There is a vast treasure trove of free Latin language pedagogical content available on the Internet, but shockingly little of it is in plaintext -- the greppable, diffable, interoperable, and eternal format that is the fundamental substrate of GNU Emacs.

In this talk I'll demonstrate how I use both local and cloud-hosted ML models to transform recorded Latin audio into glorious structured plaintext that not only encodes the text itself, but also its morphological and semantic context. Thus transformed, we can now render the text with attached metadata that enables powerful word-at-point features like linguistic details in the echo area, accurate dictionary lookups, and audio playback with precise word alignment.