L2 writing is one of the most demanding activities learners engage in, drawing on their linguistic resources (Hyland, 2003), and learners are required to establish control over these resources to construct a sophisticated and coherent written discourse. As Housen and Kuiken (2009) indicated, proficiency in writing is grounded in complexity, accuracy and fluency, the CAF framework, but complexity has been given greater importance, as it is regarded as the metric for learners’ development of linguistic resources and their responsiveness to particular communicative needs (Bulté & Housen, 2012). 

Linguistic complexity is a multidimensional construct comprising lexical, syntactic, morphological, and discourse-level components. Given their theoretical relevance and methodological soundness, lexical and syntactic complexity have been more salient in the literature as these have more standardised analytical frameworks and are more informative in languages with limited morphology. Lexical complexity comprises three dimensions, lexical diversity, lexical density, and lexical sophistication, which encompass different aspects of lexical use (Bulté & Housen, 2012). Syntactic complexity, being a multidimensional construct, comprises the variety, sophistication, and range of the grammatical structures in a text (Lu, 2010).

These dimensions of linguistic complexity have also become salient in discussions of Generative AI texts, as they are applicable to AI-generated texts. Hence, the emergence of Generative AI has introduced a new point of comparison for L2 writing performance since its incorporation into L2 writing classrooms has become more prominent. Generative AI models can generate written texts that demonstrate considerable grammaticality, lexical sophistication, etc., attributes that are mostly linked to higher proficiency in academic writing. Several studies (Chen, 2025; Fredrick & Craven, 2025; Sun & Li, 2025; Mizumoto, 2024; Zindela, 2023) compared L2 students’ and AI-generated texts in their studies, the findings of which disclosed that AI-generated texts exhibit higher lexical diversity but similar syntactic complexity, with more formulaic structures and different rhetoric and agency. There is a growing body of literature that has compared AI-generated and student-authored written texts, yet the comparisons have primarily focused on accuracy, and thus overall performance or concentrated only on specific dimensions of linguistic complexity, such as only lexical diversity, etc.  Therefore, a comparison between human-authored L2 writing and AI-generated texts, from a multidimensional perspective, on both syntactic and lexical complexity remains underresearched in the field of L2 writing. 

Thus, the present study investigates differences in syntactic and lexical complexity between AI-generated and human-authored opinion essays and whether these syntactic and lexical indices can be utilised to model the patterns of similarity and difference between the two. The following research question will be addressed in this study:

1. Are there statistically significant lexical and syntactic complexity indices between human-authored and AI-generated opinion essays?

   1. If yes, can these indices be used to model the lexical and syntactic complexity patterns of human-authored and AI-generated opinion essays?

Adopting a quantitative, corpus-based approach, two corpora consisting of 223 human-authored and 223 AI-generated opinion essays will be compared using 25 lexical complexity indices and 14 syntactic complexity indices proposed by Lu (2012). Accordingly, the present study is anticipated to provide theoretical, pedagogical, and policy-level implications for researchers in the field, teachers, and policymakers in AI literacy and education. At the theoretical level, it is expected to expand L2 writing and the CAF research by utilizing AI-generated texts as a comparative reference point. At the pedagogical and policy levels, the findings are expected to inform the shift from product-oriented evaluation to process-oriented instruction, where instructors are given a way to infer the learners’ use of writing supports and could implement support systems accordingly, that foregrounds rhetorical awareness, authorial voice, and idea construction, in line with UNESCO's (2024) AI competency framework for students.

II. Methodology This study adopts a quantitative, corpus-based comparative analysis to explore the differences in syntactic and lexical complexity between AI-generated and human-authored opinion essays, and to investigate whether these syntactic and lexical indices can model systematic differences between the two text types. II. I. Data Collection This study is a single-case, corpus-based research consisting of pre-service English language teachers enrolled in a compulsory English preparation program at Sakarya University and ChatGPT. The sample was obtained through convenience sampling, as all texts were collected from two classes in the same compulsory English preparation program during the 2024-2025 academic year. The other data source consists of opinion essays generated by ChatGPT. In preliminary unstructured observations, ChatGPT (Free version) was identified as the most widely used chatbot due to its popularity and accessibility. Within these observations, typical chatbot prompts for generating an opinion essay writing task were also noted and used to generate AI-generated texts for the corpus. Two corpora consist of 223 human-authored and 223 AI-generated opinion essays, totalling approximately 55,000 and 60,000 words, respectively. II. II. Data Analysis In this study, 446 opinion essays (223 human-authored, 223 AI-generated) will be analyzed using neoSCA (Tan, 2024), a software forked from Lu’s L2 Syntactic Complexity Analyzer (2010) and Lexical Complexity Analyzer (2012). This software combines Lu’s tools (2010,2012), which allow the calculation of 14 syntactic complexity indices, calculating the length of production unit, sentence complexity, subordination, coordination and particular structures and 25 lexical complexity indices, calculating lexical variation, lexical sophistication and lexical density, in a single application. An index redundancy analysis will be conducted to identify indices with high intercorrelation, multicollinearity, and minimal dispersion, thereby meeting the assumptions of multivariate analysis while preserving theoretical coverage of lexical and syntactic complexity. Two MANOVA will be conducted on Lexical complexity and Syntactic complexity to identify unstable indices. Discriminant function analysis will be conducted to evaluate the blind classification of AI-generated versus human-authored texts.

This study is expected to reveal statistically significant patterns in lexical and syntactic complexity indices between human-authored and AI-generated opinion essays, which adds to the findings of Chen (2025) regarding the greater lexical density of AI-generated texts, Zindela (2023) regarding the distinguishable use of syntactic structures, and Sun and Li (2025) regarding greater syntactic complexity measures of AI-generated texts. Moreover, AI-generated texts are expected to exhibit greater clausal elaboration, as shown in previous studies (Zindela, 2023; Chen & Hong, 2025). Nonetheless, these studies also revealed that AI-generated texts showed notably lower distributions of lexical diversity indices compared with human-generated texts, which imply greater internal regularity and, in turn, a lack of creativity and human authorship. At the pedagogical level, the findings are likely to show how these indices could be construed in relation to the approach to teaching L2 writing, particularly in awareness of rhetoric and voice, opinion elaboration, and lexical or syntactic item selection. Thus, the study's findings may indicate the need to expand instructional or feedback practices that place greater emphasis on the aforementioned areas, as outlined in UNESCO's AI competency framework for students (2024). Moreover, the study is anticipated to build on previous research on L2 writing by broadening the scope of the complexity–accuracy–fluency (CAF) framework to digitally mediated writing contexts, such as generative AI models. However, the present study also has several limitations. It focuses only on the specific AI model that the pre-service teachers primarily use in this study’s context, but given the variability of generative AI models, the study's findings might not be generalisable to other AI models. Additionally, this study focuses on a single writing task type, which, given the variation in prompts, may limit generalizability. Thus, future research could explore multiple genres, text types, prompts, and AI models across different institutional contexts.

Ai, H., & Lu, X. (2010). A web-based system for automatic measurement of lexical complexity. Paper presented at the 27th Annual Symposium of the Computer-Assisted Language Consortium (CALICO-10). Amherst, MA. June 8-12. Bulté, B., & Housen, A. (2012). Defining and operationalising L2 complexity. Dimensions of L2 performance and proficiency: Complexity, accuracy and fluency in SLA, 32, 21. Chen, J., & Hong, Q. (2025). Lexical Diversity and Syntactic Complexity in AI-Translated Legislative Texts. Theory and Practice in Language Studies, 15(9), 2867-2875. Chen, Y. (2025). Evaluating the Potential of ChatGPT-reformulated Essays as Written Feedback in L2 Writing. Computers and Education: Artificial Intelligence, 100500. Fredrick, D. R., & Craven, L. (2025, September). Lexical diversity, syntactic complexity, and readability: A corpus-based analysis of ChatGPT and L2 student essays. In Frontiers in Education (Vol. 10, p. 1616935). Frontiers Media SA. Housen, A., & Kuiken, F. (2009). Complexity, accuracy, and fluency in second language acquisition. Applied linguistics, 30(4), 461-473. Hyland, K. (2003). Writing and teaching writing. Second language writing, 1(2), 1-30. Lu, X. (2010). Automatic analysis of syntactic complexity in second language writing. International journal of corpus linguistics, 15(4), 474-496. Lu, X. (2012). The Relationship of Lexical Richness to the Quality of ESL Learners' Oral Narratives. The Modern Language Journal, 96(2), 190-208. Sun, S., & Li, Y. (2025). AI-generated, L2 learner, and native German writing: a comparative analysis of linguistic complexity. Glottotheory, 16(2), 127-148. Tan, L. (2024). NeoSCA (version 0.1.4) [Computer software]. GitHub. https://github.com/tanloong/neosca UNESCO. (2024). AI competency framework for students. United Nations Educational, Scientific and Cultural Organization. https://unesdoc.unesco.org/ark:/48223/pf0000391105 Zindela, N. (2023). Comparing Measures of Syntactic and Lexical Complexity in Artificial Intelligence and L2 Human-Generated Argumentative Essays. International Journal of Education and Development Using Information and Communication Technology, 19(3), 50-68.