Reorganization of the neurobiology of speech perception after sentence overlearning

Abstract

Existing models of the organization of language and the brain are fixed. They assume fixed linguistic units (like phonemes) that are combined into larger units (like words and sentences). Perceiving these units is said to be supported by a distributed but fixed set of brain regions. There is, however, little evidence for fixed linguistic units or stable language regions. Our NOLB Model is an alternative in which linguistic unit size and the corresponding neurobiology of speech perception dynamically self-organizes as a function of prior listening experience and current context. Here we evaluate this model by testing the hypothesis that increasing experience producing specific sentences will increase their linguistic unit size, creating “prefabs”. This will result in the reorganization of brain regions supporting the perception of those sentences. Based on studies of preserved formulaic speech in aphasia, we hypothesized that those would include right hemisphere regions and subcortical structures like the basal ganglia and cerebellum.

In each of two functional magnetic resonance imaging (fMRI) sessions, 12 participants “passively” listened to two sentences repeated 60 times each and 60 other novel sentences. All sentences were 2.5 seconds long and randomly presented over the course of six runs in a slow event related design (TR=700ms). In a final run, participants heard and spoke the 60 novel (non repeating) sentences. The two fMRI sessions were separated by 15 days. During this period, participants trained at home by producing the two repeated sentences 30 times each, twice a day. Finally, they completed memory recall, sentence completion and lexical decision tasks in both sessions to assess learning.

Behavioural results confirm that participants significantly overlearned the two sentences repeated at home and showed evidence that they were treated as “prefabs” during recall. For example, reaction times to complete the final word of repeated sentences decreased by over a second from session one to two. Correspondingly, we found a dramatic bilateral reduction in activity in posterior superior temporal, inferior parietal, insula, inferior frontal and ventral premotor cortices. There was an increase in cortical activity after training around the planum temporale bilaterally and a large increase in right motor and somatosensory cortices. There was also an increase in subcortical activity in the hippocampus, thalamus, caudate and cerebellum.

These results suggest that the distribution of pattern of activity supporting speech perception is not fixed but, rather, dramatically shifts as of function of individual experience with speech production. Specifically, repeated experience speaking the same word sequence seems to change the memory representation of those words to be collectively more unit like or prefabricated. This trace is subsequently used by the brain in the process of speech perception, involving a different set of brain regions than is said to support language in most contemporary neurobiological models. Results are more consistent with the described NOLB model in which linguistic units and associated brain regions are dynamically organizing. Results also warrant a more serious consideration of how overlearning and formulaic speech might be harnessed to aid in recovery from aphasia.