Multidimensional Quality Assessment of AI-Generated Paraphrases

The present study was guided by the following research questions:
RQ1: Do artificial intelligence (AI) tools exhibit significant differences in analytic scores of paraphrased texts?
RQ2: Do AI tools differ significantly in the readability of paraphrased texts?
RQ3: Do AI tools vary significantly in the linguistic complexity of paraphrased texts?

The AI tools evaluated in this investigation comprised Writefull, ChatGPT-4.1, Wordvice, and QuillBot. The dataset included analytic scores, readability indices, and measures of linguistic complexity—both lexical and syntactic—for the original texts as well as their paraphrased versions produced by the four AI tools. All original texts were reformulated by the AI tools employing an academic writing style. Higher values in analytic scores, readability indices, and complexity metrics were interpreted as indicative of greater alignment with academic writing conventions.

Given the non-normal distribution of the data, the Related-Samples Friedman’s Two-Way Analysis of Variance by Ranks was employed to conduct multiple comparisons among the tools.

The results revealed that ChatGPT-4.1 and Wordvice outperformed the other AI tools in terms of overall quality and readability. Notably, Wordvice demonstrated a particular advantage in enhancing the linguistic complexity of the paraphrased texts.