The Translation Problem

Machine translation has improved so quickly that many users assume it understands language the way people do. It does not. A translation system predicts likely word sequences; it has no experience of the world those words describe. This gap rarely matters for simple sentences, but it appears in idioms, humor, and anything that depends on shared culture. Consider a headline that calls a losing team lovable losers. A literal translation keeps the words but loses the affection, turning gentle mockery into an insult. Human translators solve this by asking what the text is doing, not only what it says: is it praising, warning, joking, or persuading? They then rebuild that function with different materials in the new language. The best evaluation of a translation is therefore not word-by-word accuracy but effect: does the reader laugh, worry, or act as the original reader would? This is why professional translators still beat machines on contracts, poems, and speeches, where tone carries as much weight as content. Machines handle volume; humans handle consequence. The sensible workflow uses both: software for the first draft, and a skilled human who knows the subject, the audience, and what is truly at stake.