A Biologically Plausible Parser

A Biologically Plausible Parser Daniel Mitropolsky Department of Computer Science Columbia University New York, NY 10027, USA Michael J. Collins Google Research New York, NY 10011, USA Christos H. Papadimitriou Department of Computer Science Columbia University New York, NY 10027, USA Abstract We describe a parser of English effectuated by biologically plausible neurons and syn- apses, and implemented through the Assembly Calculus, a recently proposed computational framework for cognitive function. We demon- strate that this device is capable of correctly parsing reasonably nontrivial sentences.1 While our experiments entail rather simple sentences in English, our results suggest that the parser can be extended beyond what we have imple- mented, to several directions encompassing much of language. For example, we present a simple Russian version of the parser, and discuss how to handle recursion, embedding, and polysemy. 1 Introduction Language is a distinguishing human function in- volving the creation, articulation, comprehension, and maintenance of hierarchically structured in- formation about the world. It is beyond doubt that language is achieved through the activity of neurons and synapses—but how? There has been extensive previous work in cognitive experiments —psycholinguistics, computational psycholin- guistics, and brain imaging—that has led to many insights into how the brain processes language (see Section 2 for an overview). However, no concrete narrative emerges yet from these advances about the precise way in which the activity of individual neurons can result in language. In particular, we are not aware of an experiment in which a rea- sonably complex linguistic phenomenon is repro- duced through simulated neurons and synapses. This is the direction pursued here. Developing an overarching computational un- derstanding of the way neurons in the human brain can make language is hindered by the state of neuroscience, which (a) predominantly 1Codeavailablehttps://www.github.com/dmitropolsky /assemblies. studies sensory and motor brain functions of an- imals other than humans; and (b) with respect to computational modeling, focuses on the level of neurons and neuronal circuits, and lacks the kind of high-level computational model that seems to be needed for understanding how high-level cogni- tive functions can emerge from neuronal activity. Very recently, a computational model of brain function, the Assembly Calculus (AC), has been proposed (Papadimitriou et al., 2020). The AC describes a dynamical system involving the fol- lowing parts and properties, well attested in the neuroscience literature: (a) brain areas with ran- dom connections between neurons; (b) a simple linear model of neuronal inputs; (c) inhibition within each area such that the top k most activated neurons fire; (d) a simple model of Hebbian plas- ticity, whereby the strength of synapses increase as neurons fire. An important object emerges from these properties: the assembly, a large set of highly interconnected excitatory neurons, all residing in the same brain area. Assemblies were hypothesized by Hebb seven decades ago (Hebb, 1949), and were identified in the brain of experimental animals two decades ago (Harris, 2005). There is a growing consensus that assemblies play a central role in the way brains work (Eichenbaum, 2018), and were recently …

Content truncated to 500 words