Trash on the Sentence Level: From Detection to Minimization

This paper extends the notion of trash from the lexical to the sentence level in semantic discourse models. Whereas words may or may not align with a semantic attractor, sentences can be judged by whether they contribute orientation or remain inert. We distinguish three trajectories: meaningful sentences converge into the attractor basin, sentences of stagnation remain practically invariant, and sentences of divergence drift away from orientation, producing semantic noise. Early detection is possible by examining the first iterations of the attractor dynamics: negligible displacement indicates stagnation, growing displacement signals divergence, and monotonic decrease marks convergence. This analysis parallels Lyapunov stability theory, where stability corresponds to meaningfulness and instability to trash. Once identified, trash sentences can be compressed, filtered, recycled, or retrospectively re-evaluated, depending on discourse needs. In this way, the model provides both a diagnostic tool and a minimization strategy for maintaining orientation and coherence in discourse.