Our research examines how the human brain codes and responds to regularity/uncertainty in ongoing input. We approach this question from the perspective of non-strategic statistical learning and focus on systems that are sensitive to regularities and that use this information to predict the near future.
The core questions we address are the following: are there systems sensitive to uncertainty independently of the factor affecting uncertainty; are there systems that respond to uncertainty in an a-modal, domain general manner; and finally, does sensitivity to uncertainty manifest itself in a monotonic relation between uncertainty and brain activity. Our work suggests the following: (1) different sources of uncertainty are tracked by different systems, (2) there is limited evidence for systems that track uncertainty in a modality general manner, and (3) responses to uncertainty are both monotonic and curvilinear.
A premise underlying our work is that tracking regularity is spontaneously triggered by environmental features, even when individuals do not explicitly track order in their environment. Our paradigms therefore rely on rapid stimulus presentation and do not require attention to the material whose order is being manipulated.
We have shown (Tobia, Iacovella, & Hasson, 2012) that different systems are sensitive to different facets of input order, such as the strength of transition constraints between input elements or their relative diversity; there was no single system that generally coded for input statistics. The same study showed that certain brain systems track the strength of transition constraints in a monotonic pattern, with increased activity to more disordered inputs, consistent with the possibility that they code for order. However, some areas show parabolic response profiles with maximal activity for highly ordered and highly disordered states, suggesting sensitivity to statistical structure, but according to a different coding principle. We have also investigated whether regularity-sensitive systems operate in a modality-specific manner (Nastase, Iacovella, & Hasson, 2014). Series varying in disorder were transposed into tonal sound sequences or sequences of simple visual shapes presented in different screen locations. There as no region that robustly tracked the degree of disorder in both auditory and visual series. Instead, the modal pattern was one of modality-specific systems that were sensitive to degree of disorder in just the auditory or just the visual modality.
In other work (Nastase, Iacovella, Davis, & Hasson, 2015) we examined effective connectivity patterns as function of input uncertainty. We documented that whole-brain functional connectivity of the anterior cingulate cortex varies systematically with input disorder: In some brain regions connectivity reduced monotonically with uncertainty, whereas in others it showed an inverse curvilinear pattern. In accordance with arguments from complexity sciences, these findings suggest that different systems are sensitive to the uncertainty of the input and the complexity of the mechanism generating the input. We have also found (Andric & Hasson, 2015) that the regularity of auditory input series has a fundamental impact on the way whole-brain networks organize, impacting the modular organization of functional networks, the partition structure and the degree distributions of such networks.
We have also identified systems sensitive to perceived changes in regularity (Tobia, Iacovella, Davis, & Hasson, 2012), finding that posterior-medial systems show specific activity profiles prior to subjective perception of changes in regularity. Interestingly, these systems have been implicated in event segmentation of movies and narratives. Other questions addressed in that work deal with the role of the hippocampus in signaling changes in regularity and the role of explicit vs. implicit attending in the process. In recent work (Cashdollar et al., 2016) we used MEG to examine the timeline of predictive processes by studying pre-stimulus activity in series were predictions were statistically licensed as compared to random series. We found that differential patterns of pre-stimulus theta-band activity, terminating immediately prior to stimulus presentaiton, correlated with participants' working memory capacity,
In a different line of work we evaluated whether neural responses to uncertainty, as well as ratings of perceived uncertainty, depend on familiarity with input tokens (Tremblay, Baroni, & Hasson, 2012). We presented regular and irregular series consisting of familiar tokens (syllables) or unfamiliar ones (bird chirps). Focusing on subdivisions of the supratemporal plane, we found that several posterior regions including planum temporale are sensitive to statistical structure for both speech and non-speech inputs. Interestingly, some regions were only sensitive to statistical structure for non-speech input, and a behavioral study suggested this might be due to different proficiency in segmenting speech and non-speech inputs. We have also shown that individual differences in sensitivity to input regularity correlate with cortical thickness in brain regions involved in higher level language comprehension (Deschamps et al., 2016). More recently (Tremblay et al., 2016) we have linked the superior temporal region to predictive processing of syllables. In a study where we presented bi-syllable pairs, we found that activity in this region tracked the frequency of the first syllable, the mutual information between the two syllables, but not the frequency of the second syllable.