Simple syntactic rules through rapid synaptic changes

Syntax is a central organizing component of human language but few models explain how it may be implemented in neurons. We combined two rapid synaptic rules to demonstrate how neurons can implement a simple grammar without accounting for the hierarchical property of syntax. Words bind to syntactic roles (e.g. “dog” as subject or object) and the roles obey ordering rules (e.g. subject → verb → object), guided by predefined syntactic knowledge. We find that, like humans, the model recalls sentences better than shuffled word-lists, and when given the permitted role orderings, and a set of words, the model can select a grammatical ordering and serialize the words to form a sentence influenced by the priming effect (e.g. producing a sentence in the passive voice after input of a different sentence also in the passive voice). The model also supports languages reliant on affixes, rather than word order, to define grammatical roles, exhibits syntactic priming and demonstrates typical patterns of aphasia when damaged. Crucially, it achieves these using an intuitive representation where words fill roles, allowing structured cognition.