Table · dataset · 2026

<p>Modality ablation across seven configurations (<i>n</i> = 25 folds; 5 repeats 5-fold GroupKFold; same splits across configurations). The four text-containing configurations (top block) form the multimodal family; the three text-free configurations (bottom block) are reported as a control to quantify the necessity of the linguistic channel. values are computed against the Text-only baseline. , , (paired <i>t</i>-test, <i>df</i> = 24; Holm–Bonferroni adjusted across the six non-reference contrasts).</p>

Listed in figshare

<p>Modality ablation across seven configurations (<i>n</i> = 25 folds; 5 repeats 5-fold GroupKFold; same splits across configurations).