Novels Mental-Model Decomposition (NoMMD): A DEL Approach to Reconstructing Mental States from Real-life Interactions

Treating the novel as a corpus of recursive social cognition, we present NoMMD, a framework that decomposes a novel into structured, per-mind mental-state records.It builds on the egocentric predicament (the axiom of fundamental disconnection of minds) which recasts the single shared model of dynamic epistemic logic (DEL) asper-mind doxastic models: no world is held in common, knowledge yields to belief, the text alone is common ground, and all representation is indexed to the mind thatholds it; from this reduction the record grammar and its typing are derived. The pipeline separates deep mentalizing from the structural processing that feeds it,mirroring the Parser-to-Engine architecture of the recursive-agent modeling which is fed with structured input in TSR framework (Tochilov, 2026). Observationbuilds the ledger; inference recovers the belief, desire, and intent behind an act and predicts the next move, with each forecast scored against the same text’scontinuation. The growing per-mind state is never overwritten; recall reads it by content, similar to a sparse distributed memory (SDM). A short original storyderived from Gruffalo (Singh et al., 2022) was decomposed; the reconstruction recovered its order-3 nested belief and forecasted each mind’s own and others’ nextmoves, conceiving the right move even where the selection slips under terror. The run over Hi-ToM and ExploreToM indicated the benchmarks’ prose inconsistencies.NoMMD can generate uncontaminated benchmarks by re-rendering (novellizing) mentalizing structures graded by recursion order. NoMMD is substrate-agnostic:consuming only a segmented record of who did and said what, it extends beyondthe novel to screenplays, transcripts, and videos reduced to text, which are closerto real, unresolved interactions.