Literal Translation Isn't Bad Translation. It's Unreviewed Translation.

People love to blame "literal translation" when a translated line lands flat. I've come to think the blame is misplaced. Literal isn't the problem. Unreviewed is.

A literal draft is honest: it shows you exactly what the source said, word for word, with nothing smoothed over and nothing hidden. That's a useful starting point. The failure happens when that draft goes straight to the audience with no one asking three questions:

1. Does this idiom exist in the target language? If not, the literal version isn't "faithful" — it's nonsense in the new language.
2. Does the register match? A line that's blunt in the source shouldn't come out polite in translation, or vice versa. Tone is meaning too.
3. Would the target audience actually say this? Not "would they understand it if they tried" — would they say it, in this scene, to this person?

Take one example I keep coming back to. A Chinese line built around "装蒜" — literally, "pretending to be garlic." Translated word for word into Spanish, you get something about garlic that no viewer would parse as an insult or a bluff. The fix isn't to avoid literal translation as a method; it's to review the literal draft, spot the idiom, and replace it with the target-language equivalent that carries the same intent. In this case, the reviewed line lands as "No te hagas el tonto conmigo" — no garlic, same challenge.

That's the whole workflow, really: draft literal, then review for idiom, register, and audience fit — in that order. Most "bad translation" I've seen isn't a failure of the first step. It's a missing second step.

So if you're using AI to translate dialogue, don't judge the tool by its first draft. Judge your process by whether a human (or a good checklist) actually reviews that draft before it ships. The draft is cheap. The review is the product.

What's your test? When a translated line feels off, do you check the idiom first, the tone first, or something else entirely?

Disclosure: I sell a short-drama line checker built around this review step — it flags literal-translation traps in Spanish and Brazilian Portuguese (full product, and there's a free 5-line sample if you want to test the format first). The point above stands with or without it: review the draft, don't just blame it.