-09-15",true,{"@type":102,"interactionType":103,"userInteractionCount":52},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"How does the proposed method detect hallucinations at the token level?","Question",{"text":112,"@type":113},"It treats hallucinations as spans over time and uses sequence labeling, scoring each token via features that fuse text statistics, NLI entailment, and language-model surprisal, then modeling temporal dynamics with a BiGRU.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"What signals are fused in the 33-dimensional feature stream?",{"text":117,"@type":113},"The feature vector fuses three families: text statistics (including novelty and context overlap dynamics), NLI entailment from DeBERTa, and language-model surprisal from TinyLlama, each enriched with temporal derivatives.",{"name":119,"@type":110,"acceptedAnswer":120},"Why does temporal modeling improve detection compared to independent per-token classifiers?",{"text":121,"@type":113},"Most errors evolve jointly over spans because hallucinated tokens become context for subsequent generation; the method captures this via temporal order and evidence propagation from confident positions to ambiguous neighbors.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},273560,1789472045,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":52,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":139,"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":129,"read_time":144,"preview_page_count":4},8796095462418,"https://ap-avatar.wpscdn.com/avatar/80000253c1241d02b47?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778826106357471780","Temporal Multi-Signal Fusion for Token-Level Hallucination Detection  \nIgor Itkin  \nIndependent Researcher  \n[ig.itkin@gmail.com](ig.itkin@gmail.com)  \nORCID: 0009-0004-9513-8463  \nPreprint, June 2026  \narXiv : 2608 . 18 1 15v 1 [cs .CL] 30 Jun 2026  \nAbstract  \nToken-level hallucination detectors score each token independently from a single signal, and fail exactly when the generating model is confidently wrong. This paper instead treats hallucination as a temporally extended span and detects it by sequence labeling: each token is scored from a 33-dimensional feature stream that fuses text statistics, Natural Language Inference (NLI) entailment, and language model surprisal, with no access to model internals. ABidirectional Gated Recurrent Unit (BiGRU) over these features reaches an AUC of 0.840 on RAGTruth (10 seeds), an 11-point gain over an independent logistic-regression baseline (p = 0 .002, Wilcoxon signed-rank) . A controlled decomposition attributes most of the gain to temporal order rather than model capacity: evidence propagates from confident positions to ambiguous neighbors within a span.  \nThe same 0.845 ceiling recurs across recurrent, state-space (Mamba), and attention architectures, locating the bottleneck in the feature set rather than the model. Because it reads only the generated text and external signals, the detector works on closed-source models, and it keeps working on text produced by language models it never saw during training, losing under 4% AUC.  \n1 Introduction  \nLarge language models generate text that is fluent, contextually appropriate, and sometimes entirely fabricated (Ji et al., 2023 ; Huang et al., 2023) . In retrieval-augmented generation systems, a single hallucinated claim can propagate through downstream decisions in healthcare, legal reasoning, or financial analysis. The core difficulty is that these models fail silently: there is no intrinsic signal that anything has gone wrong.  \nA substantial body of work addresses this problem from different angles. Internal probes train classifiers on hidden representations to separate hallucinated from grounded tokens within the model’s  \nown representation space (Kossen et al., 2024 ; Chen et al., 2024) . Uncertainty estimators measure entropy or semantic divergence across sampled generations, using the model’s own confidence as a proxy for correctness (Farquhar et al., 2024) . Natural Language Inference (NLI)-based classifiers compare generated claims against source documents using external entailment models (Kovácsand Recski, 2025 ; vLLM Team, 2025) .  \nEach of these approaches makes progress, yet each has a characteristic failure mode. Simhi et al.(2025) show that uncertainty-based detectors fail on high-confidence hallucinations, precisely the cases that matter most. Cheang et al. (2025) demonstrate that internal probes become indistinguishable when the model has partial knowledge of the subject. Dubanowska et al. (2025) find that current detectors do not generalize out-of-distribution. These failures are not coincidental: all existing methods share a common design choice that limits their effectiveness. Every approach examines each token in isolation, computing a score at one position without reference to how the signal at neighboring positions is evolving.  \nWe argue that this per-token independence assumption discards a fundamental regularity: hallucination is not a point event. When an autoregressive model departs from factual grounding, the erroneous tokens become part of the context for subsequent generation, biasing the model toward continuation of the hallucinated narrative. The result is characteristic runs of hallucinated tokens whose statistical signatures evolve jointly over time. In RAGTruth (Niu et al., 2024), we confirm a median span of 5 tokens and a persistence probability P (Ht | Ht−1) = 0 .902 versus an onset probability P (Ht | Ft−1) = 0 .006—a 150:1 ratio. An information-theoretic analysis shows that 76% of label entrop","cbCaipVqEwQWXapb","https://ap.wps.com/l/cbCaipVqEwQWXapb","pdf",545679,17,"English","# 1 Introduction\n## Research questions\n## Main contributions\n# 2 Related Work","[{\"question\":\"How does the proposed method detect hallucinations at the token level?\",\"answer\":\"It treats hallucinations as spans over time and uses sequence labeling, scoring each token via features that fuse text statistics, NLI entailment, and language-model surprisal, then modeling temporal dynamics with a BiGRU.\"},{\"question\":\"What signals are fused in the 33-dimensional feature stream?\",\"answer\":\"The feature vector fuses three families: text statistics (including novelty and context overlap dynamics), NLI entailment from DeBERTa, and language-model surprisal from TinyLlama, each enriched with temporal derivatives.\"},{\"question\":\"Why does temporal modeling improve detection compared to independent per-token classifiers?\",\"answer\":\"Most errors evolve jointly over spans because hallucinated tokens become context for subsequent generation; the method captures this via temporal order and evidence propagation from confident positions to ambiguous neighbors.\"}]","Temporal Multi-Signal Fusion for Token-Level Hallucination Detection | PDF",43]