An Empirical Verification of Coverage and Correctness for a General-Purpose Sentence Generator

Abstract

This paper describes a general-purpose sentence generation system that can achieve both broad scale coverage and high quality while aiming to be suitable for a variety of generation tasks. We measure the coverage and correctness empirically using a section of the Penn Treebank corpus as a test set. We also describe novel features that help make the generator flexible and easier to use for a variety of tasks. To our knowledge, this is the first empirical measurement of coverage reported in the literature, and the highest reported measurements of correctness.