New method learns historical Greek letterform representations

Researchers have developed a novel approach to learning diachronic representations of ancient Greek letterforms, addressing the challenge of handwriting variations across centuries. They introduced three new datasets: Hell-Char for training, and PaLit-Char and Med-Char for evaluation, covering periods from the 3rd century BCE to the 14th century CE. To handle symbolic variation, scarce data, and degradation, the team proposed a similarity-weighted supervised contrastive loss and a lacuna-driven augmentation scheme. These methods, applied to CNN and ResNet models, resulted in embeddings that effectively separate character classes and enable further analysis like clustering and prototype image generation. AI

IMPACT This research offers a transferable paradigm for representation learning in scarce, temporally evolving, and noisy conditions, potentially benefiting other historical text analysis tasks.

RANK_REASON The cluster contains an academic paper detailing a new method for representation learning on historical datasets. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• John Pavlopoulos
• PaLit-Char

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →