Reasoning and other mental operations are believed to rely onmental models. Arguments have been made that mental mod-els share representational substrate with perception. Here, wedemonstrate that a neural dynamic architecture that perceptu-ally grounds language may also support the building of men-tal models. Supplied with a sequence of simple premises thatspecify the colors of object pairs as well as their spatial rela-tion, the architecture builds a mental model of the describedscene. We show how the neural processes of the architec-ture evolve in response to both determinate and indeterminatepremises. For indeterminate premises, we demonstrate thatthe preferred mental models observed in human participantsemerge from the underlying neural dynamics.

Parsing actions entails that relations between objects are dis-covered. A pervasively neural account of this process requiresthat fundamental problems are solved: the neural pointer prob-lem, the binding problem, and the problem of generating dis-crete processing steps from time-continuous neural processes.We present a prototypical solution to these problems in a neuraldynamic model that comprises dynamic neural fields holdingrepresentations close to sensorimotor surfaces as well as dy-namic nodes holding discrete, language-like representations.Making the connection between these two types of represen-tations enables the model to parse actions as well as groundmovement phrases—all based on real visual input. We demon-strate how the dynamic neural processes autonomously gen-erate the processing steps required to parse or ground object-oriented action.

The productivity and compositionality of language and thought have often been taken as evidence that higher cognitionis a form of information processing on systems of symbols with combinatorial syntax and semantics. We present anon-symbolic neural dynamic architecture that can ground combinatorial concepts in perception, i.e., establish a linkbetween a combinatorial concept and an object in the perceptual array. The components of a combinatorial concept treeare sequentially grounded from the leaves to the root, while the output of each grounding step is passed on to the nextgrounding step by means of a mental map. This way, compositionality is an emergent property of the neural dynamics anddoes not require any form of symbolic information processing. We discuss how this process account contrasts with otherneural accounts of compositionality and conclude with implications for the modeling of higher cognition.

We present a neural dynamic architecture that grounds sen-tences in perception which combine multiple concepts throughnested spatial relations. Grounding entails that the model getsfeatures and relations as categorical inputs and matches themto objects in space-continuous neural maps which represent vi-sual input. The architecture is based on the neural principlesof dynamic field theory. It autonomously generates sequencesof processing steps in continuous time, based solely on highlyrecurrent connectivity. Simulations of the architecture showthat it can ground sentences of varying complexity. We thusaddress two major challenges in dealing with nested relations:how concepts may appear in multiple different relational roleswithin the same sentence, and how in such a scenario variousgrounding outcomes may be “tried out” in a form of hypothesistesting. We close by discussing empirical evidence for crucialassumptions and choices made when developing the architec-ture.