Excited to share Enoki — a framework for efficient, fine-grained hallucination detection.</p>\n<p>The key idea is simple: use text-anchored relational facts as a shared representation for both claim-level verification and span-level localization. This avoids costly claim-to-span alignment while staying competitive at claim level and improving fine-grained localization.</p>\n<p>We’re also releasing EnokiQA, our dual-granularity dataset. Feedback and discussions are very welcome!</p>\n","updatedAt":"2026-09-07T07:11:50.656Z","author":{"_id":"659a8f0258608c4044cf7dc5","avatarUrl":"/avatars/bb213ee4e300fc9143b544f3ecdb88e8.svg","fullname":"Julia Belikova","name":"julia-bel","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.8994452953338623},"editors":["julia-bel"],"editorAvatarUrls":["/avatars/bb213ee4e300fc9143b544f3ecdb88e8.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2609.00581","authors":[{"_id":"6a97ea99d4c640d9b759ba3b","user":{"_id":"636b867ecde3707d10999b96","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1667991162387-noauth.png","isPro":false,"fullname":"Elisei Rykov","user":"lmeribal","type":"user","name":"lmeribal"},"name":"Elisei Rykov","status":"claimed_verified","statusLastChangedAt":"2026-09-03T08:17:50.998Z","hidden":false},{"_id":"6a97ea99d4c640d9b759ba3c","name":"Timur Ionov","hidden":false},{"_id":"6a97ea99d4c640d9b759ba3d","name":"Nikolay Ivanov","hidden":false},{"_id":"6a97ea99d4c640d9b759ba3e","name":"Maksim Savkin","hidden":false},{"_id":"6a97ea99d4c640d9b759ba3f","user":{"_id":"683c210c9c6d4e639fa827a2","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/6-uRkimtBedgU4eh85e9k.png","isPro":false,"fullname":"Maksim Makarenko","user":"makamoa","type":"user","name":"makamoa"},"name":"Maksim Makarenko","status":"claimed_verified","statusLastChangedAt":"2026-09-07T08:45:04.188Z","hidden":false},{"_id":"6a97ea99d4c640d9b759ba40","name":"Alexander Panchenko","hidden":false},{"_id":"6a97ea99d4c640d9b759ba41","name":"Vasily Konovalov","hidden":false},{"_id":"6a97ea99d4c640d9b759ba42","user":{"_id":"659a8f0258608c4044cf7dc5","avatarUrl":"/avatars/bb213ee4e300fc9143b544f3ecdb88e8.svg","isPro":false,"fullname":"Julia Belikova","user":"julia-bel","type":"user","name":"julia-bel"},"name":"Julia Belikova","status":"claimed_verified","statusLastChangedAt":"2026-09-04T08:45:04.179Z","hidden":false}],"publishedAt":"2026-09-01T02:21:28.000Z","submittedOnDailyAt":"2026-09-07T00:00:00.000Z","title":"Enoki: Efficient Multi-Level Hallucination Detection","submittedOnDailyBy":{"_id":"659a8f0258608c4044cf7dc5","avatarUrl":"/avatars/bb213ee4e300fc9143b544f3ecdb88e8.svg","isPro":false,"fullname":"Julia Belikova","user":"julia-bel","type":"user","name":"julia-bel"},"summary":"Ensuring factuality remains a critical challenge for deploying LLMs in high-stakes settings. Existing hallucination detectors usually operate at a single level: claim-level methods provide interpretable factual units, while span-level methods localize unsupported text. Bridging these views is costly, as LLM-heavy pipelines require multiple decomposition and verification calls, and modular systems need additional claim-to-span alignment. We propose Enoki, an Open Information Extraction framework for multi-level hallucination detection. Enoki extracts text-anchored relational facts, verifies them against evidence, and projects unsupported facts back to hallucinated spans. This shared representation enables claim-level verification and span-level localization without requiring separate alignment. Enoki supports LLM-based, encoder-based, and rule-based extraction regimes, balancing accuracy and inference cost through a common interface. Experiments show that Enoki remains competitive with strong claim-level systems while using fewer resources and achieves superior performance on fine-grained span- and entity-level localization. We also release EnokiQA, a dual-granularity dataset with aligned claim-level verification and span-level localization annotations.","upvotes":13,"discussionId":"6a97ea9ad4c640d9b759ba43","projectPage":"https://huggingface.co/collections/s-nlp/enoki","githubRepo":"https://github.com/s-nlp/Enoki","githubRepoAddedBy":"user","ai_summary":"Enoki is an open information extraction framework that unifies claim-level verification and span-level hallucination localization through shared relational facts, reducing resource use while improving detection accuracy.","ai_keywords":["Open Information Extraction","multi-level hallucination detection","relational facts","claim-level verification","span-level localization","LLM-based extraction","encoder-based extraction","EnokiQA"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":2,"organization":{"_id":"60f1454b100ab9c285090cc2","name":"s-nlp","fullname":"s-nlp","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1662038598497-605473729d7c1d4d81b7e52b.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"643c2e8086ab6dbe34efc87d","avatarUrl":"/avatars/375cb1ac37b0b5f3fb6a05d0b2c9d031.svg","isPro":false,"fullname":"kid","user":"electrostatickid","type":"user"},{"_id":"659a8f0258608c4044cf7dc5","avatarUrl":"/avatars/bb213ee4e300fc9143b544f3ecdb88e8.svg","isPro":false,"fullname":"Julia Belikova","user":"julia-bel","type":"user"},{"_id":"636b867ecde3707d10999b96","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1667991162387-noauth.png","isPro":false,"fullname":"Elisei Rykov","user":"lmeribal","type":"user"},{"_id":"661a2d0af17e918674639599","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/661a2d0af17e918674639599/prBuSpEUbhxjsmAm9Fql-.jpeg","isPro":false,"fullname":"Davydenko Grigorii","user":"DavydenkoGr","type":"user"},{"_id":"61dedb1b2066746d68b63adb","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/61dedb1b2066746d68b63adb/PLBHQxvbcay3qDJjY7HM3.jpeg","isPro":false,"fullname":"Ivan Sviridov","user":"univanxx","type":"user"},{"_id":"66d600ec73a6b923e129f2a1","avatarUrl":"/avatars/adc6c88decbb51634554b3dfffef6cc4.svg","isPro":false,"fullname":"Alexander Ivanov","user":"alexander-andreevich","type":"user"},{"_id":"6904703c9c3d523381fe6220","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6904703c9c3d523381fe6220/tEiI8NYsSz5li4v2G31-r.png","isPro":false,"fullname":"Nail","user":"gafurovnail","type":"user"},{"_id":"643fd8f427dc46cca58c661e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/klNlGqneH-2Dzvm1fqJ57.jpeg","isPro":true,"fullname":"Parchiev Rauf","user":"parchiev","type":"user"},{"_id":"61066b8044c85d83f3cbce95","avatarUrl":"/avatars/cccc9e68916ee7ebf9807e38bf74ba9f.svg","isPro":false,"fullname":"Vitaly Bezuglyj","user":"creatorofuniverses","type":"user"},{"_id":"6705578896a78d10a30a8755","avatarUrl":"/avatars/c43b4551f9c2ed6b13cb86c72996e856.svg","isPro":false,"fullname":"Ivan Kruzhilov, Gleb Mazanov, etc","user":"BearSubj13","type":"user"},{"_id":"69a1e531f22b165988ad108b","avatarUrl":"/avatars/86cfeb4b29117bfb58c80a2e7fe94d10.svg","isPro":false,"fullname":"A V","user":"who-goes-brrrr","type":"user"},{"_id":"65cdde3046a98b4598e69d00","avatarUrl":"/avatars/9f61e0c9e5754737d3cc71c8437eacf9.svg","isPro":false,"fullname":"Ivan Poddiakov","user":"ivanpodd","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"60f1454b100ab9c285090cc2","name":"s-nlp","fullname":"s-nlp","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1662038598497-605473729d7c1d4d81b7e52b.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2609/2609.00581.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?