Abstract

Models for text generation have become focal for many research tasks andespecially for the generation of sentence corpora. However, understanding theproperties of an automatically generated text corpus remains challenging. Wepropose a set of tools that examine the properties of generated text corpora.Applying these tools on various generated corpora allowed us to gain newinsights into the properties of the generative models. As part of ourcharacterization process, we found remarkable differences in the corporagenerated by two leading generative technologies.

 

Quick Read (beta)