κρῐτῐκός - automating the OCR of critical editions of pre-modern texts

• The global corpus of pre-modern texts is small when compared with modern corpora, and notwithstanding the occasional discovery does not grow.
• The state of the art of natural-language processing is data-hungry machine learning techniques.
• Hence, if research in low-resource languages like Ancient Greek, Latin, Old English, Pali, Sanskrit and Classical Chinese is to be able to leverage these tools long-term, a strategy for maximizing the data inherent in the small corpus must be adopted.
• Like any corpus, the pre-modern corpus can be subjected to data augmentation methods like "sliding window" and shuffling, however data augmentation only gets you so far.
• The real superpower of the corpus lies in the wealth of alternative readings present in critical editions. Every single
• Alternative readings can be either editorial conjectures or differing manuscripts.