From cultural bias to critical awareness: LLM-mediated voice dialogue and intercultural competence in Chinese language learners

From cultural bias to critical awareness: LLM-mediated voice dialogue and intercultural competence in Chinese language learners | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article From cultural bias to critical awareness: LLM-mediated voice dialogue and intercultural competence in Chinese language learners Jing Sun, Liming Nie This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9140660/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract Intercultural communicative competence (ICC) is a core objective of foreign language education, yet providing learners with sufficient meaningful intercultural encounters remains a persistent challenge. Large language models (LLMs) may support such encounters through AI-mediated cultural dialogue, yet no prior study has empirically examined whether such dialogues can foster ICC development. This study addresses the gap through a quasi-experimental pretest-posttest design integrating Byram's ICC model with Kolb's experiential learning cycle. Sixty-two Chinese-as-a-Foreign-Language (CFL) learners from 17 countries participated in a six-week intervention. The experimental group (n = 32) engaged in structured voice-based cultural dialogue tasks with a Chinese-developed LLM; the control group (n = 30) received equivalent hours of conventional cultural instruction. Analysis of covariance (ANCOVA) results showed that the experimental group significantly outperformed the control group on both the Intercultural Sensitivity Scale (ISS, d = 0.82) and the Cultural Intelligence Scale (CQS, d = 0.83). Dimension-level analysis revealed a non-uniform pattern: Metacognitive CQ yielded the largest effect (d = 1.08), whilst Cognitive CQ showed the smallest (d = 0.52) and did not survive Bonferroni correction. Learner perception data indicated high ratings for cultural learning effectiveness but lower ratings for communicative authenticity. The study proposes a "problem-as-resource" strategy that transforms LLM cultural biases into material for cultivating critical cultural awareness, offering empirical evidence and reusable task templates for integrating LLMs into intercultural language education. Humanities/Cultural and media studies Social science/Cultural and media studies Social science/Education Humanities/Language and linguistics Social science/Language and linguistics Biological sciences/Psychology Social science/Psychology intercultural communicative competence large language models Chinese as a foreign language cultural dialogue quasi-experimental design critical cultural awareness Figures Figure 1 Figure 2 Figure 3 Figure 4 1 Introduction As international education expands and cross-cultural contact intensifies, the ability to communicate effectively across cultural boundaries has become an essential competence for global citizens. Intercultural communicative competence (ICC) is accordingly a core objective of foreign language education (Byram 1997 ; Deardorff 2006 ), and its development requires experiential interaction with cultural otherness and sustained critical reflection (Huang 2021 ). Yet providing learners with sufficient meaningful intercultural encounters within formal classroom constraints remains a fundamental pedagogical challenge that intensifies as student mobility increases while instructional resources remain finite. For learners of Chinese as a Foreign Language (CFL), this challenge is especially acute: the sociocultural complexity of China, the linguistic distance of Chinese, and the structural scarcity of authentic cultural interaction opportunities together create barriers that conventional instruction struggles to overcome (Huang and Cui 2025 ). Telecollaboration and virtual exchange programs have partially bridged this gap (Freiermuth and Huang 2021 ), but these approaches are constrained by partner availability, scheduling logistics, and a limited range of simulable cultural scenarios. Recent advances in large language models (LLMs) present an alternative approach. As "on-demand" dialogue partners, LLMs can simulate diverse cultural perspectives, adapt to learners' proficiency levels, and allow low-stakes experimentation with intercultural communication strategies. Liu, J. ( 2025 ) provided initial empirical evidence that an AI-enhanced language learning system can significantly improve learners' intercultural communication competence, yet that study employed a bespoke deep-learning architecture rather than generative LLM dialogue, leaving the pedagogical potential of LLM-mediated cultural interaction largely untested. At the same time, LLMs carry well-documented cultural biases: Cao et al. ( 2023 ) found that ChatGPT systematically favours American cultural values, Yuan et al. ( 2025 ) recorded stereotypical patterns across 13 cultural dimensions, and Jenks ( 2024 ) warned against treating AI-generated cultural content as objective. Wang et al. ( 2025 ) demonstrated through a quasi-experimental design that LLM-integrated flipped-classroom instruction can improve critical thinking dispositions. However, their study positioned the LLM as a "learning aid" for general critical thinking. ICC cultivation demands a fundamentally different role, that of a cultural interlocutor with whom learners negotiate intercultural meaning. Can LLMs effectively simulate cultural identities? How do their biases shape learners' intercultural cognition? These questions remain unanswered. No prior study has empirically examined whether LLM-mediated cultural dialogues can promote ICC development in language learners. Given that millions of language learners worldwide already use LLM-based tools in their daily studies, this gap has direct practical implications: educators lack evidence-based guidance on whether and how to integrate these tools into intercultural instruction. The present study addresses this gap. Grounded in Byram's (1997) ICC model and Kolb's (1984) experiential learning cycle, it designed and implemented a six-week LLM-mediated cultural dialogue intervention, addressing the following research questions: RQ1: To what extent does LLM-mediated cultural dialogue improve CFL learners' intercultural competence (compared with a control group receiving conventional cultural instruction)? RQ2: Does the LLM-mediated intervention affect the various ICC dimensions (knowledge, skills, attitudes, and critical cultural awareness) differentially? RQ3: How do CFL learners perceive the value and limitations of LLMs as intercultural learning tools? This study contributes by: (a) extending Liu, J.'s (2025) AI-enhanced ICC findings to a pedagogically grounded LLM-mediated cultural dialogue paradigm; (b) proposing a problem-as-resource strategy that converts LLM cultural biases into a classroom-actionable pathway for critical awareness cultivation; and (c) producing reusable cultural dialogue task templates grounded in the Byram ICC model and the Kolb learning cycle. 2 Literature Review 2.1 Intercultural Communicative Competence in Language Education ICC has been a central construct in applied linguistics since Byram's (1997) five-savoirs model: savoirs (knowledge), savoir comprendre (interpreting skills), savoir apprendre/faire (discovery and interaction skills), savoir être (attitudes of openness), and savoir s'engager (critical cultural awareness). This model has been widely adopted for both theoretical analysis and curriculum design (Ge 2025 ). Subsequent refinements include Deardorff's (2006) process model emphasizing ICC's developmental and iterative nature, Munezane's (2019) integration of willingness to communicate and constructive conflict resolution, and Cui and Mahfoodh's (2025) empirically weighted sociopragmatic assessment framework in the Chinese EFL context. Empirically, Huang ( 2021 ) found that explicit intercultural instruction improved EFL learners' ICC in knowledge and skills dimensions but had limited impact on attitudes and critical cultural awareness, a pattern corroborated by textbook analyses. Thi ( 2021 ) demonstrated the effectiveness of project-based assessment for ICC development over a 9-week course. These findings suggest that ICC requires sustained, experiential engagement rather than isolated knowledge transmission. A persistent challenge remains the provision of authentic intercultural interaction opportunities. In CFL contexts, this is compounded by cultural distance and the complexity of navigating concepts such as mianzi, guanxi, and keqi that lack direct equivalents in many languages (Huang and Cui 2025 ). International students in China have some access to authentic encounters, but these are often unsystematic and difficult to process without guided reflection (Lin 2025 ). 2.2 Technology-Enhanced Intercultural Learning Telecollaboration has been the dominant paradigm for technology-mediated ICC development (O'Dowd 2018 ; Liu, F. 2025 ). Freiermuth and Huang ( 2021 ) found that telecollaborative video exchanges between Japanese and Taiwanese students were mutually beneficial for cultural learning, yet the study involved only 11 participants and required extensive cross-institutional coordination, illustrating both the promise and the scalability constraints of human-to-human virtual exchange. Earlier technological alternatives, from chatbots to virtual environments, lacked the linguistic sophistication for credible intercultural dialogue. Text-based chatbots also suffer a fundamental modality deficit: authentic oral intercultural communication requires real-time processing pressure, turn-taking management, and paralinguistic cue interpretation, all core skills in Byram's (1997) interaction dimension (savoir apprendre/faire; Young 2011 ). Text interaction eliminates these demands, thus failing to train real-time intercultural communication. LLMs supporting real-time voice interaction have altered this situation, offering agents with high fluency, broad cultural knowledge, and role-play capability. The voice modality restores the time pressure and turn-taking dynamics of spoken communication, enabling practice under conditions approximating authentic exchanges. 2.3 Generative AI and Cultural Learning in Language Education Generative AI applications in language education have grown substantially since ChatGPT's release. Lo et al.'s ( 2024 ) systematic review of 70 empirical studies found that research overwhelmingly focuses on writing skills, with almost none examining cultural learning or ICC, a significant gap. In CFL contexts, Liu et al. ( 2024 ) found learners adopting AI as both tutor and conversation partner, Liu and Liu ( 2025 ) explored LLM-empowered teacher content creation, and Tahmasbi et al. ( 2025 ) demonstrated that LLM-powered role-playing enhances vocabulary acquisition. Empirical work at the GenAI–ICC intersection remains scarce. Liu, J. ( 2025 ) demonstrated that an AI-enhanced language learning system integrating deep-learning models significantly improved youths' cultural understanding, communication strategies, and cultural sensitivity, providing the first experimental evidence that AI can measurably promote ICC development. However, that system relied on a custom encoder–decoder architecture rather than generative LLM dialogue, and its intervention did not incorporate critical engagement with AI-generated cultural content. The present study extends this line of inquiry by employing LLM-mediated voice-based cultural dialogue and explicitly employing LLM cultural biases as pedagogical resources. Weng and Fu ( 2025 ) and Zheng ( 2024 ) corroborated these findings while noting that cultural communication requires deeper modeling than general language tasks. A related usability concern has received limited attention: for lower-proficiency learners, discussing deep cultural topics in the target language may impose prohibitive cognitive load. Sweller's (1988) cognitive load theory suggests that HSK Level 3 learners discussing abstract cultural concepts in Chinese face simultaneous linguistic and cultural processing demands that risk overloading working memory (Dewaele and MacIntyre 2014 ). Li et al. ( 2025 ) noted that GenAI usability in language education depends on whether the design can reduce extraneous cognitive load. This consideration directly informed the platform selection in the present study (see Section 3.3.4 ). 2.4 LLM Cultural Affordances and Biases LLM-mediated intercultural learning must contend with documented cultural biases. Cao et al. ( 2023 ) found that ChatGPT exhibits strong alignment with American cultural values while adapting poorly to other contexts, with English-language prompts systematically flattening cultural differences. Yuan et al. ( 2025 ) documented cultural stereotypes across 13 dimensions, though their Study 3 demonstrated that targeted prompt strategies can effectively reduce stereotype generation. Jenks ( 2024 ) argued that intercultural communication researchers must critically examine how LLMs "produce and circulate discourse in an ostensibly impartial way" (p. 788), highlighting the need for learners to develop critical AI literacy alongside ICC. The present study treats LLM cultural biases not merely as a limitation but as a pedagogical resource. Building on Yuan et al.'s ( 2025 ) prompt-strategy findings and Jenks's (2024) call for critical engagement, our intervention incorporates explicit activities for identifying and discussing cultural biases in LLM outputs, directly targeting the critical cultural awareness dimension of Byram's (1997) model, the dimension most resistant to change through conventional instruction (Huang 2021 ). 2.5 Research Gap and the Present Study In summary, the literature reveals a three-way intersection gap: no empirical study has examined LLM-mediated cultural dialogue for ICC development in CFL contexts. This gap is both theoretically significant (leaving a rapidly adopted technology without pedagogical grounding) and practically consequential, as CFL programmes worldwide seek scalable alternatives to resource-intensive telecollaboration. The present study addresses this gap by extending Liu, J.'s (2025) AI-enhanced ICC findings to a generative LLM dialogue paradigm, introducing the LLM as a "simulated cultural other" for intercultural practice, and applying LLM cultural bias research to pedagogy through a problem-as-resource strategy. The theoretical framework integrates Byram's (1997) ICC model with Kolb's (1984) experiential learning cycle (revised by Morris 2019 ), structuring each task as a complete cycle: concrete experience (LLM dialogue), reflective observation (group discussion), abstract conceptualization (cultural concept extraction), and active experimentation (application in subsequent tasks). 3 Method 3.1 Research Design This study adopted a quasi-experimental nonequivalent control group pretest-posttest design (Campbell and Stanley 1963 ), with an explanatory sequential mixed-methods approach (Creswell and Plano Clark 2018 ). The experimental group participated in a six-week LLM-mediated cultural dialogue intervention; the control group received conventional CFL instruction with comparable cultural content but without LLM use. Quantitative data (standardised scales) served as the primary strand; qualitative data (open-ended questionnaire responses and semi-structured interviews) provided supplementary explanation. This design aligns with recent studies in the field (Hackett et al. 2023 ; Sarwari et al. 2024 ). 3.2 Context and Participants 3.2.1 Institutional Context The study was conducted at [University], a public university in eastern China. The university enrolls international students from over 30 countries, the majority in degree programs requiring Chinese proficiency (HSK Level 4 for graduation). CFL instruction follows the International Curriculum for Chinese Language Education and is delivered through comprehensive Chinese, listening, and speaking courses. 3.2.2 Participants Participants were recruited by intact class from four intermediate-level CFL classes during the spring semester of 2026. Two classes were assigned to the experimental group (n = 32) and two to the control group (n = 30). All participants met the following inclusion criteria: (a) had passed HSK Level 3 or equivalent, ensuring sufficient Chinese proficiency to participate in structured cultural dialogue tasks conducted primarily in Chinese with English as a permissible supplement; (b) had studied in China for at least one semester; and (c) had no prior formal training in prompt engineering or structured LLM use for language learning. The demographic profile of both groups is presented in Table 1 . Table 1 Participant demographics Characteristic Experimental (n = 32) Control (n = 30) Nationality 17 countries 15 countries Regional distribution Central and West Asia (31.3%), Eastern Europe (21.9%), Africa (21.9%), East and Southeast Asia (18.8%), Other (6.3%) Central and West Asia (30.0%), Eastern Europe (23.3%), Africa (20.0%), East and Southeast Asia (20.0%), Other (6.7%) Gender Female (65.6%), Male (34.4%) Female (60.0%), Male (40.0%) Age M = 21.8, SD = 2.5 M = 22.1, SD = 2.3 Chinese proficiency HSK 3 (90.6%), HSK 4 (9.4%) HSK 3 (93.3%), HSK 4 (6.7%) Duration in China M = 2.1 semesters, SD = 0.9 M = 2.0 semesters, SD = 0.8 Baseline equivalence on gender, age, Chinese proficiency, and duration in China was confirmed through independent-samples t-tests and chi-square tests (see Section 4.2 ). A priori power analysis (GPower 3.1; Faul et al. 2007 ) indicated that detecting a medium effect (d = 0.50) for an independent-samples t-test (α = 0.05, power = 0.80) requires 26 per group. At the field-specific median between-group effect of d* = 0.70 (Plonsky and Oswald 2014 ), our per-group sample of 30 + yields power exceeding 85%. The 17-country participant diversity enhances ecological validity while limiting generalisability to any single cultural group. 3.3 The LLM-Mediated Cultural Dialogue Intervention 3.3.1 Task Design Principles The intervention was grounded in two complementary theoretical frameworks. Byram's (1997) ICC model guided the selection of intercultural themes and the operationalisation of learning objectives across four dimensions: knowledge, skills, attitudes, and critical cultural awareness. Kolb's (1984) experiential learning cycle, as revised by Morris ( 2019 ), provided the pedagogical structure for each task session. Each task followed a four-phase sequence: (1) Concrete experience: Learners engaged in LLM-mediated cultural dialogues, encountering culturally embedded communicative situations through role-play with the LLM. (2) Reflective observation: Learners discussed the dialogue experience in small groups, examining what they had learned, felt, and found challenging. (3) Abstract conceptualization: Through teacher-facilitated discussion, learners extracted cultural concepts, compared cross-cultural patterns, and constructed interpretive frameworks. (4) Active experimentation: Learners applied newly acquired cultural understanding and communicative strategies in subsequent dialogue tasks, initiating the next learning cycle. Figure 1 illustrates how the two theoretical frameworks were integrated into the intervention design. Additional design principles included: structured prompts specifying cultural contexts to mitigate the cultural-flattening effect of default prompting (Cao et al. 2023 ; Yuan et al. 2025 ); an explicit bias-identification phase in each session following Jenks's (2024) call for critical AI literacy; and scenario-based communicative tasks consistent with the task-based approach advocated for ICC development (Thi 2021 ). 3.3.2 Overview of the Six Cultural Dialogue Modules The six weekly modules, their intercultural themes, primary ICC target dimensions, and corresponding LLM task descriptions are summarised in Table 2 . Table 2 Cultural dialogue modules Week Theme ICC Dimensions LLM Dialogue Task 1 Social norms: First encounters Knowledge, Attitudes LLM plays a Chinese student; learner practices greetings, address forms, and small talk 2 Food culture and hospitality Knowledge, Skills LLM plays a Chinese host; learner navigates dining etiquette and hospitality rituals 3 Intercultural misunderstanding Skills, Attitudes LLM simulates a cross-cultural friction scenario; learner identifies sources and negotiates resolution 4 Digital China Knowledge, Critical Awareness Learner discusses Chinese digital culture and critically examines LLM's cultural representations 5 Identity and cultural adaptation Attitudes, Critical Awareness LLM plays a cross-cultural counselor; dialogue explores culture shock and identity negotiation 6 Integrative reflection Critical Awareness (all) Guided reflective dialogue reviewing the six-week journey, identifying remaining biases, and completing an intercultural growth reflection questionnaire Modules progress from knowledge-oriented (Weeks 1–2) through skills/attitudes (Weeks 3–4) to critical awareness (Weeks 5–6), following the developmental trajectory endorsed in ICC scholarship (Deardorff 2006 ). 3.3.3 Task Implementation Procedure Each weekly 90-minute session followed four standardised phases (see Supplementary Note S2 for a detailed sample lesson plan): (1) Activation (15 min): the instructor presented a cultural scenario; learners discussed in pairs. (2) LLM dialogue (30 min): learners individually engaged with the LLM following a task sheet specifying the scenario, cultural role, and guiding questions; interactions were primarily in Chinese calibrated to HSK Level 3, with English permitted as a supplementary language when learners encountered expressive difficulties; minimum 10 conversational turns. After each dialogue, learners submitted screenshots for task completion verification by the research assistant; full interaction transcripts were automatically generated and saved by Doubao's built-in ASR system, requiring no additional effort from learners. (3) Group discussion (30 min): in culturally mixed groups of 3–4, learners discussed what they learned, cross-cultural similarities/differences, surprising or uncomfortable moments, and how they might act differently in real encounters. (4) Synthesis and bias check (15 min): the instructor led a whole-class discussion and projected 2–3 LLM responses for collective identification of stereotyping, oversimplification, or factual inaccuracy, targeting Byram's (1997) critical cultural awareness dimension. 3.3.4 LLM Platform Selection and Prompt Design Platform selection. This study adopted a single-platform design using Doubao (豆包), a Chinese-developed LLM by ByteDance. Selection was based on four criteria: (a) Communicative modality authenticity: Doubao's end-to-end real-time voice model (launched January 2025) supports ultra-low-latency natural spoken conversation, creating an interaction modality approximating face-to-face encounters, given that ICC's skill dimension (savoir faire) is fundamentally realized through oral interaction (Byram 1997 ); (b) Role-playing capability: its dedicated role-playing system supports cultural identity assignment and multi-turn persona consistency, and Tu et al.'s ( 2024 ) CharacterEval benchmark showed that Chinese LLMs outperformed GPT-4 in Chinese role-playing conversation; (c) Chinese cultural knowledge: Cao et al. ( 2023 ) found that LLM cultural alignment varies by platform, and Chinese models significantly outperform international models on Chinese evaluation benchmarks such as C-Eval (Huang et al. 2023 ) and CMMLU (Li, H. et al. 2024 ); (d) Ecological validity: Doubao has over 170 million monthly active users (Xie et al. 2025 ) and is freely available as a mobile app, integrating seamlessly into international students' daily digital environment. The selection of a Chinese-developed LLM over internationally dominant alternatives (e.g., GPT-4o) was grounded in ecological validity and construct alignment: Chinese-developed LLMs are trained on substantially larger proportions of Chinese-language corpora, enabling richer representation of Chinese cultural practices and pragmatic conventions (Huang et al. 2023 ; Li, H. et al. 2024 ), while Doubao's NLP is natively optimized for Chinese tonal recognition, reducing technical friction for HSK Level 3 learners. As the most widely used LLM among Chinese-language users, Doubao directly reflects the AI tools CFL learners encounter in their daily lives in China. This approach, selecting a target-language-native AI platform, follows precedent in Kim and Su's (2024) adoption of a Korean-developed chatbot for Korean-as-a-foreign-language research. A single-platform design was adopted because different LLMs embed different cultural perspectives (Dai et al. 2025 ); standardising the platform isolates the pedagogy as the independent variable. While this design limits cross-platform generalisability, it ensures that observed effects can be attributed to the pedagogical intervention rather than to confounding differences in LLM architecture, training data, or cultural alignment. Generalisability limitations are discussed in Section 5.3 . The study used Doubao-Pro (2025 version), which supports both real-time voice dialogue and text dialogue, with a dedicated role-playing system enabling role specification, personality assignment, and context memory. The platform supports major world languages, enabling learners to clarify concepts in their L1. Prompt design. Each task used a structured prompt template with four components: role specification (cultural identity assignment), language calibration (primarily Chinese at HSK Level 3, with English as a permissible supplement), cultural context framing (authentic practices rather than stereotypes), and interaction constraints (follow-up questions, culturally specific terms with explanations). A sample prompt is provided in Appendix A; complete prompt templates for all six weekly modules are available in Supplementary Note S1. Voice data recording. Doubao's built-in ASR system generates real-time transcripts. Pre-intervention verification with 30 voice samples from five non-participant learners yielded character-level accuracy of 92.3% (SD = 3.8%) for Chinese speech, rising to 95.1% within HSK Level 3 vocabulary. English-language segments were transcribed with comparable accuracy given Doubao's multilingual ASR capability. Segments with recognition errors were manually corrected during analysis. ASR-generated transcripts constitute the interaction logs cited in this study. Interaction data presentation and management. Drawing on the multimodal data collection strategy employed by Yan and Zhang ( 2024 ) in their ChatGPT writing feedback study, experimental group learners were asked to complete a brief reflective learning journal after each weekly voice dialogue session, documenting the core topics discussed, notable cultural discoveries, and shifts in their own perspectives. Learners were encouraged to supplement their journals multimodally, including key screenshots of the dialogue interface, annotated highlights from ASR-generated transcripts, and short voice memos. This design ensured that the qualitative evidence obtained by the researchers consisted of representative interaction episodes self-selected by learners rather than unprocessed complete conversation records, reducing the complexity of data management while simultaneously capturing learners' self-interpretations of their cultural dialogue experiences through the act of purposeful selection. Findings are presented in the Results and Discussion sections through representative excerpts, each annotated with participant number, turn sequence, and week (e.g., Excerpt 1, P03, Week 3). The full interaction log dataset is available from the corresponding author upon reasonable request. 3.3.5 Control Group and Implementation Fidelity The control group received conventional CFL instruction covering comparable cultural content through traditional methods (lectures, readings, discussion, videos) without LLM use. Class hours (90 min/week), topic coverage, and instructor were held constant. All four classes were taught by the same instructor, the regular course teacher rather than a researcher. Researchers designed the intervention and collected data but did not deliver instruction. Implementation fidelity was ensured through: (a) standardised lesson plans; (b) two pre-intervention teacher training sessions; (c) blind classroom observation of two sessions per group; (d) independent learner completion of LLM dialogue phases. Experimental group learners could use Doubao outside class without restrictions (ecological validity rationale), though this created potential exposure non-equivalence (discussed in Section 5.3 ). Post-intervention interviews indicated extracurricular use ranging from none to 3–4 times per week. 3.4 Instruments Three questionnaire instruments were used, all administered in bilingual Chinese–English format. Internal consistency was assessed using Cronbach's α at both pretest and posttest, with a minimum acceptable threshold of α ≥ 0.70 (Nunnally 1978 ). 3.4.1 Intercultural Sensitivity Scale (ISS) The ISS (Chen and Starosta 2000 ) measured the affective dimension of ICC. It comprises 24 items on a 5-point Likert scale (1 = strongly disagree to 5 = strongly agree), measuring five factors: Interaction Engagement (7 items), Respect for Cultural Differences (6 items), Interaction Confidence (5 items), Interaction Enjoyment (3 items), and Interaction Attentiveness (3 items). Nine negatively worded items were reverse-coded prior to analysis (complete items are provided in Supplementary Table S1 ). The ISS is among the most widely used instruments in intercultural communication research and has been adopted by multiple JCR Q1 studies including Hackett et al. ( 2023 ), with well-established cross-cultural reliability and validity evidence. 3.4.2 Cultural Intelligence Scale (CQS) The CQS (Ang et al. 2007 ) measured the cognitive and behavioural dimensions of ICC. It comprises 20 items on a 7-point Likert scale (1 = strongly disagree to 7 = strongly agree), measuring four dimensions: Metacognitive CQ (4 items, assessing cultural awareness and strategic cultural thinking), Cognitive CQ (6 items, assessing cultural knowledge), Motivational CQ (5 items, assessing intrinsic interest and self-efficacy in intercultural interaction), and Behavioral CQ (5 items, assessing behavioural adaptation in intercultural contexts) (complete items are provided in Supplementary Table S2). The CQS has been validated in 98 countries (Ang et al. 2007 ) and was used as a core measure by Hackett et al. ( 2023 ) in their study of technology-mediated intercultural learning. Its four dimensions map onto Byram's (1997) ICC model: Metacognitive CQ corresponds to critical cultural awareness (savoir s'engager), capturing reflective cultural thinking and awareness monitoring; Cognitive CQ corresponds to knowledge (savoirs), assessing cultural facts and norms; Motivational CQ corresponds to attitudes (savoir être), measuring interest in and openness to intercultural interaction; and Behavioral CQ corresponds to skills (savoir faire), assessing behavioural adaptation in intercultural contexts. This mapping is a theoretical approximation. Metacognitive CQ captures the reflective-monitoring dimension of savoir s'engager, specifically awareness of one's own cultural assumptions and strategic adjustment of cultural thinking, but Byram's (1997) broader definition also encompasses political and ideological critique beyond the CQS subscale's scope. Results for Metacognitive CQ are therefore interpreted as capturing the reflective-monitoring component of critical cultural awareness, not its full breadth. 3.4.3 LLM Cultural Learning Perception Questionnaire (LCPQ) A self-developed LCPQ was administered post-intervention to the experimental group only, assessing learners' perceptions and experiences of LLMs as intercultural learning tools. The questionnaire comprised two parts: Part 1: Likert scale items (25 items, 5-point scale) measuring five dimensions (5 items each): perceived cultural learning effectiveness (sample item: "LLM cultural dialogues deepened my understanding of Chinese cultural customs"), LLM dialogue partner quality ("The LLM could play a credible Chinese cultural role"), critical AI literacy development ("I can identify cultural stereotypes or oversimplifications in LLM responses"), intercultural reflection facilitation ("Dialogues with the LLM prompted me to reflect on differences between my culture and Chinese culture"), and continued use intention ("I would continue using LLMs for cultural learning after the course ends"). Part 2: Open-ended questions (4 items) eliciting qualitative data on the most impactful dialogue experience, differences between AI and human cultural discussion, LLM bias encounters, and recommendations for other learners. Given the small sample (n = 32), structural validity evidence is limited to content validity (expert review, I-CVI ≥ 0.78; Lynn 1986 ) and internal consistency (see Table 3 ). Preliminary discriminant validity evidence (inter-dimension Pearson correlations r = 0.35–0.55, all p < 0.05; see Supplementary Table S4) indicated related but distinct constructs. The LCPQ was developed through initial item generation (35 items), expert review, pilot testing with 15 non-participant CFL learners (corrected item-total correlation ≥ 0.30), yielding the final 25 items (all subscale α ≥ 0.70; see Supplementary Table S3 for complete items). 3.5 Data Collection The ISS and CQS were administered to both groups at pretest (Week 0) and posttest (Week 7); the LCPQ was administered to the experimental group only at posttest. All questionnaires were administered online via Wenjuanxing. Within one week after posttest (Week 8), five experimental group learners selected through maximum variation sampling were interviewed individually (see Section 3.6.2 ). Figure 2 presents the complete research procedure and data collection timeline. 3.6 Data Analysis 3.6.1 Quantitative Analysis (RQ1 and RQ2) Quantitative data were analysed using SPSS 28. After computing descriptive statistics and confirming normality (Shapiro–Wilk), baseline equivalence was tested using independent-samples t-tests and chi-square tests. One-way ANCOVA with pretest scores as covariates tested between-group posttest differences (RQ1), with Cohen's d, partial η², and 95% CIs reported. Dimension-level ANCOVAs with Bonferroni correction examined differential responsiveness across ICC dimensions (RQ2). Paired-samples t-tests assessed within-group pre-post changes. LCPQ Likert items were analysed descriptively. 3.6.2 Qualitative Analysis (RQ3) LCPQ open-ended responses and semi-structured interview transcripts were analysed using Braun and Clarke's (2006) six-phase thematic analysis, combining deductive coding (Byram's four ICC dimensions; LLM affordances/limitations) with inductive coding for emergent themes. A second coder independently coded 30% of responses (κ ≥ 0.75 required; Landis and Koch 1977 ). Learners responded in their preferred language; Chinese responses were back-translated for verification. Five interviewees were selected through maximum variation sampling (Patton 2015 ), maximizing diversity in CQS change magnitude, cultural background, gender, and LCPQ partner quality ratings. Interviews lasted 25–40 minutes and were member-checked. Case profiles integrating each participant's quantitative trajectories and qualitative narratives provided individual-level supplements to group-level findings. 3.7 Trustworthiness and Rigor Research quality was ensured through: control group design and ANCOVA controlling for maturation and pretest differences (internal validity); established scales with cross-cultural validation evidence (construct validity); a priori power analysis, effect sizes with CIs, and Bonferroni correction (statistical conclusion validity); 17-country participant diversity with thick description (external validity); and inter-coder reliability, data triangulation, member checking, and disconfirming evidence (qualitative credibility). 3.8 Ethical Considerations This study involved routine classroom pedagogical activities and did not require formal ethics committee approval under institutional guidelines for classroom-based educational research. Protocols included: bilingual informed consent with withdrawal rights; data anonymisation with encrypted storage; multicultural advisory review of cultural topics; equitable post-study access to intervention materials for the control group; and AI ethics disclosure informing participants that LLM outputs may contain biases, with bias identification framed as a learning objective, aligning the ethical requirement with the study's problem-as-resource pedagogy. 4 Results This section presents scale reliability, baseline equivalence, descriptive statistics, and results for the three research questions. 4.1 Scale Reliability As shown in Table 3 , all coefficients exceeded the minimum threshold of α ≥ 0.70 (Nunnally 1978 ). ISS total reliability improved from 0.89 (pretest) to 0.91 (posttest); CQS from 0.91 to 0.93, consistent with reliability levels reported in its 98-country validation (Ang et al. 2007 ). The LCPQ total α reached 0.92, with subscale values ranging from 0.81 to 0.88. Table 3 Internal consistency coefficients (Cronbach's α) Scale/Subscale Pretest α Posttest α ISS Total 0.89 0.91 Interaction Engagement 0.82 0.85 Respect for Cultural Differences 0.79 0.83 Interaction Confidence 0.84 0.86 Interaction Enjoyment 0.76 0.80 Interaction Attentiveness 0.73 0.77 CQS Total 0.91 0.93 Metacognitive CQ 0.85 0.88 Cognitive CQ 0.88 0.90 Motivational CQ 0.83 0.86 Behavioral CQ 0.86 0.89 LCPQ Total — 0.92 Cultural learning effectiveness — 0.87 Dialogue partner quality — 0.81 Critical AI literacy — 0.84 Intercultural reflection — 0.86 Continued use intention — 0.88 4.2 Baseline Equivalence As shown in Table 4 , no significant differences were found on any demographic or outcome variable, confirming baseline equivalence despite intact-class assignment and satisfying ANCOVA prerequisites. Table 4 Baseline equivalence tests Variable Experimental (n = 32) Control (n = 30) Test Statistic p Age (M ± SD) 21.8 ± 2.5 22.1 ± 2.3 t(60) = − 0.49 0.624 Gender (% female) 65.6% 60.0% χ²(1) = 0.21 0.648 HSK Level 3 (%) 90.6% 93.3% Fisher's exact 1.000 Duration in China (M ± SD) 2.1 ± 0.9 2.0 ± 0.8 t(60) = 0.46 0.649 ISS pretest total (M ± SD) 83.50 ± 10.82 80.27 ± 9.45 t(60) = 1.24 0.221 CQS pretest mean (M ± SD) 3.15 ± 0.52 3.08 ± 0.48 t(60) = 0.55 0.586 4.3 Descriptive Statistics As reported in Table 5 , the experimental group showed substantially larger gains on most measures. Table 5 ISS and CQS descriptive statistics by group Measure Experimental (n = 32) Control (n = 30) Pre M(SD) Post M(SD) Δ Pre M(SD) Post M(SD) Δ ISS Total 83.50(10.82) 91.53(9.65) + 8.03 80.27(9.45) 84.20(9.00) + 3.93 Interaction Engagement 3.52(0.58) 3.95(0.50) + 0.43 3.28(0.51) 3.42(0.50) + 0.14 Respect for Cultural Diff. 3.81(0.48) 4.18(0.42) + 0.37 3.85(0.45) 3.95(0.44) + 0.10 Interaction Confidence 2.94(0.62) 3.40(0.55) + 0.46 2.88(0.55) 3.05(0.54) + 0.17 Interaction Enjoyment 3.38(0.55) 3.80(0.50) + 0.42 3.15(0.50) 3.38(0.50) + 0.23 Interaction Attentiveness 3.62(0.50) 3.80(0.46) + 0.18 3.72(0.46) 3.80(0.46) + 0.08 CQS Mean 3.15(0.52) 3.72(0.52) + 0.57 3.08(0.48) 3.30(0.48) + 0.22 Metacognitive CQ 2.73(0.65) 3.58(0.62) + 0.85 3.10(0.58) 3.28(0.58) + 0.18 Cognitive CQ 2.88(0.68) 3.38(0.65) + 0.50 2.92(0.60) 3.22(0.58) + 0.30 Motivational CQ 3.74(0.62) 4.12(0.55) + 0.38 3.38(0.55) 3.55(0.55) + 0.17 Behavioral CQ 2.81(0.58) 3.18(0.56) + 0.37 2.85(0.52) 2.96(0.52) + 0.11 The most notable dimension-level change was Metacognitive CQ (+ 0.85 experimental vs. +0.18 control). Cognitive CQ showed a more moderate between-group difference, with the control group also demonstrating meaningful gains (+ 0.30), consistent with the capacity of conventional instruction to transmit cultural knowledge. Shapiro–Wilk tests confirmed approximate normality for all variables in both groups (p > 0.05). 4.4 RQ1: Overall Impact of LLM-Mediated Cultural Dialogue on ICC 4.4.1 Within-Group Pre-Post Comparisons Paired-samples t-tests revealed that the experimental group showed highly significant gains on all measures (p < 0.001), with effect sizes from d = 0.70 to 1.25 (Table 6 ). Table 6 Paired-samples t-test results Measure Experimental Control t(31) p d t(29) p d ISS Total 5.20 < 0.001 0.92 2.53 0.017 0.46 CQS Mean 5.80 < 0.001 1.03 2.18 0.037 0.40 Metacognitive CQ 7.08 < 0.001 1.25 1.37 0.181 0.25 Cognitive CQ 4.15 < 0.001 0.73 2.72 0.011 0.50 Motivational CQ 4.10 < 0.001 0.73 1.50 0.144 0.27 Behavioral CQ 3.95 < 0.001 0.70 1.09 0.284 0.20 The control group reached significance on ISS Total (p = 0.017, d = 0.46), CQS Mean (p = 0.037, d = 0.40), and on Cognitive CQ in particular (p = 0.011, d = 0.50), indicating that conventional instruction can transmit cultural knowledge effectively but has minimal impact on metacognitive, motivational, and behavioural dimensions. 4.4.2 ANCOVA Between-Group Comparisons After controlling for pretest scores through one-way ANCOVA, between-group differences were significant for both ISS (p = 0.002) and CQS (p = 0.002), as shown in Table 7 . Table 7 ANCOVA results: between-group posttest comparisons (pretest as covariate) DV Adj. M (Exp.) Adj. M (Ctrl.) F(1, 59) p Partial η² Cohen's d [95% CI] ISS Total 90.85 84.92 10.35 0.002 0.149 0.82 [0.30, 1.34] CQS Mean 3.70 3.32 10.85 0.002 0.155 0.83 [0.31, 1.35] Partial η² values of 0.149 and 0.155 both reached Cohen's (1988) large-effect benchmark (0.14), with positive lower bounds of the 95% CI for d confirming reliable intervention effects. LLM-mediated cultural dialogue significantly outperformed conventional instruction in promoting CFL learners' intercultural sensitivity and cultural intelligence. 4.5 RQ2: Differential Response Across ICC Dimensions 4.5.1 ISS Five-Dimension ANCOVA Dimension-level ANCOVAs with Bonferroni correction revealed that four of the five ISS subscales survived the corrected threshold (Table 8 ). Table 8 ISS subscale ANCOVA results (Bonferroni-corrected α' = 0.01) ISS Dimension F(1, 59) p Partial η² Cohen's d Significant Interaction Engagement 11.15 < 0.001 0.159 0.82 Yes Respect for Cultural Diff. 8.68 0.005 0.128 0.72 Yes Interaction Confidence 7.45 0.008 0.112 0.68 Yes Interaction Enjoyment 7.85 0.007 0.117 0.70 Yes Interaction Attentiveness 4.18 0.045 0.066 0.50 No Interaction Engagement showed the largest effect (η² = 0.159, d = 0.82), consistent with the task requirement for learners to actively initiate and sustain conversational turns. Interaction Attentiveness (d = 0.50) did not reach the corrected threshold, likely because habitual attentional tendencies require longer periods to change substantially. 4.5.2 CQS Four-Dimension ANCOVA The same procedure applied to the four CQS dimensions showed that three of four remained significant after Bonferroni correction (Table 9 ). Table 9 CQS subscale ANCOVA results (Bonferroni-corrected α' = 0.0125) CQS Dimension F(1, 59) p Partial η² Cohen's d Significant Metacognitive CQ 18.05 < 0.001 0.234 1.08 Yes Cognitive CQ 4.85 0.031 0.076 0.52 No Motivational CQ 10.42 0.002 0.150 0.80 Yes Behavioral CQ 7.15 0.010 0.108 0.65 Yes Metacognitive CQ yielded the largest single effect in the entire study (η² = 0.234, d = 1.08), corresponding to the reflective-monitoring dimension of Byram's (1997) critical cultural awareness, historically the most resistant to change through conventional instruction. Cognitive CQ showed the smallest effect (d = 0.52) and did not survive Bonferroni correction, consistent with the finding that the control group also showed significant within-group gains on Cognitive CQ (p = 0.011, d = 0.50; Table 6 ). This pattern indicates that cultural knowledge can be effectively transmitted through conventional instruction, and the LLM intervention's distinctive contribution lies elsewhere, specifically in metacognitive, attitudinal, and behavioural dimensions that lectures and readings alone struggle to develop. Figure 3 visualises the effect size ranking across all nine dimensions. The top three dimensions (d ≥ 0.80), corresponding primarily to critical cultural awareness and attitudes, reached large-effect levels. Dimensions ranked fourth through seventh (d = 0.65–0.72) represented medium-to-large effects across attitudes and skills. The bottom two dimensions, Cognitive CQ (d = 0.52) and Interaction Attentiveness (d = 0.50), did not survive Bonferroni correction, revealing a theoretically meaningful pattern: the knowledge dimension, where conventional instruction is also effective, showed the weakest between-group differentiation. This pattern is consistent with Deardorff's (2006) ICC process model, in which attitudes and reflective capacities serve as the "inner ring" that drives subsequent knowledge and skill development. Overall, RQ2 reveals a non-uniform distribution of intervention effects: the LLM intervention's advantages were most pronounced in metacognitive and attitudinal dimensions that conventional instruction struggles to address, and substantially smaller in the knowledge dimension that conventional teaching can partially cover through lectures and readings. 4.6 RQ3: Learner Perceptions of LLM Cultural Learning 4.6.1 LCPQ Descriptive Statistics To examine how learners perceived the LLM-mediated cultural dialogue experience, descriptive statistics for the five LCPQ dimensions were computed (Table 10 ). The overall mean (M = 4.08) indicated a positive evaluation. Table 10 LCPQ descriptive statistics (n = 32) LCPQ Dimension M SD Min Max Rating Level Cultural learning effectiveness 4.38 0.52 3.40 5.00 High Continued use intention 4.42 0.55 3.20 5.00 High Intercultural reflection 4.28 0.48 3.40 4.80 High Critical AI literacy 3.85 0.72 3.00 5.00 Medium-High Dialogue partner quality 3.45 0.58 2.60 4.40 Medium LCPQ Total Mean 4.08 0.45 3.32 4.68 High Cultural learning effectiveness (M = 4.38) and continued use intention (M = 4.42) were highest, indicating that learners perceived substantial learning gains and were willing to continue using LLMs. Dialogue partner quality (M = 3.45) was considerably lower; learners acknowledged the LLM's knowledge breadth and patience but questioned its emotional authenticity and communicative naturalness. Critical AI literacy showed the highest individual variation (SD = 0.72), suggesting uneven development of critical awareness across learners. 4.6.2 Preliminary Observations on Individual Differences Case profile examination suggested two potentially influential factors, though these remain exploratory given sample size constraints. Personality and partner quality ratings. Self-reported extraverted learners (e.g., EXP-029, EXP-031, EXP-023) tended to rate AI partner quality lower (M = 3.07) than more introverted learners (e.g., EXP-011, EXP-005, M = 3.95), possibly because extraverts value emotional reciprocity and spontaneous humor, while introverts appreciate the judgment-free practice environment. Cultural knowledge and critical AI literacy. Learners with stronger cultural knowledge and personal experiences of being misrepresented by AI (e.g., EXP-008 critical AI literacy = 5.00, EXP-005 = 4.80) scored higher than those without such experiences (e.g., EXP-031 = 3.20, EXP-023 = 3.40), suggesting that critical AI literacy development may depend on both a knowledge foundation and triggering events. 4.6.3 Thematic Analysis of Open-Ended Responses Thematic analysis of 32 LCPQ open-ended responses and 5 semi-structured interview transcripts (inter-coder reliability κ = 0.82) yielded four core themes (see Fig. 4 for the thematic map). Theme 1: LLM as a "pressure-free intercultural practice space" (30/32, 93.8%). Nearly all learners noted that the LLM provided a mistake-tolerant environment. Sub-themes included judgment-free language practice ("Talking to AI isn't stressful — it won't laugh at my mistakes," EXP-011), adaptive difficulty calibration ("AI can explain complex cultural concepts in simple words," EXP-031), and freedom to revisit topics without social embarrassment. Theme 2: Experiential cultural knowledge acquisition (28/32, 87.5%). Learners reported gaining situated, concrete cultural knowledge through dialogue that contrasted sharply with textbook-style knowledge. EXP-005 noted: "The textbook says Chinese people value mianzi. But only through AI dialogue did I understand how mianzi works in specific situations." Multiple learners also reported that LLM dialogue prompted re-examination of their own cultural practices. Theme 3: Differences between AI and human interaction quality (32/32, 100%). All learners spontaneously compared AI and human interactions, adopting a "complementary rather than substitutive" cognitive frame. Extraverted learners particularly noted the AI's lack of humor and emotional responsiveness ("Talking to AI lacks one thing — fun," EXP-031). Most positioned the LLM as a preparation tool for learning basic cultural knowledge and strategies through AI before applying them in real intercultural encounters. Theme 4: LLM cultural bias identification and response (25/32, 78.1%). Nearly 80% of learners reported encountering at least one instance of LLM cultural bias. Sub-themes included stereotypical representations of learners' home cultures (EXP-031 objecting to "taco and sombrero" as first associations with Mexico; EXP-008 noting AI's treatment of "Chinese people" as a monolithic group ignoring Southeast Asian Chinese diversity), idealized representations of Chinese culture ("AI's version of Chinese culture is too perfect — different from what I actually see in China," EXP-011), and differentiated response strategies ranging from low-level accept and note through mid-level correct and probe to high-level critical analysis of training data composition and cultural power relations (mainly from learners with stronger cultural knowledge, e.g., EXP-008, EXP-005). Interview data corroborated these themes and revealed three additional findings: (a) learners unanimously emphasized the voice modality's contribution to communicative authenticity and real-time processing practice; (b) three interviewees reported substantial extracurricular self-directed use of Doubao for cultural exploration; (c) two high-scoring learners could pinpoint specific "awakening moments" when their critical awareness shifted, triggered by encountering personally relevant cultural misrepresentations. 4.6.4 Triangulation of Quantitative and Qualitative Findings Qualitative themes corroborated quantitative results. The largest effect size on Metacognitive CQ (quantitative) aligned with the high frequency of bias identification in Theme 4 (qualitative), pointing to a shared mechanism: repeated bias identification and discussion trained learners' metacognitive reflective capacity. The lowest LCPQ rating on dialogue partner quality (quantitative) corresponded to Theme 3's consistent AI–human comparison narratives (qualitative). The high individual variance on critical AI literacy (SD = 0.72, quantitative) was mirrored by the stratified response strategies in Theme 4c (qualitative), suggesting that critical awareness development may be linked to learners' cultural knowledge base and personal experiences. 5 Discussion 5.1 Why LLM-Mediated Cultural Dialogue Works: Contributing Factors (RQ1 and RQ2) The significant between-group effects on both ISS and CQS corroborate Liu, J.'s (2025) finding that AI-enhanced learning can promote ICC development, and extend it to a generative LLM dialogue paradigm. We identify three factors that likely contributed to the intervention's effectiveness. First, the LLM's role as a cultural interlocutor rather than a learning scaffold distinguishes this intervention from prior work (e.g., Wang et al. 2025 ). By positioning the LLM as a simulated cultural other with whom learners negotiate intercultural meaning in real time, the intervention activated experiential learning processes (Kolb 1984 ) that lectures and readings cannot replicate. The effect sizes (d = 0.82–0.83) exceeded the median in L2 intervention research (d = 0.70; Plonsky and Oswald 2014 ), suggesting that dialogic engagement with a culturally embedded AI partner provides qualitatively different learning affordances. Second, the problem-as-resource strategy transformed LLM cultural biases into metacognitive triggers. When learners encountered stereotypical representations of their home cultures or idealised portrayals of Chinese culture, they were compelled to ask "Is this fact or bias?", operations at the core of metacognitive CQ. This mechanism, combined with institutionalised reflection support (group discussion and bias checking), explains why Metacognitive CQ showed the largest effect while Cognitive CQ showed the smallest: the intervention's distinctive value lies not in knowledge transmission (which conventional instruction also achieves) but in cultivating reflective capacities that traditional methods struggle to develop. This is consistent with Jenks's (2024) argument that LLM biases can serve as cognitive scaffolds for critical cultural awareness. A methodological caveat is warranted: Metacognitive CQ captures the reflective-monitoring subcomponent of Byram's (1997) savoir s'engager but not its full scope, which also includes political and ideological critique. Third, the voice modality restored authentic communicative pressure. Unlike text-based chatbot interactions that eliminate real-time processing demands (Young 2011 ), Doubao's voice dialogue required learners to manage turn-taking, process spoken Chinese under time pressure, and respond without the luxury of editing, conditions approximating face-to-face intercultural encounters. Interview data consistently highlighted this feature as a key contributor to communicative authenticity, likely explaining the strong effects on Interaction Engagement and Interaction Confidence. Nevertheless, Behavioral CQ remained the lowest CQS dimension, confirming that simulated dialogue cannot fully substitute for the behavioural demands of real intercultural encounters; adjusting speech rate, managing silence, and deploying non-verbal cues remain beyond LLM-mediated practice. Learners consistently described the LLM as preparation for, rather than replacement of, human interaction. 5.2 Who Benefits and How: Implications for Educators, Learners, and Researchers (RQ3) The LCPQ data and qualitative findings reveal that the intervention benefits different learner profiles in different ways, carrying specific implications for three stakeholder groups. For educators, the results demonstrate that LLMs are most effective as intercultural preparation tools paired with authentic interaction opportunities. The low dialogue partner quality rating (M = 3.45) confirms that LLMs cannot replace human cultural exchange; rather, they provide a low-stakes space for learners to acquire cultural strategies before applying them in real encounters. It is essential that bias-check activities not be omitted, as the high individual variation in critical AI literacy (SD = 0.72) indicates that critical awareness does not grow spontaneously with LLM use, and learners lacking cultural knowledge or personal experience of AI misrepresentation need explicit teacher guidance. The LLM's safety alignment mechanism also constrained intercultural simulation on sensitive topics, requiring instructors to convert evasive AI behaviour into discussion material. For learners, personality and prior knowledge moderated the experience. Introverted learners valued the judgment-free environment and rated partner quality higher; extraverted learners were more dissatisfied with the AI's lack of emotional reciprocity. This suggests that LLM-mediated cultural dialogue may be particularly valuable for learners who experience anxiety in face-to-face intercultural encounters, a population that conventional interaction-based approaches may underserve. Additionally, learners with stronger cultural knowledge developed critical AI literacy more effectively, suggesting that the intervention works best when learners have a baseline cultural foundation to evaluate AI outputs against. Educators should consider sequencing: building foundational cultural knowledge before introducing LLM-mediated bias-identification tasks. For researchers, the non-uniform dimension-level pattern raises important methodological considerations. The intervention comprised multiple co-occurring components (individualised AI dialogue, voice modality, adaptive feedback, and structured reflection) that the present design cannot fully disentangle. A purely practice-quantity or novelty explanation is unlikely given the pronounced differentiation (Metacognitive CQ d = 1.08 vs. Cognitive CQ d = 0.52), but factorial designs (e.g., LLM voice vs. text vs. human partner vs. control) are needed to isolate each component's contribution. The problem-as-resource framework demonstrated here may also be applicable beyond CFL contexts; any LLM-mediated educational setting where AI outputs contain systematic biases could benefit from converting those biases into critical thinking opportunities. Future studies should incorporate objective measures (e.g., role-play assessments, discourse analysis), active placebo controls, and delayed posttests to assess sustainability. 5.3 Limitations The sample of 62 participants from a single university, predominantly at HSK Level 3 and using a single LLM platform (Doubao), limits generalisability. Both instruments are self-report measures susceptible to social desirability bias and Hawthorne effects, though the non-uniform results and the control group's modest gains partially mitigate these concerns. Finally, six weeks may be insufficient for full Behavioral CQ development, which requires sustained real-world practice. 6 Conclusion This quasi-experimental study provides empirical evidence that LLM-mediated cultural dialogue can promote ICC development in CFL learners, with medium-to-large effects on both intercultural sensitivity and cultural intelligence. The non-uniform dimension-level pattern, where Metacognitive CQ showed the largest effect while Cognitive CQ showed the smallest, indicates that the intervention's distinctive value lies in cultivating reflective capacities rather than transmitting cultural knowledge. The core contribution is a problem-as-resource strategy that transforms LLM cultural biases into pedagogical material for critical cultural awareness, extending Byram's (1997) ICC pedagogy into AI-mediated language education. Declarations Competing Interests The authors declare no competing interests. Funding [Removed for double-blind review.] Author Contribution Sun Jing: Conceptualization, Methodology, Investigation, Formal analysis, Writing – original draft. Nie Liming: Supervision, Validation, Writing – review & editing. Acknowledgement This research was supported by the National Natural Science Foundation of China (Grant No. W2412110), the Shenzhen Peacock Plan (Grant No. GDRC202515), and the Shenzhen Technology University Teaching Reform Project (Grant No. 20251016). Data Availability The datasets generated by the survey research during and/or analysed during the current study are available in the Google repository(anonymous):https://docs.google.com/spreadsheets/d/1a2C5QdXQ5PWBUiaeDk48zW8MY6OcAHRP/edit?usp=sharing&ouid=103666323466924573598&rtpof=true&sd=true Ethics Statement This study was reviewed and approved by the Academic Committee of the ****[Removed for double-blind review]. This committee oversees research involving human participants conducted within the school and ensures compliance with institutional ethical standards. All research procedures were performed in accordance with the ethical standards of the Academic committee, the 1964 Declaration of Helsinki and its later amendments. The study involved routine classroom pedagogical activities within an established curriculum and did not constitute biomedical or clinical research. Written informed consent was obtained from all participants prior to data collection. Participants were informed of the study's purpose, procedures, the voluntary nature of participation, and their right to withdraw at any time without penalty. All data were anonymized prior to analysis. References Ang S, Van Dyne L, Koh C, Ng KY, Templer KJ, Tay C, Chandrasekar NA (2007) Cultural intelligence: its measurement and effects on cultural judgment and decision making, cultural adaptation and task performance. Manage Organ Rev 3(3):335–371. https://doi.org/10.1111/j.1740-8784.2007.00082.x Braun V, Clarke V (2006) Using thematic analysis in psychology. Qualitative Res Psychol 3(2):77–101. https://doi.org/10.1191/1478088706qp063oa Byram M (1997) Teaching and assessing intercultural communicative competence. Multilingual Matters Campbell DT, Stanley JC (1963) Experimental and quasi-experimental designs for research. Houghton Mifflin Cao Y, Zhou L, Lee S, Cabello L, Chen M, Hershcovich D (2023) Assessing cross-cultural alignment between ChatGPT and human societies: an empirical study. In: Proceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP), pp 53–67. Association for Computational Linguistics. https://doi.org/10.18653/v1/2023.c3nlp-1.7 Chen GM, Starosta WJ (2000) The development and validation of the intercultural sensitivity scale. Hum Communication 3:1–15 Cohen J (1988) Statistical power analysis for the behavioral sciences, 2nd edn. Lawrence Erlbaum Associates Creswell JW, Plano Clark VL (2018) Designing and conducting mixed methods research, 3rd edn. SAGE Cui H, Mahfoodh OHA (2025) An intercultural framework for assessing sociopragmatic competence in English language education: evidence from the Chinese EFL context. International Journal of Applied Linguistics. Advance online publication Dai DW, Zhu H, Chen G (2025) How does interaction with LLM powered chatbots shape human understanding of culture? The need for Critical Interactional Competence (CritIC). Annu Rev Appl Linguist 45:28–49. https://doi.org/10.1017/S0267190525000054 Deardorff DK (2006) Identification and assessment of intercultural competence as a student outcome of internationalization. J Stud Int Educ 10(3):241–266. https://doi.org/10.1177/1028315306287002 Dewaele J-M, MacIntyre PD (2014) The two faces of Janus? Anxiety and enjoyment in the foreign language classroom. Stud Second Lang Learn Teach 4(2):237–274. https://doi.org/10.14746/ssllt.2014.4.2.5 Faul F, Erdfelder E, Lang A-G, Buchner A (2007) GPower 3: a flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behav Res Methods* 39(2):175–191. https://doi.org/10.3758/BF03193146 Freiermuth M, Huang H-C (2021) Zooming across cultures: can a telecollaborative video exchange between language learning partners further the development of intercultural competences? Foreign Lang Annals 54(1):185–206. https://doi.org/10.1111/flan.12504 Ge H (2025) The theoretical framework of compulsory English textbook evaluation based on Byram's intercultural communicative competence (ICC) in alignment with Chinese English curriculum standard (CECS) educational policy. The EUrASEANs: Journal on Global Socio-Economic Dynamics 2(51):25–40 Hackett S, Janssen J, Beach P, Perreault M, Beelen J, van Tartwijk J (2023) The effectiveness of Collaborative Online International Learning (COIL) on intercultural competence development in higher education. Int J Educational Technol High Educ 20. https://doi.org/10.1186/s41239-022-00373-3 . :Article 5 Huang L-J (2021) Developing intercultural communicative competence in foreign language classrooms – a study of EFL learners in Taiwan. Int J Intercultural Relations 83:78–90. https://doi.org/10.1016/j.ijintrel.2021.06.001 Huang Y, Bai Y, Zhu Z, Zhang J, Zhang J, Su T, Liu J, Lv C, Zhang Y, Lei J, Fu Y, Sun M, He J (2023) C-Eval: a multi-level multi-discipline Chinese evaluation suite for foundation models. In: Advances in Neural Information Processing Systems 36 (NeurIPS 2023) Huang Z, Cui N (2025) Research on the challenges, strategies and optimization paths of cultivating cross cultural communication competence in international Chinese education. J Educational Res Policies 7(6):40–43. https://doi.org/10.53469/jerp.2025.07(06).08 Jenks CJ (2024) Communicating the cultural other: trust and bias in generative AI and large language models. Appl Linguistics Rev 16(3):787–795. https://doi.org/10.1515/applirev-2024-0051 Kim A, Su Y (2024) The effects of chatbot-mediated L2 practice opportunities on Korean as a foreign language learners' willingness to communicate. System 125. https://doi.org/10.1016/j.system.2024.103425 . Article 103425 Kolb DA (1984) Experiential learning: experience as the source of learning and development. Prentice-Hall Landis JR, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics 33(1):159–174. https://doi.org/10.2307/2529310 Li B, Tan YL, Wang C, Lowell V (2025) Two years of innovation: a systematic review of empirical generative AI research in language learning and teaching. Computers Education: Artif Intell 9. https://doi.org/10.1016/j.caeai.2025.100445 . Article 100445 Li H, Zhang Y, Koto F, Yang Y, Zhao H, Gong Y, Duan N, Baldwin T (2024) CMMLU: measuring massive multitask language understanding in Chinese. In: Findings of the Association for Computational Linguistics: ACL 2024, pp 11260–11285. https://doi.org/10.18653/v1/2024.findings-acl.671 Lin S (2025) The cultivation of international students' intercultural communicative competence in teaching Chinese as a foreign language. Vocat Educ 14(1):78–85 Liu D, Liu X (2025) Exploring the empowerment of international Chinese language teachers through large language models. J Contemp Educational Res 9(2):1–12 Liu F (2025) Second language teaching and learning through virtual exchange [Book review]. Innov Lang Learn Teach 19(1):1–3. https://doi.org/10.1080/17501229.2025.2451729 Liu G, Darvin R, Ma C (2024) Exploring AI-mediated informal digital learning of English (AI-IDLE): a mixed-method investigation of Chinese EFL learners' AI adoption and experiences. Comput Assist Lang Learn 38(8):1632–1660. https://doi.org/10.1080/09588221.2024.2312130 Liu J (2025) Exploring the impact of artificial intelligence-enhanced language learning on youths' intercultural communication competence. Humanit Social Sci Commun 12. https://doi.org/10.1057/s41599-025-06033-x . :Article 1757 Lo CK, Yu PLH, Xu S, Ng D, Jong MS-Y (2024) Exploring the application of ChatGPT in ESL/EFL education and related research issues: a systematic review of empirical studies. Smart Learn Environ 11. https://doi.org/10.1186/s40561-024-00319-2 . Article 34 Lynn MR (1986) Determination and quantification of content validity. Nurs Res 35(6):382–385 Morris T (2019) Experiential learning – a systematic review and revision of Kolb's model. Interact Learn Environ 28(8):1064–1077. https://doi.org/10.1080/10494820.2019.1570279 Munezane Y (2019) A new model of intercultural communicative competence: bridging language classrooms and intercultural communicative contexts. Stud High Educ 46(8):1664–1681. https://doi.org/10.1080/03075079.2019.1698537 Nunnally JC (1978) Psychometric theory, 2nd edn. McGraw-Hill O'Dowd R (2018) From telecollaboration to virtual exchange: state-of-the-art and the role of UNICollaboration in moving forward. J Virtual Exch 1:1–23. https://doi.org/10.14705/rpnet.2018.jve.1 Patton MQ (2015) Qualitative research & evaluation methods: integrating theory and practice, 4th edn. SAGE Plonsky L (2013) Study quality in SLA: an assessment of designs, analyses, and reporting practices in quantitative research. Stud Second Lang Acquisition 35(4):655–687. https://doi.org/10.1017/S0272263113000399 Plonsky L, Oswald FL (2014) How big is big? Interpreting effect sizes in L2 research. Lang Learn 64(4):878–912. https://doi.org/10.1111/lang.12079 Sarwari A, Javed MN, Adnan HM, Abdul Wahab MN (2024) Assessment of the impacts of artificial intelligence (AI) on intercultural communication among postgraduate students in a multicultural university environment. Sci Rep 14. https://doi.org/10.1038/s41598-024-74565-w . Article 23108 Sweller J (1988) Cognitive load during problem solving: effects on learning. Cogn Sci 12(2):257–285. https://doi.org/10.1207/s15516709cog1202_4 Tahmasbi A, Esrafilian M, Wright J, Jeong S, Bera A (2025) Game master LLM: task-based role-playing for natural slang learning. Preprint at https://arxiv.org/abs/2511.15504 Thi H-T (2021) Project-based assessment in teaching intercultural communication competence for foreign language students in higher education: a case study. Eur J Educational Res 10(2):933–944. https://doi.org/10.12973/eu-jer.10.2.933 Tu Q, Fan S, Tian Z, Shen T, Shang S, Gao X, Yan R (2024) CharacterEval: a Chinese benchmark for role-playing conversational agent evaluation. In: Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp 11836–11850. https://doi.org/10.18653/v1/2024.acl-long.638 Wang P, Yin K, Zhang M, Dong Y, Wang T (2025) The effect of incorporating large language models into the teaching on critical thinking disposition: an AI + Constructivism Learning Theory attempt. Educ Inform Technol 30:11625–11647. https://doi.org/10.1007/s10639-024-13244-3 Weng Z, Fu Y (2025) Generative AI in language education: bridging divide and fostering inclusivity. Int J Technol Educ 8(1):1–22 Xie T, Zhou Y, Yu J (2025) Ce-LLMs: status and trends of education-specific large language models developed in China. Future Educational Res 3:505–525. https://doi.org/10.1002/fer3.70008 Yan D, Zhang S (2024) L2 writer engagement with automated written corrective feedback provided by ChatGPT: a mixed-method multiple case study. Humanit Social Sci Commun 11(1):1–14. https://doi.org/10.1057/s41599-024-03543-y Young RF (2011) Interactional competence in language learning, teaching, and testing. In: Hinkel E (ed) Handbook of research in second language teaching and learning, vol 2. Routledge, pp 426–443 Yuan H, Che Z, Zhang Y, Li S, Yuan X, Huang L, Hu X, Peng K, Luo S (2025) The cultural stereotype and cultural bias of ChatGPT. J Pac Rim Psychol 19:1–15. https://doi.org/10.1177/18344909251315797 Zheng H (2024) ChatGPT integration of English education: implications for English language learners' cross-cultural communication. J Educ Humanit Social Sci 27:128–135 Additional Declarations No competing interests reported. Supplementary Files FromculturalbiastocriticalawarenessSupplementaryInformation.docx AppendixA.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 05 May, 2026 Reviews received at journal 30 Apr, 2026 Reviewers agreed at journal 11 Apr, 2026 Reviewers agreed at journal 09 Apr, 2026 Reviewers agreed at journal 01 Apr, 2026 Reviewers invited by journal 30 Mar, 2026 Editor assigned by journal 30 Mar, 2026 Editor invited by journal 30 Mar, 2026 Submission checks completed at journal 29 Mar, 2026 First submitted to journal 29 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9140660","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":614807555,"identity":"fc31c923-21c5-4f34-95be-be2737bd52ce","order_by":0,"name":"Jing Sun","email":"","orcid":"","institution":"China Jiliang University","correspondingAuthor":false,"prefix":"","firstName":"Jing","middleName":"","lastName":"Sun","suffix":""},{"id":614807557,"identity":"526c6748-be00-4cee-949b-5fa8f8396507","order_by":1,"name":"Liming Nie","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAwUlEQVRIiWNgGAWjYBACxvmHDz+Q+CEhx0+0FuYZbGkGlj0WxpINxGphn8FjIFHBVpFocIBYLbyzGwwMbvBIJBgfT97A8KNiG2EtknMOJDycYSGRZ3bmWQFjz5nbhLUYNiQcMJbgkSg2u5FjwMzYRoQW+wOJDdJ/2CQSN88gVgvjjGQGCQmglg0SRGvpOcZmINkjYSwB9MtBovzC2N7/GRiVdXL87ckbH/yoIEILEkggPmoQWkjVMQpGwSgYBSMEAAAK7z4vDiIeMAAAAABJRU5ErkJggg==","orcid":"","institution":"Shenzhen Technology University","correspondingAuthor":true,"prefix":"","firstName":"Liming","middleName":"","lastName":"Nie","suffix":""}],"badges":[],"createdAt":"2026-03-16 17:23:21","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9140660/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9140660/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105902193,"identity":"488ecebb-fc59-497f-908a-1d5cf4fb6762","added_by":"auto","created_at":"2026-04-01 09:38:02","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1720039,"visible":true,"origin":"","legend":"\u003cp\u003eTheoretical framework integrating Byram’s (1997) ICC model and Kolb’s (1984) experiential learning cycle.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-9140660/v1/c10e7926734e3c5eefac8b43.png"},{"id":105902196,"identity":"43761570-5ed2-4ef8-83d0-dcd7da7085f9","added_by":"auto","created_at":"2026-04-01 09:38:02","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2003801,"visible":true,"origin":"","legend":"\u003cp\u003eResearch procedure and data collection timeline.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9140660/v1/c827330aa39074d3485882ce.png"},{"id":106093046,"identity":"2d96310f-261a-4888-871b-7a983274052b","added_by":"auto","created_at":"2026-04-03 11:33:08","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":2295968,"visible":true,"origin":"","legend":"\u003cp\u003eCohen’s d effect sizes for the LLM-mediated intervention across nine ICC dimensions.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-9140660/v1/5216b9f5ddc40ad145fa0dcc.png"},{"id":106093179,"identity":"a4348e11-c26d-4281-8309-d20d89f9e270","added_by":"auto","created_at":"2026-04-03 11:35:43","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":3408950,"visible":true,"origin":"","legend":"\u003cp\u003eThematic map of learner perceptions of LLM-mediated cultural learning (RQ3).\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-9140660/v1/ece563c6755c194ccbd3faaa.png"},{"id":108005741,"identity":"8e7ebbde-45b7-42d0-9d29-8003b53ea5d2","added_by":"auto","created_at":"2026-04-28 12:47:21","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":11805732,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9140660/v1/21626792-1ead-4e4f-a35f-31b4e45f7b43.pdf"},{"id":105905242,"identity":"44ec603a-28a2-4d97-9abd-31c241efa147","added_by":"auto","created_at":"2026-04-01 10:11:43","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":53215,"visible":true,"origin":"","legend":"","description":"","filename":"FromculturalbiastocriticalawarenessSupplementaryInformation.docx","url":"https://assets-eu.researchsquare.com/files/rs-9140660/v1/afb79fb0b7073385f60d5560.docx"},{"id":105902195,"identity":"027651dc-70bc-4778-b5d2-4fb886ce65ff","added_by":"auto","created_at":"2026-04-01 09:38:02","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":15196,"visible":true,"origin":"","legend":"","description":"","filename":"AppendixA.docx","url":"https://assets-eu.researchsquare.com/files/rs-9140660/v1/4945225157da6e88d8c6d204.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"From cultural bias to critical awareness: LLM-mediated voice dialogue and intercultural competence in Chinese language learners","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eAs international education expands and cross-cultural contact intensifies, the ability to communicate effectively across cultural boundaries has become an essential competence for global citizens. Intercultural communicative competence (ICC) is accordingly a core objective of foreign language education (Byram \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e1997\u003c/span\u003e; Deardorff \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2006\u003c/span\u003e), and its development requires experiential interaction with cultural otherness and sustained critical reflection (Huang \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Yet providing learners with sufficient meaningful intercultural encounters within formal classroom constraints remains a fundamental pedagogical challenge that intensifies as student mobility increases while instructional resources remain finite. For learners of Chinese as a Foreign Language (CFL), this challenge is especially acute: the sociocultural complexity of China, the linguistic distance of Chinese, and the structural scarcity of authentic cultural interaction opportunities together create barriers that conventional instruction struggles to overcome (Huang and Cui \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Telecollaboration and virtual exchange programs have partially bridged this gap (Freiermuth and Huang \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), but these approaches are constrained by partner availability, scheduling logistics, and a limited range of simulable cultural scenarios.\u003c/p\u003e \u003cp\u003eRecent advances in large language models (LLMs) present an alternative approach. As \"on-demand\" dialogue partners, LLMs can simulate diverse cultural perspectives, adapt to learners' proficiency levels, and allow low-stakes experimentation with intercultural communication strategies. Liu, J. (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) provided initial empirical evidence that an AI-enhanced language learning system can significantly improve learners' intercultural communication competence, yet that study employed a bespoke deep-learning architecture rather than generative LLM dialogue, leaving the pedagogical potential of LLM-mediated cultural interaction largely untested. At the same time, LLMs carry well-documented cultural biases: Cao et al. (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) found that ChatGPT systematically favours American cultural values, Yuan et al. (\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) recorded stereotypical patterns across 13 cultural dimensions, and Jenks (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) warned against treating AI-generated cultural content as objective.\u003c/p\u003e \u003cp\u003eWang et al. (\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) demonstrated through a quasi-experimental design that LLM-integrated flipped-classroom instruction can improve critical thinking dispositions. However, their study positioned the LLM as a \"learning aid\" for general critical thinking. ICC cultivation demands a fundamentally different role, that of a cultural interlocutor with whom learners negotiate intercultural meaning. Can LLMs effectively simulate cultural identities? How do their biases shape learners' intercultural cognition? These questions remain unanswered.\u003c/p\u003e \u003cp\u003eNo prior study has empirically examined whether LLM-mediated cultural dialogues can promote ICC development in language learners. Given that millions of language learners worldwide already use LLM-based tools in their daily studies, this gap has direct practical implications: educators lack evidence-based guidance on whether and how to integrate these tools into intercultural instruction. The present study addresses this gap. Grounded in Byram's (1997) ICC model and Kolb's (1984) experiential learning cycle, it designed and implemented a six-week LLM-mediated cultural dialogue intervention, addressing the following research questions:\u003c/p\u003e \u003cp\u003eRQ1: To what extent does LLM-mediated cultural dialogue improve CFL learners' intercultural competence (compared with a control group receiving conventional cultural instruction)?\u003c/p\u003e \u003cp\u003eRQ2: Does the LLM-mediated intervention affect the various ICC dimensions (knowledge, skills, attitudes, and critical cultural awareness) differentially?\u003c/p\u003e \u003cp\u003eRQ3: How do CFL learners perceive the value and limitations of LLMs as intercultural learning tools?\u003c/p\u003e \u003cp\u003eThis study contributes by: (a) extending Liu, J.'s (2025) AI-enhanced ICC findings to a pedagogically grounded LLM-mediated cultural dialogue paradigm; (b) proposing a problem-as-resource strategy that converts LLM cultural biases into a classroom-actionable pathway for critical awareness cultivation; and (c) producing reusable cultural dialogue task templates grounded in the Byram ICC model and the Kolb learning cycle.\u003c/p\u003e"},{"header":"2 Literature Review","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Intercultural Communicative Competence in Language Education\u003c/h2\u003e \u003cp\u003eICC has been a central construct in applied linguistics since Byram's (1997) five-savoirs model: savoirs (knowledge), savoir comprendre (interpreting skills), savoir apprendre/faire (discovery and interaction skills), savoir \u0026ecirc;tre (attitudes of openness), and savoir s'engager (critical cultural awareness). This model has been widely adopted for both theoretical analysis and curriculum design (Ge \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Subsequent refinements include Deardorff's (2006) process model emphasizing ICC's developmental and iterative nature, Munezane's (2019) integration of willingness to communicate and constructive conflict resolution, and Cui and Mahfoodh's (2025) empirically weighted sociopragmatic assessment framework in the Chinese EFL context.\u003c/p\u003e \u003cp\u003eEmpirically, Huang (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) found that explicit intercultural instruction improved EFL learners' ICC in knowledge and skills dimensions but had limited impact on attitudes and critical cultural awareness, a pattern corroborated by textbook analyses. Thi (\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) demonstrated the effectiveness of project-based assessment for ICC development over a 9-week course. These findings suggest that ICC requires sustained, experiential engagement rather than isolated knowledge transmission.\u003c/p\u003e \u003cp\u003eA persistent challenge remains the provision of authentic intercultural interaction opportunities. In CFL contexts, this is compounded by cultural distance and the complexity of navigating concepts such as mianzi, guanxi, and keqi that lack direct equivalents in many languages (Huang and Cui \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). International students in China have some access to authentic encounters, but these are often unsystematic and difficult to process without guided reflection (Lin \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Technology-Enhanced Intercultural Learning\u003c/h2\u003e \u003cp\u003eTelecollaboration has been the dominant paradigm for technology-mediated ICC development (O'Dowd \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Liu, F. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Freiermuth and Huang (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) found that telecollaborative video exchanges between Japanese and Taiwanese students were mutually beneficial for cultural learning, yet the study involved only 11 participants and required extensive cross-institutional coordination, illustrating both the promise and the scalability constraints of human-to-human virtual exchange.\u003c/p\u003e \u003cp\u003eEarlier technological alternatives, from chatbots to virtual environments, lacked the linguistic sophistication for credible intercultural dialogue. Text-based chatbots also suffer a fundamental modality deficit: authentic oral intercultural communication requires real-time processing pressure, turn-taking management, and paralinguistic cue interpretation, all core skills in Byram's (1997) interaction dimension (savoir apprendre/faire; Young \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). Text interaction eliminates these demands, thus failing to train real-time intercultural communication. LLMs supporting real-time voice interaction have altered this situation, offering agents with high fluency, broad cultural knowledge, and role-play capability. The voice modality restores the time pressure and turn-taking dynamics of spoken communication, enabling practice under conditions approximating authentic exchanges.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Generative AI and Cultural Learning in Language Education\u003c/h2\u003e \u003cp\u003eGenerative AI applications in language education have grown substantially since ChatGPT's release. Lo et al.'s (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) systematic review of 70 empirical studies found that research overwhelmingly focuses on writing skills, with almost none examining cultural learning or ICC, a significant gap. In CFL contexts, Liu et al. (\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) found learners adopting AI as both tutor and conversation partner, Liu and Liu (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) explored LLM-empowered teacher content creation, and Tahmasbi et al. (\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) demonstrated that LLM-powered role-playing enhances vocabulary acquisition.\u003c/p\u003e \u003cp\u003eEmpirical work at the GenAI\u0026ndash;ICC intersection remains scarce. Liu, J. (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) demonstrated that an AI-enhanced language learning system integrating deep-learning models significantly improved youths' cultural understanding, communication strategies, and cultural sensitivity, providing the first experimental evidence that AI can measurably promote ICC development. However, that system relied on a custom encoder\u0026ndash;decoder architecture rather than generative LLM dialogue, and its intervention did not incorporate critical engagement with AI-generated cultural content. The present study extends this line of inquiry by employing LLM-mediated voice-based cultural dialogue and explicitly employing LLM cultural biases as pedagogical resources. Weng and Fu (\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) and Zheng (\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) corroborated these findings while noting that cultural communication requires deeper modeling than general language tasks.\u003c/p\u003e \u003cp\u003eA related usability concern has received limited attention: for lower-proficiency learners, discussing deep cultural topics in the target language may impose prohibitive cognitive load. Sweller's (1988) cognitive load theory suggests that HSK Level 3 learners discussing abstract cultural concepts in Chinese face simultaneous linguistic and cultural processing demands that risk overloading working memory (Dewaele and MacIntyre \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Li et al. (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) noted that GenAI usability in language education depends on whether the design can reduce extraneous cognitive load. This consideration directly informed the platform selection in the present study (see Section \u003cspan refid=\"Sec17\" class=\"InternalRef\"\u003e3.3.4\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 LLM Cultural Affordances and Biases\u003c/h2\u003e \u003cp\u003eLLM-mediated intercultural learning must contend with documented cultural biases. Cao et al. (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) found that ChatGPT exhibits strong alignment with American cultural values while adapting poorly to other contexts, with English-language prompts systematically flattening cultural differences. Yuan et al. (\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) documented cultural stereotypes across 13 dimensions, though their Study 3 demonstrated that targeted prompt strategies can effectively reduce stereotype generation. Jenks (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) argued that intercultural communication researchers must critically examine how LLMs \"produce and circulate discourse in an ostensibly impartial way\" (p. 788), highlighting the need for learners to develop critical AI literacy alongside ICC.\u003c/p\u003e \u003cp\u003eThe present study treats LLM cultural biases not merely as a limitation but as a pedagogical resource. Building on Yuan et al.'s (\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) prompt-strategy findings and Jenks's (2024) call for critical engagement, our intervention incorporates explicit activities for identifying and discussing cultural biases in LLM outputs, directly targeting the critical cultural awareness dimension of Byram's (1997) model, the dimension most resistant to change through conventional instruction (Huang \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Research Gap and the Present Study\u003c/h2\u003e \u003cp\u003eIn summary, the literature reveals a three-way intersection gap: no empirical study has examined LLM-mediated cultural dialogue for ICC development in CFL contexts. This gap is both theoretically significant (leaving a rapidly adopted technology without pedagogical grounding) and practically consequential, as CFL programmes worldwide seek scalable alternatives to resource-intensive telecollaboration. The present study addresses this gap by extending Liu, J.'s (2025) AI-enhanced ICC findings to a generative LLM dialogue paradigm, introducing the LLM as a \"simulated cultural other\" for intercultural practice, and applying LLM cultural bias research to pedagogy through a problem-as-resource strategy.\u003c/p\u003e \u003cp\u003eThe theoretical framework integrates Byram's (1997) ICC model with Kolb's (1984) experiential learning cycle (revised by Morris \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), structuring each task as a complete cycle: concrete experience (LLM dialogue), reflective observation (group discussion), abstract conceptualization (cultural concept extraction), and active experimentation (application in subsequent tasks).\u003c/p\u003e \u003c/div\u003e"},{"header":"3 Method","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Research Design\u003c/h2\u003e \u003cp\u003eThis study adopted a quasi-experimental nonequivalent control group pretest-posttest design (Campbell and Stanley \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e1963\u003c/span\u003e), with an explanatory sequential mixed-methods approach (Creswell and Plano Clark \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). The experimental group participated in a six-week LLM-mediated cultural dialogue intervention; the control group received conventional CFL instruction with comparable cultural content but without LLM use. Quantitative data (standardised scales) served as the primary strand; qualitative data (open-ended questionnaire responses and semi-structured interviews) provided supplementary explanation. This design aligns with recent studies in the field (Hackett et al. \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Sarwari et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Context and Participants\u003c/h2\u003e \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1 Institutional Context\u003c/h2\u003e \u003cp\u003eThe study was conducted at [University], a public university in eastern China. The university enrolls international students from over 30 countries, the majority in degree programs requiring Chinese proficiency (HSK Level 4 for graduation). CFL instruction follows the International Curriculum for Chinese Language Education and is delivered through comprehensive Chinese, listening, and speaking courses.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003ch2\u003e3.2.2 Participants\u003c/h2\u003e \u003cp\u003eParticipants were recruited by intact class from four intermediate-level CFL classes during the spring semester of 2026. Two classes were assigned to the experimental group (n\u0026thinsp;=\u0026thinsp;32) and two to the control group (n\u0026thinsp;=\u0026thinsp;30). All participants met the following inclusion criteria: (a) had passed HSK Level 3 or equivalent, ensuring sufficient Chinese proficiency to participate in structured cultural dialogue tasks conducted primarily in Chinese with English as a permissible supplement; (b) had studied in China for at least one semester; and (c) had no prior formal training in prompt engineering or structured LLM use for language learning. The demographic profile of both groups is presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eParticipant demographics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCharacteristic\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExperimental (n\u0026thinsp;=\u0026thinsp;32)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eControl (n\u0026thinsp;=\u0026thinsp;30)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNationality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e17 countries\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e15 countries\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRegional distribution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCentral and West Asia (31.3%), Eastern Europe (21.9%), Africa (21.9%), East and Southeast Asia (18.8%), Other (6.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCentral and West Asia (30.0%), Eastern Europe (23.3%), Africa (20.0%), East and Southeast Asia (20.0%), Other (6.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFemale (65.6%), Male (34.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFemale (60.0%), Male (40.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eM\u0026thinsp;=\u0026thinsp;21.8, SD\u0026thinsp;=\u0026thinsp;2.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eM\u0026thinsp;=\u0026thinsp;22.1, SD\u0026thinsp;=\u0026thinsp;2.3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChinese proficiency\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHSK 3 (90.6%), HSK 4 (9.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHSK 3 (93.3%), HSK 4 (6.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDuration in China\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eM\u0026thinsp;=\u0026thinsp;2.1 semesters, SD\u0026thinsp;=\u0026thinsp;0.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eM\u0026thinsp;=\u0026thinsp;2.0 semesters, SD\u0026thinsp;=\u0026thinsp;0.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eBaseline equivalence on gender, age, Chinese proficiency, and duration in China was confirmed through independent-samples t-tests and chi-square tests (see Section \u003cspan refid=\"Sec31\" class=\"InternalRef\"\u003e4.2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eA priori power analysis (GPower 3.1; Faul et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2007\u003c/span\u003e) indicated that detecting a medium effect (d\u0026thinsp;=\u0026thinsp;0.50) for an independent-samples t-test (α\u0026thinsp;=\u0026thinsp;0.05, power\u0026thinsp;=\u0026thinsp;0.80) requires 26 per group. At the field-specific median between-group effect of d* = 0.70 (Plonsky and Oswald \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), our per-group sample of 30\u0026thinsp;+\u0026thinsp;yields power exceeding 85%. The 17-country participant diversity enhances ecological validity while limiting generalisability to any single cultural group.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.3 The LLM-Mediated Cultural Dialogue Intervention\u003c/h2\u003e \u003cdiv id=\"Sec14\" class=\"Section3\"\u003e \u003ch2\u003e3.3.1 Task Design Principles\u003c/h2\u003e \u003cp\u003eThe intervention was grounded in two complementary theoretical frameworks. Byram's (1997) ICC model guided the selection of intercultural themes and the operationalisation of learning objectives across four dimensions: knowledge, skills, attitudes, and critical cultural awareness. Kolb's (1984) experiential learning cycle, as revised by Morris (\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), provided the pedagogical structure for each task session. Each task followed a four-phase sequence: (1) Concrete experience: Learners engaged in LLM-mediated cultural dialogues, encountering culturally embedded communicative situations through role-play with the LLM. (2) Reflective observation: Learners discussed the dialogue experience in small groups, examining what they had learned, felt, and found challenging. (3) Abstract conceptualization: Through teacher-facilitated discussion, learners extracted cultural concepts, compared cross-cultural patterns, and constructed interpretive frameworks. (4) Active experimentation: Learners applied newly acquired cultural understanding and communicative strategies in subsequent dialogue tasks, initiating the next learning cycle.\u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates how the two theoretical frameworks were integrated into the intervention design. Additional design principles included: structured prompts specifying cultural contexts to mitigate the cultural-flattening effect of default prompting (Cao et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Yuan et al. \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e); an explicit bias-identification phase in each session following Jenks's (2024) call for critical AI literacy; and scenario-based communicative tasks consistent with the task-based approach advocated for ICC development (Thi \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003e3.3.2 Overview of the Six Cultural Dialogue Modules\u003c/h2\u003e \u003cp\u003eThe six weekly modules, their intercultural themes, primary ICC target dimensions, and corresponding LLM task descriptions are summarised in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCultural dialogue modules\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeek\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTheme\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eICC Dimensions\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLLM Dialogue Task\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSocial norms: First encounters\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eKnowledge, Attitudes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLLM plays a Chinese student; learner practices greetings, address forms, and small talk\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFood culture and hospitality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eKnowledge, Skills\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLLM plays a Chinese host; learner navigates dining etiquette and hospitality rituals\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIntercultural misunderstanding\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSkills, Attitudes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLLM simulates a cross-cultural friction scenario; learner identifies sources and negotiates resolution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDigital China\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eKnowledge, Critical Awareness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLearner discusses Chinese digital culture and critically examines LLM's cultural representations\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIdentity and cultural adaptation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAttitudes, Critical Awareness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLLM plays a cross-cultural counselor; dialogue explores culture shock and identity negotiation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIntegrative reflection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCritical Awareness (all)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGuided reflective dialogue reviewing the six-week journey, identifying remaining biases, and completing an intercultural growth reflection questionnaire\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eModules progress from knowledge-oriented (Weeks 1\u0026ndash;2) through skills/attitudes (Weeks 3\u0026ndash;4) to critical awareness (Weeks 5\u0026ndash;6), following the developmental trajectory endorsed in ICC scholarship (Deardorff \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2006\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section3\"\u003e \u003ch2\u003e3.3.3 Task Implementation Procedure\u003c/h2\u003e \u003cp\u003eEach weekly 90-minute session followed four standardised phases (see Supplementary Note S2 for a detailed sample lesson plan): (1) Activation (15 min): the instructor presented a cultural scenario; learners discussed in pairs. (2) LLM dialogue (30 min): learners individually engaged with the LLM following a task sheet specifying the scenario, cultural role, and guiding questions; interactions were primarily in Chinese calibrated to HSK Level 3, with English permitted as a supplementary language when learners encountered expressive difficulties; minimum 10 conversational turns. After each dialogue, learners submitted screenshots for task completion verification by the research assistant; full interaction transcripts were automatically generated and saved by Doubao's built-in ASR system, requiring no additional effort from learners. (3) Group discussion (30 min): in culturally mixed groups of 3\u0026ndash;4, learners discussed what they learned, cross-cultural similarities/differences, surprising or uncomfortable moments, and how they might act differently in real encounters. (4) Synthesis and bias check (15 min): the instructor led a whole-class discussion and projected 2\u0026ndash;3 LLM responses for collective identification of stereotyping, oversimplification, or factual inaccuracy, targeting Byram's (1997) critical cultural awareness dimension.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section3\"\u003e \u003ch2\u003e3.3.4 LLM Platform Selection and Prompt Design\u003c/h2\u003e \u003cp\u003ePlatform selection. This study adopted a single-platform design using Doubao (豆包), a Chinese-developed LLM by ByteDance. Selection was based on four criteria: (a) Communicative modality authenticity: Doubao's end-to-end real-time voice model (launched January 2025) supports ultra-low-latency natural spoken conversation, creating an interaction modality approximating face-to-face encounters, given that ICC's skill dimension (savoir faire) is fundamentally realized through oral interaction (Byram \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e1997\u003c/span\u003e); (b) Role-playing capability: its dedicated role-playing system supports cultural identity assignment and multi-turn persona consistency, and Tu et al.'s (\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) CharacterEval benchmark showed that Chinese LLMs outperformed GPT-4 in Chinese role-playing conversation; (c) Chinese cultural knowledge: Cao et al. (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) found that LLM cultural alignment varies by platform, and Chinese models significantly outperform international models on Chinese evaluation benchmarks such as C-Eval (Huang et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and CMMLU (Li, H. et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e); (d) Ecological validity: Doubao has over 170\u0026nbsp;million monthly active users (Xie et al. \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) and is freely available as a mobile app, integrating seamlessly into international students' daily digital environment.\u003c/p\u003e \u003cp\u003eThe selection of a Chinese-developed LLM over internationally dominant alternatives (e.g., GPT-4o) was grounded in ecological validity and construct alignment: Chinese-developed LLMs are trained on substantially larger proportions of Chinese-language corpora, enabling richer representation of Chinese cultural practices and pragmatic conventions (Huang et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Li, H. et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), while Doubao's NLP is natively optimized for Chinese tonal recognition, reducing technical friction for HSK Level 3 learners. As the most widely used LLM among Chinese-language users, Doubao directly reflects the AI tools CFL learners encounter in their daily lives in China. This approach, selecting a target-language-native AI platform, follows precedent in Kim and Su's (2024) adoption of a Korean-developed chatbot for Korean-as-a-foreign-language research.\u003c/p\u003e \u003cp\u003eA single-platform design was adopted because different LLMs embed different cultural perspectives (Dai et al. \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2025\u003c/span\u003e); standardising the platform isolates the pedagogy as the independent variable. While this design limits cross-platform generalisability, it ensures that observed effects can be attributed to the pedagogical intervention rather than to confounding differences in LLM architecture, training data, or cultural alignment. Generalisability limitations are discussed in Section \u003cspan refid=\"Sec47\" class=\"InternalRef\"\u003e5.3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe study used Doubao-Pro (2025 version), which supports both real-time voice dialogue and text dialogue, with a dedicated role-playing system enabling role specification, personality assignment, and context memory. The platform supports major world languages, enabling learners to clarify concepts in their L1.\u003c/p\u003e \u003cp\u003ePrompt design. Each task used a structured prompt template with four components: role specification (cultural identity assignment), language calibration (primarily Chinese at HSK Level 3, with English as a permissible supplement), cultural context framing (authentic practices rather than stereotypes), and interaction constraints (follow-up questions, culturally specific terms with explanations). A sample prompt is provided in Appendix A; complete prompt templates for all six weekly modules are available in Supplementary Note S1.\u003c/p\u003e \u003cp\u003eVoice data recording. Doubao's built-in ASR system generates real-time transcripts. Pre-intervention verification with 30 voice samples from five non-participant learners yielded character-level accuracy of 92.3% (SD\u0026thinsp;=\u0026thinsp;3.8%) for Chinese speech, rising to 95.1% within HSK Level 3 vocabulary. English-language segments were transcribed with comparable accuracy given Doubao's multilingual ASR capability. Segments with recognition errors were manually corrected during analysis. ASR-generated transcripts constitute the interaction logs cited in this study.\u003c/p\u003e \u003cp\u003eInteraction data presentation and management. Drawing on the multimodal data collection strategy employed by Yan and Zhang (\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) in their ChatGPT writing feedback study, experimental group learners were asked to complete a brief reflective learning journal after each weekly voice dialogue session, documenting the core topics discussed, notable cultural discoveries, and shifts in their own perspectives. Learners were encouraged to supplement their journals multimodally, including key screenshots of the dialogue interface, annotated highlights from ASR-generated transcripts, and short voice memos. This design ensured that the qualitative evidence obtained by the researchers consisted of representative interaction episodes self-selected by learners rather than unprocessed complete conversation records, reducing the complexity of data management while simultaneously capturing learners' self-interpretations of their cultural dialogue experiences through the act of purposeful selection. Findings are presented in the Results and Discussion sections through representative excerpts, each annotated with participant number, turn sequence, and week (e.g., Excerpt 1, P03, Week 3). The full interaction log dataset is available from the corresponding author upon reasonable request.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e \u003ch2\u003e3.3.5 Control Group and Implementation Fidelity\u003c/h2\u003e \u003cp\u003eThe control group received conventional CFL instruction covering comparable cultural content through traditional methods (lectures, readings, discussion, videos) without LLM use. Class hours (90 min/week), topic coverage, and instructor were held constant. All four classes were taught by the same instructor, the regular course teacher rather than a researcher. Researchers designed the intervention and collected data but did not deliver instruction. Implementation fidelity was ensured through: (a) standardised lesson plans; (b) two pre-intervention teacher training sessions; (c) blind classroom observation of two sessions per group; (d) independent learner completion of LLM dialogue phases.\u003c/p\u003e \u003cp\u003eExperimental group learners could use Doubao outside class without restrictions (ecological validity rationale), though this created potential exposure non-equivalence (discussed in Section \u003cspan refid=\"Sec47\" class=\"InternalRef\"\u003e5.3\u003c/span\u003e). Post-intervention interviews indicated extracurricular use ranging from none to 3\u0026ndash;4 times per week.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Instruments\u003c/h2\u003e \u003cp\u003eThree questionnaire instruments were used, all administered in bilingual Chinese\u0026ndash;English format. Internal consistency was assessed using Cronbach's α at both pretest and posttest, with a minimum acceptable threshold of α\u0026thinsp;\u0026ge;\u0026thinsp;0.70 (Nunnally \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e1978\u003c/span\u003e).\u003c/p\u003e \u003cdiv id=\"Sec20\" class=\"Section3\"\u003e \u003ch2\u003e3.4.1 Intercultural Sensitivity Scale (ISS)\u003c/h2\u003e \u003cp\u003eThe ISS (Chen and Starosta \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2000\u003c/span\u003e) measured the affective dimension of ICC. It comprises 24 items on a 5-point Likert scale (1\u0026thinsp;=\u0026thinsp;strongly disagree to 5\u0026thinsp;=\u0026thinsp;strongly agree), measuring five factors: Interaction Engagement (7 items), Respect for Cultural Differences (6 items), Interaction Confidence (5 items), Interaction Enjoyment (3 items), and Interaction Attentiveness (3 items). Nine negatively worded items were reverse-coded prior to analysis (complete items are provided in Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). The ISS is among the most widely used instruments in intercultural communication research and has been adopted by multiple JCR Q1 studies including Hackett et al. (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), with well-established cross-cultural reliability and validity evidence.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e \u003ch2\u003e3.4.2 Cultural Intelligence Scale (CQS)\u003c/h2\u003e \u003cp\u003eThe CQS (Ang et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2007\u003c/span\u003e) measured the cognitive and behavioural dimensions of ICC. It comprises 20 items on a 7-point Likert scale (1\u0026thinsp;=\u0026thinsp;strongly disagree to 7\u0026thinsp;=\u0026thinsp;strongly agree), measuring four dimensions: Metacognitive CQ (4 items, assessing cultural awareness and strategic cultural thinking), Cognitive CQ (6 items, assessing cultural knowledge), Motivational CQ (5 items, assessing intrinsic interest and self-efficacy in intercultural interaction), and Behavioral CQ (5 items, assessing behavioural adaptation in intercultural contexts) (complete items are provided in Supplementary Table S2).\u003c/p\u003e \u003cp\u003eThe CQS has been validated in 98 countries (Ang et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2007\u003c/span\u003e) and was used as a core measure by Hackett et al. (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) in their study of technology-mediated intercultural learning. Its four dimensions map onto Byram's (1997) ICC model: Metacognitive CQ corresponds to critical cultural awareness (savoir s'engager), capturing reflective cultural thinking and awareness monitoring; Cognitive CQ corresponds to knowledge (savoirs), assessing cultural facts and norms; Motivational CQ corresponds to attitudes (savoir \u0026ecirc;tre), measuring interest in and openness to intercultural interaction; and Behavioral CQ corresponds to skills (savoir faire), assessing behavioural adaptation in intercultural contexts. This mapping is a theoretical approximation. Metacognitive CQ captures the reflective-monitoring dimension of savoir s'engager, specifically awareness of one's own cultural assumptions and strategic adjustment of cultural thinking, but Byram's (1997) broader definition also encompasses political and ideological critique beyond the CQS subscale's scope. Results for Metacognitive CQ are therefore interpreted as capturing the reflective-monitoring component of critical cultural awareness, not its full breadth.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section3\"\u003e \u003ch2\u003e3.4.3 LLM Cultural Learning Perception Questionnaire (LCPQ)\u003c/h2\u003e \u003cp\u003eA self-developed LCPQ was administered post-intervention to the experimental group only, assessing learners' perceptions and experiences of LLMs as intercultural learning tools. The questionnaire comprised two parts:\u003c/p\u003e \u003cp\u003ePart 1: Likert scale items (25 items, 5-point scale) measuring five dimensions (5 items each): perceived cultural learning effectiveness (sample item: \"LLM cultural dialogues deepened my understanding of Chinese cultural customs\"), LLM dialogue partner quality (\"The LLM could play a credible Chinese cultural role\"), critical AI literacy development (\"I can identify cultural stereotypes or oversimplifications in LLM responses\"), intercultural reflection facilitation (\"Dialogues with the LLM prompted me to reflect on differences between my culture and Chinese culture\"), and continued use intention (\"I would continue using LLMs for cultural learning after the course ends\").\u003c/p\u003e \u003cp\u003ePart 2: Open-ended questions (4 items) eliciting qualitative data on the most impactful dialogue experience, differences between AI and human cultural discussion, LLM bias encounters, and recommendations for other learners.\u003c/p\u003e \u003cp\u003eGiven the small sample (n\u0026thinsp;=\u0026thinsp;32), structural validity evidence is limited to content validity (expert review, I-CVI\u0026thinsp;\u0026ge;\u0026thinsp;0.78; Lynn \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e1986\u003c/span\u003e) and internal consistency (see Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Preliminary discriminant validity evidence (inter-dimension Pearson correlations r\u0026thinsp;=\u0026thinsp;0.35\u0026ndash;0.55, all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05; see Supplementary Table S4) indicated related but distinct constructs. The LCPQ was developed through initial item generation (35 items), expert review, pilot testing with 15 non-participant CFL learners (corrected item-total correlation\u0026thinsp;\u0026ge;\u0026thinsp;0.30), yielding the final 25 items (all subscale α\u0026thinsp;\u0026ge;\u0026thinsp;0.70; see Supplementary Table S3 for complete items).\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Data Collection\u003c/h2\u003e \u003cp\u003eThe ISS and CQS were administered to both groups at pretest (Week 0) and posttest (Week 7); the LCPQ was administered to the experimental group only at posttest. All questionnaires were administered online via Wenjuanxing. Within one week after posttest (Week 8), five experimental group learners selected through maximum variation sampling were interviewed individually (see Section \u003cspan refid=\"Sec26\" class=\"InternalRef\"\u003e3.6.2\u003c/span\u003e). Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e presents the complete research procedure and data collection timeline.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003e3.6 Data Analysis\u003c/h2\u003e \u003cdiv id=\"Sec25\" class=\"Section3\"\u003e \u003ch2\u003e3.6.1 Quantitative Analysis (RQ1 and RQ2)\u003c/h2\u003e \u003cp\u003eQuantitative data were analysed using SPSS 28. After computing descriptive statistics and confirming normality (Shapiro\u0026ndash;Wilk), baseline equivalence was tested using independent-samples t-tests and chi-square tests. One-way ANCOVA with pretest scores as covariates tested between-group posttest differences (RQ1), with Cohen's d, partial η\u0026sup2;, and 95% CIs reported. Dimension-level ANCOVAs with Bonferroni correction examined differential responsiveness across ICC dimensions (RQ2). Paired-samples t-tests assessed within-group pre-post changes. LCPQ Likert items were analysed descriptively.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec26\" class=\"Section3\"\u003e \u003ch2\u003e3.6.2 Qualitative Analysis (RQ3)\u003c/h2\u003e \u003cp\u003eLCPQ open-ended responses and semi-structured interview transcripts were analysed using Braun and Clarke's (2006) six-phase thematic analysis, combining deductive coding (Byram's four ICC dimensions; LLM affordances/limitations) with inductive coding for emergent themes. A second coder independently coded 30% of responses (κ\u0026thinsp;\u0026ge;\u0026thinsp;0.75 required; Landis and Koch \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e1977\u003c/span\u003e). Learners responded in their preferred language; Chinese responses were back-translated for verification.\u003c/p\u003e \u003cp\u003eFive interviewees were selected through maximum variation sampling (Patton \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2015\u003c/span\u003e), maximizing diversity in CQS change magnitude, cultural background, gender, and LCPQ partner quality ratings. Interviews lasted 25\u0026ndash;40 minutes and were member-checked. Case profiles integrating each participant's quantitative trajectories and qualitative narratives provided individual-level supplements to group-level findings.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section2\"\u003e \u003ch2\u003e3.7 Trustworthiness and Rigor\u003c/h2\u003e \u003cp\u003e Research quality was ensured through: control group design and ANCOVA controlling for maturation and pretest differences (internal validity); established scales with cross-cultural validation evidence (construct validity); a priori power analysis, effect sizes with CIs, and Bonferroni correction (statistical conclusion validity); 17-country participant diversity with thick description (external validity); and inter-coder reliability, data triangulation, member checking, and disconfirming evidence (qualitative credibility).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec28\" class=\"Section2\"\u003e \u003ch2\u003e3.8 Ethical Considerations\u003c/h2\u003e \u003cp\u003e This study involved routine classroom pedagogical activities and did not require formal ethics committee approval under institutional guidelines for classroom-based educational research. Protocols included: bilingual informed consent with withdrawal rights; data anonymisation with encrypted storage; multicultural advisory review of cultural topics; equitable post-study access to intervention materials for the control group; and AI ethics disclosure informing participants that LLM outputs may contain biases, with bias identification framed as a learning objective, aligning the ethical requirement with the study's problem-as-resource pedagogy.\u003c/p\u003e \u003c/div\u003e"},{"header":"4 Results","content":"\u003cp\u003eThis section presents scale reliability, baseline equivalence, descriptive statistics, and results for the three research questions.\u003c/p\u003e \u003cdiv id=\"Sec30\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Scale Reliability\u003c/h2\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, all coefficients exceeded the minimum threshold of α\u0026thinsp;\u0026ge;\u0026thinsp;0.70 (Nunnally \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e1978\u003c/span\u003e). ISS total reliability improved from 0.89 (pretest) to 0.91 (posttest); CQS from 0.91 to 0.93, consistent with reliability levels reported in its 98-country validation (Ang et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). The LCPQ total α reached 0.92, with subscale values ranging from 0.81 to 0.88.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eInternal consistency coefficients (Cronbach's α)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eScale/Subscale\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePretest α\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePosttest α\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eISS Total\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Engagement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRespect for Cultural Differences\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Confidence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Enjoyment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.80\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Attentiveness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.77\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCQS Total\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetacognitive CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCognitive CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMotivational CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBehavioral CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLCPQ Total\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCultural learning effectiveness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDialogue partner quality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCritical AI literacy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntercultural reflection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eContinued use intention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec31\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Baseline Equivalence\u003c/h2\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, no significant differences were found on any demographic or outcome variable, confirming baseline equivalence despite intact-class assignment and satisfying ANCOVA prerequisites.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBaseline equivalence tests\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExperimental (n\u0026thinsp;=\u0026thinsp;32)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eControl (n\u0026thinsp;=\u0026thinsp;30)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTest Statistic\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (M\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e21.8\u0026thinsp;\u0026plusmn;\u0026thinsp;2.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22.1\u0026thinsp;\u0026plusmn;\u0026thinsp;2.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003et(60)\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.624\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender (% female)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e65.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e60.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eχ\u0026sup2;(1)\u0026thinsp;=\u0026thinsp;0.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.648\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHSK Level 3 (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e90.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e93.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFisher's exact\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDuration in China (M\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.1\u0026thinsp;\u0026plusmn;\u0026thinsp;0.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003et(60)\u0026thinsp;=\u0026thinsp;0.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.649\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eISS pretest total (M\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e83.50\u0026thinsp;\u0026plusmn;\u0026thinsp;10.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e80.27\u0026thinsp;\u0026plusmn;\u0026thinsp;9.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003et(60)\u0026thinsp;=\u0026thinsp;1.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.221\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCQS pretest mean (M\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.15\u0026thinsp;\u0026plusmn;\u0026thinsp;0.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.08\u0026thinsp;\u0026plusmn;\u0026thinsp;0.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003et(60)\u0026thinsp;=\u0026thinsp;0.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.586\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec32\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Descriptive Statistics\u003c/h2\u003e \u003cp\u003eAs reported in Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, the experimental group showed substantially larger gains on most measures.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eISS and CQS descriptive statistics by group\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMeasure\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eExperimental (n\u0026thinsp;=\u0026thinsp;32)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003eControl (n\u0026thinsp;=\u0026thinsp;30)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePre M(SD)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePost M(SD)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eΔ\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePre M(SD)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePost M(SD)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eΔ\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eISS Total\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e83.50(10.82)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e91.53(9.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;8.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e80.27(9.45)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e84.20(9.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e+\u0026thinsp;3.93\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Engagement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.52(0.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.95(0.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.28(0.51)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.42(0.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e+\u0026thinsp;0.14\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRespect for Cultural Diff.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.81(0.48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.18(0.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.85(0.45)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.95(0.44)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e+\u0026thinsp;0.10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Confidence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.94(0.62)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.40(0.55)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.88(0.55)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.05(0.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e+\u0026thinsp;0.17\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Enjoyment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.38(0.55)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.80(0.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.15(0.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.38(0.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e+\u0026thinsp;0.23\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Attentiveness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.62(0.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.80(0.46)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.72(0.46)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.80(0.46)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e+\u0026thinsp;0.08\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCQS Mean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.15(0.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.72(0.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.08(0.48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.30(0.48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e+\u0026thinsp;0.22\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetacognitive CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.73(0.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.58(0.62)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.10(0.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.28(0.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e+\u0026thinsp;0.18\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCognitive CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.88(0.68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.38(0.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.92(0.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.22(0.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e+\u0026thinsp;0.30\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMotivational CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.74(0.62)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.12(0.55)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.38(0.55)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.55(0.55)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e+\u0026thinsp;0.17\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBehavioral CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.81(0.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.18(0.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.85(0.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2.96(0.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e+\u0026thinsp;0.11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe most notable dimension-level change was Metacognitive CQ (+\u0026thinsp;0.85 experimental vs. +0.18 control). Cognitive CQ showed a more moderate between-group difference, with the control group also demonstrating meaningful gains (+\u0026thinsp;0.30), consistent with the capacity of conventional instruction to transmit cultural knowledge. Shapiro\u0026ndash;Wilk tests confirmed approximate normality for all variables in both groups (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec33\" class=\"Section2\"\u003e \u003ch2\u003e4.4 RQ1: Overall Impact of LLM-Mediated Cultural Dialogue on ICC\u003c/h2\u003e \u003cdiv id=\"Sec34\" class=\"Section3\"\u003e \u003ch2\u003e4.4.1 Within-Group Pre-Post Comparisons\u003c/h2\u003e \u003cp\u003ePaired-samples t-tests revealed that the experimental group showed highly significant gains on all measures (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), with effect sizes from d\u0026thinsp;=\u0026thinsp;0.70 to 1.25 (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePaired-samples t-test results\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMeasure\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eExperimental\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003et(31)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ed\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003et(29)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ed\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eISS Total\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.46\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCQS Mean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.037\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetacognitive CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.181\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCognitive CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.011\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.50\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMotivational CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.144\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.27\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBehavioral CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.284\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe control group reached significance on ISS Total (p\u0026thinsp;=\u0026thinsp;0.017, d\u0026thinsp;=\u0026thinsp;0.46), CQS Mean (p\u0026thinsp;=\u0026thinsp;0.037, d\u0026thinsp;=\u0026thinsp;0.40), and on Cognitive CQ in particular (p\u0026thinsp;=\u0026thinsp;0.011, d\u0026thinsp;=\u0026thinsp;0.50), indicating that conventional instruction can transmit cultural knowledge effectively but has minimal impact on metacognitive, motivational, and behavioural dimensions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec35\" class=\"Section3\"\u003e \u003ch2\u003e4.4.2 ANCOVA Between-Group Comparisons\u003c/h2\u003e \u003cp\u003eAfter controlling for pretest scores through one-way ANCOVA, between-group differences were significant for both ISS (p\u0026thinsp;=\u0026thinsp;0.002) and CQS (p\u0026thinsp;=\u0026thinsp;0.002), as shown in Table\u0026nbsp;\u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eANCOVA results: between-group posttest comparisons (pretest as covariate)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAdj. M (Exp.)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAdj. M (Ctrl.)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eF(1, 59)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePartial η\u0026sup2;\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eCohen's d [95% CI]\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eISS Total\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e90.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e84.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e10.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.149\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.82 [0.30, 1.34]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCQS Mean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e10.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.155\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.83 [0.31, 1.35]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003ePartial η\u0026sup2; values of 0.149 and 0.155 both reached Cohen's (1988) large-effect benchmark (0.14), with positive lower bounds of the 95% CI for d confirming reliable intervention effects. LLM-mediated cultural dialogue significantly outperformed conventional instruction in promoting CFL learners' intercultural sensitivity and cultural intelligence.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec36\" class=\"Section2\"\u003e \u003ch2\u003e4.5 RQ2: Differential Response Across ICC Dimensions\u003c/h2\u003e \u003cdiv id=\"Sec37\" class=\"Section3\"\u003e \u003ch2\u003e4.5.1 ISS Five-Dimension ANCOVA\u003c/h2\u003e \u003cp\u003eDimension-level ANCOVAs with Bonferroni correction revealed that four of the five ISS subscales survived the corrected threshold (Table\u0026nbsp;\u003cspan refid=\"Tab8\" class=\"InternalRef\"\u003e8\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab8\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 8\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eISS subscale ANCOVA results (Bonferroni-corrected α' = 0.01)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eISS Dimension\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eF(1, 59)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePartial η\u0026sup2;\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCohen's d\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSignificant\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Engagement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.159\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRespect for Cultural Diff.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e8.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.005\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.128\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Confidence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.008\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.112\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Enjoyment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.007\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.117\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction Attentiveness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.045\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.066\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eInteraction Engagement showed the largest effect (η\u0026sup2; = 0.159, d\u0026thinsp;=\u0026thinsp;0.82), consistent with the task requirement for learners to actively initiate and sustain conversational turns. Interaction Attentiveness (d\u0026thinsp;=\u0026thinsp;0.50) did not reach the corrected threshold, likely because habitual attentional tendencies require longer periods to change substantially.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec38\" class=\"Section3\"\u003e \u003ch2\u003e4.5.2 CQS Four-Dimension ANCOVA\u003c/h2\u003e \u003cp\u003eThe same procedure applied to the four CQS dimensions showed that three of four remained significant after Bonferroni correction (Table\u0026nbsp;\u003cspan refid=\"Tab9\" class=\"InternalRef\"\u003e9\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab9\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 9\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCQS subscale ANCOVA results (Bonferroni-corrected α' = 0.0125)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCQS Dimension\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eF(1, 59)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePartial η\u0026sup2;\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCohen's d\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSignificant\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetacognitive CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e18.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.234\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCognitive CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.031\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.076\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMotivational CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e10.42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.150\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBehavioral CQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.010\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.108\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eMetacognitive CQ yielded the largest single effect in the entire study (η\u0026sup2; = 0.234, d\u0026thinsp;=\u0026thinsp;1.08), corresponding to the reflective-monitoring dimension of Byram's (1997) critical cultural awareness, historically the most resistant to change through conventional instruction. Cognitive CQ showed the smallest effect (d\u0026thinsp;=\u0026thinsp;0.52) and did not survive Bonferroni correction, consistent with the finding that the control group also showed significant within-group gains on Cognitive CQ (p\u0026thinsp;=\u0026thinsp;0.011, d\u0026thinsp;=\u0026thinsp;0.50; Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). This pattern indicates that cultural knowledge can be effectively transmitted through conventional instruction, and the LLM intervention's distinctive contribution lies elsewhere, specifically in metacognitive, attitudinal, and behavioural dimensions that lectures and readings alone struggle to develop.\u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e visualises the effect size ranking across all nine dimensions. The top three dimensions (d\u0026thinsp;\u0026ge;\u0026thinsp;0.80), corresponding primarily to critical cultural awareness and attitudes, reached large-effect levels. Dimensions ranked fourth through seventh (d\u0026thinsp;=\u0026thinsp;0.65\u0026ndash;0.72) represented medium-to-large effects across attitudes and skills. The bottom two dimensions, Cognitive CQ (d\u0026thinsp;=\u0026thinsp;0.52) and Interaction Attentiveness (d\u0026thinsp;=\u0026thinsp;0.50), did not survive Bonferroni correction, revealing a theoretically meaningful pattern: the knowledge dimension, where conventional instruction is also effective, showed the weakest between-group differentiation. This pattern is consistent with Deardorff's (2006) ICC process model, in which attitudes and reflective capacities serve as the \"inner ring\" that drives subsequent knowledge and skill development.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eOverall, RQ2 reveals a non-uniform distribution of intervention effects: the LLM intervention's advantages were most pronounced in metacognitive and attitudinal dimensions that conventional instruction struggles to address, and substantially smaller in the knowledge dimension that conventional teaching can partially cover through lectures and readings.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec39\" class=\"Section2\"\u003e \u003ch2\u003e4.6 RQ3: Learner Perceptions of LLM Cultural Learning\u003c/h2\u003e \u003cdiv id=\"Sec40\" class=\"Section3\"\u003e \u003ch2\u003e4.6.1 LCPQ Descriptive Statistics\u003c/h2\u003e \u003cp\u003eTo examine how learners perceived the LLM-mediated cultural dialogue experience, descriptive statistics for the five LCPQ dimensions were computed (Table\u0026nbsp;\u003cspan refid=\"Tab10\" class=\"InternalRef\"\u003e10\u003c/span\u003e). The overall mean (M\u0026thinsp;=\u0026thinsp;4.08) indicated a positive evaluation.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab10\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 10\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eLCPQ descriptive statistics (n\u0026thinsp;=\u0026thinsp;32)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLCPQ Dimension\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eM\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMin\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMax\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eRating Level\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCultural learning effectiveness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e5.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eContinued use intention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e5.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntercultural reflection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCritical AI literacy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e5.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMedium-High\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDialogue partner quality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMedium\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLCPQ Total Mean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eCultural learning effectiveness (M\u0026thinsp;=\u0026thinsp;4.38) and continued use intention (M\u0026thinsp;=\u0026thinsp;4.42) were highest, indicating that learners perceived substantial learning gains and were willing to continue using LLMs. Dialogue partner quality (M\u0026thinsp;=\u0026thinsp;3.45) was considerably lower; learners acknowledged the LLM's knowledge breadth and patience but questioned its emotional authenticity and communicative naturalness. Critical AI literacy showed the highest individual variation (SD\u0026thinsp;=\u0026thinsp;0.72), suggesting uneven development of critical awareness across learners.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec41\" class=\"Section3\"\u003e \u003ch2\u003e4.6.2 Preliminary Observations on Individual Differences\u003c/h2\u003e \u003cp\u003eCase profile examination suggested two potentially influential factors, though these remain exploratory given sample size constraints.\u003c/p\u003e \u003cp\u003ePersonality and partner quality ratings. Self-reported extraverted learners (e.g., EXP-029, EXP-031, EXP-023) tended to rate AI partner quality lower (M\u0026thinsp;=\u0026thinsp;3.07) than more introverted learners (e.g., EXP-011, EXP-005, M\u0026thinsp;=\u0026thinsp;3.95), possibly because extraverts value emotional reciprocity and spontaneous humor, while introverts appreciate the judgment-free practice environment.\u003c/p\u003e \u003cp\u003eCultural knowledge and critical AI literacy. Learners with stronger cultural knowledge and personal experiences of being misrepresented by AI (e.g., EXP-008 critical AI literacy\u0026thinsp;=\u0026thinsp;5.00, EXP-005\u0026thinsp;=\u0026thinsp;4.80) scored higher than those without such experiences (e.g., EXP-031\u0026thinsp;=\u0026thinsp;3.20, EXP-023\u0026thinsp;=\u0026thinsp;3.40), suggesting that critical AI literacy development may depend on both a knowledge foundation and triggering events.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec42\" class=\"Section3\"\u003e \u003ch2\u003e4.6.3 Thematic Analysis of Open-Ended Responses\u003c/h2\u003e \u003cp\u003eThematic analysis of 32 LCPQ open-ended responses and 5 semi-structured interview transcripts (inter-coder reliability κ\u0026thinsp;=\u0026thinsp;0.82) yielded four core themes (see Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e for the thematic map).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTheme 1: LLM as a \"pressure-free intercultural practice space\" (30/32, 93.8%). Nearly all learners noted that the LLM provided a mistake-tolerant environment. Sub-themes included judgment-free language practice (\"Talking to AI isn't stressful \u0026mdash; it won't laugh at my mistakes,\" EXP-011), adaptive difficulty calibration (\"AI can explain complex cultural concepts in simple words,\" EXP-031), and freedom to revisit topics without social embarrassment.\u003c/p\u003e \u003cp\u003eTheme 2: Experiential cultural knowledge acquisition (28/32, 87.5%). Learners reported gaining situated, concrete cultural knowledge through dialogue that contrasted sharply with textbook-style knowledge. EXP-005 noted: \"The textbook says Chinese people value mianzi. But only through AI dialogue did I understand how mianzi works in specific situations.\" Multiple learners also reported that LLM dialogue prompted re-examination of their own cultural practices.\u003c/p\u003e \u003cp\u003eTheme 3: Differences between AI and human interaction quality (32/32, 100%). All learners spontaneously compared AI and human interactions, adopting a \"complementary rather than substitutive\" cognitive frame. Extraverted learners particularly noted the AI's lack of humor and emotional responsiveness (\"Talking to AI lacks one thing \u0026mdash; fun,\" EXP-031). Most positioned the LLM as a preparation tool for learning basic cultural knowledge and strategies through AI before applying them in real intercultural encounters.\u003c/p\u003e \u003cp\u003eTheme 4: LLM cultural bias identification and response (25/32, 78.1%). Nearly 80% of learners reported encountering at least one instance of LLM cultural bias. Sub-themes included stereotypical representations of learners' home cultures (EXP-031 objecting to \"taco and sombrero\" as first associations with Mexico; EXP-008 noting AI's treatment of \"Chinese people\" as a monolithic group ignoring Southeast Asian Chinese diversity), idealized representations of Chinese culture (\"AI's version of Chinese culture is too perfect \u0026mdash; different from what I actually see in China,\" EXP-011), and differentiated response strategies ranging from low-level accept and note through mid-level correct and probe to high-level critical analysis of training data composition and cultural power relations (mainly from learners with stronger cultural knowledge, e.g., EXP-008, EXP-005).\u003c/p\u003e \u003cp\u003eInterview data corroborated these themes and revealed three additional findings: (a) learners unanimously emphasized the voice modality's contribution to communicative authenticity and real-time processing practice; (b) three interviewees reported substantial extracurricular self-directed use of Doubao for cultural exploration; (c) two high-scoring learners could pinpoint specific \"awakening moments\" when their critical awareness shifted, triggered by encountering personally relevant cultural misrepresentations.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec43\" class=\"Section3\"\u003e \u003ch2\u003e4.6.4 Triangulation of Quantitative and Qualitative Findings\u003c/h2\u003e \u003cp\u003eQualitative themes corroborated quantitative results. The largest effect size on Metacognitive CQ (quantitative) aligned with the high frequency of bias identification in Theme 4 (qualitative), pointing to a shared mechanism: repeated bias identification and discussion trained learners' metacognitive reflective capacity. The lowest LCPQ rating on dialogue partner quality (quantitative) corresponded to Theme 3's consistent AI\u0026ndash;human comparison narratives (qualitative). The high individual variance on critical AI literacy (SD\u0026thinsp;=\u0026thinsp;0.72, quantitative) was mirrored by the stratified response strategies in Theme 4c (qualitative), suggesting that critical awareness development may be linked to learners' cultural knowledge base and personal experiences.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"5 Discussion","content":"\u003cdiv id=\"Sec45\" class=\"Section2\"\u003e \u003ch2\u003e5.1 Why LLM-Mediated Cultural Dialogue Works: Contributing Factors (RQ1 and RQ2)\u003c/h2\u003e \u003cp\u003eThe significant between-group effects on both ISS and CQS corroborate Liu, J.'s (2025) finding that AI-enhanced learning can promote ICC development, and extend it to a generative LLM dialogue paradigm. We identify three factors that likely contributed to the intervention's effectiveness.\u003c/p\u003e \u003cp\u003eFirst, the LLM's role as a cultural interlocutor rather than a learning scaffold distinguishes this intervention from prior work (e.g., Wang et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). By positioning the LLM as a simulated cultural other with whom learners negotiate intercultural meaning in real time, the intervention activated experiential learning processes (Kolb \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e1984\u003c/span\u003e) that lectures and readings cannot replicate. The effect sizes (d\u0026thinsp;=\u0026thinsp;0.82\u0026ndash;0.83) exceeded the median in L2 intervention research (d\u0026thinsp;=\u0026thinsp;0.70; Plonsky and Oswald \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), suggesting that dialogic engagement with a culturally embedded AI partner provides qualitatively different learning affordances.\u003c/p\u003e \u003cp\u003eSecond, the problem-as-resource strategy transformed LLM cultural biases into metacognitive triggers. When learners encountered stereotypical representations of their home cultures or idealised portrayals of Chinese culture, they were compelled to ask \"Is this fact or bias?\", operations at the core of metacognitive CQ. This mechanism, combined with institutionalised reflection support (group discussion and bias checking), explains why Metacognitive CQ showed the largest effect while Cognitive CQ showed the smallest: the intervention's distinctive value lies not in knowledge transmission (which conventional instruction also achieves) but in cultivating reflective capacities that traditional methods struggle to develop. This is consistent with Jenks's (2024) argument that LLM biases can serve as cognitive scaffolds for critical cultural awareness. A methodological caveat is warranted: Metacognitive CQ captures the reflective-monitoring subcomponent of Byram's (1997) savoir s'engager but not its full scope, which also includes political and ideological critique.\u003c/p\u003e \u003cp\u003eThird, the voice modality restored authentic communicative pressure. Unlike text-based chatbot interactions that eliminate real-time processing demands (Young \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2011\u003c/span\u003e), Doubao's voice dialogue required learners to manage turn-taking, process spoken Chinese under time pressure, and respond without the luxury of editing, conditions approximating face-to-face intercultural encounters. Interview data consistently highlighted this feature as a key contributor to communicative authenticity, likely explaining the strong effects on Interaction Engagement and Interaction Confidence. Nevertheless, Behavioral CQ remained the lowest CQS dimension, confirming that simulated dialogue cannot fully substitute for the behavioural demands of real intercultural encounters; adjusting speech rate, managing silence, and deploying non-verbal cues remain beyond LLM-mediated practice. Learners consistently described the LLM as preparation for, rather than replacement of, human interaction.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec46\" class=\"Section2\"\u003e \u003ch2\u003e5.2 Who Benefits and How: Implications for Educators, Learners, and Researchers (RQ3)\u003c/h2\u003e \u003cp\u003eThe LCPQ data and qualitative findings reveal that the intervention benefits different learner profiles in different ways, carrying specific implications for three stakeholder groups.\u003c/p\u003e \u003cp\u003eFor educators, the results demonstrate that LLMs are most effective as intercultural preparation tools paired with authentic interaction opportunities. The low dialogue partner quality rating (M\u0026thinsp;=\u0026thinsp;3.45) confirms that LLMs cannot replace human cultural exchange; rather, they provide a low-stakes space for learners to acquire cultural strategies before applying them in real encounters. It is essential that bias-check activities not be omitted, as the high individual variation in critical AI literacy (SD\u0026thinsp;=\u0026thinsp;0.72) indicates that critical awareness does not grow spontaneously with LLM use, and learners lacking cultural knowledge or personal experience of AI misrepresentation need explicit teacher guidance. The LLM's safety alignment mechanism also constrained intercultural simulation on sensitive topics, requiring instructors to convert evasive AI behaviour into discussion material.\u003c/p\u003e \u003cp\u003eFor learners, personality and prior knowledge moderated the experience. Introverted learners valued the judgment-free environment and rated partner quality higher; extraverted learners were more dissatisfied with the AI's lack of emotional reciprocity. This suggests that LLM-mediated cultural dialogue may be particularly valuable for learners who experience anxiety in face-to-face intercultural encounters, a population that conventional interaction-based approaches may underserve. Additionally, learners with stronger cultural knowledge developed critical AI literacy more effectively, suggesting that the intervention works best when learners have a baseline cultural foundation to evaluate AI outputs against. Educators should consider sequencing: building foundational cultural knowledge before introducing LLM-mediated bias-identification tasks.\u003c/p\u003e \u003cp\u003eFor researchers, the non-uniform dimension-level pattern raises important methodological considerations. The intervention comprised multiple co-occurring components (individualised AI dialogue, voice modality, adaptive feedback, and structured reflection) that the present design cannot fully disentangle. A purely practice-quantity or novelty explanation is unlikely given the pronounced differentiation (Metacognitive CQ d\u0026thinsp;=\u0026thinsp;1.08 vs. Cognitive CQ d\u0026thinsp;=\u0026thinsp;0.52), but factorial designs (e.g., LLM voice vs. text vs. human partner vs. control) are needed to isolate each component's contribution. The problem-as-resource framework demonstrated here may also be applicable beyond CFL contexts; any LLM-mediated educational setting where AI outputs contain systematic biases could benefit from converting those biases into critical thinking opportunities. Future studies should incorporate objective measures (e.g., role-play assessments, discourse analysis), active placebo controls, and delayed posttests to assess sustainability.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec47\" class=\"Section2\"\u003e \u003ch2\u003e5.3 Limitations\u003c/h2\u003e \u003cp\u003eThe sample of 62 participants from a single university, predominantly at HSK Level 3 and using a single LLM platform (Doubao), limits generalisability. Both instruments are self-report measures susceptible to social desirability bias and Hawthorne effects, though the non-uniform results and the control group's modest gains partially mitigate these concerns. Finally, six weeks may be insufficient for full Behavioral CQ development, which requires sustained real-world practice.\u003c/p\u003e \u003c/div\u003e"},{"header":"6 Conclusion","content":"\u003cp\u003eThis quasi-experimental study provides empirical evidence that LLM-mediated cultural dialogue can promote ICC development in CFL learners, with medium-to-large effects on both intercultural sensitivity and cultural intelligence. The non-uniform dimension-level pattern, where Metacognitive CQ showed the largest effect while Cognitive CQ showed the smallest, indicates that the intervention's distinctive value lies in cultivating reflective capacities rather than transmitting cultural knowledge. The core contribution is a problem-as-resource strategy that transforms LLM cultural biases into pedagogical material for critical cultural awareness, extending Byram's (1997) ICC pedagogy into AI-mediated language education.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting Interests\u003c/h2\u003e \u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003e[Removed for double-blind review.]\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eSun Jing: Conceptualization, Methodology, Investigation, Formal analysis, Writing \u0026ndash; original draft. Nie Liming: Supervision, Validation, Writing \u0026ndash; review \u0026amp; editing.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThis research was supported by the National Natural Science Foundation of China (Grant No. W2412110), the Shenzhen Peacock Plan (Grant No. GDRC202515), and the Shenzhen Technology University Teaching Reform Project (Grant No. 20251016).\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets generated by the survey research during and/or analysed during the current study are available in the Google repository(anonymous):https://docs.google.com/spreadsheets/d/1a2C5QdXQ5PWBUiaeDk48zW8MY6OcAHRP/edit?usp=sharing\u0026amp;amp;ouid=103666323466924573598\u0026amp;amp;rtpof=true\u0026amp;amp;sd=true\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was reviewed and approved by the Academic Committee of the ****[Removed for double-blind review]. This committee oversees research involving human participants conducted within the school and ensures compliance with institutional ethical standards.\u003c/p\u003e\n\u003cp\u003eAll research procedures were performed in accordance with the ethical standards of the Academic committee, the 1964 Declaration of Helsinki and its later amendments. The study involved routine classroom pedagogical activities within an established curriculum and did not constitute biomedical or clinical research.\u003c/p\u003e\n\u003cp\u003eWritten informed consent was obtained from all participants prior to data collection. Participants were informed of the study\u0026apos;s purpose, procedures, the voluntary nature of participation, and their right to withdraw at any time without penalty. All data were anonymized prior to analysis.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAng S, Van Dyne L, Koh C, Ng KY, Templer KJ, Tay C, Chandrasekar NA (2007) Cultural intelligence: its measurement and effects on cultural judgment and decision making, cultural adaptation and task performance. Manage Organ Rev 3(3):335\u0026ndash;371. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.1740-8784.2007.00082.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1740-8784.2007.00082.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBraun V, Clarke V (2006) Using thematic analysis in psychology. Qualitative Res Psychol 3(2):77\u0026ndash;101. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1191/1478088706qp063oa\u003c/span\u003e\u003cspan address=\"10.1191/1478088706qp063oa\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eByram M (1997) Teaching and assessing intercultural communicative competence. Multilingual Matters\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCampbell DT, Stanley JC (1963) Experimental and quasi-experimental designs for research. Houghton Mifflin\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCao Y, Zhou L, Lee S, Cabello L, Chen M, Hershcovich D (2023) Assessing cross-cultural alignment between ChatGPT and human societies: an empirical study. In: Proceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP), pp 53\u0026ndash;67. Association for Computational Linguistics. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18653/v1/2023.c3nlp-1.7\u003c/span\u003e\u003cspan address=\"10.18653/v1/2023.c3nlp-1.7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen GM, Starosta WJ (2000) The development and validation of the intercultural sensitivity scale. Hum Communication 3:1\u0026ndash;15\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCohen J (1988) Statistical power analysis for the behavioral sciences, 2nd edn. Lawrence Erlbaum Associates\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCreswell JW, Plano Clark VL (2018) Designing and conducting mixed methods research, 3rd edn. SAGE\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCui H, Mahfoodh OHA (2025) An intercultural framework for assessing sociopragmatic competence in English language education: evidence from the Chinese EFL context. International Journal of Applied Linguistics. Advance online publication\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDai DW, Zhu H, Chen G (2025) How does interaction with LLM powered chatbots shape human understanding of culture? The need for Critical Interactional Competence (CritIC). Annu Rev Appl Linguist 45:28\u0026ndash;49. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1017/S0267190525000054\u003c/span\u003e\u003cspan address=\"10.1017/S0267190525000054\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeardorff DK (2006) Identification and assessment of intercultural competence as a student outcome of internationalization. J Stud Int Educ 10(3):241\u0026ndash;266. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/1028315306287002\u003c/span\u003e\u003cspan address=\"10.1177/1028315306287002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDewaele J-M, MacIntyre PD (2014) The two faces of Janus? Anxiety and enjoyment in the foreign language classroom. Stud Second Lang Learn Teach 4(2):237\u0026ndash;274. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.14746/ssllt.2014.4.2.5\u003c/span\u003e\u003cspan address=\"10.14746/ssllt.2014.4.2.5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFaul F, Erdfelder E, Lang A-G, Buchner A (2007) GPower 3: a flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behav Res Methods* 39(2):175\u0026ndash;191. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3758/BF03193146\u003c/span\u003e\u003cspan address=\"10.3758/BF03193146\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFreiermuth M, Huang H-C (2021) Zooming across cultures: can a telecollaborative video exchange between language learning partners further the development of intercultural competences? Foreign Lang Annals 54(1):185\u0026ndash;206. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/flan.12504\u003c/span\u003e\u003cspan address=\"10.1111/flan.12504\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGe H (2025) The theoretical framework of compulsory English textbook evaluation based on Byram's intercultural communicative competence (ICC) in alignment with Chinese English curriculum standard (CECS) educational policy. The EUrASEANs: Journal on Global Socio-Economic Dynamics 2(51):25\u0026ndash;40\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHackett S, Janssen J, Beach P, Perreault M, Beelen J, van Tartwijk J (2023) The effectiveness of Collaborative Online International Learning (COIL) on intercultural competence development in higher education. Int J Educational Technol High Educ 20. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s41239-022-00373-3\u003c/span\u003e\u003cspan address=\"10.1186/s41239-022-00373-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. :Article 5\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang L-J (2021) Developing intercultural communicative competence in foreign language classrooms \u0026ndash; a study of EFL learners in Taiwan. Int J Intercultural Relations 83:78\u0026ndash;90. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ijintrel.2021.06.001\u003c/span\u003e\u003cspan address=\"10.1016/j.ijintrel.2021.06.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang Y, Bai Y, Zhu Z, Zhang J, Zhang J, Su T, Liu J, Lv C, Zhang Y, Lei J, Fu Y, Sun M, He J (2023) C-Eval: a multi-level multi-discipline Chinese evaluation suite for foundation models. In: Advances in Neural Information Processing Systems 36 (NeurIPS 2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang Z, Cui N (2025) Research on the challenges, strategies and optimization paths of cultivating cross cultural communication competence in international Chinese education. J Educational Res Policies 7(6):40\u0026ndash;43. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.53469/jerp.2025.07(06).08\u003c/span\u003e\u003cspan address=\"10.53469/jerp.2025.07(06).08\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJenks CJ (2024) Communicating the cultural other: trust and bias in generative AI and large language models. Appl Linguistics Rev 16(3):787\u0026ndash;795. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1515/applirev-2024-0051\u003c/span\u003e\u003cspan address=\"10.1515/applirev-2024-0051\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim A, Su Y (2024) The effects of chatbot-mediated L2 practice opportunities on Korean as a foreign language learners' willingness to communicate. System 125. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.system.2024.103425\u003c/span\u003e\u003cspan address=\"10.1016/j.system.2024.103425\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Article 103425\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKolb DA (1984) Experiential learning: experience as the source of learning and development. Prentice-Hall\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLandis JR, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics 33(1):159\u0026ndash;174. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2307/2529310\u003c/span\u003e\u003cspan address=\"10.2307/2529310\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi B, Tan YL, Wang C, Lowell V (2025) Two years of innovation: a systematic review of empirical generative AI research in language learning and teaching. Computers Education: Artif Intell 9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.caeai.2025.100445\u003c/span\u003e\u003cspan address=\"10.1016/j.caeai.2025.100445\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Article 100445\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi H, Zhang Y, Koto F, Yang Y, Zhao H, Gong Y, Duan N, Baldwin T (2024) CMMLU: measuring massive multitask language understanding in Chinese. In: Findings of the Association for Computational Linguistics: ACL 2024, pp 11260\u0026ndash;11285. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18653/v1/2024.findings-acl.671\u003c/span\u003e\u003cspan address=\"10.18653/v1/2024.findings-acl.671\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin S (2025) The cultivation of international students' intercultural communicative competence in teaching Chinese as a foreign language. Vocat Educ 14(1):78\u0026ndash;85\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu D, Liu X (2025) Exploring the empowerment of international Chinese language teachers through large language models. J Contemp Educational Res 9(2):1\u0026ndash;12\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu F (2025) Second language teaching and learning through virtual exchange [Book review]. Innov Lang Learn Teach 19(1):1\u0026ndash;3. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/17501229.2025.2451729\u003c/span\u003e\u003cspan address=\"10.1080/17501229.2025.2451729\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu G, Darvin R, Ma C (2024) Exploring AI-mediated informal digital learning of English (AI-IDLE): a mixed-method investigation of Chinese EFL learners' AI adoption and experiences. Comput Assist Lang Learn 38(8):1632\u0026ndash;1660. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/09588221.2024.2312130\u003c/span\u003e\u003cspan address=\"10.1080/09588221.2024.2312130\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu J (2025) Exploring the impact of artificial intelligence-enhanced language learning on youths' intercultural communication competence. Humanit Social Sci Commun 12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1057/s41599-025-06033-x\u003c/span\u003e\u003cspan address=\"10.1057/s41599-025-06033-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. :Article 1757\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLo CK, Yu PLH, Xu S, Ng D, Jong MS-Y (2024) Exploring the application of ChatGPT in ESL/EFL education and related research issues: a systematic review of empirical studies. Smart Learn Environ 11. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s40561-024-00319-2\u003c/span\u003e\u003cspan address=\"10.1186/s40561-024-00319-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Article 34\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLynn MR (1986) Determination and quantification of content validity. Nurs Res 35(6):382\u0026ndash;385\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorris T (2019) Experiential learning \u0026ndash; a systematic review and revision of Kolb's model. Interact Learn Environ 28(8):1064\u0026ndash;1077. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/10494820.2019.1570279\u003c/span\u003e\u003cspan address=\"10.1080/10494820.2019.1570279\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMunezane Y (2019) A new model of intercultural communicative competence: bridging language classrooms and intercultural communicative contexts. Stud High Educ 46(8):1664\u0026ndash;1681. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/03075079.2019.1698537\u003c/span\u003e\u003cspan address=\"10.1080/03075079.2019.1698537\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNunnally JC (1978) Psychometric theory, 2nd edn. McGraw-Hill\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eO'Dowd R (2018) From telecollaboration to virtual exchange: state-of-the-art and the role of UNICollaboration in moving forward. J Virtual Exch 1:1\u0026ndash;23. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.14705/rpnet.2018.jve.1\u003c/span\u003e\u003cspan address=\"10.14705/rpnet.2018.jve.1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePatton MQ (2015) Qualitative research \u0026amp; evaluation methods: integrating theory and practice, 4th edn. SAGE\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePlonsky L (2013) Study quality in SLA: an assessment of designs, analyses, and reporting practices in quantitative research. Stud Second Lang Acquisition 35(4):655\u0026ndash;687. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1017/S0272263113000399\u003c/span\u003e\u003cspan address=\"10.1017/S0272263113000399\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePlonsky L, Oswald FL (2014) How big is big? Interpreting effect sizes in L2 research. Lang Learn 64(4):878\u0026ndash;912. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/lang.12079\u003c/span\u003e\u003cspan address=\"10.1111/lang.12079\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSarwari A, Javed MN, Adnan HM, Abdul Wahab MN (2024) Assessment of the impacts of artificial intelligence (AI) on intercultural communication among postgraduate students in a multicultural university environment. Sci Rep 14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-024-74565-w\u003c/span\u003e\u003cspan address=\"10.1038/s41598-024-74565-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Article 23108\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSweller J (1988) Cognitive load during problem solving: effects on learning. Cogn Sci 12(2):257\u0026ndash;285. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1207/s15516709cog1202_4\u003c/span\u003e\u003cspan address=\"10.1207/s15516709cog1202_4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTahmasbi A, Esrafilian M, Wright J, Jeong S, Bera A (2025) Game master LLM: task-based role-playing for natural slang learning. Preprint at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/2511.15504\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/2511.15504\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThi H-T (2021) Project-based assessment in teaching intercultural communication competence for foreign language students in higher education: a case study. Eur J Educational Res 10(2):933\u0026ndash;944. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.12973/eu-jer.10.2.933\u003c/span\u003e\u003cspan address=\"10.12973/eu-jer.10.2.933\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTu Q, Fan S, Tian Z, Shen T, Shang S, Gao X, Yan R (2024) CharacterEval: a Chinese benchmark for role-playing conversational agent evaluation. In: Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp 11836\u0026ndash;11850. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18653/v1/2024.acl-long.638\u003c/span\u003e\u003cspan address=\"10.18653/v1/2024.acl-long.638\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang P, Yin K, Zhang M, Dong Y, Wang T (2025) The effect of incorporating large language models into the teaching on critical thinking disposition: an AI\u0026thinsp;+\u0026thinsp;Constructivism Learning Theory attempt. Educ Inform Technol 30:11625\u0026ndash;11647. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10639-024-13244-3\u003c/span\u003e\u003cspan address=\"10.1007/s10639-024-13244-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWeng Z, Fu Y (2025) Generative AI in language education: bridging divide and fostering inclusivity. Int J Technol Educ 8(1):1\u0026ndash;22\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXie T, Zhou Y, Yu J (2025) Ce-LLMs: status and trends of education-specific large language models developed in China. Future Educational Res 3:505\u0026ndash;525. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/fer3.70008\u003c/span\u003e\u003cspan address=\"10.1002/fer3.70008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan D, Zhang S (2024) L2 writer engagement with automated written corrective feedback provided by ChatGPT: a mixed-method multiple case study. Humanit Social Sci Commun 11(1):1\u0026ndash;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1057/s41599-024-03543-y\u003c/span\u003e\u003cspan address=\"10.1057/s41599-024-03543-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYoung RF (2011) Interactional competence in language learning, teaching, and testing. In: Hinkel E (ed) Handbook of research in second language teaching and learning, vol 2. Routledge, pp 426\u0026ndash;443\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYuan H, Che Z, Zhang Y, Li S, Yuan X, Huang L, Hu X, Peng K, Luo S (2025) The cultural stereotype and cultural bias of ChatGPT. J Pac Rim Psychol 19:1\u0026ndash;15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/18344909251315797\u003c/span\u003e\u003cspan address=\"10.1177/18344909251315797\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng H (2024) ChatGPT integration of English education: implications for English language learners' cross-cultural communication. J Educ Humanit Social Sci 27:128\u0026ndash;135\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"intercultural communicative competence, large language models, Chinese as a foreign language, cultural dialogue, quasi-experimental design, critical cultural awareness","lastPublishedDoi":"10.21203/rs.3.rs-9140660/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9140660/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIntercultural communicative competence (ICC) is a core objective of foreign language education, yet providing learners with sufficient meaningful intercultural encounters remains a persistent challenge. Large language models (LLMs) may support such encounters through AI-mediated cultural dialogue, yet no prior study has empirically examined whether such dialogues can foster ICC development. This study addresses the gap through a quasi-experimental pretest-posttest design integrating Byram's ICC model with Kolb's experiential learning cycle. Sixty-two Chinese-as-a-Foreign-Language (CFL) learners from 17 countries participated in a six-week intervention. The experimental group (n\u0026thinsp;=\u0026thinsp;32) engaged in structured voice-based cultural dialogue tasks with a Chinese-developed LLM; the control group (n\u0026thinsp;=\u0026thinsp;30) received equivalent hours of conventional cultural instruction. Analysis of covariance (ANCOVA) results showed that the experimental group significantly outperformed the control group on both the Intercultural Sensitivity Scale (ISS, d\u0026thinsp;=\u0026thinsp;0.82) and the Cultural Intelligence Scale (CQS, d\u0026thinsp;=\u0026thinsp;0.83). Dimension-level analysis revealed a non-uniform pattern: Metacognitive CQ yielded the largest effect (d\u0026thinsp;=\u0026thinsp;1.08), whilst Cognitive CQ showed the smallest (d\u0026thinsp;=\u0026thinsp;0.52) and did not survive Bonferroni correction. Learner perception data indicated high ratings for cultural learning effectiveness but lower ratings for communicative authenticity. The study proposes a \"problem-as-resource\" strategy that transforms LLM cultural biases into material for cultivating critical cultural awareness, offering empirical evidence and reusable task templates for integrating LLMs into intercultural language education.\u003c/p\u003e","manuscriptTitle":"From cultural bias to critical awareness: LLM-mediated voice dialogue and intercultural competence in Chinese language learners","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-01 09:37:46","doi":"10.21203/rs.3.rs-9140660/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-05T13:29:36+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-30T15:21:48+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"145061867193426153243113335908535609637","date":"2026-04-11T14:04:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"171405498878827751661684262774214318265","date":"2026-04-09T14:19:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"61107422733291971925816195576045320332","date":"2026-04-01T14:16:58+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-30T09:03:59+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-30T09:00:51+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-30T05:20:48+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-29T16:50:06+00:00","index":"","fulltext":""},{"type":"submitted","content":"Humanities and Social Sciences Communications","date":"2026-03-29T16:45:14+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"bb4ba127-36df-440e-9799-844ff340af71","owner":[],"postedDate":"April 1st, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-05T13:29:36+00:00","index":41,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-30T15:21:48+00:00","index":40,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":65419538,"name":"Humanities/Cultural and media studies"},{"id":65419539,"name":"Social science/Cultural and media studies"},{"id":65419540,"name":"Social science/Education"},{"id":65419541,"name":"Humanities/Language and linguistics"},{"id":65419542,"name":"Social science/Language and linguistics"},{"id":65419543,"name":"Biological sciences/Psychology"},{"id":65419544,"name":"Social science/Psychology"}],"tags":[],"updatedAt":"2026-04-01T09:37:47+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-01 09:37:46","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9140660","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9140660","identity":"rs-9140660","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}