Translation
A book should sound like itself in every language.
We translate between Arabic, English, Spanish, and French — mostly books, mostly for publishers who will be judged on the result. The register you wrote in is the register you get back.
What makes book translation different
A document can be translated a page at a time. A book cannot. The same term has to land the same way in chapter twelve as in chapter one; a name transliterated one way on page 3 cannot drift by page 200; a voice that is formal at the start cannot loosen because a different pass handled the end. So we build the glossary and the name table before the first chapter is drafted, and every later pass is bound by them.
How it runs
AI does the labour. A human makes the call.
Nothing here is hidden. This is the actual pipeline, in the order it runs.
- 01
Read the whole book first
The book is parsed into chapters and blocks, and every name, term, honorific, and quoted source is extracted into a glossary before any translation happens. This is what keeps chapter twelve consistent with chapter one.
- 02
Draft against the glossary
Each chapter is drafted with the glossary and the surrounding context in view, so decisions carry forward instead of being re-made per chunk.
- 03
Revise for literature, not accuracy alone
A second pass reads the draft beside the source and rewrites for rhythm, register, and idiom. Accurate but wooden prose fails this step.
- 04
Score every segment, re-run the weak ones
A quality model scores each segment against its source. Anything that scores low goes back for another pass rather than quietly shipping.
- 05
A human reads it
Every flagged decision — a Qur'anic quotation, an ambiguous pronoun, a term with two defensible renderings — is surfaced as a decision to make, not buried. On Professional and Premium a person reads the whole text.
What you receive
- Translated book as .docx, styles intact
- A PDF proof for reading
- The glossary and name table used, as a file you keep
- A list of every flagged decision and how it was resolved
- SRT and VTT subtitle files, where the source was video
- TMX export of the translation memory on request
Questions we actually get
How do you handle Qur'anic quotations?
We never machine-translate the Qur'an. Quotations are detected, matched to their reference, and rendered using an established translation — Sahih International for English by default, configurable per project. Every instance is flagged so you can check it.
What happens to honorifics?
They are rendered as Arabic glyphs — ﷺ and its companions — in every target language, rather than spelled out. That is the house convention and it is applied consistently across the whole book.
Is my book used to train a model?
No. We use no-training API tiers, and document text is never written to logs. Your text is used to produce your book and nothing else.
Can you match the style of a book we already published?
Yes. Send the earlier book and we build the style memory from it, so the new translation reads as part of the same series rather than a fresh start.
What if I disagree with a choice?
Tell us the term and the preferred rendering. It goes into the glossary and is applied everywhere in the book, not just where you spotted it.
Do you do certified translation?
Yes, for documents rather than books, at a 25% surcharge over the tier price. The rate is on the pricing page.
Price your book.
Upload it and the calculator counts the words and gives you the number.