On the Distinction Between Interpretation and Formal Continuity: Why human-LLM interaction works -- and why it often doesn't

Abstract

Large language models (LLMs) often exhibit puzzling behaviors that remain difficult to explain within conventional accounts of representation, information processing, or prediction alone. Context can exert disproportionate influence over outcomes, small interventions can reorganize entire discussions, conceptual drift can occur despite the preservation of coherence, and interactions may appear simultaneously insightful and misguided. Building upon previous studies of framework activation, possibility formation, and conceptual development in human-LLM interaction, this paper argues that these phenomena become intelligible once a distinction is made between interpretation and formal continuity. Human participants primarily contribute interpretative orientation, while language models primarily contribute the preservation and extension of formally organized conceptual trajectories. Productive interaction depends upon the synchronization of these complementary activities within a shared conceptual framework. From this perspective, phenomena such as framework activation, conceptual drift, progressive determination of meaning, and the coexistence of characteristic strengths and limitations emerge as consequences of the same underlying structure. More broadly, the paper argues that human–LLM interaction provides an unusual opportunity to examine the relationship between meaning and formal organization because processes that are ordinarily intertwined within human cognition become partially distinguishable. The significance of LLMs may therefore lie not only in what they can do, but also in what they reveal about the semiotic processes through which meaning becomes possible.

Keywords