How L2 selves shape attitudes toward AI-assisted language learning: the mediating roles of intrinsic and extrinsic motivation in cross-cultural Asian contexts

Introduction: This study investigates the relationship between the L2 motivational self-system and attitudes toward AI-enhanced second language (L2) learning, and tests whether intrinsic and extrinsic motivation mediate these relationships among students learning Chinese or English as a foreign language in Thailand, Macau, and Mainland China.

Methods: Questionnaire data were collected from 361 participants. Correlational analysis, hierarchical regression, and a bootstrapping approach were used to test mediational relations.

Results: Both the ought-to L2 self and the ideal L2 self were significantly associated with attitudes toward AI-enhanced L2 learning, with the ideal L2 self showing the stronger association. In the overall sample, and in the Thai and Mainland Chinese samples analyzed separately, intrinsic and extrinsic motivation statistically fully mediated the association between the ought-to L2 self and attitudes toward AI-enhanced L2 learning. In the Macau sample, extrinsic motivation was more involved than intrinsic motivation in transmitting this effect. For the ideal L2 self, intrinsic and extrinsic motivation only partially mediated the relationship with attitudes toward AI-enhanced L2 learning.

Discussion: Positive attitudes toward AI-enhanced L2 learning are not formed merely through future self-representations; they are also linked to learners’ intrinsic and extrinsic motivation. By examining the interaction between the L2 motivational self-system and intrinsic and extrinsic motivation across different regions, this study extends L2 motivation research to the area of AI-enhanced language learning.

The notion of motivation has long been considered a crucial determinant of success in second and foreign language acquisition because of its direct influence on learners’ efforts, persistence, and achievement (Dörnyei, 2001; Thohir, 2017). Early and more recent works on the subject have shown that motivation in language learning consists of self-related, social, affective, and situational dimensions rather than being a monolithic construct defined by an internal state (Csizér and Dörnyei, 2005a, 2005b). In this vein, the L2 Motivational Self System theory gained prominence for emphasizing the pivotal role of individuals’ ideal L2 self (ILS) and ought-to L2 self (OLS) in predicting language learning behavior (Khany and Amiri, 2016; Papi, 2010).

Meanwhile, artificial intelligence is currently revolutionizing the sphere of language instruction. Contemporary findings confirm the potential for positive changes to be brought about by AI assistance in terms of increased engagement, satisfaction, motivation, and effectiveness, but only if individual learners perceive their learning experiences positively based on emotions, self-efficacy beliefs, and learning situations (Yuan and Liu, 2024; Zheng et al., 2024; Zhou and Hou, 2024). In particular, attitude toward AI-assisted L2 learning (ATA) becomes critical in this context since the degree to which people embrace, employ, and benefit from AI technology depends on their attitude (Chen and Xu, 2026; Qin et al., 2026; Wang et al., 2025; Wu et al., 2024).

While a considerable body of evidence exists on L2 motivation, technology acceptance, and AI-assisted language learning, relatively little is known regarding the translation of core self-guides under the L2 Motivational Self System theory into positive attitudes towards AI-assisted L2 learning via intrinsic motivation (IM) and extrinsic motivation (EM). Such an investigation becomes especially timely and relevant in the context of diverse multilingual environments in Asia with various learning targets, different educational, cultural, and technological settings (Chen et al., 2026; Xu et al., 2022). Thus, this study will focus on Thailand, Macau, and Mainland China in order to explore whether ought-to L2 self and ideal L2 self influence attitude toward AI-assisted L2 learning mediated by motivation.

Literature review

Motivation and the L2 motivational self system

The investigation of motivation within SLA moved from relatively general social-psychological approaches to more dynamic and self-referential frameworks (Dörnyei, 2001). For example, in an influential study, Csizér and Dörnyei (2005a) showed that language learners could be distinguished by their motivational profile, thus illustrating internal complexity and subjectivity of motivation. In another structural analysis, Csizér and Dörnyei (2005b) proposed that integrative factors play a crucial role in the motivational system and can mediate other aspects of motivation that influence behavior. Those two studies laid the groundwork for our understanding of the complex and multifaceted nature of L2 motivation, though they predated the emergence of AI-assisted education and focused on adolescent Europeans.

Continuing the trend, the L2 Motivational Self System uses the framework of future self-guides, namely, the ideal L2 self and the ought-to L2 self, to explain motivational dynamics. Papi (2010) found that both guides contributed to motivated effort but concluded that the ideal L2 self was positively correlated with better results, whereas the ought-to L2 self increased anxiety. Similarly, Khany and Amiri (2016) noted that the ideal L2 self played the critical role in driving the motivational process, whereas the ought-to L2 self acted within narrower limits. Kormos and Csizér (2010) confirmed that self-image was a powerful predictor of motivated effort, whereas Kormos and Csizér (2013) found that future self-guides were the basis of self-regulated behavior and autonomous learning. Recently, Yang and Lian (2023) demonstrated the indirect effect of the ideal L2 self on pragmatic performance through willingness to communicate, whereas Wu (2024) noted that the ideal L2 self was a powerful predictor of motivation both directly and indirectly through learning experience and enjoyment, although the ought-to L2 self played a comparatively limited role. Overall, all those studies demonstrate the effectiveness of self-guided approaches in explaining motivation, albeit focusing predominantly on effort, behavior, anxiety, and willingness to communicate, rather than attitudes to AI assistance (Huang et al., 2024).

Intrinsic and extrinsic motivation in foreign language learning

Both intrinsic and extrinsic motivations continue to be the key concepts in language learning research. According to Thohir (2017), intrinsic motivation was found to be more significant when it comes to effective and persistent language learning, but extrinsic motivation is also very widespread since many activities related to language learning involve certain goals and requirements. Moreover, according to Carrió-Pastor and Mestre (2014), motivation works in interaction with the students’ beliefs, task design, and rich technological experience of the learners. Finally, Pourfeiz (2016) showed reciprocal relations between attitudes toward foreign language learning and academic motivation, proving the close connection between intrinsic and extrinsic approaches and learners’ evaluations.

However, motivation does not exist in isolation and can be affected or weakened by external and interpersonal factors. Djafri and Wimbarti (2018) showed that students’ motivation was unrelated to their foreign language anxiety, but the teachers’ behavior had a significant impact on it. Similarly, in a literature review on praise, Peng (2021) suggested that positive communication behaviors of instructors could promote learner motivation and engagement. Such information is important since it demonstrates that motivational processes occur against the background of the whole educational environment (Yu et al., 2022). Still, the abovementioned sources primarily address classroom education and not AI-assisted one.

Mediating mechanisms between motivation and L2 outcomes

One recurrent theme that can be observed in modern scholarly articles is the indirect impact of motivation on results through the intervention of emotions, behaviors, or self-regulatory mechanisms. According to Zhang et al. (2020), enjoyment acts as a mediator between motivation and performance in second foreign language acquisition. Moreover, Yang and Lian (2023) found willingness to communicate as a mediator between the ideal L2 self and pragmatic production. According to Wu (2024), mediators of motivation and its impact on behavior include learning experiences and foreign language enjoyment. The importance of combining intrinsic and extrinsic motivation, vivid goals, and feedback has also been stressed (Huo et al., 2025; Ni and Xu, 2025).

Such studies are important due to their focus on how motivation affects other factors, as opposed to the direct impact of motivation. Yet, most of the mediation research done until now highlights enjoyment, willingness to communicate, self-regulation, or engagement as mediators between the effect of self-guides and attitudes towards learning. It can be argued, however, that the same applies to intrinsic and extrinsic motivation in the context of AI-based learning.

Motivation, attitudes, and AI-assisted L2 learning

Several recent AI-related studies also consistently point out the significance of motivation as the factor explaining learners’ behaviors towards AI technology. Zheng et al. (2024) proved that self-determination motivation was positively correlated with both behavioral intention and use of generative AI for EFL learning purposes and could also serve as the moderator of technology acceptance. Similarly, Wu et al. (2024) elaborated on the development and validation of an attitude scale regarding AI-assisted L2 learning, indicating that attitudes were indeed measurable, relatively positive, and correlated with learners’ proficiency. Wang et al. (2025) demonstrated a direct relationship between acceptance and behavioral engagement, which occurred indirectly via motivational aspects. Qin et al. (2026) established that motivation mediated the link between AI acceptance and proficiency, whereas Liang et al. (2026) found that L2 selves were related to AI adoption mediated by perceived usefulness and ease of use.

Another set of AI-related research questions concerns the analysis of motivation and emotional processes in AI-mediated language learning. Liu et al. (2024) discovered that the ideal L2 self positively predicted both enjoyment and participation in informal digital learning assisted by AI. At the same time, it was the ought-to L2 self that positively predicted participation mediated by enjoyment. Analogously, Li and Yan (2026) proved that enjoyment mediated the effect of the ideal L2 self on participation in informal digital learning. Moreover, participation in this type of learning could be hindered by anxiety. Wang, Gao, and Reynolds (2026) demonstrated that the ideal L2 self was highly correlated with positive emotions and behavioral engagement, while the ought-to L2 self played only a minor role in these processes. Huang and Mizumoto (2024) found that the use of ChatGPT in classes led to a stronger activation of the ideal L2 self and a more favorable learning experience as compared to the ought-to L2 self. Rahimi et al. (2025) suggested that ChatGPT-assisted personalized learning could help motivate and regulate learners’ behavior based on their self-experience.

Other AI studies widen the scope, emphasizing the role of contextual resources in L2 learning. Pan and Zhang (2025) reported that teacher affective support, AI literacy, and self-efficacy could foster the development of L2 writing motivation in an AI-mediated context. Chen (2025) discovered that motivation served as the mediator of the relationship between resilience and engagement in AI-mediated CSL. Zhou and Hou (2024) demonstrated the effects of AI use on behavioral, cognitive, and emotional engagement while also warning about its potential negative consequences. Guo et al. (2025) highlighted the effects of autonomy support and familiarity on collaborative motivation in AI-assisted activities, while Min and Jiazhu (2025) revealed that interactivity perception could predict motivated learning. Wang and Guo (2026) underlined the relevance of use behavior and task-technology fit to English learning success, whereas Rad et al. (2026) identified conditional shaping of AI use intentions by agency-related appraisals. Finally, Hao (2026) argued that the ideal L2 self was more related to positive affect than the ought-to L2 self in an AI-mediated context.

Thus, these studies yield a few conclusions. First, the correlation of AI-assisted L2 learning with motivation, attitude, emotion, and engagement can be traced. Second, the ideal L2 self functions more favorably and more consistently in relation to AI technology. Third, mediation mechanisms play a significant role. However, there are still some problems with this literature.

Evaluation of the literature and research gaps

Existing literature is theoretically and methodologically rich and includes theories such as L2MSS, SDT, TAM, UTAUT2, expectancy-value theory, and engagement frameworks. Studies including Dörnyei (2001), Csizér and Dörnyei (2005a, 2005b), and Papi (2010) form a solid theoretical background for self-motivation. Recent studies associated with AI include works by Zheng et al. (2024), Wu et al. (2024), Liu et al. (2024), Wang et al. (2025), and Liang et al. (2026). These studies constitute a valuable resource for formulating the present research.

However, there are four key gaps in this literature review. First, while many articles address topics related to motivation or AI acceptance, fewer studies analyze attitudes toward AI-assisted L2 learning specifically. Second, while the ideal self and ought-to self have been found to predict effort, enjoyment, anxiety, engagement, and self-regulation in previous research, intrinsic motivation and extrinsic motivation have not been studied to determine whether they mediate the influence of self-guides on attitude formation. Third, most research associated with AI focuses on only one country or linguistic group (Chinese EFL learners). Finally, it was shown in some recent studies that the role of ought-to self in motivating students in AI conditions might not be as prominent, direct, and positive as expected.

To fill in these gaps, the present research will address the attitudes of Thai, Macanese, and Mainland Chinese students who are studying Chinese or English as a foreign language to AI-assisted L2 learning. The main dependent variable in the study will be the attitude of these learners to AI-assisted L2 learning. Furthermore, the study will attempt to find out whether the impact of ideal L2 self and ought-to L2 self on attitude is mediated by intrinsic and extrinsic motivation (see Figure 1).

Based on this rationale, the study asks:

RQ1: How are ought-to L2 self and ideal L2 self related to attitude toward AI-assisted L2 learning? And what are the differences in students learning Chinese/English as a foreign language in Thailand, Macau, and Mainland China?

RQ2: Do intrinsic and extrinsic motivation play mediating roles between ought-to L2 self as well as ideal L2 self and attitude toward AI-assisted L2 learning among those students in the three areas?

Data collection

During the questionnaire design phase, consideration was given to the participants being L2 learners from different countries/regions. Therefore, the items were presented in both Chinese and English. To ensure linguistic fluency, conceptual accuracy, and comprehensibility, two researchers participated in the translation and proofreading of the items, ultimately reaching a consensus on the final versions. This study recruited participants through convenience sampling at eight universities: three in Thailand, two in Macau, and three in Mainland China. A total of 400 questionnaires were distributed via course instructors, of which 361 valid responses were retained after excluding 39 incomplete or invalid responses. The final sample comprised 113 learners from Thailand, 103 from Macau, and 145 from Mainland China. Participants met the following inclusion criteria: (a) they were currently enrolled as university students, (b) they were studying Chinese or English as a foreign language, and (c) they had prior experience with AI-assisted learning tools. Participants in Thailand were learners of Chinese, whereas those in Macau and Mainland China were learners of English. Although the questionnaire was administered in Chinese and English rather than Thai, all Thai participants were university students formally enrolled in Chinese language programs, for whom Chinese was the main medium of study. Item wording was kept simple and was reviewed by a Thai–Chinese bilingual researcher to confirm comprehensibility. No comprehension problems were reported. For ethical considerations, all participants were informed that their personal information would be used solely for research purposes. The research themes and objectives were explained in detail, and participants carefully reviewed the informed consent form before proceeding. They were also explicitly notified that they could discontinue the questionnaire at any time, without any consequences. All questionnaires were distributed and collected via the Tencent Questionnaire online platform. Upon completion of the data collection phase, all responses were exported into Excel files.

Measures

Both ought-to L2 self and ideal L2 self were evaluated by using a five-point Likert scale that was revised from the L2 Motivational Self System scales (Dörnyei and Ushioda, 2009, pp. 90–97). This instrument was employed to examine learners’ perceptions of their future language-related identities and motivational drivers in the context of using AI for L2 learning. The measurement items included statements such as: ‘I must learn a second language because I think my parents would be disappointed if I didn't’ and ‘I can imagine myself living abroad and having discussions in a second language.’ The reliability analysis demonstrated strong internal consistency, with Cronbach's α of 0.88 for the ought-to L2 self scales and 0.92 for the ideal L2 self scales.

The intrinsic motivation scales and the extrinsic motivation scales were assessed by using a modified version of the Online Chinese Learning Motivation Scale (Xu et al., 2022). The measures aimed to evaluate the participants’ internal drive and external incentives for AI-assisted L2 learning. Two example items were: ‘Using AI tools and their interactive features are to stimulate my interest in the language’ and ‘I use AI for second language learning because I believe it benefits my personal development.’ The Cronbach's α for intrinsic motivation scales was 0.89, and for extrinsic motivation scales was 0.85.

An adapted version of the GenAI-Assisted EFL Learning scales (Wang et al., 2025) was employed to assess the attitude toward AI-assisted L2 learning. An example item was: ‘I find the process of using AI for a second language learning highly enjoyable.’ The Cronbach's α was 0.94.

Data analysis

This study employed SPSS 27 for data analysis following a structured procedure. Initially, descriptive statistics were computed to examine the fundamental demographic and learning characteristics of students learning Chinese/English as a foreign language across three areas: Thailand, Macau, and Mainland China. This preliminary analysis provided a comprehensive overview of the sample distribution and key variables. Subsequently, hierarchical regression analysis was implemented to systematically investigate the predictive relationships between variables, including ought-to L2 self, ideal L2 self, intrinsic motivation, extrinsic motivation, and learners’ attitudes toward AI-assisted L2 learning. Finally, this study utilized the Bootstrap technique within PROCESS 4.1 to assess mediation effects. If the 95% confidence interval (CI) derived from 5,000 Bootstrap samples does not include zero, the indirect effect is considered significant. The model's total and indirect effects were reported with comprehensive statistical indices, encompassing variance explained (R2), unstandardized coefficients (B), Standard Error of the standardized coefficient regression coefficients (SE β), p-values, and 95% CIs.

Descriptive statistics

This study analyzed the basic background information of 361 L2 learners from Thailand (N = 113), Macau (N = 103), and Mainland China (N = 145). This demographic data helps to establish a fundamental understanding of the sample. As presented in Table 1, the participant pool was predominantly female (84.0%), with males accounting for 16.1%. This gender imbalance was consistently observed across all three regions, reflecting a common sampling pattern in second language acquisition research. In terms of age distribution, most students (95.0%) were concentrated in the 18–25 range, indicating that our findings primarily reflect the characteristics of traditional university-aged learners. Notably, the Thai cohort demonstrated a wider age distribution, including a minority of learners aged 26 and above (approximately 14.2%), while the Macau and Mainland China groups consisted almost exclusively of 18–25-year-olds. This demographic variation suggests that the Thai data may incorporate perspectives from learners with more diverse life experiences.

Table 1. Demographic statistics.

Regarding experience with AI learning tools, most students (88.9%) reported having used such tools for three months or longer. This indicates that participants were generally familiar with AI-assisted learning, and their survey responses were informed by substantial practical experience, thereby enhancing the credibility of their feedback. However, notable regional differences emerged in usage duration: nearly all Mainland Chinese students (97.9%) were long-term users, a proportion that was also high among Macau students (91.3%), while substantially lower among Thai students (75.2%). These differences in experience suggest varying degrees of technological integration in educational contexts across regions, an important contextual factor that should be considered when interpreting subsequent findings.

Hierarchical regression analysis

Table 2 presents the Pearson correlation coefficients for five variables among the total sample and students from three different regions. From the overall sample perspective, all variables exhibit significant positive correlations. Notably, the strongest correlations are observed between ATA and IM, and between IM and EM, both reaching 0.60 and being significant at the p < 0.001 level. In contrast, the lowest correlations are found for OLS with both ATA and ILS, both at 0.21, also significant at the p < 0.001 level. Regional analyses reveal interesting disparities. For Thai students, the correlation between ATA and IM reaches a strong level (r = 0.76, p < 0.001). Similarly, IM and EM also show a strong correlation (r = 0.75, p < 0.001). However, there is no significant correlation between OLS and ILS. Analysis of Macau students shows a significant moderate correlation between ATA and IM (r = 0.56, p < 0.001). In contrast, correlations between OLS and both ILS and IM do not reach a significant level. Among Mainland Chinese students, the correlation coefficient between EM and IM is 0.55, significantly positive at the p < 0.001 level, indicating a moderate degree of significant positive correlation. OLS shows a significant weak correlation with both ATA (r = 0.20, p < 0.05) and IM (r = 0.21, p < 0.05).

Table 2. Correlation analysis.

*p ＜ 0.05, **p ＜ 0.01, ***p ＜ 0.001.

Mediating effects of intrinsic and extrinsic motivation towards AI-assisted L2 learning

Through the Bootstrap sampling method, the mediating roles of IM and EM between the OLS and ATA were examined. As shown in Table 3, parallel mediation models were established for the total sample and the three regional subsamples, and a systematic analysis of the total, direct, and indirect effects was conducted.

Table 3. Total and direct effects of model A.

From the perspective of the total sample, when the mediating variables were not included, the OLS had a significant positive association on attitudes towards ATA, with a total effect value of 0.15. However, when both IM and EM were introduced into the model as mediators, the direct effect of the OLS on attitudes dropped to −0.01 and became statistically non-significant. At the same time, the effects of the two mediating paths were highly significant: the path coefficient for intrinsic motivation was 0.38, and for extrinsic motivation, it was 0.29.

Among the Thailand student group, a pattern highly consistent with the total sample was observed. The total effect of the OLS on attitudes was 0.28. After introducing the mediating variables, its direct effect became non-significant. Analysis of the mediating paths revealed that both IM and EM played key roles, with path coefficients as high as 0.48 and 0.34, respectively.

The data for Macau students presented a unique characteristic. In the baseline model, the direct effect of the OLS on ATA was already non-significant, with a total effect value of only 0.07. However, after introducing the motivation variables, both IM and EM demonstrated strong predictive power on ATA, with path coefficients of 0.42 and 0.27, but the relationship between OLS and IM showed no significant correlation.

Finally, for the Mainland China student subsample, the initial model indicated a significant positive total effect of the OLS on ATA, with a total effect value of 0.15. However, after incorporating IM and EM, this direct effect became non-significant.

The results confirm that IM and EM fully mediated the relationship between the OLS and ATA among the total sample, as well as among students from Thailand and Mainland China. For students from Macau, while the OLS showed no direct effect on ATA, the indirect association operated only through EM. Although IM is a strong predictor of ATA, it is not a significant mediator between OLS and ATA, as the link from OLS to IM is not significant. The proposed mediation model is visually summarized in Figure 2.

Next, we examined the mediating roles of IM and EM between the ILS and ATA. Table 4 presents the results of the parallel mediation model analysis with the ILS as the independent variable. Consistent with the previous analysis, we investigated the total sample and the three regional subsamples separately.

Table 4. Total and direct effects of model B.

First, from the overall sample analysis, the ILS demonstrated a strong predictive power over ATA. When mediators were not included, the total effect value was 0.35 and reached statistical significance. After introducing intrinsic and extrinsic motivation as mediators into the model, the direct effect of the ILS on ATA decreased to 0.15 but remained statistically significant. At the same time, the mediating paths of both intrinsic and extrinsic motivation were highly significant, with path coefficients of 0.36 and 0.24, respectively.

Among the Thailand student group, the total effect of the ILS on ATA was 0.45, indicating a strong relationship. After introducing the mediating variables, the direct effect decreased to 0.14 but remained significant. Analysis of the mediating paths revealed that both IM and EM played important mediating roles, with path coefficients of 0.30 and 0.30, respectively.

Regarding the data for Macau students, the ILS showed a very strong direct predictive relationship with ATA, with a total effect value as high as 0.39. After introducing intrinsic and extrinsic motivation, the direct effect decreased to 0.26 but remained highly significant. Meanwhile, both mediating paths were statistically significant, with intrinsic motivation exhibiting a path coefficient of 0.39 and extrinsic motivation a coefficient of 0.17.

Finally, for the Mainland China student subsample, the analysis indicated a significant positive total effect of the ILS on ATA, with an effect value of 0.32. After incorporating IM and EM, this direct effect remained significant, with an effect value of 0.13. At the same time, the effects of both IM and EM on ATA were significant, with path coefficients of 0.31 and 0.20, respectively.

It suggests that IM and EM partially mediated the relationship between the ILS and ATA among the total sample, students from Thailand, Macau, and Mainland China. The conceptual model is illustrated in Figure 3.

These findings offer an unequivocal answer to RQ1. Ought-to L2 self and ideal L2 self correlated positively with attitudes toward AI-assisted L2 learning, with the latter being a stronger and more stable predictor of attitudes across the entire sample and among each regionally grouped subgroup. These results are congruent with existing research within the L2MSS framework, according to which the predictive power of the ideal L2 self is generally greater compared to that of the ought-to L2 self (Khany and Amiri, 2016; Papi, 2010; Wang et al., 2026; Wu, 2024). They correspond to the findings in AI-related research that ideal images of one's self in the context of using technology in learning lead to greater positive emotions and behavioral engagement (Dou et al., 2025; Hao et al., 2025; Huang and Mizumoto, 2024; Li and Yan, 2026; Liu et al., 2024). Finally, the weaker role of externally motivated ought-to L2 self predictors is consistent with research stating that externally driven self-guides are not always effective or may even inhibit certain positive behaviors (Hao, 2026; Papi, 2010).

The current findings also contribute to addressing RQ2. Intrinsic and extrinsic motivation acted as complete mediators of the relationship between ought-to L2 self and attitudes toward AI-assisted L2 learning in the overall sample and Thailand and Mainland China groups. In Macau, however, only extrinsic motivation mediated the relationship since the role of ought-to L2 self for predicting intrinsic motivation turned out to be non-significant. Theoretically, these findings extend mediation studies conducted in SLA as they demonstrate that motivation itself can become an intermediary of self-guided predictors of behavior or attitude changes. Previously, research has focused on enjoyment, willingness to communicate, personality, or self-regulation (Li and Xu, 2025; Wu, 2024; Yang and Lian, 2023; Zhang et al., 2020). Given the cross-sectional design, all mediation findings should be interpreted as statistical mediation of associations rather than causal pathways.

For the ideal L2 self, partial mediation of its effect on attitudes was observed. Thus, ideal self-guides affected attitudes through both direct effects and mediation. Theoretically, the results are significant since they indicate that ideal self-guides have a greater degree of internalization, which allows influencing attitudes independently from certain motivational intermediaries. In addition, this pattern matches those observed in other studies in which ideal L2 self predicted both directly and indirectly through intermediaries positive outcomes of digital or AI-assisted learning (Liang et al., 2026; Liu et al., 2024; Wang et al., 2026).

The other contribution of the current research consists of the cross-regional analysis of motivational predictors. As expected, there is great similarity among the three regions in the relationships observed between the studied variables. At the same time, it becomes clear that regional differences are related to different educational environments affecting the transformation of self-guides into motivation and attitudes. This observation is supported by the literature on SLA focusing on the role of various contextual variables, such as teachers’ behaviors, autonomy support, AI literacy, and socio-cultural factors in motivation formation (Djafri and Wimbarti, 2018; Guo et al., 2025; Pan and Zhang, 2025; Peng, 2021; Xu and Wang, 2024). Specifically, the case of Macau highlights the possibility that externally driven motives cannot be internalized enough to promote interest-based motivation in AI-assisted learning.

Limitations of this study include its cross-sectional nature (causal interpretation requires longitudinal designs); self-reports used in questionnaire design (self-reported attitudes and motivations may not fully reflect actual behavior); the fact that the sample includes three regions of China and Southeast Asia; and the focus on attitudes, rather than other possible outcomes, such as proficiency or persistence.

Possible topics of future research include longitudinal studies examining the interplay between self-guides, motivation, and attitudes as it develops with greater exposure to AI tools. Mediators in the form of other relevant constructs, such as enjoyment, anxiety, self-efficacy, AI literacy, or autonomy, should be introduced into the equation (Li and Yan, 2026; Liu et al., 2024; Pan and Zhang, 2025; Sukjairungwattana et al., 2025; Xu, 2025; Zheng et al., 2024). Comparative studies involving different linguistic settings and other geographical locations will help to understand whether the patterns discussed here are region-specific or general. Finally, future studies need to make the link between attitudes and actual behavior/engagement/proficiency.

The current research has revealed that the attitudes learners hold toward AI-based L2 learning are strongly associated with not only the L2 motivational self system but also intrinsic and extrinsic motivation. While the ideal L2 self proved to be a better predictor in predicting attitudes toward AI-based L2 learning across all regions involved, the ought-to L2 self was related to attitudes more indirectly via motivation. Intrinsic and extrinsic motivation were found to completely mediate the relationship between the ought-to L2 self and learner attitude and partially mediate that between the ideal L2 self and attitude among all participants. The results obtained are a valuable contribution to the existing scientific literature on AI-assisted language learning as they show that future selves of learners and their intrinsic and extrinsic motivation play an important role in determining positive attitudes toward the use of AI-based technology for L2 learning.