Comparison of Large Language Models and Traditional Neural Networks in Optical Character Recognition for Old Alphabets [Articol]

Abstract

This study compares large language models (LLMs) and traditional neural networks (TNNs) in Optical Character Recognition (OCR) for historical alphabets. While deep learning has advanced OCR technology, recognizing old scripts remains challenging due to their complexity. LLMs, with vision capabilities, offer a novel approach by integrating visual and linguistic understanding. This research evaluates the accuracy and robustness of both models on a dataset of an ancient alphabet, highlighting the potential of LLMs to improve OCR in historical linguistics and digital preservation. The findings provide valuable insights for applying modern AI to the preservation of historical texts.