The project has achieved most of its objectives and milestones for the period (considering the delay to the COVID-19 pandemic, and the early termination of the grant), with relatively minor deviations.
The project pursued two general aims: 1) to characterize the process of proceduralization to ascertain whether it entails a transformation of declarative representations or rather mere deeper declarative processing; and 2) To elucidate the dynamics of these emergent action-oriented representations in critical brain regions. Given the delays explained above, we could only achieve the first general aim. Our fMRI study (WP1) reveals that instruction implementation recruits not only declarative but also procedural representations of instructed contents.
In WP1 we had human participants perform an instruction following task inside an MRI scanner. Importantly, our task allowed to tag, within each trial, instructions that were relevant for future behavior and instructions that were irrelevant. To evaluate brain activity during the implementation stage of our task we created canonical templates of procedural and reclarative representations. That is, in two separate tasks, we tried to approximate to process-pure measures of procedural and declarative coding formats. Then, we tracked to what extent these templates were reinstantiated during implementation in the main task. First, we found that templates of the relevant instructions of the trial were reinstantiated to a larger extent than irrelevant ones, suggesting that our tracking procedure was efficient. Second, contrary to a hard interpretation of the serial-coding hypothesis, both declarative and procedural representations seem to explain unique parts of neural activity in relevant brain regions. Thus, it does not seem to the case that only procedural information explains brain activity during the implementation stage, and rather, some declarative information might be needed as well. However, we did find some evidence that suggests a more crucial role of procedural representations. Specifically, the strength of these representations predicted behavioral performance: the more an instruction was coded in a procedural format, the faster and more efficient participants would later on execute that instruction, Importantly, this was not the case for declarative information. We interpret this result in the context of output gating. Similar to the idea of an input gate that limits what information enters working memory, some models propose an additional output gate that determines what information will drive behavior. We believe implementation might be a particular instance of output gating that engages relevant brain regions to transfer relevant content into a state that is optimal for behavior.
