Neural substrates of a symbolic action grammar in primate frontal cortex

At the core of intelligence is the capacity to solve new problems. In turn, problem-solving has been hypothesized to depend on operations resembling symbolic grammars, with two components: discrete units (symbols) and rules for recombining symbols into new composite representations (syntax). Whether and how grammars are implemented in neural substrates remains unknown. We have established a research program to study the mechanisms of grammatically structured action sequencing. At its core is a task in which macaques learn and generalize symbolic action grammars for drawing geometric figures. These grammars consist of action symbols—recombinable units of motor behavior (e.g., a “circle” stroke)—and syntactic rules (e.g., “repeat circles N times”). Building on this drawing task, we identified a neural population encoding action symbols, localized to a specific area of frontal cortex. First, we found behavioral evidence for three critical properties that indicate actions have an underlying symbolic representation: (i) invariance over motor parameters; (ii) categorical structure, reflecting discrete classes of action; and (iii) recombination into novel sequences. In simultaneous recordings across motor, premotor, and prefrontal cortex, we found that planning-related activity in ventral premotor cortex (PMv) encodes actions in a manner that also reflects these three symbolic properties. These findings reveal a representation of action symbols localized to PMv, and therefore identify a putative neural substrate for symbolic cognitive operations. In this talk, I will mainly present this task paradigm and our discovery of action symbols. I will also briefly discuss early answers to the logical next question: what are the neural substrates of syntax?