[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"article-doc:docs\u002Fconcepts\u002F6.live-translation":3},"---\ntitle: Live Translation\ndescription: Live translation is different from translating a finished transcript. It must balance responsiveness, linguistic context, translation quality, and stability.\nstatus: preview\n---\nLive translation is different from translating a finished transcript.\n\nWith a finished transcript, the system already knows:\n\n- the complete sentence\n- punctuation\n- sentence boundaries\n- surrounding context\n- what comes next\n\nDuring live speech, none of those conditions are guaranteed.\n\n## Translation while meaning is still forming\n\nConsider:\n\n```text\nI don't think we should...\nI don't think we should launch...\nI don't think we should launch tomorrow.\n```\n\nA translation system receiving the first fragment must decide whether to produce output immediately or wait for additional context.\n\nProducing too early can cause repeated corrections.\n\nWaiting too long creates noticeable latency.\n\nLive translation therefore requires balancing:\n\n- responsiveness\n- linguistic context\n- translation quality\n- stability\n\n## Source and translated context\n\nIn live applications, source text and translated text are related streams.\n\nA change in the interpretation of the source may also affect the translation.\n\nApplications should therefore avoid assuming that translation is simply a second independent transcript.\n\nInstead:\n\n```text\nlive speech\n     ↓\nsource context\n     ↓\ntranslation context\n```\n\nBoth continue evolving while the conversation progresses.\n\n## Designed for delivery\n\nThe objective of live translation is not merely to eventually produce a correct translated transcript.\n\nIt is to allow someone to **follow speech in another language while it is happening**.\n\nThat distinction affects how the entire system should be designed.\n",1790059118949]