Science learning requires students to build new mental models of imperceptible mechanisms (photosynthesis, circadian rhythms, atmospheric pressure, etc.). Since mechanisms are structurally complex and dynamic, building such mental models requires mentally simulating novel structures, their state changes, and higher-order transformations (transpiration, oscillation, liquid levels, etc.). These mental simulations also need to be intertwined with a series of external representations (ERs), including formal terms (stomata, guard cells, mass points, damping, etc.), schematic structures (figures, graphs, etc.), and mathematical notations (equations, vectors, etc.). Students' later encounters with these ERs activate the dynamic mental model of the mechanism. To help learners build such many-layered and dynamic mental models of mechanisms, teachers narrate, and act out, the structures, state changes, transformations, and related ERs. These cohere together to "constitute" (bring into being) the mechanism models. Based on classroom teaching data, we present a theoretical account of the cognitive mechanisms involved in this complex teaching-to-build process, extending the enactive simulation theory of language and distributed cognition theory. Since teaching narratives seek to approximate scientific mechanisms presented in textbooks, we extend this account to academic language, to understand how textbook descriptions embed mechanisms. We close with some theoretical and pedagogical implications of these two accounts.