The Relational Formation of Possibility: Recursive Determination and the Hidden Logic of Large Language Models (LLMs)

Large language models (LLMs) are typically described as probabilistic systems that select outputs from a distribution over possible continuations. This paper argues that such accounts are insufficient. Instead, it develops a hierarchical relational ontology in which determination proceeds through the progressive organization of relational constraints. Within this framework, possibility is not pre-defined but relationally formed through recursive continuation across multiple levels. Drawing on a semiotic account of relational ordering, determination is understood as the generation, stabilization, and re-engagement of distinctions through the interplay of synchronization, recursion, and return. Large language models are then interpreted as computational models of semiotic agency: systems that operate on the structured availability of relational form without enacting the processes through which it is produced. This explains their capacity for coherent, context-sensitive output as a function of hierarchical constraint rather than probabilistic selection. At the same time, it clarifies a fundamental limitation: they generate formal determination without interpretive unity, giving rise to an ethical imperative that emerges in their use rather than their operation. More broadly, the framework developed here may be read as a formal articulation of relational ontology in which determination emerges through the hierarchical organization of relational constraints.