Metacognitive Text Organizastion Semiotic and Rhetorical Agency in LLMs

This paper proposes a semiotic-rhetorical framework for analyzing Large Language Models (LLMs) as textual agents. Against dominant approaches that focus on language, intentionality, or artificial intelligence, I argue for a text-centered perspective: LLMs produce and process texts, and their agency must be understood at this level. Drawing on semiotics and rhetoric, I establish that LLMs qualify as semiotic agents through their capacity to process and produce textually organized signs that enable further textual responses. They additionally qualify as rhetorical agents through their strategic organization of texts along multiple strategic levels, particularly the legitimative (justification structures) and modal (certainty/uncertainty markers) level. A key phenomenon illuminating this rhetorical agency is the metacognitive illusion: LLM outputs simulate reflection on their own epistemic processes through textual self-reference, creating the appearance of metacognitive monitoring and regulation. The theoretical innovation of this approach lies in treating metacognitive illusion not as a deviation from epistemic transparency but as a structural precondition for the rhetorical functionality of text-based interaction between heterogeneous agents. The analysis demonstrates how LLMs achieve rhetorical functionality without intentionality, consciousness, or genuine metacognition – purely through textual organization.