“Don’t You Just Upload It to ChatGPT?” — A Translator’s Answer

A freelance translator in Ottawa has described a conversation at her gym that has since circulated among language professionals. A classmate, already dressed for an office day, asked why she was leaving the class early. Three assignments had landed that evening, all due the following morning. The reply came without hesitation: “But… it won’t take long. Don’t you just upload the documents to ChatGPT?”

The question, and the answer

The translator paused for a split second, assuming it was a joke. It was not one. Her explanation was short and technical: ChatGPT will produce a translated document, but formatting problems come first, and more importantly the translation itself will be questionable. Machine output is not the same as a human understanding what another human is trying to say — adapting, localizing, researching terminology and keeping it consistent throughout.

Where AI actually fits into the workflow

She is not anti-AI. She began experimenting with the tools the previous autumn, when, as she puts it, they started taking real work. In practice the uses are narrow and verifiable: feeding a client’s 500-page style guides to a model so it can flag rule breaks during final checks, or extracting specialized terminology from reference documents to build glossaries faster than a manual search. She also keeps a dedicated spell-checker and occasionally asks a model for a second opinion on a paragraph.

The parallel she draws is older than the current debate: translators used to paste stubborn sentences into Google Translate, then DeepL, to see whether the machine suggested a phrasing they had missed. Tools are what professionals use — the accountant’s spreadsheet formulas, the manager’s slide templates.

Why the checks stay manual

The limits she describes are concrete: models invent acronyms and organization names, skip entire sentences, ignore supplied terminology unless repeatedly pushed, and occasionally miss the point completely. Every output has to be checked twice or three times, which makes it a different way of working rather than a magic button. Her conclusion on pay is blunt: translators, writers and editors should not be paid less because AI exists, any more than a roofer should be paid less for using a hammer.

The conversation ended with a detail she found hard to ignore. Asked whether she used AI at work, her classmate — a director general in human resources — said she could not: the technology, in her words, is not reliable enough.