Optical Character Recognition for Decipherment

At this week’s lab meeting, Logan will survey applications of optical character recognition (OCR) to decipherment. An early step in any decipherment is to understand the cipher’s “alphabet”, i.e. the set of distinct characters it employs. Inaccurate identification of the underlying script can obscure frequency and adjacency information which decipherers (both human and algorithmic) rely on to crack the cipher. We will discuss four papers which address the task of identifying the underlying character inventories of enciphered or otherwise unknown scripts, finishing with a recent publication from our lab and comments on ongoing followup experiments:

• Yin et al 2018, Decipherment of Historical Manuscript Images
• Corazza et al. 2022a, Unsupervised deep learning supports reclassification of Bronze age cypriot writing system
• Corazza et al. 2022b, Contextual Unsupervised Clustering of Signs for Ancient Writing Systems
• Kelley et al. 2022, Image-aware language modeling for proto-Elamite

Wednesday, 29 March at 12pm – click to add to calendar.

This will be a hybrid meeting. The zoom link will be posted on zulip.