Generative Topolinguistics: Bidirectional interfaces for emergent language topologies
Iulia Ionescu, Jenn Leung, Yannis Siglidis · 2025
The experimental framework set out for generative topolinguistics seeks to investigate the sociality of meaning construction in artificial cognitive systems. While the semanticity of artificial linguistic systems is an emerging area of research, our work explores how the tokenization of language could produce new interfaces for the exploration of sociolinguistic phenomena. Generative topolinguistics presents a perspective on artificial sociality in simulated environments, employing a functionalist framework to capture its structure through token interactions inside the high-dimensional vector spaces of modern LLMs. In our model, language functions geometrically while sociality functions topologically, with changes in the topology of movement in semantic space interpreted as social behavior. Through the proposal of a bidirectional interface for large language models, we speculate how structural manipulations of semantic space could lead to the emergence of various sociolinguistic features that scaffold toward interpretable higher-order social phenomena.