Assessing the quality of ChatGPT’s generated output in light of human-written texts: A corpus study based on textual parameters

Research output: Contribution to journalResearch articleContributedpeer-review

Abstract

This contribution has an exploratory nature, marking the initial phase of a broader research project aimed at achieving both descriptive and theoretical objectives. The primary goal is to evaluate the ‘quality’ of texts produced by Language Model Models (LLMs). Two key aspects are examined: the quality of generated texts in comparison to human-authored texts and the identification of distinctive features characterizing this emerging text typology. The analysis is centered on textual parameters, encompassing various phenomena related to text segmentation and three dimensions of text organization (the referential-thematic dimension, the logico-argumentative dimension, and the polyphonic-enunciative dimension). Results of different case studies based on a self-assemble corpus of biographies generated by ChatGPT-3.5 and published on Wikipedia are presented.

Details

Original languageEnglish
Pages (from-to)179-210
JournalChimera : journal of romance corpora and linguistic studies
Volume10
Publication statusPublished - 1 Nov 2023
Peer-reviewedYes

Keywords

Research priority areas of TU Dresden

Library keywords