Combining Specialised and General-Purpose AI — First Results and Challenges from "Neo-Latin in the Modern World"

Neo-Latin and New Ancient Greek texts meet cutting-edge AI in this FWF-funded project at the University of Innsbruck. The digital platform uses Transkribus Smart Extract Models to handle layout analysis, transcription, and Named Entity Recognition in a single automated pipeline — a controllable, structured alternative to LLMs for the core recognition work. Large language models then take over downstream, post-correcting text and enabling intelligent querying of the resulting structured corpus. A practical demonstration of how specialised and general-purpose AI can work together without one replacing the other.