AI-GFCA students produce fewer L2 errors than AI-C students, with the advantage varying by error type and task¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r1–r2 · q2 · i2–i3
Currency · newest study 2025 · not yet checked against its sources · not reviewed by a person
Design limits · Darmawansah 2025: groups were assigned but students were analysed as independent · Darmawansah 2025 (2): groups were assigned but students were analysed as independent (read from the study by two models, 2026-10-29)

Subclaims¶

q2 i2 At T1 the AI-GFCA group produced fewer morpho-syntactic errors than the AI-C group, while lexical errors were relatively equal between groups. → Darmawansah 2025 q2 i3 At T2 the AI-GFCA group produced fewer morpho-syntactic errors than the AI-C group, while phonological errors were relatively equal between groups. → Darmawansah 2025 (2)

Evidence¶

Darmawansah 2025¶

Darmawansah, D., Hwang, G.-J., Lin, C.-J., & Febiyani, F. (2025). An artificial intelligence-supported GFCA learning model to enhance L2 students' role-play performance, English speaking and interaction mindset. Education Tech Research Dev (2025) 73:1451–1479. https://doi.org/10.1007/s11423-025-10453-6

q2 · i2 · causal · r1

Mann–Whitney tests of expert-identified L2 errors (morpho-syntactic, lexical, phonological) in role-play task T1; morpho-syntactic errors favored AI-GFCA with "r = 0.30", and lexical errors showed no significant difference.

"students in the AI-GFCA group (Mdn = 0, 1) produced fewer morpho-syntactic errors and phonological errors compared to students in the AI-C group (Mdn = 2, 1) with U = 165.50, z = − 2.068, p = 0.039, and r = 0.30, and U = 157.00, z = − 2.535, p = 0.019, and r = 0.35, respectively."

Darmawansah 2025 (2)¶

Darmawansah, D., Hwang, G.-J., Lin, C.-J., & Febiyani, F. (2025). An artificial intelligence-supported GFCA learning model to enhance L2 students' role-play performance, English speaking and interaction mindset. Education Tech Research Dev (2025) 73:1451–1479. https://doi.org/10.1007/s11423-025-10453-6

q2 · i3 · causal · r2

Mann–Whitney tests of L2 errors in role-play task T2; morpho-syntactic errors favored AI-GFCA with "r = 0.38", while phonological errors showed no significant difference.

"students in the AI-GFCA group (Mdn = 1, 2) produced fewer morpho-syntactic errors and lexical errors compared to students in the AI-C group (Mdn = 0, 1) with U = 145.00, z = − 2.569, p = 0.01, and r = 0.38, and U = 158.00, z = − 2.396, p = 0.017, and r = 0.35, respectively."

Discussion¶

Related Claims¶

• AI-GFCA students show better pronunciation, fluency and IELTS speaking predictor scores than AI-C students, but intonation does not differ — related
• Students report morpho-syntactic errors most in corrective feedback, followed by lexical then phonological errors — related
• Students using the AI-GFCA learning model outperform AI-supported conventional-learning peers in role-play performance at both measured tasks — related