A Syntactic Comparison between Human and AI-Generated Sentences: Exploring Patterns and Deviations

Abstract

This study explores the syntactic complexity, grammatical correctness, and structural variety in the essays by Pakistani ESL learners as compared to the AI (ChatGPT) essays. Three main types of analysis, syntactic complexity, error and fluency analysis and syntax tree pattern variation was conducted on nine essays from each group (Human and AI). The present study employed a mixed approach to data analysis. Using Coh-Metrix, a computational text analysis tool the syntactic complexity, grammatical accuracy and cohesion measures were taken whereas the syntax tree pattern analysis was conducted manually by the researcher. The findings reveal the AI generated essays exhibit higher complexity, fewer errors and varied clause structures, relative to human essays by 5th semester BS English learners, which are moderate and with scattered grammatical errors and less complex sentence patterns. These results indicate the developmental needs of ESL students in learning to master and have command over complex sentence structures and improve fluency in writing. This study has implications for the application of AI-based resources in ESL writing teaching by helping to narrow the gap and to develop advanced language skills.