Dongqi Pu
My name in Oracle bone script: 
Doctoral researcher in Computational Linguistics
Campus C7.4, Saarland University, 66123, Germany
dongqi.me [AT] gmail.com
ChatGPT vs Human-authored Text: Insights into Controllable Text Summarization and Sentence Style Transfer
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 style (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.
Citation:
ChatGPT vs Human-authored Text: Insights into Controllable Text Summarization and Sentence Style Transfer (Pu & Demberg, ACL 2023)