FAQs
The questions that come up most, from translation teams and engineers alike.
Why rule-based, and not a neural model?
Neural translation needs a large body of parallel text to train on, and for most of the world's minority languages that simply doesn't exist. Rule-based generation doesn't need a training corpus — it needs a linguist to describe how the language actually works, once. That description is also useful on its own: it's a real, checkable grammar of the language, not a black box.
Where does the source material come from?
The starting point is the Greek New Testament and Hebrew Old Testament, read alongside established lexicons, and encoded into a shared semantic representation built specifically for this project — one designed to hold onto meaning precisely enough to regenerate it in a very different language, and expressive enough to cover the range of content across both testaments. Encoding both source languages into one shared representation, rather than keeping them separate, also means each target language needs only a single transfer grammar — not one per source language it might translate from.
How is this different from a general translation tool?
A general-purpose translator predicts likely words in the target language from patterns in training data. This works the other way around: a person first builds a detailed semantic representation of what the text actually means — its participants, its structure, the distinctions the source language marks — and the software then generates a draft from that representation using the target language's own grammar rules. A native speaker takes it from there.
How much of the Bible can it translate?
Any passage that has a semantic representation built for it. Coverage is expanding toward the full Bible — see Remaining work for where that stands today — and the same approach extends beyond Scripture itself, toward commentaries, study materials, and other Christian literature.
What about poetry?
Meter and rhyme rarely survive translation intact, so the priority is meaning over form: the parallelism that carries so much of Hebrew poetry is preserved deliberately, and metaphors are handled contextually — sometimes as a simile instead, when that's what actually lands in the target language.
What does it take to set up a new language?
Roughly 40–50 hours to get comfortable with the software itself, then another 150–200 hours to configure a new target language's grammar and produce a first sample book. A background in linguistics or software helps, but neither is required — see How does it work? for what that setup actually involves.