Sound symbolization with deep learning

Abstract

This Final Year Project explores the intricate relationship between phonemes and visual shapes, inspired by the Bouba-Kiki effect. The effect reveals consistent associations between specific sounds and certain visual forms, such as rounded phonemes correlating with circular shapes and angular phonemes evoking jagged forms. Leveraging linguistic research, a rule-based system and a machine learning model were developed to map pseudowords to geometric shapes. The project evaluates the creativity, accuracy, and alignment of these approaches with sound-symbolism principles. Additionally, the study addresses challenges such as integrating linguistic theory into computational methods, creating synthetic datasets, and designing robust evaluation frameworks. This comprehensive report outlines the methodology, implementation, results, and implications of this work, providing a foundation for further exploration of sound symbolization and its applications in computational linguistics and artificial intelligence.