Automated Comparative Analysis of Visual and Textual Representations of Logographic Writing Systems in Large Language Models

Abstract The complex nature of logographic writing systems, characterized by their visually intricate characters and context-dependent meanings, presents unique challenges for computational models designed primarily for alphabetic scripts. Understanding the ability of LLMs to process logographic scripts across visual and textual input modalities is essential for advancing their application in multilingual contexts. The novel approach presented in this study systematically compares the performance of LLMs when interpreting logographic characters as both visual and textual data, offering new insights into the semantic consistency and accuracy of model outputs across these modalities. The findings reveal critical disparities in performance, particularly highlighting the models' tendency to favor textual inputs, which suggests the need for further refinement in multimodal processing capabilities. Through detailed analysis of error patterns, semantic similarity, and computational complexity, the research demonstrates the importance of developing more robust and versatile LLM architectures capable of effectively managing the inherent complexities of logographic writing systems. The conclusions drawn from the study not only provide a deeper understanding of the limitations of current LLMs but also set the stage for future innovations in the field, aiming to enhance the models' ability to generalize across diverse linguistic structures and input types.