Commentary on Neuroemergentism: A framework for studying cognition and the brain. The neurocomputations of neuroemergentism: Long-term memory plus reinforcement learning = language?

摘要

Hernandez and colleagues proposed a new framework, neuroemergentism, according to which complex functions (such as language) may originate from the interactions of existing neural mechanisms that have evolved for potentially different reasons, and whose recycling within a new network results in apparent functional specialization. Here, I will consider neuroemergentism from a computational perspective. In particular, I will examine how this approach could provide a way to make sense of linguistic abilities within two accepted, general-purpose neurocomputational frameworks: rational (Bayesian) retrieval of information from long-term memory, and reinforcement learning. These two frameworks might provide a natural way to account for the two most important components of language, lexicon and grammar, and they are consistent with the established neural substrates of linguistic processes. In agreement with the requirements of Hernandez and colleagues, this hypothesis is also compatible, in principle, with language genetics and language development.