{"record": {"aliases": ["Not Truly Multilingual: Script Consistency as a Missing Dimension in VLM Evaluation"], "artifacts": [{"id": "arxiv:2606.17188", "kind": "paper", "url": "https://arxiv.org/abs/2606.17188"}, {"id": "gh:prabhjotschugh/not-truly-multilingual-pumvr", "kind": "repo", "url": "https://github.com/prabhjotschugh/Not-Truly-Multilingual-PuMVR"}], "authors": ["Prabhjot Singh", "Bhushan Pawar", "Madhu Reddiboina", "Rajvee Sheth"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.17188", "source_url": "https://arxiv.org/abs/2606.17188"}, "categories": ["General AI", "Multimodal", "Reasoning"], "description": {"en": "A multilingual visual-reasoning benchmark comprising parallel Punjabi image-text instances in Gurmukhi, Shahmukhi, and Roman scripts."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.17188", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.17188", "title": "PuMVR"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.17188", "modality": null, "name": "PuMVR", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_not-truly-multilingual-script-consistency-_fd5037c9", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "28e22c7559675fc020531cc7fd02910555904f0434c375f838d6f15bab604b91", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T17:15:50.996573Z", "source_url": "https://arxiv.org/abs/2606.17188"}, "publisher": null, "released": "2026-06-15", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.17188", "source_url": "https://arxiv.org/abs/2606.17188"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-17188", "source": "claire_radar", "source_benchmark_id": "2606.17188", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["Not Truly Multilingual: Script Consistency as a Missing Dimension in VLM Evaluation"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Multimodal", "attentionForecast": {"confidence": "Low", "horizon": "7d", "reason": "Its focus on multilingual VLM evaluation and script consistency may draw focused evaluation-community interest.", "score": 3}, "capabilities": ["Reasoning"], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T17:15:50.996573Z", "inputHash": "00f0c03647bcb9136c94dd3ea44b11a1c1b4fae11a6daaed552a62f4f37c8fdb", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "We introduce PuMVR (Punjabi Multimodal Visual Reasoning), a benchmark", "canonicalNameSource": "abstract", "decisionReason": "The supplied text provides a formal benchmark name, repeatable parallel task instances, explicit accuracy and Script Consistency Rate scoring, and a public data-and-code path.", "evaluationEvidence": {"reuse": "Data and code are available at: https://github.com/prabhjotschugh/Not-Truly-Multilingual-PuMVR.", "scoring": "We propose the Script Consistency Rate (SCR)", "task": "1,000 strictly parallel image-text instances across Punjabi's three active scripts: Gurmukhi, Shahmukhi, and Roman"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T17:15:50.996573Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A multilingual visual-reasoning benchmark comprising parallel Punjabi image-text instances in Gurmukhi, Shahmukhi, and Roman scripts.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce PuMVR (Punjabi Multimodal Visual Reasoning), a benchmark of 1,000 strictly parallel image-text instances across Punjabi's three active scripts: Gurmukhi, Shahmukhi, and Roman."}, "familyId": "bmf_dfb40df17735", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_not-truly-multilingual-script-consistency-_fd5037c9", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/prabhjotschugh/Not-Truly-Multilingual-PuMVR", "data": null, "paper": "https://arxiv.org/abs/2606.17188", "pdf": "https://arxiv.org/pdf/2606.17188", "project": null, "report": "https://arxiv.org/abs/2606.17188"}, "metrics": [], "motivation": "Current multilingual evaluations for Vision-Language Models (VLMs) assume a one-to-one mapping between language and orthography, overlooking billions of users of multi-script languages.", "name": "PuMVR", "oneLine": "A multilingual visual-reasoning benchmark comprising parallel Punjabi image-text instances in Gurmukhi, Shahmukhi, and Roman scripts.", "paperTitle": "Not Truly Multilingual: Script Consistency as a Missing Dimension in VLM Evaluation", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.17188", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-06-15", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Prabhjot Singh", "Bhushan Pawar", "Madhu Reddiboina", "Rajvee Sheth"], "categories": ["cs.CV", "cs.CL"], "id": "2606.17188", "title": "Not Truly Multilingual: Script Consistency as a Missing Dimension in VLM Evaluation", "type": "arxiv", "url": "https://arxiv.org/abs/2606.17188"}, "sourceUpdatedAt": "2026-08-05T08:40:48Z", "topics": ["Multimodal", "Reasoning"], "venueAttempts": [], "whyItMatters": "It measures whether visual-language performance remains consistent when identical Punjabi tasks are rendered in different scripts."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
