Bayesian Selection of Grammar Productions for the Language of Thought

MARIANO SIGMAN; SALLES, ALEJO; SERGIO ROMANO; SANTIAGO FIGUEIRA

New York

Conferencia; Conference on Cognitive Computational Neuroscience 2017; 2019

Cognitive Computational Neuroscience

Probabilistic proposals of Language of Thoughts (LoTs) canexplain learning across different domains as statistical inference over a compositionally structuredhypothesis space. This space is usually formalized with a grammar where productions are fixed byresearchers based on intuitions about the domain. In this work we show a generic method to selectthe set of productions of a LoT grammar from a broad repertoire of productions by an inferentialprocess starting from experimental data, and then we test it in a specific LoT for the domain ofgeometry. A generic method like this could help prove the claim that the resulting grammars areactually suitable and could be a way to select the correct set of productions when it is not clearwhat a proper set should be.