Same-category words had slightly more similar cross-list SEEG patterns in 18 catFR1 subjects, with lexical confounds unresolved

We tested whether semantic category relates to similarity in word-encoding SEEG patterns in 18 subjects from the catFR1 categorized free-recall task (session 0). For each word, we used archived band-power patterns across bipolar pairs from 0 to 1.6 s after word onset, aligned using events sample/fs time. After removing list and serial-position effects, same-category words from different lists had slightly more similar patterns than different-category words matched on list distance. The preregistered high-frequency activity (70–150 Hz) result was +0.0033 in mean Fisher z difference (95% bootstrap CI [+0.0008, +0.0060], dz = 0.56, 12/18 subjects positive, exact one-sided sign-flip p = 0.013). Combining all five bands gave +0.0035 (95% CI [+0.0015, +0.0055], dz = 0.77, 15/18 positive, p = 0.0025; Holm-adjusted p = 0.005). The effect was small; within-subject permutation p < 0.05 occurred in only 2/18 subjects. The lag-matched within-list HFA result was not significant (p = 0.052). Recalled and unrecalled pairs did not differ, with a wide confidence interval. Category is confounded with lexical and visual word properties, so these findings show category-related pattern similarity, not abstract semantic coding. The analysis used only 18 subjects with archived features, session 0, and a 0–1.6 s window; electrode coverage varied, and brain regions and time windows were not tested.

SEEG · word encoding · semantic category · pattern similarity · categorized free recall · catFR1 · high-frequency activity · reproducible analysis