Using Sparse Parameter Estimation for Semantic Parsing

This paper addresses the problem of semantic parsing, by which natural language sentences are translated into a form which conveys their underlying meaning. Semantic parsing involves a parameter estimation process, which is a convex optimization problem. The optimization formulation of previous approaches often requires huge amount of time to converge due to the high dimensional feature space. In this paper we introduce a fast semantic parsing framework which uses l1norm regularized learning to get a sparse model and better convergence speed. Experiments demonstrate overall higher performance of our semantic parsing system using inverse λ, generalization and l1 regularization. Regularized parameter updating shows significantly improvement on the learning speed and reduced model size.