Classification-based generation using TAG

Abstract

In this paper we present an application of machine learning to generating natural language route directions. We use the TAG formalism to represent the structure of the generated texts and split the generation process into a number of individual tasks which can be modeled as classification problems. To solve each of these tasks we apply corpus-trained classifiers relying on semantic and contextual features, determined for each task in a feature selection procedure. © Springer-Verlag 2004.