OCR

Optical Character Recognition: the technique that converts images and scans into selectable text. For documents not "born digital", it's the step before the language model: without OCR, the model has no text to work on.