ChatGPT vs Human-authored Text: Insights into Controllable Text Summarization and Sentence Style Transfer

TL;DR

This paper systematically compares ChatGPT's performance in controllable text summarization and style transfer tasks to human-authored texts, highlighting differences in stylistic variation, faithfulness, and factual accuracy.

Contribution

It provides a detailed analysis of ChatGPT's ability to adapt to different audiences and styles, revealing limitations in stylistic diversity and factual correctness compared to humans.

Findings

Humans exhibit greater stylistic variation than ChatGPT.

ChatGPT's generated texts differ from human samples in word type distribution.

Factual errors and hallucinations occur when ChatGPT adapts to specific styles.

Abstract

Large-scale language models, like ChatGPT, have garnered significant media attention and stunned the public with their remarkable capacity for generating coherent text from short natural language prompts. In this paper, we aim to conduct a systematic inspection of ChatGPT's performance in two controllable generation tasks, with respect to ChatGPT's ability to adapt its output to different target audiences (expert vs. layman) and writing styles (formal vs. informal). Additionally, we evaluate the faithfulness of the generated text, and compare the model's performance with human-authored texts. Our findings indicate that the stylistic variations produced by humans are considerably larger than those demonstrated by ChatGPT, and the generated texts diverge from human samples in several characteristics, such as the distribution of word types. Moreover, we observe that ChatGPT sometimes incorporates factual errors or hallucinations when adapting the text to suit a specific style.