From Translator to AI Translation Engineer

Why would someone who's been a literary translator and a publisher for over a decade develop an AI book translation software? I keep answering various formulations of this question whenever I talk about BookTranslate.ai and how it came to be.

I was still in high school when I first encountered the ideas of classical liberalism and free market economics through English works. Born and raised in post-soviet Hungary, these ideas were thoroughly foreign to me, but I immediately felt they explain so much of what is wrong with my country.

But when I looked for Hungarian publications on the subject to recommend to friends and family, I've had my fateful encounter with the vicious economic realities of publishing. I couldn't find a single book in libraries nor in bookstores that would present these ideas of liberty to the Hungarian audience. None were published ‐ of course - under soviet censorship and half a century of communism has purged any favorable thought towards capitalism from our culture so thoroughly that no publisher could find a market for such books.

The old method of translation is exceedingly expensive, which raises the minimum number of copies a publisher has to sell to justify a project. This means: niche works, innovative new ideas, less popular authors, topics without mass appeal all become that much harder to reach a global audience.

This is bad enough for fiction. But when it comes to scientific or philosophical works, the stakes become civilizational. New ideas, innovations, new methods of science, medicine, technology, the progress of moral and political philosophy - all start out as a fringe minority's conviction who are often hated and exiled by the majority before their ideas conquer the world.

The economic realities of translation place a cruel limit on the progress of human civilization - and this is barely addressed when AI and translation is brought up. We talk a lot about the seen plight of the translators, but the unseen economic prison of old workflows which traps ideas and prevents them from reaching those who could bring about change is never mentioned.

I've spent much of my twenties translating and publishing the literature of liberty to Hungarian - the old fashioned way, with the manual translation workflow that barely differs from that of a translator from a century or two ago. I tried all the pre-AI machine translation options to make my work faster and nothing worked; it merely changed the nature of the work to post-editing, without changing the time it took to complete it.

When ChatGPT became a global phenomenon a few years ago, I immediately began experimenting - being a software engineer myself - whether it can be used to reduce the costs of translation. The initial results showed promise but it took well over a year of obsessive engineering to create BookTranslate.ai, which is by far the most complex piece of literary AI engineering I've ever seen.

With this tech, a book that would have taken me a month to translate now takes a weekend - achieving the same quality that I expected from my own work as a translator.

So why have I built it?

Because encountering the catastrophic economic realities of literary translation has shocked me for life.

Because I have a radical belief in human progress - and I believe all progress begins as ideas.

Because ideas are important and they must reach the furthest corners of the world so every nation and every specialized field could debate and consider and learn from them.

Because translation is too important to be left to humans alone.