Machines translate sentences. Books are not sentences.

Machine translation is genuinely good now, which is exactly what makes this question harder than it used to be. It is fast, effectively free, and produces output that looks correct. For understanding a document, handling correspondence, or getting the gist of something, it is excellent.

For publishing a book it remains unsuitable, and the reasons are structural rather than a matter of waiting for the technology to improve.

What machine translation does well

Worth being fair about this, because dismissing it entirely is not honest and would not help you decide.

It handles factual, functional text reliably. Instructions, specifications, correspondence, and anything where the goal is transferring information rather than producing an effect. It is fast enough to translate an entire book in minutes and consistent in the narrow sense that it will render the same phrase the same way each time.

It is also genuinely useful inside a professional process. Translators use it as a starting point for straightforward passages, freeing attention for the parts that need judgment. The distinction is between a tool used by someone who can evaluate its output and a substitute for that person.

The five things it structurally cannot do

These are not gaps waiting to be closed by a better model. They follow from what the task involves.

1. Hold a voice across a book. It translates sentence by sentence with no memory of how it rendered your narrator forty pages ago. The register drifts, and a reader experiences that as the narrator subtly changing personality.
2. Decide what to sacrifice. Every translation loses something. Choosing what to give up requires knowing what a passage is for, and no sentence-level process has access to that.
3. Resolve ambiguity deliberately. English is often ambiguous on purpose. Spanish frequently forces a choice about gender, formality, or number. The machine picks one silently and you never learn a choice was made.
4. Handle formality as a stance. The tú and usted decision runs through an entire book and encodes the relationship between narrator and reader. It is a single authorial decision, not a per-sentence one.
5. Rebuild an effect. A joke depending on English wordplay has to become a different joke in Spanish. That is invention, not conversion, and it requires knowing what the original was doing.

The first is a literal rendering of two English idioms. It is grammatically correct Spanish that no Spanish speaker would write. The second says what the sentence meant. Multiply that decision by every idiom in a novel and you have the difference between the two approaches.

Where post-editing sits

The obvious middle path is to run the machine translation and pay someone to fix it. This is a real service, widely used, and it is worth understanding where it works.

For technical and functional text it is efficient and sensible. For a book it tends to disappoint, for a reason that is not obvious in advance: fixing prose is often slower than writing it. An editor working through machine output is constantly deciding whether each sentence is acceptable, subtly wrong, or actively misleading, and that evaluation costs more attention than translating the sentence would have.

There is also an anchoring effect. Working from an existing rendering pulls the result toward that rendering. The output ends up correct and flat, because nobody rebuilt the paragraph from the intention, they repaired it from the draft.

The questions only a person asks

A useful way to see the difference is to notice what a human translator asks you during the work. These questions are the job, and nothing that translates sentence by sentence generates any of them.

• "Is this character being rude here, or just direct?" The answer changes the register throughout their dialogue.
• "Should the reader already know this, or is it being introduced?" Spanish sometimes forces a definite or indefinite article where English was neutral.
• "This reference will mean nothing to a Mexican reader. Do you want an equivalent or a footnote?" An authorial decision, not a linguistic one.
• "Your narrator addresses the reader directly in chapter two. Formal or informal?" Once decided, it governs the whole book.
• "This joke does not work. Can I write a different one here?" The only honest options are to invent or to lose it.
• "You have used this phrase as a motif. Should I keep it identical each time?" A pattern invisible at sentence level.

Every one of those has consequences beyond the sentence it appears in, and every one requires knowing what the book is trying to do. A process that never asks them has not resolved them; it has answered them silently, differently each time, and moved on.

What readers actually notice

Readers rarely identify a translation as machine-produced. What they report is vaguer and more damaging: that the book feels flat, that something is off, or that they lost interest without being able to say where.

That vagueness is the problem. A clear error can be corrected. A book that is technically accurate and lifeless generates reviews saying it did not hold their attention, and those reviews sit on your product page permanently, attributed to your writing rather than to the process.

It is worth remembering who the audience for a translated edition is. They are reading in their own language, with the sensitivity to register that comes with that, and they are precisely the people most likely to notice prose that reads as processed.

How to evaluate a translation you cannot read

A genuine difficulty: if you do not read Spanish, how do you judge what you receive? Authors often feel they simply have to trust the result, and there are better options than that.

That fourth question is the useful one. Native speakers will almost always say a competent translation is correct, because it is. Asking whether it reads as originally written is a different question and gets a much more informative answer.

Translating into a native language

One detail that matters more than it sounds: professional translators work into their native language, not out of it.

Someone who learned Spanish as an adult can understand your English perfectly and still produce Spanish that is correct and subtly foreign. Native-language production is where idiom, rhythm, and register live, and readers detect the absence without being able to name it, in exactly the way they detect machine output.

This is why our translations are handled by translators working into their native language in both directions, and it is a reasonable question to ask of any provider.

Deciding for your book

A reasonable test: does anyone read your book for the experience of reading it, or purely for the information? If it is purely informational and the tone genuinely does not matter, machine translation with a careful check may be adequate.

For fiction, memoir, or any nonfiction where your voice is part of why people read you, it is not. The whole reason a reader chose your book over a summary is the way you say things, and that is precisely what does not survive.

There is also a sequencing point worth making. A translated edition needs its own formatting, because Spanish runs longer than English and pagination changes with it, and whether your book suits a Spanish edition at all is a separate question worth settling before the translation decision. Neither of those is affected by which translation approach you choose, but both affect whether the second edition is worth producing.

Translation is $0.09 per word, in either direction, with a turnaround of 5 days from a final manuscript. You can see the figure for your word count before speaking to anyone, and what a real translation involves is worth reading alongside this. Our translation service covers both directions, and you keep 100 percent of your rights and royalties on both editions.