The lead translator
Your translator works on the book, and their name stays on it. The machine writes the draft, and the translator decides every sentence.
Universities and research centers
Academic Arabic translation for universities and research centers, with your own translators and reviewers
For research centers taking Arabic sources into English for international publication, and for universities and institutions translating textbooks, encyclopedias and multi-volume series into Arabic. Your translators and peer reviewers work on Tarjum with one glossary, a translation memory and a file for the printer, in both directions.
A chapter you choose, worked on by your translator and assessed by your two reviewers on your center’s own form.
Your center’s procedures stay as they are
We don’t change who translates, who reviews and who approves. We give them one place to work.
Two peer reviewers
Each reviewer reads the chapter through a private link with no account, comments, and records a decision under their own name. The grade stays on your center’s form.
A deciding reviewer
If the two reviewers disagree, you send the link to a third, who reads the same pages.
The board
The board sees the pages as they will print, and the approval of the translation stays in its hands.
No AI tool takes part in peer review. Each peer reviewer sees the text alone.
The last volume knows what was decided in the first
The translation memory gathers every volume in your center’s library, and the glossary carries from one volume to the next. You follow every title on one board.
Volume oneAn example series of three volumes
GlossaryTranslationPeer reviewLayout
Volume twoWith the second reviewer
GlossaryTranslationPeer reviewLayout
Volume threeDraws on what was decided in the first two
GlossaryTranslationPeer reviewLayout
What an academic text needs
Scanned sources
Old editions and image-only PDFs come back as text you can work on.
Footnotes in their places
Every footnote stays tied to its mark in the text.
References in their language
The bibliography stays in its own language, and you can ask for it to be translated.
An index rebuilt
The index is rebuilt at layout, on the pages of the translation.
Tables, figures and equations
Tables stay tables, equations keep their numbers, and figures appear in the language of the translation.
One term
Your committee decides a term once, and it holds through the whole book.
Research projects in both directions
Textbooks, encyclopedias, edited volumes, conference proceedings and research reports, from translation to the print file.
For international publication
From Arabic into English, with pages that take the direction of English. Publishers’ policies on machine-assisted translation differ, so check your publisher’s policy before you start.
Textbooks and reference works in Arabic
From English into Arabic, with one glossary and subject review. A foreign text needs its publisher’s permission, so the permission is checked first.
Where does your licensed text go?
We say it plainly, before you ask.
The text’s path
Files are stored on servers in Germany. To translate, the text is sent to AI providers, and we will name them if you ask.
Rights, title by title
Your contract with a foreign publisher may not cover sending the book to an AI service. So we check the rights with you before any title is uploaded, and the final word belongs to your legal adviser.
Disclosure
We give you a statement of what the machine did, who reviewed and who assessed, to print in the book.
Custom models and fine-tuning on your texts
We are open to working with research centers on building a custom model, or on fine-tuning one on your texts. We offer this as research we do together with you. There is no ready-made product to buy.
Most centers don’t start here. Their terms and earlier translations, used inside the workflow, are enough, and that works today. A custom model is something we test with you, and we recommend it only if it does better on texts it has not seen before.
Rights check
The right to publish a translation is not necessarily the right to train on it, so we start here.
Aligning source and translation
We align the sentences, and set aside texts the model never sees, to serve as the test.
Glossary and memory first
We measure what they add on their own, before any training.
Trying the model
We adapt an open model, named in the contract, if there are enough texts.
The report
For each direction separately, with your reviewers’ reading beside the automatic metrics.
The results are yours
You keep the texts, the test and the results, and you may publish them.
If you have few texts, the results can get worse, and we will tell you so. In a published study (Vieira and colleagues, 2024, from English into other languages), results got worse with one to two thousand segments and improved in most cases from 5,000. We know of no published study that measures this on Arabic books, which is why we test.
Questions from centers and universities
Will Tarjum replace our translators?
No. Your translator stays the translator, with their name on the book and their decision on every sentence.
Do publishers accept machine-assisted translation?
Policies differ from one publisher to another, so check the policy of the publisher you have in mind before you start. We give you the disclosure statement.
Does it suit a large team?
Invite whoever you want to each book, set their role in it, and send links to reviewers outside your center.
We start with one chapter your center assesses
Choose a chapter from a book you have the right to work on, one you expect to show the weak spots. Your translator works on it, your two reviewers assess it, and then you decide.
Prefer to try it yourself? Start free trial: 14 days and 10,000 words, no card needed.