Translators Put ChatGPT to the Test: What It Can and Cannot Do

Two weeks after ChatGPT's release, translator communities worldwide subjected it to intensive testing. Comparisons and usage notes flooded social media, and a relatively clear capability map began to emerge.

Where it excels: polishing and rewriting — turning stiff literal translations into idiomatic prose is ChatGPT's strength; style transfer — recasting a passage as formal, colloquial or marketing register; terminology explanation and background lookup — it explains specialist concepts like an assistant; and short-form transcreation — for taglines and slogans it produces multiple candidate versions in one breath.

Where it falls short: factual verification — it fabricates plausible-looking references and figures; long-document consistency — translate a document in chunks and terminology and style drift; precision in specialist domains — in legal and medical texts it occasionally delivers wrong equivalents in fluent language; and data privacy — pasting client documents into an online chat box itself crosses the confidentiality line.

The industry consensus after two weeks: ChatGPT is a powerful "translator's assistant", but not yet a "translator". Whoever learns to harness it gains efficiency first — and harnessing it starts with knowing exactly where its abilities end.