Classifier Combination for Grammar-Guided Sentence Recognition

Abstract

In this paper we consider a category of classification tasks, where the classification results are sentences of words subject to a given grammar. The particular nature of grammar-guided sentence recognition makes classifier combination rules known from the literature not applicable any longer. We propose a conceptually new approach to classifier combination that consists of three main components: class set reduction, inconsistency localization, and resolution. The proposed algorithm represents a framework for classifier combination in grammar-guided sentence recognition that is applicable to a variety of different tasks. Experimental results will be shown for the task of spoken email command recognition, where an acoustic and a visual classifier are combined.