Episode Details

Sixty percent of the time, student evaluators using an industry rubric judged human literary translations as no better than machine output. Switch to a simple side-by-side preference test, and those same readers chose the human version 80 to 100 percent of the time. The rubric was trained to miss what the humans were doing. This episode traces translation from the Epic of Gilgamesh and Jerome's Vulgate through Tyndale, Luther, Baghdad's House of Wisdom, and Xuan Zang's twenty-year pilgrimage — then asks what happens when the field's own automatic metrics and LLM judges systematically penalize creative, culturally appropriate human choices. Puns, idioms, and culturally loaded content remain the machine's weakest ground. If translation is an art rather than plumbing, the human translator's job shifts from production to judgment — and the grading infrastructure needs rebuilding before it trains the next generation of tools to optimize for the wrong thing. Episode #891661 — open it directly at myweirdprompts.com/891661