Measuring and Modifying the Readability of English Texts with GPT-4

TL;DR

This study evaluates GPT-4's ability to assess and modify English text readability, finding it correlates well with human judgments and can influence text difficulty, though with some variability and limitations.

Contribution

It provides empirical evidence that GPT-4 can reliably estimate and alter text readability, outperforming traditional formulas and psycholinguistic indices.

Findings

GPT-4 estimates correlate highly with human judgments

GPT-4 can reliably modify text difficulty

Variability in human perception remains significant

Abstract

The success of Large Language Models (LLMs) in other domains has raised the question of whether LLMs can reliably assess and manipulate the readability of text. We approach this question empirically. First, using a published corpus of 4,724 English text excerpts, we find that readability estimates produced ``zero-shot'' from GPT-4 Turbo and GPT-4o mini exhibit relatively high correlation with human judgments (r = 0.76 and r = 0.74, respectively), out-performing estimates derived from traditional readability formulas and various psycholinguistic indices. Then, in a pre-registered human experiment (N = 59), we ask whether Turbo can reliably make text easier or harder to read. We find evidence to support this hypothesis, though considerable variance in human judgments remains unexplained. We conclude by discussing the limitations of this approach, including limited scope, as well as the validity of the ``readability'' construct and its dependence on context, audience, and goal.