Skip to content

Latest commit

 

History

History
10 lines (7 loc) · 998 Bytes

README.md

File metadata and controls

10 lines (7 loc) · 998 Bytes
Raw

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.