Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster E in Poster Session E: Thursday, August 6, 10:30 am – 12:15 pm, Kimmel Center, Shorin & Rosenthal Rooms

Decoding Second-Language Proficiency from Shared Network Representations Using PCA and ICA

Onila R. N. M. Don¹, Lucy L.W. Owen¹; ¹University of Montana

Presenter: Onila R. N. M. Don

In bilingual neuroimaging, conclusions about second- language (L2) proficiency depend on how neural activity is represented. Although L2–L1 contrasts are common, subtraction may discard individual-difference information if proficiency is expressed in absolute L2 engagement or network coordination. We analyzed task fMRI from adult language learners performing semantic animacy judgments in L1 and L2 (Gurunandan et al., 2019), using shared PCA/ICA representations with leakage-aware decoding and permutation testing (Valente et al., 2021; Varoquaux et al., 2017). At K = 20, PCA L2-only static features showed the clearest marginal decoding signal, whereas ICA static summaries were weaker but improved with targeted connectivity. Sensitivity analyses across K ∈ {5,10,15,20,25,30} showed that the signal was not tied to one component number. Overall, the strongest evidence favoured L2-centred and connectivity-augmented representations; Δ(L2 − L1) did not show a robust advantage. A paired comparison showed a consistent PCA L2-only advantage over Δ(L2 − L1), although label permutation was marginal. Given the small sample, we interpret this as representational analysis, not a generalizable classifier.

Topic Area: Methods, Tools, Theory & Neural Coding