Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Decoding of vocalized sentences, melodies, and emotions in the left and right perisylvian cortex of healthy adults
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Kelly Martin1, David Plaut1,2, Marlene Behrmann1,2; 1University of Pittsburgh, 2Carnegie Mellon University
We examined three types of auditory information processing that are typically lateralized—sentence processing to the left hemisphere (LH), and vocal emotion and melody processing to the right hemisphere (RH)—to understand the degree of participation and overlap among involved regions in both hemispheres. A female speaker recorded each of eight ten-syllables sentences in four emotions (angry, overjoyed, proud, and sad), four melodies sung a capella, and in a neutral tone. In an fMRI experiment, ten healthy, right-handed, English-monolingual adults (5 female, average age 22.5) heard each sentence 30 times, each emotion 24 times, each melody 24 times, and each neutral sentence 6 times (in six runs balanced across information type). On four additional catch trials per run, participants heard a one-back identical repeat and responded via button-press. Two first-level general linear models (GLMs) modeled the instances of each sentence and of each vocal tone, respectively, for each participant. Three whole-brain decoding analyses were performed on the two resulting sets of unsmoothed, native-space beta maps: one classifier was trained/tested on all sentence labels, a second was trained/tested on all emotion and neutral labels, and a third was trained/tested on all melody and neutral labels. For group analysis, a second-level GLM was run on normalized, smoothed accuracy maps to determine where the group’s classification-minus-chance (AMC) values were significantly above zero. Finally, a region-of-interest (ROI) analysis was performed using 32 10mm-radius spheres spread symmetrically across left and right perisylvian cortices. Confusion matrices were extracted for each ROI in native space using the classification parameters from whole-brain analyses, and then compared mean accuracies between categories and between regions for all participants. Whole-brain classification maps revealed one small cluster in left hemisphere (LH) region A4, with accuracy-minus-chance (AMC) values that survived voxelwise but not cluster-level correction. Results for melodies and emotions were both robust, spanning much of the temporal lobe with a concentration in anterior and middle areas. In the ROI analysis, sentences across vocal tones were only classifiable in LH A4. Vocal tone was classifiable across all sentences in LH A4, RH A4, anterior-middle STS and bilateral posterior-middle STG. Melody was classifiable in LH posterior-middle STG and emotion-only was classified in LH STG, RH posterior STS, RH middle STS and RH anterior-middle insula. Among the regions identified for both emotions and melodies, only LH posterior-middle STG showed a mean difference, in favor of melodies. Among all regions with AMC values greater than zero, the only hemispheric differences were in anterior-middle and middle STS, which were greater in the RH than LH for melodies. In summary, left A4 was a key region for sentence processing, but also for vocal tone processing. There was no region that processed sentences that did not also process one or both vocal tone categories. We identified additional areas, mostly in the RH, that processed both vocal tone categories, or selectively processed either melodies or emotions. These findings show that lateralization is not binary and that regions in the dominant/non-dominant hemisphere can process different dimensions of information that occur simultaneously.
Topic Areas: Methods, Speech Perception