Natural language syntax complies with the free-energy principle

Abstract

Natural language syntax yields an unbounded array of hierarchically structured expressions. We claim that these are used in the service of active inference in accord with the free-energy principle (FEP). While conceptual advances alongside modelling and simulation work have attempted to connect speech segmentation and linguistic communication with the FEP, we extend this program to the underlying computations responsible for generating elementary syntactic objects. We argue that recently proposed principles of economy in language design—such as “minimal search” and “least effort” criteria from theoretical syntax—adhere to the FEP. This permits a greater degree of explanatory power to the FEP—with respect to higher language functions—and presents linguists with a grounding in first principles of notions pertaining to computability. More generally, we explore the possibility of migrating certain topics in linguistics over to the domain of fields that investigate the FEP, such as complex polysemy. We aim to align concerns of linguists with the normative model for organic self-organisation associated with the FEP, marshalling evidence from theoretical linguistics and psycholinguistics to ground core principles of efficient syntactic computation within active inference.

The Bayesian brain and cooperative communication in schizophrenia

“…Language, by its very nature, provides these shared tools at several levels of the hierarchical priors (phonetics, semantics, syntax). 25 Other features, such as culture, shared identity and common context, also provide higher-level alignment for shared message passing. 26 This sort of prior alignment is said to be species-specific and adaptive, making cooperative communication less demanding and more likely to occur in our social world.…”

Models of the Translation Process and the Free Energy Principle

“…It does not (explicitly) address aspects of meaning construction, and can accommodate representationalist and nonrepresentationalist views. FEP/active inference has been used to elicit different kinds of processes (e.g., neural and symbolic) across different domains and it has been instrumental to support representationalist [55] and non-representationalist [64] approaches in cognitive sciences. FEP/active inference is thus suited for modelling horizontal/priming/s-mode processes, as well as vertical/reflective/i-mode translation processes under a representationalist, 'classical', and/or non-representationalist approaches to translational cognition.…”

“…As discussed, e.g., in (Parr et al [29], Smith et al [49]), free energy can be factorised in different ways. In Equation (2), it is the "trade-off between accuracy and complexity" [55]. Minimising free energy corresponds to minimising complexity while maximising the production accuracy.…”

A future without a past: Philosophical consequences of Merge

“…One of Mukherji's main complaints is that "there is very little connection between the formal and empirical studies of human language and related aspects of human cognition and the rest of the topics pursued in cognitive science" (p. 54). There have been few attempts to migrate the concerns of theoretical linguistics to other domains of the cognitive sciences (see Murphy, Holmes, & Friston, 2022). Mukherji's charge is legitimate, but given his philosophical interests there is limited discussion of some of the more recent developments in cognitive neuroscience and experimental psychology that bear on these topics.…”