Understanding the Properties of Generated Corpora

TL;DR

This paper introduces tools to analyze the properties of automatically generated text corpora, revealing significant differences between leading generative models and enhancing understanding of their outputs.

Contribution

It presents novel tools for analyzing generated text corpora and applies them to compare different generative technologies, providing new insights into their properties.

Findings

Significant differences in generated corpora by different models

Tools reveal detailed corpus properties

Enhanced understanding of generative model outputs

Abstract

Models for text generation have become focal for many research tasks and especially for the generation of sentence corpora. However, understanding the properties of an automatically generated text corpus remains challenging. We propose a set of tools that examine the properties of generated text corpora. Applying these tools on various generated corpora allowed us to gain new insights into the properties of the generative models. As part of our characterization process, we found remarkable differences in the corpora generated by two leading generative technologies.