Fixed word context tests did not confirm shared prediction gains in six iEEG patients and 22 fMRI adults

We tested whether recent word context adds neural prediction beyond sound, timing, lexical features and current word vectors. Public iEEG and fMRI came from separate human studies and stories. Three patients supplied development. Fixed tests used six other patients and 22 fMRI adults. A 20-word GloVe context feature entered purged, nested ridge models. Mean iEEG gain was +0.0067% of null loss. Its 95% person interval was -0.0355% to +0.0593%. The fMRI mean was -0.0142% (interval -0.1583% to +0.1663%). Person and common-block intervals also crossed zero. Gains were positive in 4/6 and 5/22 people. A required shuffled-dictionary comparison failed. The past-over-future claim also failed. Both shared Holm p values were 1. No positive cross-modal neural finding is claimed. The result limits this fixed model and regional sample. It does not show absent semantic processing or a common causal mechanism. The package preserves code, public numerical inputs, source versions and terms, all scores, controls, figures and repeat steps. Original-source rebuilds matched one compact record per modality exactly. All 28 primary person scores passed a separate arithmetic check.

human neuroscience · intracranial EEG · fMRI · language context · semantic dissimilarity · negative result · encoding models · GloVe