clarity · spectral grammar analysis

Grammar as a measurable signal.

Discourse structure is analyzed as a weighted graph. The graph Laplacian's eigenvalue spectrum measures how sentences link to each other, with no language model involved. It measures the text, not you — and connectivity is not coherence: repetition raises it. That last clause was written as a caveat and turned out to be the finding.

This line used to say eigenvalue bands (θ α β γ) correlate with EEG oscillation bands. That was a hypothesis stated as a result, and it has never been tested on brain data — the one EEG file on the machine that produced it is a joint-agency dataset, not language, and the correlation once cited here came from a script that generates its own data. Tested on real parse trees instead, the eigenvalue reading tracks whether a sentence parses, not what it means, and held out against sentences we did not write it performs at chance. · the whole result