Using complementary multi-modal neuroimaging methods (functional magnetic resonance imaging (fMRI) and electrocorticography (ECoG)) in conjunction with rigorous behavioral approaches, we will examine the role of multiple cortico-striatal and sensory cortical networks in the acquisition and automatization of novel non- speech and speech categories in the mature adult brain. We test the scientific premise of a dual-learning systems (DLS) model by probing neural function using fMRI or ECoG during the process of feedback-dependent category learning. In contrast to popular single-learning system (SLS) approaches, DLS posits that two neurally- dissociable cortico-striatal systems are critical to speech learning: an explicit, sound-to-rule cortico-striatal system, that maps sounds onto rules, and an implicit, sound-to-reward cortico-striatal system that implicitly associates sounds with actions that lead to immediate reward. Per DLS, the two systems contribute to the emerging expertise of the learner. Via closed loops, the highly plastic cortico-striatal systems ?train? key less labile temporal lobe networks to categorize information by validated rules or rewards. Once categories are learned to the point of automaticity, cortico-striatal networks are no longer required to mediate behavior. Instead, abstract categorical information within the temporal cortex drives highly accurate speech categorization.
In Aim 1. 1, we use fMRI to examine the relative dominance of the two cortico-striatal networks in learning multidimensional non-speech category structures that are experimenter-constrained to either rely on rules (rule- based, RB), or on implicit integration of multidimensional cues (information-integration, II). We predict that key regions of the sound-to-rule network, the prefrontal cortex (PFC), hippocampus, and caudate nucleus show greater activation during RB, relative to II learning; in contrast, key regions within the sound-to-reward network, the putamen and the ventral striatum show greater activation during II, relative to RB learning.
In Aims 1. 2 and 1.3, we leverage the temporal precision of ECoG measurements from high-density grids in temporal, PFC, and Hippocampal regions to examine the extent to which temporal lobe representational changes during RB learning are an outcome of error-monitoring processes within the PFC and hippocampus.
In Aim 2, we probe neural function using fMRI or ECoG to assess network and representational changes during the acquisition of non- native supra-segmental and segmental categories to native-like performance levels. We predict that early ?novice? speech acquisition involves sound-to-rule mapping; later ?experienced? involves sound-to-reward mapping. In contrast, only cortical networks are active at the point of ?native-like automaticity? in categorization. Using innovative single-trial classification and network-level decoding analyses on ECoG data, we examine learning-induced changes in speech representation within the temporal lobe. Further, we examine the extent to which error monitoring processes within the PFC and the hippocampus drive emergent temporal lobe representations of novel speech categories.
This proposal will examine the neurobiological mechanisms underlying the representation, acquisition, and automatization of speech signals. The proposal will employ complementary invasive and non-invasive neuroimaging methods to examine the premise that neurally dissociable networks operate at different stages of learning expertise. The proposed studies will yield foundational knowledge and advance our understanding of the role of the striatum in mediating neuroplasticity in the mature, adult brain.
|Feng, Gangyi; Gan, Zhenzhong; Wang, Suiping et al. (2018) Task-General and Acoustic-Invariant Neural Representation of Speech Categories in the Human Brain. Cereb Cortex 28:3241-3254|
|Reetzke, Rachel; Xie, Zilong; Llanos, Fernando et al. (2018) Tracing the Trajectory of Sensory Plasticity across Different Stages of Speech Learning in Adulthood. Curr Biol 28:1419-1427.e4|