Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
Language selectivity may be highly localized: Evidence from univariate and multivoxel analyses
Poster A111 in Poster Session A, Tuesday, October 24, 10:15 am - 12:00 pm CEST, Espace Vieux-Port
Rebecca Belisle1, Terri Scott2, Tyler Perrachione1; 1Boston University, 2University of California, San Francisco
A longstanding question in cognitive neuroscience is the extent to which language and domain general cognitive processes are functionally segregated in the brain. Many recent studies using within-subject language localizer tasks claim that there is strict functional segregation of language-selective brain areas. However, these studies typically consider only the voxels that are maximally responsive to language vs. domain general tasks—usually only the top 10%. It remains unclear how the other 90% of voxels are responding to language vs. domain-general tasks, especially whether domain specificity changes in a sharp or gradual manner. We obtained functional MRI scans from 24 healthy, neurotypical young adults, who performed two fMRI tasks: an auditory language localizer (involving passive listening to intact speech or incomprehensible degraded speech) and a visual-spatial working memory task (involving remembering sequences of 6 (hard) or 3 (easy) locations in a grid). Language-selective responses were defined as the magnitude of the intact > degraded contrast from the language localizer, and domain general-selective responses were defined as the hard > easy contrast from the working memory task. In within-subject analyses, we measured the relative language- vs. domain general-selectivity of voxels within bilateral frontal and temporal anatomical regions of interest (aROIs), including IFG, MFG, STG, and MTG. We sampled voxels in these regions as a function of their uniformly distributed deciles of selectivity to the language task and analyzed their selectivity via three metrics: (1) univariate response magnitude within each decile to the contrasts of interest, (2) multivoxel pattern analysis (MVPA) of the response profile within each decile across different runs of the task, and (3) MVPA of response profile magnitudes within decile across the two tasks. All aROIs showed a high degree of localized selectivity for language based on univariate response magnitudes. The top decile of voxels was strongly language selective, with selectivity dropping off dramatically for the second decile and below. The bottom decile also often showed a sharp selectivity inversion, favoring degraded > intact speech. Response to the working memory task showed the opposite pattern, with strongest response in the lowest language-selective decile and relatively weak response in all other deciles. The MVPA analysis of within-task response profiles likewise showed strikingly circumscribed patterns of response consistency in only the highest (and lowest) deciles. The top language-selective decile in all aROIs had very high between-run voxelwise correlations, whereas the intermediate eight deciles showed essentially uncorrelated activity patterns between runs. Interestingly, the bottom decile also showed a high degree of between-run correlation, consistent with its sharper selectivity for the inverse contrast. The between-task MVPA analysis showed near-zero correlations in the top and intermediate deciles of language selectivity and negative correlations in the lowest deciles, where the degraded > intact contrast had a similar response profile to the hard > easy working memory contrast. These results suggest that frontal and temporal regions have sharply circumscribed language-selective and domain general-selective regions, where maximally selective language regions show much stronger responses and much more consistent response patterns than intermediately selective areas.
Topic Areas: Methods, Control, Selection, and Executive Processes