A Semiotic Account of the Hierarchical Relational Organization of Large Language Models (LLMs): Implications for AI Development, Use and Evaluation

Abstract

This publication develops a semiotic and relational framework for understanding how large language models (LLMs) form and preserve organized continuations across training, inference, prompting, retrieval, evaluation and agentic action. Its specific concern is not the general theory of signs but how linguistic, conceptual, inferential, evidential and task relations become operative as hierarchical constraints on model continuation. Its central distinction is between information that is merely available to a model and information, frameworks, sources, or task constraints that actually govern continuation. This distinction makes it possible to understand hallucination, sycophancy, framework substitution, retrieval failure, conceptual memory loss, and agent drift as related failures of constraint preservation rather than as isolated defects. The publication consequently shifts attention from increasing generative capacity alone toward constructing broader systems in which formal continuation remains answerable to evidence, task identity, conceptual framework, external tests, and human purposes. The framework begins from the premise that the tokens processed by LLMs function as relationally ordered signs and that the texts on which language models are trained are also composed of signs whose uses have been shaped by human conceptual, inferential, referential and practical relations. It examines is how formal traces of those relations become available to an LLM and how they govern—or fail to govern—subsequent continuation. The module therefore investigates the formal organization of sign-mediated continuation while maintaining a categorical distinction between that organization and human interpretation. Developed through a structured interaction with ChatGPT, the module is intended to offer AI users and developers a complementary conceptual and diagnostic lens through which to consider large language model behaviour, design, use and evaluation. It presents a theoretical framework developed from a semiotic perspective and elaborated in the author’s related work (doi.org/10.5281/zenodo.21281882).

Keywords