Machine translation

Mova translates without a neural network at answer time: a recogniser of the language, a tree, frames, a world, then a meta-world shared by all languages and authors, and back down through the standard stack of the target language, with each author as a small delta over it. These cards cover the work this rests on: interlingua and transfer, tree-to-tree translation and synchronous grammars, rule-based systems, generation from trees and meaning, style and author transfer, Bible translation, and how translation is measured.

Each card names the ideas we take for the translator and what they did on the verse-aligned Bibles, where every block must come back byte for byte.

310 pages

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/mtrans/