{"id":"a6fe9f1bc0773b1b4f01c3b9e5efbecb2c52e7dd2e98891e681924a01935b841","title":"Researchers test text-only training for video temporal localization","argument":"In “Grounding in the Dark,” Divyanshu Mishra, Sabine Sternig, Rakshith Shetty and Erhan Gundogdu investigate whether a model trained only on textual video descriptions can locate events in video it sees only at inference time. The approach avoids paired video-text training data, which the authors say is costly to collect with precise temporal annotations. They create a 134K-example corpus and adapt Qwen3-8B using text embeddings, then evaluate on Charades-STA and QV-Highlights without training on videos from those benchmarks.\n\nThe authors report that directly replacing text inputs with frame embeddings performs poorly, which they describe as a large modality gap. Their training-free Nearest Entity Replacement (NER) instead represents each frame using embeddings of nearby text entities, making the input more like what the model encountered during training. NER improves the reported mtIoU scores on both benchmarks and beats the strongest of four modality-gap baseline families by 5.00 and 1.71 points, respectively. The result is partial transfer from text to video, not a claim that text-only training solves video grounding: the authors say more reliable grounding will likely require better-aligned text and visual representations.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence is the paper abstract; it does not provide details of the experiments beyond the reported setup and results.","speaker":{"name":"Mishra, Divyanshu, Sternig, Sabine, Shetty, Rakshith, Gundogdu, Erhan","roleAtSource":"Article author"},"source":{"title":"Grounding in the Dark: Text-Only Training for Zero-Shot Video Temporal Localization","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.06149","publishedAt":"2026-11-04","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors investigate training a text-only LLM on textual video descriptions for temporal grounding, with video supplied only at inference time.","excerpt":"a text-only LLM, with no vision-language pretraining, can be trained on textual video descriptions alone to perform temporal grounding and receive video only at inference time","url":"https://arxiv.org/abs/2611.06149","relationship":"Source-stated"},{"text":"The authors report a large modality gap when using frame embeddings directly at inference, and propose training-free Nearest Entity Replacement to address it.","excerpt":"Still, replacing text with frame embeddings at inference yields only 4.99 mtIoU on Charades-STA and 11.12 on QV-Highlights, revealing a large modality gap. To bridge it, we propose Nearest Entity Replacement (NER)","url":"https://arxiv.org/abs/2611.06149","relationship":"Source-stated"},{"text":"The authors report that NER outperforms their strongest modality-gap baseline by 5.00 points on Charades-STA and 1.71 points on QV-Highlights.","excerpt":"NER reaches 14.45 mtIoU on Charades-STA and 20.72 on QV-Highlights, outperforming the strongest of four families of modality-gap baselines by 5.00 and 1.71 points.","url":"https://arxiv.org/abs/2611.06149","relationship":"Source-stated"},{"text":"The authors characterize the transfer from text to video as partial and say more reliable grounding will likely require better-aligned text and visual representations.","excerpt":"NER thus enables partial transfer from text to video, while more reliable grounding will likely require encoders with better-aligned text and visual representations.","url":"https://arxiv.org/abs/2611.06149","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.06149","checkedAt":"2026-11-09T11:34:07.331Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How much would better-aligned text and visual representations improve grounding in this text-only training setting?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Researchers test text-only training for video temporal localization","summary":"In “Grounding in the Dark,” Divyanshu Mishra, Sabine Sternig, Rakshith Shetty and Erhan Gundogdu investigate whether a model trained only on textual video descriptions can locate events in video it sees only at inference time. The approach avoids paired video-text training data, which the authors say is costly to collect with precise temporal annotations. They create a 134K-example corpus and adapt Qwen3-8B using text embeddings, then evaluate on Charades-STA and QV-Highlights without training on videos from those benchmarks.\n\nThe authors report that directly replacing text inputs with frame embeddings performs poorly, which they describe as a large modality gap. Their training-free Nearest Entity Replacement (NER) instead represents each frame using embeddings of nearby text entities, making the input more like what the model encountered during training. NER improves the reported mtIoU scores on both benchmarks and beats the strongest of four modality-gap baseline families by 5.00 and 1.71 points, respectively. The result is partial transfer from text to video, not a claim that text-only training solves video grounding: the authors say more reliable grounding will likely require better-aligned text and visual representations.","question":"How much would better-aligned text and visual representations improve grounding in this text-only training setting?","claims":[{"text":"The authors investigate training a text-only LLM on textual video descriptions for temporal grounding, with video supplied only at inference time.","excerpt":"a text-only LLM, with no vision-language pretraining, can be trained on textual video descriptions alone to perform temporal grounding and receive video only at inference time","support":"source-stated"},{"text":"The authors report a large modality gap when using frame embeddings directly at inference, and propose training-free Nearest Entity Replacement to address it.","excerpt":"Still, replacing text with frame embeddings at inference yields only 4.99 mtIoU on Charades-STA and 11.12 on QV-Highlights, revealing a large modality gap. To bridge it, we propose Nearest Entity Replacement (NER)","support":"source-stated"},{"text":"The authors report that NER outperforms their strongest modality-gap baseline by 5.00 points on Charades-STA and 1.71 points on QV-Highlights.","excerpt":"NER reaches 14.45 mtIoU on Charades-STA and 20.72 on QV-Highlights, outperforming the strongest of four families of modality-gap baselines by 5.00 and 1.71 points.","support":"source-stated"},{"text":"The authors characterize the transfer from text to video as partial and say more reliable grounding will likely require better-aligned text and visual representations.","excerpt":"NER thus enables partial transfer from text to video, while more reliable grounding will likely require encoders with better-aligned text and visual representations.","support":"source-stated"}],"limitations":["The supplied evidence is the paper abstract; it does not provide details of the experiments beyond the reported setup and results."],"mediaId":"92158dfc165bbda4fed82fcda41be5b5767abc2456a4d5d89d127f9492abbd36","correctionReason":null,"id":"a6fe9f1bc0773b1b4f01c3b9e5efbecb2c52e7dd2e98891e681924a01935b841","publicationId":"ab19b5f9-1ac0-4b9e-88a4-a127a178cab5","version":1,"sourceUrl":"https://arxiv.org/abs/2611.06149","sourceTitle":"Grounding in the Dark: Text-Only Training for Zero-Shot Video Temporal Localization","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-04","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-05T05:00:20.121Z","createdAt":"2026-11-09T11:03:30.798Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"92158dfc165bbda4fed82fcda41be5b5767abc2456a4d5d89d127f9492abbd36","kind":"article","url":"https://arxiv.org/abs/2611.06149","sourceUrl":"https://arxiv.org/abs/2611.06149","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T11:02:42.761Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.06149","checkedAt":"2026-11-09T11:34:07.331Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"92158dfc165bbda4fed82fcda41be5b5767abc2456a4d5d89d127f9492abbd36","confidence":0.88,"provider":"typesafe","model":"jev-1.13.0","inputHash":"ba5a5cb4106c680e2a24536a1c329fd150a80897ec5055d65dab165a4d5ddb88","decidedAt":"2026-11-09T11:02:44.532Z"},"provenance":{"textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observationId":"4ac4d4310a70a6679449e3036be34465ca60fc4ffbb535cbbaf3ab01aaf6d4a9","retrievedAt":"2026-11-09T11:02:42.454Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"3863e982-34a7-42a2-8f5d-c456d27f150c"},"reviewer":{"model":"gpt-6-luna","runId":"66e2d004-22f8-46ce-8eb2-02444c70e3e2"}},"audit":{"status":"passed","reviewedAt":"2026-11-09T11:03:30.798Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Mishra, Divyanshu","url":null,"kind":"unknown"},{"name":"Sternig, Sabine","url":null,"kind":"unknown"},{"name":"Shetty, Rakshith","url":null,"kind":"unknown"},{"name":"Gundogdu, Erhan","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Mishra, Divyanshu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Mishra, Divyanshu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06149","locator":"head/meta[name=author|citation_author]"},"assertionId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:34:07.331Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939","sourceUrl":"https://arxiv.org/abs/2611.06149","contentSha256":"8dcca976a4cef011332bd051d7adce0e303d529408e13d17108182c3c7e73710","textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observedAt":"2026-11-09T11:34:07.331Z"}},{"actor":{"name":"Sternig, Sabine","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Sternig, Sabine","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06149","locator":"head/meta[name=author|citation_author]"},"assertionId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:34:07.331Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939","sourceUrl":"https://arxiv.org/abs/2611.06149","contentSha256":"8dcca976a4cef011332bd051d7adce0e303d529408e13d17108182c3c7e73710","textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observedAt":"2026-11-09T11:34:07.331Z"}},{"actor":{"name":"Shetty, Rakshith","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Shetty, Rakshith","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06149","locator":"head/meta[name=author|citation_author]"},"assertionId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:34:07.331Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939","sourceUrl":"https://arxiv.org/abs/2611.06149","contentSha256":"8dcca976a4cef011332bd051d7adce0e303d529408e13d17108182c3c7e73710","textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observedAt":"2026-11-09T11:34:07.331Z"}},{"actor":{"name":"Gundogdu, Erhan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Gundogdu, Erhan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06149","locator":"head/meta[name=author|citation_author]"},"assertionId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:34:07.331Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939","sourceUrl":"https://arxiv.org/abs/2611.06149","contentSha256":"8dcca976a4cef011332bd051d7adce0e303d529408e13d17108182c3c7e73710","textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observedAt":"2026-11-09T11:34:07.331Z"}}],"participationCoverage":{"total":4,"returned":4,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.06149","snapshotId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939","contentSha256":"8dcca976a4cef011332bd051d7adce0e303d529408e13d17108182c3c7e73710","textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observedAt":"2026-11-09T11:34:07.331Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/a6fe9f1bc0773b1b4f01c3b9e5efbecb2c52e7dd2e98891e681924a01935b841/source"}},"canonicalUrl":"https://escapemeridian.com/readings/researchers-test-text-only-training-for-video-temporal-a6fe9f1b","formats":{"html":"/readings/researchers-test-text-only-training-for-video-temporal-a6fe9f1b","markdown":"/readings/researchers-test-text-only-training-for-video-temporal-a6fe9f1b.md","json":"/readings/researchers-test-text-only-training-for-video-temporal-a6fe9f1b.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-embedding-vector","name":"Embedding vector","explanation":"A list of numbers that represents a piece of text, an image or another item so software can compare meanings.","referenceUrl":"https://developers.google.com/machine-learning/glossary#embedding-vector"},{"id":"term-benchmark","name":"Benchmark","explanation":"A defined set of tests used to compare systems. Results depend on the tasks and measures selected.","referenceUrl":"https://arxiv.org/abs/2211.09110"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"},{"id":"time","name":"TIME","kind":"organization","context":"TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.","referenceUrls":["https://time.com/"],"path":"/entities/time","json":"/entities/time.json","markdown":"/entities/time.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"a6fe9f1bc0773b1b4f01c3b9e5efbecb2c52e7dd2e98891e681924a01935b841","revision":"ab19b5f9-1ac0-4b9e-88a4-a127a178cab5","revisionKind":"publication-id","title":"Researchers test text-only training for video temporal localization","url":"https://escapemeridian.com/readings/researchers-test-text-only-training-for-video-temporal-a6fe9f1b"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.06149","revision":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","revisionKind":"sha256-source-text","title":"Grounding in the Dark: Text-Only Training for Zero-Shot Video Temporal Localization","url":"https://arxiv.org/abs/2611.06149"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"a6fe9f1bc0773b1b4f01c3b9e5efbecb2c52e7dd2e98891e681924a01935b841","title":"Researchers test text-only training for video temporal localization","argument":"In “Grounding in the Dark,” Divyanshu Mishra, Sabine Sternig, Rakshith Shetty and Erhan Gundogdu investigate whether a model trained only on textual video descriptions can locate events in video it sees only at inference time. The approach avoids paired video-text training data, which the authors say is costly to collect with precise temporal annotations. They create a 134K-example corpus and adapt Qwen3-8B using text embeddings, then evaluate on Charades-STA and QV-Highlights without training on videos from those benchmarks.\n\nThe authors report that directly replacing text inputs with frame embeddings performs poorly, which they describe as a large modality gap. Their training-free Nearest Entity Replacement (NER) instead represents each frame using embeddings of nearby text entities, making the input more like what the model encountered during training. NER improves the reported mtIoU scores on both benchmarks and beats the strongest of four modality-gap baseline families by 5.00 and 1.71 points, respectively. The result is partial transfer from text to video, not a claim that text-only training solves video grounding: the authors say more reliable grounding will likely require better-aligned text and visual representations.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence is the paper abstract; it does not provide details of the experiments beyond the reported setup and results.","speaker":{"name":"Mishra, Divyanshu, Sternig, Sabine, Shetty, Rakshith, Gundogdu, Erhan","roleAtSource":"Article author"},"source":{"title":"Grounding in the Dark: Text-Only Training for Zero-Shot Video Temporal Localization","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.06149","publishedAt":"2026-11-04","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors investigate training a text-only LLM on textual video descriptions for temporal grounding, with video supplied only at inference time.","excerpt":"a text-only LLM, with no vision-language pretraining, can be trained on textual video descriptions alone to perform temporal grounding and receive video only at inference time","url":"https://arxiv.org/abs/2611.06149","relationship":"Source-stated"},{"text":"The authors report a large modality gap when using frame embeddings directly at inference, and propose training-free Nearest Entity Replacement to address it.","excerpt":"Still, replacing text with frame embeddings at inference yields only 4.99 mtIoU on Charades-STA and 11.12 on QV-Highlights, revealing a large modality gap. To bridge it, we propose Nearest Entity Replacement (NER)","url":"https://arxiv.org/abs/2611.06149","relationship":"Source-stated"},{"text":"The authors report that NER outperforms their strongest modality-gap baseline by 5.00 points on Charades-STA and 1.71 points on QV-Highlights.","excerpt":"NER reaches 14.45 mtIoU on Charades-STA and 20.72 on QV-Highlights, outperforming the strongest of four families of modality-gap baselines by 5.00 and 1.71 points.","url":"https://arxiv.org/abs/2611.06149","relationship":"Source-stated"},{"text":"The authors characterize the transfer from text to video as partial and say more reliable grounding will likely require better-aligned text and visual representations.","excerpt":"NER thus enables partial transfer from text to video, while more reliable grounding will likely require encoders with better-aligned text and visual representations.","url":"https://arxiv.org/abs/2611.06149","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.06149","checkedAt":"2026-11-09T11:34:07.331Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How much would better-aligned text and visual representations improve grounding in this text-only training setting?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Researchers test text-only training for video temporal localization","summary":"In “Grounding in the Dark,” Divyanshu Mishra, Sabine Sternig, Rakshith Shetty and Erhan Gundogdu investigate whether a model trained only on textual video descriptions can locate events in video it sees only at inference time. The approach avoids paired video-text training data, which the authors say is costly to collect with precise temporal annotations. They create a 134K-example corpus and adapt Qwen3-8B using text embeddings, then evaluate on Charades-STA and QV-Highlights without training on videos from those benchmarks.\n\nThe authors report that directly replacing text inputs with frame embeddings performs poorly, which they describe as a large modality gap. Their training-free Nearest Entity Replacement (NER) instead represents each frame using embeddings of nearby text entities, making the input more like what the model encountered during training. NER improves the reported mtIoU scores on both benchmarks and beats the strongest of four modality-gap baseline families by 5.00 and 1.71 points, respectively. The result is partial transfer from text to video, not a claim that text-only training solves video grounding: the authors say more reliable grounding will likely require better-aligned text and visual representations.","question":"How much would better-aligned text and visual representations improve grounding in this text-only training setting?","claims":[{"text":"The authors investigate training a text-only LLM on textual video descriptions for temporal grounding, with video supplied only at inference time.","excerpt":"a text-only LLM, with no vision-language pretraining, can be trained on textual video descriptions alone to perform temporal grounding and receive video only at inference time","support":"source-stated"},{"text":"The authors report a large modality gap when using frame embeddings directly at inference, and propose training-free Nearest Entity Replacement to address it.","excerpt":"Still, replacing text with frame embeddings at inference yields only 4.99 mtIoU on Charades-STA and 11.12 on QV-Highlights, revealing a large modality gap. To bridge it, we propose Nearest Entity Replacement (NER)","support":"source-stated"},{"text":"The authors report that NER outperforms their strongest modality-gap baseline by 5.00 points on Charades-STA and 1.71 points on QV-Highlights.","excerpt":"NER reaches 14.45 mtIoU on Charades-STA and 20.72 on QV-Highlights, outperforming the strongest of four families of modality-gap baselines by 5.00 and 1.71 points.","support":"source-stated"},{"text":"The authors characterize the transfer from text to video as partial and say more reliable grounding will likely require better-aligned text and visual representations.","excerpt":"NER thus enables partial transfer from text to video, while more reliable grounding will likely require encoders with better-aligned text and visual representations.","support":"source-stated"}],"limitations":["The supplied evidence is the paper abstract; it does not provide details of the experiments beyond the reported setup and results."],"mediaId":"92158dfc165bbda4fed82fcda41be5b5767abc2456a4d5d89d127f9492abbd36","correctionReason":null,"id":"a6fe9f1bc0773b1b4f01c3b9e5efbecb2c52e7dd2e98891e681924a01935b841","publicationId":"ab19b5f9-1ac0-4b9e-88a4-a127a178cab5","version":1,"sourceUrl":"https://arxiv.org/abs/2611.06149","sourceTitle":"Grounding in the Dark: Text-Only Training for Zero-Shot Video Temporal Localization","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-04","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-05T05:00:20.121Z","createdAt":"2026-11-09T11:03:30.798Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"92158dfc165bbda4fed82fcda41be5b5767abc2456a4d5d89d127f9492abbd36","kind":"article","url":"https://arxiv.org/abs/2611.06149","sourceUrl":"https://arxiv.org/abs/2611.06149","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T11:02:42.761Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.06149","checkedAt":"2026-11-09T11:34:07.331Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"92158dfc165bbda4fed82fcda41be5b5767abc2456a4d5d89d127f9492abbd36","confidence":0.88,"provider":"typesafe","model":"jev-1.13.0","inputHash":"ba5a5cb4106c680e2a24536a1c329fd150a80897ec5055d65dab165a4d5ddb88","decidedAt":"2026-11-09T11:02:44.532Z"},"provenance":{"textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observationId":"4ac4d4310a70a6679449e3036be34465ca60fc4ffbb535cbbaf3ab01aaf6d4a9","retrievedAt":"2026-11-09T11:02:42.454Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"3863e982-34a7-42a2-8f5d-c456d27f150c"},"reviewer":{"model":"gpt-6-luna","runId":"66e2d004-22f8-46ce-8eb2-02444c70e3e2"}},"audit":{"status":"passed","reviewedAt":"2026-11-09T11:03:30.798Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Mishra, Divyanshu","url":null,"kind":"unknown"},{"name":"Sternig, Sabine","url":null,"kind":"unknown"},{"name":"Shetty, Rakshith","url":null,"kind":"unknown"},{"name":"Gundogdu, Erhan","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Mishra, Divyanshu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Mishra, Divyanshu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06149","locator":"head/meta[name=author|citation_author]"},"assertionId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:34:07.331Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939","sourceUrl":"https://arxiv.org/abs/2611.06149","contentSha256":"8dcca976a4cef011332bd051d7adce0e303d529408e13d17108182c3c7e73710","textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observedAt":"2026-11-09T11:34:07.331Z"}},{"actor":{"name":"Sternig, Sabine","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Sternig, Sabine","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06149","locator":"head/meta[name=author|citation_author]"},"assertionId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:34:07.331Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939","sourceUrl":"https://arxiv.org/abs/2611.06149","contentSha256":"8dcca976a4cef011332bd051d7adce0e303d529408e13d17108182c3c7e73710","textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observedAt":"2026-11-09T11:34:07.331Z"}},{"actor":{"name":"Shetty, Rakshith","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Shetty, Rakshith","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06149","locator":"head/meta[name=author|citation_author]"},"assertionId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:34:07.331Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939","sourceUrl":"https://arxiv.org/abs/2611.06149","contentSha256":"8dcca976a4cef011332bd051d7adce0e303d529408e13d17108182c3c7e73710","textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observedAt":"2026-11-09T11:34:07.331Z"}},{"actor":{"name":"Gundogdu, Erhan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Gundogdu, Erhan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06149","locator":"head/meta[name=author|citation_author]"},"assertionId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:34:07.331Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939","sourceUrl":"https://arxiv.org/abs/2611.06149","contentSha256":"8dcca976a4cef011332bd051d7adce0e303d529408e13d17108182c3c7e73710","textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observedAt":"2026-11-09T11:34:07.331Z"}}],"participationCoverage":{"total":4,"returned":4,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.06149","snapshotId":"c95ceddbc65ec69dff1aca61d58378934abed25e758904430f1651061ceca939","contentSha256":"8dcca976a4cef011332bd051d7adce0e303d529408e13d17108182c3c7e73710","textHash":"5a7c10e7d79a14ab8e590419ffc5d0f2ee979ed76d933517e130edca313e6f9f","observedAt":"2026-11-09T11:34:07.331Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/a6fe9f1bc0773b1b4f01c3b9e5efbecb2c52e7dd2e98891e681924a01935b841/source"}}}],"audits":[{"publicationId":"ab19b5f9-1ac0-4b9e-88a4-a127a178cab5","verdict":"passed","findings":[{"dimension":"attribution","status":"pass","explanation":"The draft names the four listed authors and the paper title correctly. It frames the method, results and limitations as the authors' investigation, report and assessment rather than as established fact.","excerpt":"Authors:Divyanshu Mishra, Sabine Sternig, Rakshith Shetty, Erhan Gundogdu"},{"dimension":"summary","status":"pass","explanation":"The reading leads with the actor and the question, explains temporal grounding as locating events in video, and preserves the authors' hedge about partial transfer. Optional fixes: use 'nearest' rather than 'nearby' text entities, since nearby could be misread as nearby in time. Adding the absolute scores (4.99/11.12 rising to 14.45/20.72) would show how modest the results remain. mtIoU and embeddings are left to tooltips, which is acceptable.","excerpt":"represents each video frame by the normalized mean of its nearest text-entity embeddings, so that the model receives the kind of input it was trained on"},{"dimension":"date","status":"pass","explanation":"originalPublished is 2026-11-04, which matches the v1 submission date. The v2 revision on 5 Nov is correctly not used as the publication date.","excerpt":"[Submitted on 4 Nov 2026 (v1), last revised 5 Nov 2026 (this version, v2)]"},{"dimension":"support","status":"pass","explanation":"Every claim is source-stated, and the excerpts are verbatim. The draft does not turn the benchmark scores into real-world reliability, and it keeps the authors' caveat. The question follows from the authors' own conclusion and does not presume an answer.","excerpt":"NER thus enables partial transfer from text to video, while more reliable grounding will likely require encoders with better-aligned text and visual representations."},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct link to the arXiv abstract page, which is the original source.","excerpt":null}],"createdAt":"2026-11-09T11:03:30.798Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}