I am grateful to the authors of the commentary articles for identifying productive points of pressure for any neurocomputational account of syntax: whether semantic interpretation can proceed without a full syntactic derivation; how dynamical motifs are selected and coordinated; whether proposed mechanisms generalize across languages and modalities; how evolution may have repurposed older memory circuitry; and how neural dynamics distinguish types and tokens. Here, I clarify that ROSE is an architecture for implementing hierarchical syntactic computation, bringing with it no commitment that its full code is obligatory for every act of meaning construction. The commentaries suggest a number of compelling concrete extensions: task-dependent gating of S/E; factorized R/O subspaces; representation-relative rather than language-specific complexity measures; comparative tests of neural reuse; and occurrence-sensitive phase or state-space addresses. These thoughtful extensions sharpen the central aim of ROSE to formulate multiscale, falsifiable links between formal properties of language and neural dynamics.
In a previous discussion paper, we conducted a systematic review of 13 implicit memory studies that reported fMRI activity in the hippocampus. It was determined that each of these studies suffered from at least one of the following confounds that could produce hippocampal activity: explicit memory contamination, imbalanced attentional states, imbalanced stimuli between conditions, or differential novelty. Thus, we concluded that implicit memory is not associated with the hippocampus. Six commentaries on that discussion paper were received from Hannula (2024), Henke and Ruch (2024), Rosenthal (2024), Spaak (2024), Thakral et al. (2024), and Züst (2024). In this response, we address several issues to clarify our theoretical and experimental framework and maintain that when confounds are excluded, there is no evidence for hippocampal involvement during implicit memory. This discourse highlights the need for conceptual clarity, methodological consistency, and the critical distinction between neural engagement and behavioral expression.
Slotnick (2026) provides a large number of simulations to demonstrate that statistical power in fMRI can be improved by including the sample size N when calculating an appropriate cluster extent threshold for thresholding statistical maps. I argue that the problems acknowledged by Slotnick can instead be solved using threshold free cluster enhancement (TFCE) and a permutation test, which together apply a large number of cluster forming thresholds and implicitly model the sample size as well as the spatial autocorrelation. Furthermore, I briefly mention some other approaches for increasing statistical power in fMRI.
fMRI research is highly prolific but raises multiple concerns. Many competing statistical methods and respective packages are available using different assumptions, none of which applies equally well to all settings. However, the most fundamental concerns are not about the statistical machinery, but about issues of reproducibility, utility, and even construct validity. One can probe how much the field would benefit by statistical refinements, the conduct of larger studies and/or improved reproducibility practices. Alternatively, maybe fMRI research should largely be abandoned with focus shifting toward developing imaging methods with construct validity for granular neuronal activity and higher potential for clinical utility.
