Generative AI in Collaborative Language Projects Boosts Student Achievement and Autonomy

In an explanatory sequential mixed-methods investigation, researchers evaluated how integrating generative artificial intelligence into collaborative project-based language learning influences language performance and autonomous learning capacities among undergraduate students studying English as a foreign language. Conducted within a higher education setting, the study examined whether structured, peer-mediated engagement with generative AI tools over a full academic semester yielded measurable differences compared with conventional classroom instruction. The primary findings demonstrated that undergraduates participating in AI-assisted collaborative projects achieved significantly greater improvements in overall English language achievement—most notably in writing and vocabulary development—alongside substantial advancements across all measured dimensions of learner autonomy, particularly in motivation and self-regulation.

The study was motivated by persistent pedagogical challenges in higher education foreign language programs, where instructional time is often constrained and expectations for student proficiency remain elevated. While previous scholarship has highlighted the benefits of student-centered designs and learner autonomy, empirical research on generative artificial intelligence has largely remained focused on short-term interactions, individual tool use, or learner self-reports. Standalone AI tools frequently lack an overarching pedagogical architecture, leaving unclear how generative systems function when situated inside collaborative learning environments. By embedding generative AI within project-based language learning, the researchers sought to evaluate an instructional model that combines technological scaffolding with socially structured inquiry, shared problem-solving, and learner agency.

To investigate these dynamics, the researchers designed a quasi-experimental pretest-posttest control group framework accompanied by a qualitative evaluation phase. The participant cohort comprised 53 female first-year undergraduate students majoring in computer science at a Saudi Arabian university, enrolled in a required four-credit English language course. Participants had an average age of 18.20 years, with the majority demonstrating language proficiency at the A2 or B1 levels on the Oxford Placement Test. Two intact classes were allocated through convenience sampling into an experimental group of 28 students and a control group of 25 students. While participants reported moderate prior familiarity with AI applications for independent study, none possessed previous experience utilizing generative AI within collaborative, project-based coursework, establishing comparable baseline familiarity across cohorts.

The instructional intervention spanned 12 weeks divided into three sequential phases consisting of orientation, project implementation, and evaluation. Both cohorts followed the identical institutional English curriculum covering reading comprehension, grammar, vocabulary, and writing over equivalent contact hours. Control students engaged in traditional individual and pair classroom activities with regular instructor feedback but without project work or generative AI integration. Conversely, experimental students worked in self-selected teams of three to four members on projects aligned with Saudi Vision 2030, such as bilingual travel chatbots, narrated presentations on e-governance featuring AI-generated avatars, and virtual cultural tours. To ensure parity in total instructional exposure, project activities directly replaced standard in-class exercises and homework assignments.

During the orientation phase, experimental learners received formal training on digital platforms including ChatGPT, Microsoft Copilot, Canva AI, Narakeet, Grammarly, and QuillBot, emphasizing prompt engineering, critical evaluation of machine-generated text, and ethical guidelines. Over the subsequent ten-week project phase, teams navigated iterative cycles of planning, drafting, revision, and multimodal production. ChatGPT was applied to brainstorm ideas, simulate dialogues, and model sentence structures, while tools such as QuillBot and Copilot supported syntactic and lexical refinements. Selective translation through systems like DeepL assisted students in adapting Arabic reference sources into academic English. Throughout the intervention, the instructor monitored progress through digital platforms and held structured tutorial meetings, ensuring fidelity while requiring teams to formally acknowledge AI integration in their project artifacts.

Data collection utilized parallel-form English language achievement tests and a researcher-developed 16-item Learner Autonomy Questionnaire administered before and after the 12-week intervention. Achievement assessments followed institutional examination blueprints measuring reading, vocabulary, grammar, and writing, with writing samples rated independently by calibrated evaluators. The autonomy instrument captured five distinct dimensions: awareness, motivation, intentionality, self-regulation, and critical ability. Following the post-test, experimental participants submitted written reflection reports detailing their collaborative experiences and interactions with digital tools. Quantitative outcomes were analyzed using linear mixed-effects modeling in SPSS, treating the interaction between time and group as the primary test of differential intervention effects.

Quantitative results revealed statistically significant differential gains favoring the experimental group across overall English achievement and specific linguistic subskills. Linear mixed-effects models indicated significant Time by Group interactions for reading, vocabulary, writing, and total achievement scores. Estimated marginal means showed that the experimental cohort achieved positive gains in reading, grammar, vocabulary, writing, and overall score, whereas the control cohort demonstrated modest increases in grammar and vocabulary alongside score declines in reading and writing. In contrast, the interaction term for grammar performance did not reach statistical significance, indicating that grammatical development progressed at comparable rates across both groups as part of general institutional course coverage.

Analysis of learner autonomy measures showed statistically significant advantages for the experimental cohort across every evaluated domain. Linear mixed-effects models demonstrated significant interaction effects for learner awareness, motivation, intentionality, self-regulation, critical ability, and total autonomy. Estimated marginal means indicated that students in the experimental group experienced marked score improvements across all autonomy subscales, with the most pronounced development occurring in motivation and self-regulation. While control participants also registered modest increases in autonomy scores, their gains were substantially smaller, demonstrating that collaborative project work integrated with generative AI was associated with more robust self-directed learning behaviors and strategic academic control.

Inductive thematic analysis of the reflection reports submitted by experimental students yielded two overarching themes: collaborative dynamics and perceptions of generative AI integration. Participants reported that peer collaboration, shared decision-making, and collective consensus were essential for resolving task complexities and overcoming initial interpersonal or technical discomfort. Working collaboratively helped students refine prompt phrasing, troubleshoot unexpected software behaviors, and re-evaluate digital outputs. Students described generative AI as enhancing learning efficiency and creative expression, noting that iterative drafting reduced cognitive strain and expanded writing fluency. Concurrently, reflections documented emerging metacognitive behaviors, such as vocabulary tracking and independent goal setting, while highlighting critical awareness regarding the hazards of excessive dependence on automated tools.

In discussing these findings, the authors emphasized that generative AI functioned as a mediational scaffold within a socially mediated educational environment rather than an isolated teaching agent. The observed improvements in productive language skills align with sociocultural theories of learning, wherein student agency is fostered through structured dialogue, collaborative inquiry, and authentic communication tasks. The lack of differential improvement in grammar suggested that explicit form-focused instruction may require targeted pedagogical designs rather than open-ended project work alone. Furthermore, the modest gains observed in critical evaluative ability underscored that technology integration does not automatically cultivate critical appraisal unless tasks explicitly demand systematic source comparison and analytical validation.

The researchers noted several methodological constraints that frame the interpretation of their findings. The sample was restricted to 53 female students from intact classes within a single university, reflecting contextual institutional norms that constrain direct generalizability to co-educational or diverse student populations. Additionally, because the intervention merged collaborative project structures with digital tool access and teacher guidance, isolating the unique contribution of generative AI apart from the project methodology was not feasible. The absence of automated digital tracking logs prevented direct measurement of individual AI interaction intensity, and psychometric validation of the autonomy questionnaire remains exploratory. The authors concluded that while generative AI holds substantial promise for language learning, its educational utility depends on sound pedagogical designs that emphasize peer collaboration, critical reflection, and deliberate instructional scaffolding.

Subject of Research: GenAI-supported collaborative project-based language learning

Article Title: GenAI-supported collaborative project-based language learning: effects on EFL undergraduates’ achievement and autonomy

Article References: Abdelhalim, S. M., & Almaneea, M. O. (2026). GenAI-supported collaborative project-based language learning: effects on EFL undergraduates’ achievement and autonomy. Asian-Pacific Journal of Second and Foreign Language Education, 11(1), Article 50. https://doi.org/10.1186/s40862-026-00430-8

Image Credits: AI Generated

DOI: 10.1186/s40862-026-00430-8

Keywords: Generative AI, Project-Based Language Learning, EFL Education, Learner Autonomy, English Language Achievement, Higher Education, Computer-Assisted Language Learning, Educational Technology, GenAI-supported, collaborative, project-based, language