{"paper":{"title":"VIDA: A dataset for Visually Dependent Ambiguity in Multimodal Machine Translation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A dataset of 2,500 translation instances shows that chain-of-thought fine-tuning helps models use visual evidence to resolve ambiguities more consistently.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Chris Biemann, Jingheng Pan, Liang Ding, Longyue Wang, Weihua Luo, Xintong Wang","submitted_at":"2026-05-03T19:55:06Z","abstract_excerpt":"Ambiguity resolution is a key challenge in multimodal machine translation (MMT), where models must genuinely leverage visual input to map an ambiguous expression to its intended meaning. Although prior work has proposed disambiguation-oriented benchmarks probing the role of vision, we observe that existing benchmarks remain limited by task-format mismatch, narrow ambiguity coverage, or insufficient visual-dependency validation. Moreover, existing ambiguity evaluations are not well suited to diverse ambiguity types in open-ended translation. To address these limitations, we present VIDA (Visual"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments with two state-of-the-art Large Vision Language Models under vanilla inference, supervised fine-tuning (SFT), and our chain-of-thought SFT (CoT-SFT) show that while SFT improves overall translation quality, CoT-SFT yields more consistent gains in disambiguation accuracy, especially on out-of-distribution subsets, indicating a stronger generalization for resolving diverse ambiguity types.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The 2,500 instances are accurately annotated such that visual evidence is genuinely required to resolve each ambiguous span, and the LLM-as-a-judge classifier reliably measures correct span-level disambiguation without its own biases or errors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"VIDA provides 2,500 visually-dependent ambiguous MT instances and LLM-judge metrics; chain-of-thought SFT improves disambiguation accuracy over standard SFT, especially out-of-distribution.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A dataset of 2,500 translation instances shows that chain-of-thought fine-tuning helps models use visual evidence to resolve ambiguities more consistently.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"45d78a49e96f52ec49566ea228aa1360ac9e806f57d6c7e59f784d555163599a"},"source":{"id":"2605.02035","kind":"arxiv","version":2},"verdict":{"id":"da5efd4c-ebc8-44bd-9337-ebb4d4fafd59","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T19:29:55.976722Z","strongest_claim":"Experiments with two state-of-the-art Large Vision Language Models under vanilla inference, supervised fine-tuning (SFT), and our chain-of-thought SFT (CoT-SFT) show that while SFT improves overall translation quality, CoT-SFT yields more consistent gains in disambiguation accuracy, especially on out-of-distribution subsets, indicating a stronger generalization for resolving diverse ambiguity types.","one_line_summary":"VIDA provides 2,500 visually-dependent ambiguous MT instances and LLM-judge metrics; chain-of-thought SFT improves disambiguation accuracy over standard SFT, especially out-of-distribution.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The 2,500 instances are accurately annotated such that visual evidence is genuinely required to resolve each ambiguous span, and the LLM-as-a-judge classifier reliably measures correct span-level disambiguation without its own biases or errors.","pith_extraction_headline":"A dataset of 2,500 translation instances shows that chain-of-thought fine-tuning helps models use visual evidence to resolve ambiguities more consistently."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.02035/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T16:39:38.586407Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T04:31:22.159965Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T16:46:45.028012Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"6bd1cb08b84c182faf38354fa4024e5ef63580432218a28d66528a0d0b313047"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"5b995bf02bef8dd03296afbe046b5d0e1e12a55d5deead78cb0cac7ca1047b94"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}