[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-273437-105":59,"doc-detail-273437-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","the-null-token-knows-reducing-message-free-hallucination-in-asr-and-nmt","The Null Token Knows - Reducing Message-Free Hallucination in ASR and NMT","","Modern encoder–decoder systems can generate fluent text even when the input contains no recoverable message. This work studies message-free failure modes in ASR and NMT by analyzing reserved null-token behavior and whether end-of-generation scoring provides an abstention signal. Across multiple speech recognizers and translation models, the study audits native null-token scores and logit shifts. For Whisper, decoder-state probes and supervised row edits are compared with external gating. Raising null-token scores can suppress fabrication, but aggressive intervention may delete valid speech or shorten legitimate translations, motivating evaluation by suppression and deletion costs.",{"@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/the-null-token-knows-reducing-message-free-hallucination-in-asr-and-nmt/273437/",{"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/the-null-token-knows-reducing-message-free-hallucination-in-asr-and-nmt/273437.png","ImageObject",300,407,{"name":92,"@type":93},"Lucas Martin","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-29","2026-09-15",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",18,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What problem does the paper address in ASR and NMT?","Question",{"text":113,"@type":114},"Encoder–decoder systems may produce fluent output even when the input contains no recoverable message, leading to message-free hallucination.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"What is the null token used for in this study?",{"text":118,"@type":114},"The study uses reserved null-token behavior to test whether end-of-generation scoring provides a usable abstention signal and how it relates to fabrication vs deletion.",{"name":120,"@type":111,"acceptedAnswer":121},"What tradeoff do the authors find when increasing the null-token score?",{"text":122,"@type":114},"Increasing the null-token score can sharply suppress fabrication, but aggressive intervention can also delete valid speech or shorten legitimate translations.","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},273437,1789471432,{"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":130,"read_time":145,"preview_page_count":4},8796095360427,"https://ap-avatar.wpscdn.com/davatar_994ba38a5ba835b3df7d355c54d3ed8d","The Null Token Knows: Reducing Message-Free Hallucination in ASR and NMT  \nKirill Borodin∗1,2,3, Vasiliy Kudryavtsev 1,2,3 , Ivan Viakhirev4,5 , Grach Mkrtchian 1,3  \n1Lab260, Moscow, Russia 2BitmanagerAI, Dubai, UAE 3MTUCI, Moscow, Russia  \n4ITMO, Saint Petersburg, Russia 5FRC RAS, Saint Petersburg, Russia  \n[kborodin.research@gmail.com](kborodin.research@gmail.com)  \narXiv : 2608 . 15940v2 [cs .CL] 18 Aug 2026  \nAbstract  \nModern encoder–decoder systems can produce fluent text even when their input contains no recoverable message. We study this failure in ASR and NMT through the models’ reserved null tokens, asking whether the score for ending generation already carries a usable abstention signal. Across speech recognizers and translation models, we audit native null-token scores and scalar logit shifts. In Whisper, we additionally probe decoder states and compare supervised row edits with conventional external gates. The evaluated models often expose a useful abstention signal, but stock decoding does not reliably act on it. Raising the null-token score can sharply suppress fabrication, but aggressive intervention also deletes valid speech or shortens legitimate translations. These findings turn the null token into a diagnostic lens on hallucination and motivate evaluating abstention methods by both suppression and deletion costs, rather than by hallucination reduction alone.  \n1 Introduction  \nAttention-based encoder–decoder (AED) systems can fabricate fluent output from inputs that carry no message. Whisper (Radford et al. 2023) transcribes confident sentences from silence, room tone, and music, with documented downstream harms (Koenecke et al. 2024; Frieske and Shi 2024); neural translators can likewise return fluent text for degenerate, message-free sources (Lee et al. 2018; Guerreiro, Voita, and Martins 2023; Raunak, Menezes, and Junczys-Dowmunt 2021) . This failure does not reliably disappear with scale: across six Whisper checkpoints, our developmental endpoint is non-monotonic even as WER falls (Table 13) (Viakhirev, Borodin, and Mkrtchian 2026) .  \nOperational systems counter this failure with native nospeech scores, VAD/endpointing, and reject options (Radford et al. 2023; FireRed Team 2026; Chow 1970) . Other work re-trains attention heads (Wang et al. 2025), steers encoder features through sparse autoencoders (Aparin et al. 2026b), or distills against a clean teacher (Tripathi et al. 2025) . Yet each evaluated family already exposes a reserved output coordinate: EOT in AED decoders, EOS in translation, and the semantically different alignment blank in CTC/RNN-T (Graves et al. 2006; Graves 2012). Because Whisper encoders also represent non-speech events (Gong et al. 2023), we ask a  \n∗Corresponding author.  \nFigure 1: Developmental clip-stratified probes separate nonspeech from speech after the first Whisper decoder block. The layer-0 embedding is at chance; the gray band is the permutation baseline. Folds are not grouped by source corpus, so this does not establish source-independent encoding.  \nnarrower question: what battery-specific separation appears in decoder states and native scores, and what fabrication– deletion tradeoff follows from editing the reserved coordinate?  \nOn the developmental battery, frozen decoder states linearly separate non-speech from speech after the first decoder block (Figure 1), and native null-token scores distinguish the two conditions even when the null token loses the output decision. We therefore use readout edits only as instruments for moving the abstention operating point. A scalar bias shifts constant prior odds; the analytic graft and trained row are supervised, state-dependent variants of one output coordinate. Their results do not establish an advantage over a scalar or external gate. Prior shift is a conditional model for the scalar case (Saerens, Latinne, and Decaestecker 2002; Lipton, Wang, and Smola 2018); we test where its operatingpoint predictions hold.  \nOur evi","cbCainjp9Tv188jQ","https://ap.wps.com/l/cbCainjp9Tv188jQ","pdf",874462,22,"English","# Abstract\n# 1 Introduction\n# 2 Background and Related Work\n# 3 Method (inferred)\n# 4 Battery Separation and Readout (inferred)\n# 5 Checkpoint Audit (inferred)\n# 6 Boundary Tests (inferred)\n# 7 Additional Boundary and Safety Notes (inferred)","[{\"question\":\"What problem does the paper address in ASR and NMT?\",\"answer\":\"Encoder–decoder systems may produce fluent output even when the input contains no recoverable message, leading to message-free hallucination.\"},{\"question\":\"What is the null token used for in this study?\",\"answer\":\"The study uses reserved null-token behavior to test whether end-of-generation scoring provides a usable abstention signal and how it relates to fabrication vs deletion.\"},{\"question\":\"What tradeoff do the authors find when increasing the null-token score?\",\"answer\":\"Increasing the null-token score can sharply suppress fabrication, but aggressive intervention can also delete valid speech or shorten legitimate translations.\"}]","The Null Token Knows - Reducing Message-Free Hallucination in ASR and NMT | PDF",55]