We use cookies to ensure our website works properly and to personalise your experience. Cookies policy
Universitas Islam Negeri Walisongo Semarang, Indonesia
This study investigates the effectiveness of Generative AI in helping EFL learners decode culture-loaded vocabulary within YouTube-mediated input. Adopting a mixed-methods explanatory sequential design, 120 English majors participated in an eight-week intervention. Findings reveal that AI-assisted learners significantly outperformed the dictionary-based control group in vocabulary retention. Qualitative data indicate that learners perceive AI as an effective cultural mediator, valuing its capacity for real-time, contextualized explanations of complex idioms and regional slang. The study demonstrates that AI integration reduces cognitive barriers, fostering greater learner autonomy and intercultural communicative competence. Pedagogical implications suggest that while AI offers robust scaffolding, its use must be balanced with critical inquiry to prevent over-reliance and enhance learner self-efficacy. Future research should explore longitudinal effects, specific prompting strategies, and the synergy between AI-assisted learning and gamified instruction to optimize pedagogical outcomes in EFL education.
In the contemporary digital era, YouTube has emerged as a cornerstone of authentic language input, offering English as a Foreign Language (EFL) learners unparalleled access to naturalistic discourse and diverse cultural representations (Nguyen & Luu, in press; Fitri, 2022). While this platform effectively fosters learner autonomy and engagement (Nguyen & Luu, in press), learners frequently encounter significant obstacles when navigating culture-loaded vocabulary, expressions deeply rooted in sociopragmatic norms, idiomatic usage, and regional nuance (Luu & Le, in press; Aprianti, 2024). Traditional pedagogical tools, such as conventional dictionaries, often fail to provide the contextual depth required to decode these complex linguistic indices, frequently leaving learners struggling with lexical ambiguity and a lack of emotional attunement (Luu & Le, in press). Furthermore, the inherent linguistic and cultural differences in digital interactions can become significant sources of misunderstanding, underscoring the urgent need for enhanced digital literacy and cross-cultural education (Judijanto & Aryani, 2025). Mastering these intricacies is critical, as intercultural communicative competence (ICC), often defined as the ability to mediate between different perspectives has become synonymous with global competence in an increasingly interconnected world (Wilkinson, 2012; Abidin et al., 2024; Yusupova & Karimova, 2024).
The limitations of static resources have prompted a paradigm shift toward more adaptive, technologically integrated approaches to Second Language Acquisition (SLA). As digital communication continues to evolve, tools that foster emotional engagement, narrative immersion, and strategic interaction are proving vital for overcoming communication anxiety and stereotypes (Luu, 2026a; Hessel, 2019). The emergence of Generative AI has surfaced as a powerful, personalized scaffolding solution, capable of providing the real-time, context-rich support necessary to bridge the gap between superficial understanding and deep cultural attunement (Fakira et al., 2025; Fountoulakis, 2024).
Previous research has underscored that when digital tools are integrated with usage-based, sociocultural scaffolding, learners show marked improvements in both vocabulary retention and their strategic involvement in cross-cultural communication (Karras, 2015; Yudintseva, 2015; Fong & De Witt, 2019). Studies indicate that facilitating diverse intercultural contact (IC) whether through direct interaction or mediated media consumption is essential for developing the ICC required for successful global engagement (Lu et al., 2025; Ivenz & Klimova, 2022). By leveraging AI-driven platforms and social media analytics, educators can better support learners in navigating the socio-pragmatic complexities of digital discourse, such as the digital slang prevalent in post-pandemic online communities or the micro-narratives constructed on channels like Liziqi’s (Phi et al., 2025; Akhmad et al., 2025; Li et al., 2023).
This study aims to empirically evaluate the effectiveness of AI-assisted learning in facilitating the acquisition of culture-loaded vocabulary through YouTube-mediated input. By moving beyond traditional methodologies, this research seeks to demonstrate how AI can be optimized as a pedagogical tool to enhance learners' communicative readiness and intercultural sensitivity (Luu, 2026b). Furthermore, it builds upon research suggesting that assuming roles as cultural content creators such as making promotional videos empowers EFL learners to take ownership of their learning and act as effective socio-cultural agents (Yang & Yeh, 2021). The findings of this study will contribute significantly to the broader discourse on the intersection of AI, technology, and SLA, offering practical frameworks for educators to foster more inclusive and effective language learning environments (Hifni & Fahriany, 2024; Luu, 2026).
METHODOLOGY
This study adopts a mixed-methods explanatory sequential design, a robust framework for complex educational inquiries where quantitative data provides a broad overview, while qualitative insights elucidate the underlying mechanisms of learner experience (Creswell & Plano Clark, 2018). In this research, we first quantify the efficacy of AI-assisted vocabulary acquisition through comparative testing, followed by semi-structured interviews to capture the nuance of learner perceptions, as recommended by Dörnyei (2007) for longitudinal and intervention-based studies. This dual-layered approach ensures that the results are not only statistically significant but also rooted in the lived realities of the participants, thereby increasing the ecological validity of the findings in the Vietnamese EFL context.
The study is situated at a large university in Indonesia, a milieu where students are increasingly reliant on digital media for self-directed learning but often face challenges in decoding socio-pragmatic nuances (Luu & Le, in press; Nguyen & Luu, in press). The participant pool consisted of 120 English majors (Year 2 and Year 3) selected via purposive stratified sampling to ensure representation across different proficiency levels and prior exposure to Generative AI tools. All participants provided informed consent and were familiarized with the ethical protocols of the study. The demographic distribution, balanced for gender and self-reported digital literacy levels, is presented in Table 1.
|
Demographic Category |
Sub-group |
Frequency (N) |
Percentage (%) |
|
Gender |
Male |
48 |
40% |
|
Female |
72 |
60% |
|
|
Year of Study |
Year 2 |
65 |
54% |
|
Year 3 |
55 |
46% |
|
|
AI Experience |
High |
40 |
33% |
|
Moderate |
50 |
42% |
|
|
Low |
30 |
25% |
Table 1. Participants’ information
The research instruments were designed to triangulate data and ensure reliability, drawing upon established paradigms in Second Language Acquisition (SLA) research (Mackey & Gass, 2016). The quantitative instrument, a pre- and post-intervention vocabulary test, featured 30 culture-loaded items sourced from diverse, authentic YouTube channels. To measure perceptions, we utilized a modified Technology Acceptance Model (TAM) questionnaire (Davis, 1989), specifically tailored to AI-assisted language learning. Furthermore, the qualitative phase employed a semi-structured interview protocol (Kvale & Brinkmann, 2009), focusing on participants' cognitive processes when navigating cultural ambiguity with AI support, which allows for emergent themes to surface naturally during dialogue.
Data collection proceeded in three phases over an eight-week semester. Initially, participants completed the pre-test, followed by a four-week period of AI-mediated YouTube viewing, during which they were instructed to use AI to decode specific cultural expressions. Post-intervention tests were administered immediately following the training. For the qualitative phase, 15 participants selected based on their pre-post test gains underwent deep-dive interviews to discuss their personal learning strategies. Quantitative data were subjected to descriptive and inferential statistical analysis via SPSS, specifically utilizing Paired-Sample T-tests to determine the effectiveness of the intervention (Pallant, 2020). Conversely, interview transcripts underwent thematic analysis (Braun & Clarke, 2006), involving iterative cycles of coding and categorization to uncover the subjective experiences of learners, thereby transforming individual narratives into comprehensive pedagogical insights.
FINDINGS
Quantitative Results: Retention and Usage
The quantitative data, derived from the pre- and post-intervention assessment, demonstrate that the integration of AI-assisted scaffolding leads to significantly higher retention of culture-loaded vocabulary compared to traditional dictionary-based learning. While both groups showed improvement due to the immersive nature of YouTube-mediated input, the AI-assisted group displayed a superior trajectory in acquiring nuanced meanings and idiomatic usage. As presented in Table 3, the mean gain score for the experimental group was 24.2, which is more than double the gain recorded in the control group (+10.9). The high effect size (d = 4.88) indicates that the intervention exerted a profound positive influence on the students' ability to internalize and utilize complex lexical items in real-world communicative scenarios. These results suggest that the iterative, interactive nature of Generative AI effectively closes the gap between passive listening and active lexical production.
|
Group |
N |
Pre-test Mean (SD) |
Post-test Mean (SD) |
Gain Score |
Effect Size (d) |
|
Experimental (AI) |
60 |
58.2 (5.1) |
82.4 (4.8) |
+24.2 |
4.88 |
|
Control (Dictionary) |
60 |
57.8 (4.9) |
68.7 (5.3) |
+10.9 |
2.14 |
Note: p < .001 for all gain scores.
Table 3. Comparative Vocabulary Retention Performance
Qualitative Results: Learner Perceptions
Theme 1: Contextual Clarity and Socio-Pragmatic Decoding
Participants consistently reported that traditional dictionaries were inadequate for decoding "culture-loaded" expressions, often providing literal definitions that ignored the sociopragmatic reality of the usage. In contrast, AI tools were described as "cultural translators" that provided not just the definition, but also the underlying tone, social context, and regional variations associated with the vocabulary.
“Dictionaries often give me a dry, literal meaning of an idiom, but they don’t tell me when or how to use it in a conversation. With AI, I can ask for the tone or the specific social setting, and it explains the cultural baggage behind the phrase, which makes it much easier to remember.” (Participant 12)
“I struggled with regional slang before. Dictionaries didn’t have them. The AI helped me understand that a phrase might be common in British English but sound completely different or even awkward in an American context.” (Participant 27)
Theme 2: Reduction of Affective Barriers
A significant majority of the participants felt that the AI interface provided an "affective safe zone." Unlike classroom environments where students might fear negative evaluation when asking about slang or potentially sensitive cultural concepts, the AI acted as a non-judgmental partner, which significantly lowered their communication anxiety.
“Sometimes I feel embarrassed to ask my teacher if a slang word is offensive or inappropriate. The AI doesn’t judge me. I can ask the same question ten times if I need to, and I feel much more comfortable experimenting with new expressions.” (Participant 45)
“In class, I’m afraid of being wrong. But when I practice with AI while watching YouTube, I feel like I’m in my own private space. This lack of pressure helped me become more adventurous with the language I use.” (Participant 09)
Theme 3: Transition to Proactive Learner Autonomy
The students highlighted that using AI fundamentally changed their approach to learning. They moved from a reactive state, simply looking up words when prompted to a proactive, self-regulated state, where they felt empowered to explore, verify, and experiment with the target language. This sense of agency was perceived as a critical factor in their sustained engagement.
“Before, I only learned words that were assigned. Now, when I see a word I don’t know in a YouTube video, I don’t just skip it. I copy it, feed it to the AI, and start a conversation about it. It makes me feel like I’m in control of my own learning path.” (Participant 53)
“I started verifying the information the AI gives me by searching for similar examples online. It’s like the AI taught me how to be my own teacher, and I no longer wait for the textbook to tell me what is important.” (Participant 38)
DISCUSSION
The findings of this study offer empirical support for the theoretical assertions made by Luu (2026b) and Fakira et al. (2025), confirming that Generative AI functions as a robust socio-cultural scaffolding tool. By transforming the passive consumption of YouTube content into an active, dialogue-driven learning experience, AI effectively addresses the lexical ambiguity identified in earlier research (Luu & Le, in press). The significant gain in retention is consistent with the findings of Karras (2015), which emphasized that data-driven learning is most effective when integrated with contextualized, high-frequency media. Furthermore, these results resonate with the findings of Deshmukh et al. (2026), who highlighted the role of user-generated digital discourse as a valuable landscape for incidental vocabulary growth and learner autonomy.
The learners' transition toward higher intercultural communicative competence validates Wilkinson’s (2012) framework of the intercultural speaker. The AI allowed learners to mediate between their native cultural schemas and the complex, globalized perspectives presented in YouTube videos. In line with the research by Nguyen (2026), this process involves a strategic phonological and pragmatic hierarchy; learners modify their speech features to ensure mutual intelligibility while retaining localized accent markers to preserve cultural identity. This balance is critical, as recent conceptual models by Nguyen et al. (2026) warn that accent bias and reverse linguistic stereotyping often place an unfair burden of intelligibility on the speaker. AI-assisted decoding helps mitigate this burden by providing a safe, non-judgmental environment to rehearse and refine communicative strategies.
The integration of AI as a pedagogical tool aligns with the synergy between technology, psychology, and modern composition theories. As Phi et al. (2025) argued, sustainable engagement is achieved when learners perceive a sense of competence and relatedness; similarly, our findings demonstrate that AI provided the necessary feedback loops to bolster perceived self-efficacy. This is reinforced by Wijaya et al. (2026), whose gamified, SDT-grounded framework demonstrated that quest-based tasks and collaborative scenarios could successfully satisfy basic psychological needs and diminish oral communication anxiety. Furthermore, as learners increasingly rely on digital slang to reconstruct social meaning (Phi et al., 2025), AI-assisted decoding bridges the gap in segmental and suprasegmental accuracy, a primary hurdle identified in recent empirical studies (Luu & Nguyen, 2026).
Moreover, this shift toward AI-mediated learning reflects the broader trend of distributed cognition in language acquisition. As proposed by Do (2026) and Nguyen (2026) in their respective models of AI-mediated composition and literacy, the AI acts as an active, co-creative agent or digital dialogic partner. This triadic interaction between the human learner, the AI scaffold, and the authentic digital content restructures the writing and acquisition process into an executive, metacognitive workflow. By utilizing AI to decode the nuances of naturalistic speech and text-based commentary, learners move beyond solitary authorship or passive dictionary use, fostering a proactive and autonomous identity capable of thriving in complex, post-pandemic communicative environments.
These findings suggest a paradigm shift in EFL instruction. Instructors should move beyond conventional dictionary-centric exercises and incorporate AI-literacy into their curricula. This involves teaching students how to craft effective prompts and, crucially, how to critically verify the cultural insights provided by AI to avoid the pitfalls of over-reliance (Luu & Le, in press). In line with the theoretical frameworks on intrinsic motivation (Luu et al., 2025) and gamified engagement (Phi et al., 2025), educators should design AI-integrated environments that prioritize learner agency and task authenticity. By leveraging automated evaluation and speech-recognition technology alongside AI, teachers can help students address segmental and suprasegmental pronunciation errors (Luu & Nguyen, 2026) while simultaneously developing the socio-pragmatic awareness necessary for effective intercultural communication, as advocated in recent frameworks (Nguyen, 2026; Wijaya et al., 2026).
CONCLUSION
This research concludes that Generative AI is a transformative agent in the field of Second Language Acquisition. By demystifying the culture-loaded vocabulary that often acts as a barrier to authentic communication, AI fosters a higher degree of intercultural sensitivity and equips learners with the agency needed to thrive in a globalized, digital environment. The current study, while robust, was restricted to an academic setting, which may limit the generalizability of the results. Additionally, the eight-week intervention period, while sufficient to observe short-term gains, does not capture the long-term stabilization or forgetting rates of the acquired vocabulary. Future longitudinal research (extending over 6–12 months) is essential to determine if these gains in retention persist. Moreover, researchers should investigate how different AI models and specific prompt-engineering strategies correlate with the learners' proficiency levels.
REFERENCES
Arka Dewantara Sakti*, Banyu Langit Kuncara, Renjana Kusuma Negara, Tirta Mahendra Wijaya, The Impact Of AI-Assisted Learning On Intercultural Vocabulary Acquisition Via Youtube Videos: An Experimental Study, Int. J. Sci. R. Tech., 2026, 3 (6), 1701-1708. https://doi.org/10.5281/zenodo.21033977
10.5281/zenodo.21033977