{"paper":{"title":"DIA-HARM: Dialectal Disparities in Harmful Content Detection Across 50 English Dialects","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Disinformation detectors show reduced performance on non-Standard American English dialects.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Adaku Uchendu, Ali Al-Lawati, Dongwon Lee, Jason Lucas, Matt Murtagh, Uchendu Uchendu","submitted_at":"2026-04-07T01:43:48Z","abstract_excerpt":"Harmful content detectors, particularly disinformation classifiers, are predominantly developed and evaluated on Standard American English (SAE), leaving their robustness to dialectal variation unexplored. We present DIA-HARM, the first benchmark for evaluating disinformation detection robustness across 50 English dialects spanning U.S., British, African, Caribbean, and Asia-Pacific varieties. Using Multi-VALUE's linguistically grounded transformations, we introduce D-CUBE (Dialectal Disinformation Detection Corpus), a core corpus component of DIA-HARM comprising 195K samples derived from esta"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"These findings demonstrate that current disinformation detectors may systematically disadvantage hundreds of millions of non-SAE speakers worldwide.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The Multi-VALUE transformations produce dialectal variants that accurately reflect real-world usage and preserve the original disinformation label without introducing confounding artifacts that independently affect model performance.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DIA-HARM reveals that human-written dialectal English degrades disinformation detector F1 by 1.4-3.6% while AI-generated dialectal content stays stable, with multilingual models generalizing better than monolingual ones.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Disinformation detectors show reduced performance on non-Standard American English dialects.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"46db342ff9b6cf0b240c782adf595863a283d8b41c492b3b604aa627d04cc3b4"},"source":{"id":"2604.05318","kind":"arxiv","version":2},"verdict":{"id":"a0716a5b-bee0-41aa-8128-e061161c7c68","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T19:54:06.000936Z","strongest_claim":"These findings demonstrate that current disinformation detectors may systematically disadvantage hundreds of millions of non-SAE speakers worldwide.","one_line_summary":"DIA-HARM reveals that human-written dialectal English degrades disinformation detector F1 by 1.4-3.6% while AI-generated dialectal content stays stable, with multilingual models generalizing better than monolingual ones.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The Multi-VALUE transformations produce dialectal variants that accurately reflect real-world usage and preserve the original disinformation label without introducing confounding artifacts that independently affect model performance.","pith_extraction_headline":"Disinformation detectors show reduced performance on non-Standard American English dialects."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.05318/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"f36025ff4f6cf2e5fcfb318468941ed99e24987c37487f6b66ea7d055f9e2cfb"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}