Evaluating an AI model for cross-linguistic language learning: a randomized controlled trial

Learning related languages presents challenges due to cross-linguistic influence (CLI), often undertreated in traditional pedagogy. AI tutors, especially multimodal ones, offer potential for supplementary cross-linguistic support, yet require empirical validation. This study addressed this gap via a one-semester randomized controlled trial evaluating an AI tutor integrated into traditional instruction. 162 Mandarin L1 university students were randomized to: AI supplement for simultaneous English/German learning (AI Group), traditional English-only instruction (English-Only Group), or traditional German-only instruction (German-Only Group), with equivalent study time. Post-intervention assessments measured language proficiency (overall and specific skills), learning efficiency, cognitive load profiles, and affective experiences (emotions, motivation). Results revealed the AI Group achieved significantly superior English proficiency across all skills compared to the English-Only Group (p < 0.001), while attaining German proficiency comparable to the German-Only Group across skills (p > 0.05). This demonstrates remarkable learning efficiency, effectively enabling dual-language acquisition in the time usually needed for one. Furthermore, the AI Group benefited from a more optimal cognitive load balance (lower extraneous, higher germane load) and reported significantly enhanced affective outcomes, including higher motivation, greater satisfaction, more positive emotions, and lower stress compared to relevant controls. In conclusion, AI tutors designed for cross-linguistic comparison, when supplementing traditional teaching, represent a highly efficient, effective, and engaging blended learning methodology for related languages, significantly boosting L2 learning while providing comparable L3 support and improving the overall learner journey.