Generative Simple Dictionary Transformer (GSDT) Architecture: For Natural Language Processing without Pre-Training

Abstract The Generative Simple Dictionary Transformer (GSDT) model introduces a novel approach to language understanding in artificial intelligence (AI). By starting with a limited "Simple Dictionary" of core words and leveraging the principles of Neural Matrix Synaptic Resonance Networks (NM-SRNs), GSDT aims to achieve efficient and robust language comprehension with reduced computational cost and enhanced explainability. This paper details the GSDT architecture, including its core components, learning mechanisms, and potential applications. We explore the integration of a "universal machine-learning dictionary" (UMLDic) as a learning tool and discuss the synergistic relationship between GSDT and Large Language Models (LLMs). The GSDT architecture represents a promising step toward developing AI systems that can learn and understand language with human-like efficiency and adaptability.