Empowering language learners with AI: insights from positive psychology in an EFL context

Abstract

Introduction:

The integration of artificial intelligence (AI) into language education has generated growing interest; however, its effects on learners' language proficiency and affective experiences remain underexplored. Grounded in the framework of positive psychology, this study examined the use of the Pi AI-powered platform in a semester-long English language course for upper-intermediate learners. As an exploratory small-scale classroom-based study, it aimed to provide preliminary insights rather than broadly generalizable conclusions. Specifically, it investigated the platform's impact on students' language proficiency, motivation, anxiety, enjoyment, and engagement, as well as their perceptions of its advantages and disadvantages.

Methods:

The study was conducted with an intact class of 16 upper-intermediate EFL students at a private language institute. A mixed-methods approach was employed, drawing on multiple sources of data: students' exam scores, a focus group discussion, individual interviews, learner narratives, and instructor observations. Data were collected throughout one academic semester during the integration of the AI platform into regular classroom practice.

Results:

Findings revealed improvements in students' speaking and listening skills over the period of AI integration. Participants also reported reduced anxiety and increased motivation, enjoyment, and engagement in language learning activities. Students valued the platform's personalized and low-stakes opportunities for practice. However, they also identified limitations, including insufficient conversational complexity and limited diversity of accents.

Discussion:

The findings suggest that AI can function as a valuable complementary tool to traditional language instruction by supporting personalized practice and fostering positive affective experiences. Nevertheless, concerns regarding over-reliance on AI and restricted linguistic exposure indicate the importance of balanced and pedagogically guided integration. Future research should examine long-term outcomes, broader learner populations, and strategies for optimizing AI-supported language learning environments.

1 Introduction

In language education, AI-powered tools such as chatbots, personalized learning platforms, and natural language processing applications have recently gained prominence. These technologies offer learners opportunities for real-time interaction, immediate feedback, and adaptive learning, which were previously unattainable in traditional settings (Chassignol et al., 2018). AI's potential to simulate conversational exchanges, assess grammatical accuracy, and provide pronunciation assistance allows language learners to practice autonomously while benefiting from individualized support.

Positive psychology, a relatively new field dedicated to studying human strengths and well-being, has received considerable attention in educational research over the past two decades (Seligman and Csikszentmihalyi, 2000). Its principles, including fostering positive emotions, building resilience, and enhancing motivation are particularly relevant in language learning, where students often face challenges such as anxiety, low self-confidence, and fear of making mistakes. In the context of language learning, the principles of positive psychology aim to create a supportive and emotionally enriching learning environment (MacIntyre et al., 2019; MacIntyre and Mercer, 2014).

AI technologies can serve as powerful tools to implement the principles of positive psychology in language classrooms. For instance, AI platforms can provide personalized encouragement, foster goal-setting, and adapt feedback to learners' emotional and cognitive states, creating a more nurturing environment (Dewaele et al., 2018). AI-driven analytics can also help educators identify students’ strengths and areas for improvement, enabling targeted interventions that promote growth and resilience. Moreover, AI's capacity for self-regulation, dynamic assessment and interactive learning can enhance students' intrinsic motivation, aligning with positive psychology's emphasis on cultivating enjoyment and engagement in the learning process (Csikszentmihalyi, 1990).

However, despite growing interest in AI-supported language learning, limited research has examined how such tools simultaneously shape both language proficiency and learners' affective experiences within specific classroom contexts. The present study aims to examine the role of an AI-powered platform in supporting English language proficiency development. Conducted within the framework of positive psychology, this mixed-method classroom-based study not only investigates the outcomes of AI integration in terms of students' achievement but also explores how it influences various aspects of students' emotional state. In addition, the study examines learners' perception of the advantages and limitations of using of AI for language learning. By addressing both the cognitive and affective dimensions of language learning, the study aims to provide an evaluative account of how AI technologies can contribute to fostering proficiency and well-being simultaneously. In doing so, the study offers a contextually grounded account of how AI may support both linguistic development and learner experience.

2 Review of the literature

2.1 AI and second language proficiency

AI-driven applications such as intelligent tutoring systems, speech recognition tools, and adaptive learning platforms have been shown to support vocabulary acquisition (Chen et al., 2021b), improve pronunciation and fluency (Thi-Nhu Ngo et al., 2024), and enhance writing skills through automated feedback (Lv et al., 2021). Taken together, these studies suggest that AI is most consistently effective when it provides frequent practice, adaptive feedback, and opportunities for individualized engagement with language. Researchers have also highlighted the role of AI in fostering listening comprehension through interactive audio-based platforms that provide instant feedback (Golonka et al., 2014). Despite these advances, some studies caution against dependence on AI tools, emphasizing that their efficacy often depends on their design and implementation (Chen et al., 2022).

AI systems also facilitate collaborative language learning by integrating features that support interaction and communication among learners. Platforms employing AI chatbots, for example, enable learners to engage in simulated conversational practice, helping them improve not only grammatical accuracy but also pragmatic competence (Nazeer et al., 2024; Sadikovna et al., 2024). Furthermore, virtual reality environments augmented by AI have emerged as a promising innovation for L2 learning. These immersive spaces allow learners to participate in realistic language scenarios, fostering the development of communicative competence in contextually rich settings (Parmaxi, 2020; Qiu et al., 2024). Research has shown that such AI-enhanced experiences can significantly reduce language anxiety and increase learners' confidence in speaking (Godwin-Jones, 2021). However, the evidence remains uneven across language skills, with oral communication and affective outcomes often discussed separately rather than examined in relation to one another.

Advanced AI systems often require significant financial investment and technological infrastructure, which may not be available in under-resourced contexts (Eynon and Malmberg, 2020). Furthermore, while these tools are designed to complement traditional instruction, they cannot fully replace the nuanced guidance and cultural sensitivity provided by human educators (Celik et al., 2022). As the field continues to evolve, fostering collaboration between educators, technologists, and researchers is essential to ensure that AI applications are both effective and inclusive.

2.2 Perceptions of AI in language learning

Students' perceptions of AI tools for language learning are generally positive but nuanced. Several studies (e.g., Cai et al., 2023; Godwin-Jones, 2021) report that learners appreciate the flexibility and accessibility of AI platforms, which allow them to practice autonomously and at their own pace. Additionally, AI-driven gamified applications have been identified as effective in increasing learners' motivation and engagement, making language learning more interactive and enjoyable (Al-Dosakee and Ozdamli, 2021). However, students also express concerns about the limited contextual understanding and rigidity of some AI systems, which can lead to repetitive or simplistic feedback (Mun, 2024; Petrov, et al., 2024). This tension between accessibility and pedagogical limitation suggests that learner perceptions of AI are shaped not only by convenience, but also by the perceived authenticity and usefulness of interaction.

When focusing on cognitive aspects, students often highlight the role of AI in improving their linguistic accuracy and understanding of complex language structures. For example, automated writing evaluators like Grammarly and Criterion are noted for their ability to identify grammatical errors and provide corrective feedback, which helps students refine their written communication skills (Lv et al., 2021). Moreover, AI tools that incorporate natural language processing can enhance reading comprehension by offering instant definitions, translations, and contextual explanations of unfamiliar vocabulary (Chen et al., 2022). Research also indicates that learners value the use of adaptive learning platforms, such as Duolingo and Lingvist, which use spaced repetition algorithms to optimize vocabulary retention and reinforce grammar rules (Kaur et al., 2023). However, some experts (e.g., Cai et al., 2023) have warned against reliance on AI feedback, as it does not provide opportunities for developing critical thinking and problem-solving skills.

Nevertheless, concerns about technical challenges, cost, and the potential for over-dependence on technology persist. Importantly, studies have pointed to the need for adequate training and support to help teachers effectively integrate AI tools into their pedagogical practices (Asiri, 2019). In sum, while both learners and teachers recognize the cognitive benefits of AI tools, they advocate for their integration into a well-rounded pedagogical approach that balances automated feedback with meaningful human interaction.

Despite the growing body of research, significant gaps remain in understanding the diverse effects of AI on L2 learning. Existing studies have largely focused on specific skills or isolated applications, offering limited insight into how AI platforms affect overall language proficiency. Furthermore, the majority of these studies have been conducted in Western societies or East Asian contexts, where technological infrastructure is advanced, and attitudes toward AI and virtual tools are generally favorable (Chen et al., 2022). However, findings from these regions are often generalized universally, overlooking cultural and contextual factors that may influence the efficacy and acceptance of AI tools. In developing countries such as Iran, access to AI platforms may be constrained by financial or technological limitations, and attitudes toward the virtual world may vary significantly due to cultural or generational differences (Eynon and Malmberg, 2020). Investigating AI integration in such underrepresented contexts can offer novel insights into its adaptability and limitations, as well as strategies for implementation. Classroom-based studies are particularly valuable in this regard because they allow researchers to examine how AI functions within actual pedagogical routines rather than in isolated or decontextualized tasks.

Moreover, while the cognitive dimensions of language learning have been extensively studied, affective and motivational aspects—essential for sustained engagement and success—remain largely underexplored. Most research has emphasized AI's ability to enhance linguistic accuracy, fluency, and comprehension through advanced feedback mechanisms and adaptive learning systems (Chen et al., 2021a; Thi-Nhu Ngo et al., 2024). However, only a few studies have examined how AI tools influence learners' emotional engagement, confidence, and intrinsic motivation (Mun, 2024). For example, Mun (2024) and Petrov et al. (2024) noted that AI platforms could increase learner motivation, but their findings were not widely replicated. Similarly, Zheng (2024) observed reduced anxiety in AI-driven reading practice, but these studies often lack longitudinal data to assess the sustainability of such benefits. Addressing this imbalance by incorporating affective dimensions into research frameworks could lead to a more comprehensive understanding of AI's impact on L2 education.

2.3 The need for studies framed within positive psychology

To address these gaps, further research is required to evaluate the effects of AI integration on students' language proficiency comprehensively, incorporating both cognitive and affective dimensions. Studies grounded in the principles of positive psychology can be particularly informative in this regard, as they emphasize fostering learners' well-being, motivation, and resilience (MacIntyre et al., 2019; MacIntyre and Mercer, 2014). Such a perspective is especially relevant when evaluating AI-mediated learning, since the value of these tools cannot be judged solely by performance outcomes but must also be understood in relation to learners’ emotional experience. By integrating AI with pedagogical strategies that prioritize learners' psychological needs, educators can create environments that promote both skill acquisition and emotional growth. For example, AI platforms designed to provide not only error correction but also positive reinforcement and encouragement could align with positive psychology's focus on promoting a growth mindset (Dweck, 2006). Relying on the principles of positive psychology, researchers can employ innovative pedagogical strategies that maximize the potential of AI not only as a technical aid but also as a catalyst for learners’ holistic development. Such an approach could bridge existing research gaps, offering a more detailed understanding of AI's role in supporting language learning across diverse educational contexts.

Guided by the principles of positive psychology, the present study examines how the Pi platform affects students' emotional states, including their motivation, anxiety, enjoyment, and overall engagement in learning. By exploring these affective dimensions, the study seeks to address gaps in existing literature that often overlook the emotional and psychological impacts of AI tools. Moreover, the research provides a balanced account of the students' perceptions of the platform, including its advantages and disadvantages, as derived from their direct experiences. This evaluative approach contributes valuable insights into the potential of AI technologies to support both cognitive and affective aspects of language learning in similar educational contexts. In this study, “joy” refers to learners' experience of enjoyment and intrinsic satisfaction during language learning activities, as conceptualized within positive psychology.

3 Method

This study adopts a mixed-method design to capture both observable performance trends and the affective dimensions of learner experience, integrating qualitative data (narratives, interviews, focus group discussions, and observations) with quantitative indicators (midterm and final exam scores).

3.1 Setting and participants

The study was conducted in an upper-intermediate English class in the context of a private language institute in Iran. The participants included 16 students aged 17 to 26 years old, all from middle-class backgrounds. Most participants were university students aiming to improve their English skills for academic or occupational purposes. To ensure homogeneity in language proficiency, the institute administered a standardized proficiency test before enrollment, confirming that all participants were at the same level of proficiency. The class followed the curriculum outlined in Summit 2, a widely used textbook at this level, and was structured as an intensive course spanning approximately three months. Over the semester, students attended 35 sessions, each lasting 90 min. Classes were scheduled three times a week, providing consistent and immersive exposure to English.

The course aimed to develop all four language skills—listening, speaking, reading, and writing—with a particular emphasis on oral skills. Speaking and listening activities were thus prioritized, with more than 50 min of each session dedicated to these areas. Activities included guided discussions, role-plays, and listening exercises that required students to comprehend and respond to spoken English in real-time. Reading was another significant focus of the course. Each week, students were introduced to one reading text, typically ranging from 400 to 600 words in length. These texts were selected to align with the students' proficiency level and to expose them to a variety of topics and vocabulary. Reading activities involved comprehension questions, vocabulary building, and group discussions based on the texts. Writing, although part of the curriculum, was primarily assigned as homework. Students were given structured prompts to practice composing paragraphs or short essays, with periodic feedback provided during class. The intensive structure and balanced curriculum provided a suitable context for exploring the potential of the Pi AI-powered platform to support language development, particularly in enhancing oral skills.

3.2 Description of the AI tool: Pi, your personal AI

The study employed Pi, Your Personal AI, an advanced AI-powered platform designed to have one-to-one communication through interactive, real-time prompts and responses. Although not originally developed to offer pedagogical services, the tool leverages conversational AI technology to provide personalized and adaptive interactions, making it particularly suitable for developing students' speaking and listening skills. In this study, Pi was introduced as an educational aid to support English language learning, focusing on improving students' oral communication abilities.

At the beginning of the semester, Pi was presented to the students and the participants were instructed to install the application on their mobile devices, ensuring accessibility for both in-class and out-of-class activities. In the classroom, Pi was primarily used as a preparatory tool for oral communication activities. For example, before engaging in face-to-face conversations or role-play tasks with their peers, students practiced speaking and listening using Pi. This initial interaction helped build their confidence, refine their responses, and provided opportunities to use the linguistic structures and vocabulary relevant to the upcoming activities. Each student used the tool individually on their mobile phones, which enabled personalized feedback and autonomous practice within the classroom setting. Outside the classroom, students were encouraged to use Pi to further enhance their communicative competence. The platform provided an opportunity for extended practice, enabling students to engage in conversations at their convenience. This dual use of Pi in- and out-of-class allowed for a comprehensive evaluation of its impact on students’ L2 development, as well as their perceptions of AI-mediated language practice. In practice, students used the platform during approximately 2–3 sessions per week for short activities (5–10 min per session). Out-of-class use was encouraged but remained voluntary. Engagement was monitored through classroom observation and student self-reports rather than system-generated usage logs.

3.3 Data collection procedure

To gain a comprehensive understanding of the effects of the Pi AI-powered platform, data were collected through multiple sources including narrative frames, individual interviews, focus group discussion, students' midterm and final exam scores, and teacher observations.

3.3.1 Narrative frames

3.3.2 Individual interviews

Midway through the semester, between sessions 17 and 18, individual semi-structured interviews were conducted with all 16 students to delve deeper into their emotional states and perceptions of the Pi platform. The interviews were specifically designed to explore affective factors such as anxiety, motivation, enjoyment, and self-confidence, as these elements play a pivotal role in sustaining engagement and fostering effective language learning. Students were encouraged to reflect on how their use of the Pi platform influenced these dimensions and to provide detailed, personal descriptions of their experiences.

3.3.3 Focus group discussion

3.3.4 Students' exam scores and teacher observations

Two complementary data sources – students’ scores on their midterm and final exam and the instructor's observational notes – were utilized to assess language development and provide contextualized insights into the impact of the Pi platform. The midterm and final exams, designed and administered by the language institute, assessed students' proficiency across all four language skills. These exams included both oral and written components, ensuring a comprehensive evaluation of students' abilities. Scores were assigned by the instructor following standardized assessment criteria used by the institute. Comparing the midterm and final scores allowed for an analysis of students' progress over the semester and provided quantitative evidence of their language development.

In addition to exam scores, the instructor maintained systematic observational notes throughout the semester. As recommended in the literature on classroom-based qualitative research (Creswell, 2013; Merriam and Tisdell, 2015), these notes documented key aspects of classroom interactions, including students’ engagement with the Pi platform, their participation in speaking and listening activities, and their overall attitudes toward learning. Observational data were mainly collected after class sessions to capture spontaneous behaviors, patterns of interaction, and notable events that reflected students' responses to the AI tool and its integration into the curriculum.

This approach is widely recommended in the literature for exploring relatively unexplored domains, as it enhances the validity and reliability of findings through triangulation (Creswell, 2013; Denzin, 2012).

3.4 Data analysis

The narrative frames, interview transcripts, focus group discussions, exam scores, and observational notes were systematically reviewed and coded to identify recurring themes and patterns relevant to the research questions. A thematic analysis approach was initially employed, allowing for the extraction of key insights while remaining sensitive to the differences across individual experiences. To ensure accuracy and credibility, data were cross-verified through the integration of findings from multiple sources (Denzin, 2012).

To ensure a rigorous and transparent analytical process, the qualitative data from narrative frames, interviews, and focus group transcripts were analyzed using a systematic thematic analysis approach (Braun and Clarke, 2006). The analysis was conducted iteratively in several stages.

First, an initial round of open coding was performed on all transcripts to generate descriptive codes that captured key concepts in the participants' responses. This stage produced 48 distinct open codes. Subsequently, these initial codes were reviewed, compared, and organized into broader, more conceptual categories through axial coding. This process resulted in 10 axial codes that grouped related ideas. Finally, by examining the relationships between these axial codes and reflecting on the research questions, four overarching themes were identified that encapsulated the core findings related to students' affective experiences (RQ2) and their evaluations of the platform (RQ3).

To illustrate this process and enhance transparency, Table 2 summarizes the evolution of codes for the key themes related to RQ2 and RQ3, providing examples of raw data that informed each stage of analysis.

This structured approach to thematic analysis, coupled with constant comparison and peer debriefing sessions with a colleague familiar with qualitative methodology, ensured the credibility and confirmability of the findings (Guba & Lincoln, 1994).

In the results presented below, direct quotes from the students are included in order to vividly illustrate the participants' perceptions. The findings are organized in accordance with the research questions, emphasizing both cognitive and emotional dimensions of the participants' engagement with the AI tool.

In addition, quantitative data from midterm and final exam scores were analyzed using descriptive statistics (means and standard deviations) to examine changes between midterm and final scores; however, given the small sample size, these results are interpreted cautiously as exploratory. The instructor assumed multiple roles in this study, including teacher, data collector, and focus group moderator. While this is common in classroom-based research, it introduces potential bias. To enhance credibility, several measures were adopted, including anonymization of participant data, systematic coding procedures, and peer debriefing with a colleague familiar with qualitative research.

4 Findings

4.1 The effect of Pi on students' L2 proficiency

To address the first research question, students' midterm and final exam scores were analyzed as indicators of their language proficiency development over the semester. Analysis of the exam scores revealed, the average scores for both exams were calculated, revealing relatively high overall performance compared to the average scores of similar courses at the institute. Notably, the scores for speaking and listening skills were consistently higher than those for reading and writing. This can be partially attributed to the course design, which prioritized the development of oral skills. However, the enhanced performance in these areas also suggests a positive association between AI-supported practice and improvements in these areas, although causal relationships cannot be established within the present design (Table 1 summarizes the data collection methods).

The data from other sources corroborated these findings, offering rich qualitative insights that support the observed trends in exam performance. In their narrative responses, students frequently emphasized how their interaction with Pi improved their ability to engage in L2 communication. For instance, one student reflected, “Using Pi helped me to feel more confident, especially in speaking…because it was like practicing with a partner who never gets tired and that who never judges me…I could repeat sentences till I got them right without feeling shy.” Another student noted, “It's like you have [an] English-speaking friend who is always there to help you. This made me use English more outside the class, and I think [it] really helped me.”

One participant highlighted its role in “training [them] to think and respond quickly in English,” adding that “it was great for learning to prepare for unexpected questions, by giving some sentences that can be always used in different situations…which helped me a lot during face-to-face discussions in class.”. In addition to building fluency, some students pointed out that Pi helped them identify specific areas for improvement. As one explained, “The instant feedback on pronunciation was important for me. Because I understood how I was mispronouncing certain words, especially the stress of words, and corrected them over time.” These comments suggest that students recognized the platform not only as a practice tool but also as a valuable resource for self-assessment and targeted improvement.

The convergence of multiple data sources regarding the impact of Pi on students' L2 proficiency suggests the potential of AI tools to facilitate targeted skill development, particularly in areas where students need real-time practice and personalized feedback.

4.2 The impact of Pi on students' emotional state

The results related to the second research question, exploring students' perceptions of the Pi AI-powered platform's impact on their motivation, anxiety, joy, and engagement, revealed an overwhelmingly positive response, as reflected in the individual interviews and in the instructor's observational notes. Students frequently emphasized how the platform significantly reduced their anxiety, particularly by serving as a preparatory tool for in-class face-to-face conversations. Participant 3 explained, “Pi was like a rehearsal stage for me. By practicing with it, I felt more confident about what to say in class and less worried about making mistakes in front of others.” Another remarked, “The more I practiced with Pi, the more I got prepared for speaking [tasks]. It made me feel less nervous because I already had an idea of what to expect.”

Students also noted that while initially, they found Pi's accent difficult to follow, they quickly got used to it, which further contributed to reducing their anxiety. One participant shared, “There is an option for choosing your preferred voice and accent, but among those available, it was strange to listen to Pi's rather mechanical language…after a couple of hours working with it, I got used to it, and this familiarity made me more comfortable using English overall.” This predictability in interaction helped them focus on refining their responses rather than grappling with uncertainty, which in turn, played a key role in alleviating their speaking-related anxiety. These patterns can be tentatively interpreted through concepts drawn from positive psychology and sociocultural theory. The students' experience of reduced anxiety and increased willingness to experiment aligns with Fredrickson's (2001) broaden-and-build theory, which posits that positive emotions broaden individuals' cognitive and behavioral repertoires, building lasting personal resources. Additionally, the high levels of engagement and enjoyment reported by students resonate with Csikszentmihalyi's (1990) concept of flow, where learners become fully immersed in an optimally challenging and rewarding activity—a state closely linked to intrinsic motivation in language learning (MacIntyre and Gregersen, 2012).

The creative engagement observed, particularly through identity exploration, connects strongly to sociocultural perspectives on language learning. For instance, the student who adopted a “Canadian persona” illustrates Norton's (2013) concept of imagined identities, where learners use symbolic tools and resources to access imagined communities and empower their L2 selves. This AI-mediated role-play served as a form of mediated action, where the technology acted as a tool (mediator) that expanded the learners' capacity to experiment with new identities and social contexts, thereby enhancing agency and motivation.

Beyond reducing anxiety, students asserted that the use of Pi made the class more enjoyable compared to traditional language classes. Several participants remarked on how the interactive and novel nature of the AI platform motivated them to engage more actively, both during class and at home. One student stated, “The class was far more exciting than the usual classes. Using Pi was both like a break, a sort of playing a game through which I was learning without feeling the pressure of studying.” Another noted, “Pi pushed me to practice more often…even outside of class. Now that the course is over, I use it whenever I have some free time, and I enjoy talking about random topics, such as cryptocurrency.”

Interestingly, one student shared a unique perspective on their interactions with the Pi AI-powered platform, explaining how he used it as an opportunity to adopt a new identity. This student explained, “When I talk with Pi, I use a different name, and I’ve created a whole new personality for myself…I have a different job and even a different nationality, Canadian. It makes learning like a fun role-play game, and it helps me imagine using the language in real-world scenarios. This creative approach not only added an element of enjoyment to his learning experience but also seemed to foster greater engagement and motivation. By adopting this new identity, the student was able to immerse himself in different contexts and practice language use in various contexts, enhancing his self-confidence and language fluency.

In line with previous studies, these findings point to the potential of AI tools to enhance learners' engagement, motivation, and enjoyment through providing innovative and learner-centered experiences (e.g., Chen et al., 2022; Zheng, 2024). The platform's adaptability and low-pressure practice environment encouraged students to experiment with language use, fostering a sense of accomplishment and increasing their willingness to participate in learning activities.

Although the overall perceptions of the Pi AI-powered platform were positive, some students expressed reservations about its extensive use. In their interviews, some students expressed their concerns about becoming overly reliant on the tool, fearing that frequent use might lead to what they described as “addicted reliance.” One student noted, “I don't have any problems with using Pi—it's helpful—but I’m worried I might get dependent on it too much and lose the confidence to speak without it.” Another interviewee added, “It's a good tool, but sometimes I need to work [communicate] with real people, not just an AI, so I don't become socially isolated.”

Two further students similarly commented on their preference for human interactions, emphasizing the unique benefits of conversing with classmates and the instructor. One participant stated, “Talking with my classmates is more enjoyable because it is natural and spontaneous…We laugh together, share ideas, and learn from each other in a way that Pi can't.” Another student remarked, “Engaging in conversations with the teacher is more productive for me. The instructor can give detailed feedback and explain things in a way Pi can't because you know when you hear something from a person you remember much better than when you hear it from a programmed machine.” These concerns highlight the need to balance the use of AI tools with traditional classroom interactions, ensuring that students benefit from diverse forms of engagement. While Pi proved to be a valuable supplementary tool, these comments emphasize the irreplaceable value of human interaction in fostering social and collaborative dimensions of language learning.

4.3 Students' overall perception of Pi

In response to the third research question, students' evaluations of the Pi platform revealed a mix of positive and negative opinions. The positive feedback primarily highlighted the flexibility and benefits the platform offered for language learning. Several students expressed appreciation for the opportunity to practice speaking at any time. One participant in the focus group stated, “Pi is always available…whenever I want, and it's a great way to improve my speaking without feeling pressed by time.” This constant availability was particularly valued by students who found it difficult to engage in face-to-face conversations outside of class. Another student shared in the discussion, “I like it cause I can choose what to talk about…It makes [my] speaking [experience] more enjoyable cause I can practice topics that interest me rather than just focusing on the lessons.”

Moreover, students emphasized the customized learning experience provided by Pi. The platform's ability to adapt to their proficiency level and offer personalized feedback was frequently mentioned. One student reflected in the focus group, “Sometimes it is like it [Pi] knows exactly what I need [to work on], and asks me questions that are right for me, it gives me sentences that are useful…like a teacher who is always ready to guide me.” Another comment from a narrative emphasized the sense of autonomy, with one student noting, “I don't have to wait for the teacher to give me feedback. Pi is like a knowledgeable person that always gives me support.”

A further advantage frequently highlighted by the participants was the non-judgmental and low-pressure environment created by Pi. Many appreciated the fact that they could speak freely without the fear of being judged. As one focus group participant put it, “When I use it, [it]'s not I have to worry about my mistakes in front of others. With Pi, I feel like I can make mistakes and learn from them without feeling embarrassed.” Another participant echoed this sentiment, stating, “Pi never gets tired, and it's always there to help. You have an unlimited conversation with a partner who never judges you.” These aspects of the AI tool were particularly valued by students who had struggled with language anxiety in traditional settings.

On the other hand, several students raised concerns about the limitations of the Pi AI-powered platform, particularly in terms of its repetitiveness and the lack of real-life communication complexities. One common criticism was that the interactions with the AI felt monotonous and lacked variety, which made the experience less engaging. A participant in the focus group noted, “The conversations with Pi are always similar. It's the same feedback…and the same types of responses. It gets boring after a short time [because] there's no real variation.” Another student commented that, “It's like going through the same exercises over and over again, and it doesn't feel like real communication.”

Another concern raised by students was the absence of the complexities and unpredictability found in real-world conversations, such as background noise, slang, or spontaneous mistakes. As one participant observed, “In real life, conversations are not always clear. There's noise, and people speak in different ways, and they make mistakes, even natives. Pi doesn't have that. Everything feels too perfect and clean.” The lack of exposure to diverse accents also came up as a limitation.

Additionally, students expressed doubts about the long-term effectiveness of the instruction provided by Pi. Several participants indicated that while the tool might offer useful practice in the short term, the feedback it generates is often not memorable. One student stated, “The responses from Pi don't stick with me. I’m given instructions, but I can't sink in, and [I] forget them quickly.” Another student remarked, “It's just following a fixed path with no real joke, something that makes learning interesting…It doesn't feel like a real, and dynamic conversation.” This sentiment reflects the perception that the learning experience provided by Pi lacks the depth and humanistic qualities of natural, unscripted communication. The predictable nature of the AI's responses also detracts from its potential for fostering meaningful, long-term language acquisition.

5 Discussion and conclusions

The findings of this study suggest that the use of the Pi AI-powered platform was associated with improvements in students' English language proficiency, particularly in speaking and listening skills. These results are in line with prior research emphasizing the value of AI-assisted tools in language learning (e.g., Mohamed, 2024; Zhai and Wibowo, 2023). The design of the course, which devoted substantial time to oral skills, likely amplified these effects. However, the platform's ability to provide non-judgmental, personalized, and consistent practice opportunities stands out as a key facilitator of improved performance, addressing challenges noted in traditional language instruction such as limited practice time and performance anxiety (Chen, 2022).

In discussing these findings within the framework of positive psychology, the reduction in students' anxiety and the boost in motivation and enjoyment support the pedagogical use of Pi. Pi's role as a preparatory tool was particularly valued, with students expressing how practicing with it helped them feel more confident during in-class interactions. This reduction in anxiety can be understood within the framework of positive psychology, which highlights the importance of creating low-pressure environments to foster emotional well-being and improve learning outcomes (MacIntyre and Gregersen, 2012). The predictability of the AI's interactions also contributed to a sense of control, enabling students to focus on refining their language skills rather than dealing with unpredictable communication challenges.

In addition to its emotional benefits, the participants highlighted the motivational aspects of using Pi. Many found the experience of engaging with AI more enjoyable and interactive than traditional classes, which aligns with studies that emphasize the novelty and gamified aspects of AI platforms in sustaining learner interest (Dewaele et al., 2018). Furthermore, some students demonstrated creative engagement by adopting alternate personas during their interactions with Pi, a phenomenon that resonates with sociocultural theories emphasizing the role of imagined identities in language learning (Norton, 2013). This innovative use of AI allowed students to explore diverse linguistic and cultural contexts, enhancing their sense of agency and imagination. Such findings highlight Pi's potential to serve as a platform for personalized and learner-driven exploration, particularly in contexts like Iran where access to authentic English communication is limited.

Despite these strengths, some students raised serious concerns about the excessive use of Pi, highlighting the risk of becoming overly reliant on AI and the absence of humanistic qualities in its interactions. These reservations echo critiques in the literature, which caution against the potential for AI tools to oversimplify the complexities of real-world communication (Zhai and Wibowo, 2023). Furthermore, students expressed a preference for the richness of human interactions, which provide emotional depth and spontaneous exchanges that AI cannot replicate. These comments reinforce the importance of maintaining a balance between AI-based learning and traditional classroom practices. While Pi proved to be an effective supplementary tool, the findings also point to the irreplaceable value of human engagement in fostering social, emotional, and collaborative dimensions of language learning.

On the positive side, students emphasized Pi's flexibility, personalized support, and ability to provide a non-judgmental practice environment.

Despite these advantages, students also mentioned significant limitations of the platform, particularly in replicating the complexities of real-life communication. They critiqued Pi's interactions as repetitive, overly structured, and lacking the variability of authentic language use. These observations resonate with critiques by Zhai and Wibowo (2023) who argue that while AI systems can simulate conversation, they often fail to capture the spontaneity and richness of human interaction. Students' concerns about the absence of features like background noise, use of slang, and diverse accents highlight the platform's limited capacity to prepare learners for the sociolinguistic challenges of real-world communication. This is consistent with sociocultural perspectives that emphasize the importance of exposing learners to authentic, diverse language inputs to develop comprehensive communicative competence (Canagarajah, 2013).

Furthermore, concerns about the long-term effectiveness of Pi's feedback and its predictable nature suggest that students viewed the platform as a useful but superficial tool for language practice. Along with existing research (e.g., Godwin-Jones, 2022) these findings criticize AI-based platforms for their heavy reliance on algorithmic patterns that lack the depth needed for sustainable learning. Students' preference for dynamic human interactions over AI-mediated practice point to the irreplaceable value of social engagement in language learning, as emphasized by Vygotskian theories of learning through collaboration and interaction (Lantolf et al., 2015). These insights suggest that while Pi offers valuable support, it cannot fully substitute for the context-rich feedback provided by human instructors and peers.

The findings also hold contextual significance for Iranian EFL learners, many of whom face limited exposure to English-speaking environments outside the classroom. Pi's 24/7 availability enabled students to practice more extensively, overcoming contextual barriers such as limited access to native speakers or immersive language experiences. The platform's role as a bridge between classroom instruction and independent learning is noteworthy, providing students with both structured and spontaneous opportunities to engage with the language. Nevertheless, as some of the students noted, over-reliance on AI may create new challenges, such as reduced interaction with human interlocutors, social isolation or a potential dependence on technology. This reflects the previous findings by Cai et al. (2023) who warned that reliance on AI feedback may reduce opportunities for developing critical thinking and problem-solving skills, as AI systems tend to prioritize efficiency over deeper cognitive engagement. Consistent with the principles of blended learning approaches, these findings highlight the need for balanced integration of AI tools, ensuring they complement rather than replace human-centered learning practices.

By addressing students' linguistic and affective needs, Pi provided an accessible, adaptive, and anxiety-free environment that empowered learners to practice language skills more autonomously. However, while the platform contributed to personalized practice and reduction of language anxiety, it fell short in replicating the complexities of authentic communication and fostering deeper, long-term learning. These findings reaffirm the dual role of technology in education: as a powerful tool that complements traditional teaching methods, and as a tool with inherent restrictions that must be critically acknowledged and addressed. This balance is essential as language educators and researchers continue to explore innovative strategies for enhancing learner engagement and proficiency in diverse contexts.

First, the integration of AI tools such as Pi can significantly enhance the learning experience, particularly in EFL contexts where face-to-face interaction opportunities are limited. Educators should employ these tools to provide personalized practice opportunities, reduce learners' anxiety, and encourage consistent engagement outside the classroom. However, it is crucial to complement AI-mediated learning with rich, dynamic classroom interactions that foster authentic communication and sociocultural competence. Designing curricula that balance AI-supported activities with collaborative, human-centered approaches can maximize the benefits of both worlds.

Table 3 outlines specific classroom strategies derived from the findings of this study, designed to leverage the strengths of AI tools like Pi while safeguarding the irreplaceable value of human connection.

Implementing such a framework ensures that AI acts as a complementary tool within a socio-interactive pedagogy. Its use becomes purposeful—scaffolding, personalizing, and reflecting—rather than supplementary or substitutive, thereby mitigating risks of over-reliance while amplifying its benefits for affective and cognitive development.

Additionally, teacher training programs should include sessions on how to effectively integrate AI technologies into language teaching to address their strengths and limitations.

Finally, the limitations of the study need to be acknowledged. The findings of this study should be interpreted with caution. The small sample size (N = 16), the single institutional setting, and the focus on one proficiency level limit the generalizability of the results. In addition, affective constructs such as motivation, anxiety, and enjoyment were explored through qualitative self-report rather than validated psychometric instruments. Future research may benefit from larger samples, longitudinal designs, and the integration of standardized measurement tools. Therefore, this preliminary exploratory classroom-based study should be viewed as a starting point for future, more comprehensive research designs that can address these limitations, overcome the methodological constraints of the present study, and provide stronger and more generalizable evidence regarding the role of AI-supported language learning.

References

• 1

  Al-DosakeeK.OzdamliF. (2021). Gamification in teaching and learning languages: a systematic literature review. Rev. Rom. Educ. Multidimens.13 (2), 559–577. 10.18662/rrem/13.2/436

  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  AsiriM. J. (2019). Do teachers’ attitudes, perception of usefulness, and perceived social influences predict their behavioral intentions to use gamification in EFL classrooms? Evidence from the Middle East. Int. J. Educ. Pract.7 (3), 112–122. 10.18488/journal.61.2019.73.112.122

  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  BarkhuizenG. (2014). Revisiting narrative frames: an instrument for investigating language teaching and learning. System47, 12–27. 10.1016/j.system.2014.09.014

  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  BraunV.ClarkeV. (2006). Using thematic analysis in psychology. Qual. res. psych. 3 (2), 77–101.

  • Google Scholar

• 5

  CaiQ.LinY.YuZ. (2023). Factors influencing learner attitudes towards ChatGPT-assisted language learning in higher education. Int. J. Hum. Comput. Interact.40 (22), 7112–7126. 10.1080/10447318.2023.2261725

  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  CanagarajahS. (2013). Translingual Practice: Global Englishes and Cosmopolitan Relations. London: Routledge.

  • Google Scholar
  • View reference in article

• 7

  CelikI.DindarM.MuukkonenH.JärveläS. (2022). The promises and challenges of artificial intelligence for teachers: a systematic review of research. TechTrends66 (4), 616–630. 10.1007/s11528-022-00715-y

  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  ChassignolM.KhoroshavinA.KlimovaA.BilyatdinovaA. (2018). Artificial intelligence trends in education: a narrative overview. Procedia. Comput. Sci.136, 16–24. 10.1016/j.procs.2018.08.233

  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  ChenX.ZouD.ChengG.XieH. (2021a). “Artificial intelligence-assisted personalized language learning: systematic review and co-citation analysis,” in 2021 International Conference on Advanced Learning Technologies (ICALT), (IEEE), 241–245. 10.1109/ICALT52272.2021.00079

  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  ChenX.ZouD.XieH.ChengG. (2021b). Twenty years of personalized language learning. Educ. Technol. Soc.24 (1), 205–222.

  • Google Scholar
  • View reference in article

• 11

  ChenX.ZouD.XieH.ChengG.LiuC. (2022). Two decades of artificial intelligence in education. Educ. Technol. Soc.25 (1), 28–47.

  • Google Scholar
  • View reference in article

• 12

  ChenY. (2022). Effects of technology-enhanced language learning on reducing EFL learners’ public speaking anxiety. Comput. Assist. Lang. Learn.37 (4), 789–813. 10.1080/09588221.2022.2055083

  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  CreswellJ. W. (2013). Qualitative Inquiry & Research Design: Choosing among Five Approaches. USA: SAGE.

  • Google Scholar
  • View reference in article

• 14

  CsikszentmihalyiM. (1990). Flow–The Psychology of Optimal Experience. Harper & Row.

  • Google Scholar
  • View reference in article

• 15

  DenzinN. K. (2012). Triangulation 2.0*. J. Mix. Methods. Res.6 (2), 80–88. 10.1177/1558689812437186

  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  DewaeleJ.-M.WitneyJ.SaitoK.DewaeleL. (2018). Foreign language enjoyment and anxiety: the effect of teacher and learner variables. Lang. Teach. Res.22 (6), 676–697. 10.1177/1362168817692161

  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  DweckC. S. (2006). Mindset: The new Psychology of Success. Random House.

  • Google Scholar
  • View reference in article

• 18

  EynonR.MalmbergL. (2020). Lifelong learning and the internet: who benefits most from learning online?Br. J. Educ. Technol.52 (2), 569–583. 10.1111/bjet.13041

  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  FredricksonB. L. (2001). The role of positive emotions in positive psychology: The broaden-and-build theory of positive emotions. Am psychol. 56 (3), 218.

  • Pubmed Abstract
  • Google Scholar

• 20

  Godwin-JonesR. (2021). Evolving technologies for language learning. Lang. Learn. Technol.25 (3), 6–26. 10.64152/10125/44747

  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  Godwin-JonesR. (2022). Partnering with AI: Intelligent writing assistance and instructed language learning. Lang. Learn. Technol.26, pp. 5–24. 10.64152/10125/73474

  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  GolonkaE. M.BowlesA. R.FrankV. M.RichardsonD. L.FreynikS. (2014). Technologies for foreign language learning: a review of technology types and their effectiveness. Comput. Assist. Lang. Learn.27 (1), 70–105. 10.1080/09588221.2012.700315

  • CrossRef
  • Google Scholar
  • View reference in article

• 23

  GubaE. G.LincolnY. S. (1994). Competing paradigms in qualitative research. Handbook qual. res.2 (163-194), 105.

  • Google Scholar
  • View reference in article

• 24

  KaurP.KumarH.KaushalS. (2023). Technology-assisted language learning adaptive systems: a comprehensive review. Int. J. Cogn. Comput. Eng.4, 301–313. 10.1016/j.ijcce.2023.09.002

  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  KruegerR. A.CaseyM. A. (2015). “Focus group interviewing,” in Handbook of Practical Program Evaluation, eds. NewcomerK. E.HatryH. P.WholeyJ. S. (John Wiley & Sons), 506–534. 10.1002/9781119171386.ch20

  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  LvX.RenW.XieY. (2021). The effects of online feedback on ESL/EFL writing: a meta-analysis. Asia-Pac. Educ. Res.30 (6), 643–653. 10.1007/s40299-021-00594-6

  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  MacIntyreP. D.GregersenT. (2012). Emotions that facilitate language learning: the positive-broadening power of the imagination. Stud. Second Lang. Learn. Teach.2 (2), 193–213. 10.14746/ssllt.2012.2.2.4

  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  MacIntyreP. D.GregersenT.MercerS. (2019). Setting an agenda for positive psychology in SLA: theory, practice, and research. Mod. Lang. J.103 (1), 262–274. 10.1111/modl.12544

  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  MacIntyreP. D.MercerS. (2014). Introducing positive psychology to SLA. Stud. Second Lang. Learn. Teach.4 (2), 153–172. 10.14746/ssllt.2014.4.2.2

  • CrossRef
  • Google Scholar
  • View reference in article

• 30

  MerriamS. B.TisdellE. J. (2015). Qualitative Research: A Guide to Design and Implementation. John Wiley & Sons.

  • Google Scholar
  • View reference in article

• 31

  MohamedA. M. (2024). Exploring the potential of an AI-based chatbot (ChatGPT) in enhancing English as a foreign language (EFL) teaching: perceptions of EFL faculty members. Educ. Inf. Technol.29 (3), 3195–3217. 10.1007/s10639-023-11917-z

  • CrossRef
  • Google Scholar
  • View reference in article

• 32

  MunC. Y. (2024). EFL Learners’ English writing feedback and their perception of using ChatGPT. STEM J.25 (2), 26–39. 10.16875/stem.2024.25.2.26

  • CrossRef
  • Google Scholar
  • View reference in article

• 33

  NazeerI.KhanN. M.NawazA.RehmanJ. (2024). An experimental analysis of pragmatic competence in human-ChatGPT conversations. Pak. J. Humanit. Soc. Sci.12 (1), 424–435. 10.52131/pjhss.2024.v12i1.2061

  • CrossRef
  • Google Scholar
  • View reference in article

• 34

  NortonB. (2013). Identity and Language Learning: Extending the Conversation. 2nd ed.UK: Multilingual Matters. 10.21832/9781783090563

  • CrossRef
  • Google Scholar
  • View reference in article

• 35

  ParmaxiA. (2020). Virtual reality in language learning: a systematic review and implications for research and practice. Interact. Learn. Environ.31 (1), 172–184. 10.1080/10494820.2020.1765392

  • CrossRef
  • Google Scholar
  • View reference in article

• 36

  PetrovA.WuZ.MallahanJ.ElwoodK. (2024). “Examining college students’ perceptions of AI chatbots as a source of feedback in language learning,” in Society for Information Technology & Teacher Education International Conference (Association for the Advancement of Computing in Education (AACE)), 851–860.

  • Google Scholar

• 37

  QiuX. B.ShanC.YaoJ.FuQ. K. (2024). The effects of virtual reality on EFL learning: a meta-analysis. Educ. Inf. Technol.29 (2), 1379–1405. 10.1007/s10639-023-11738-0

  • CrossRef
  • Google Scholar

• 38

  SadikovnaS. G.MansurovnaV. F.SalixovnaR. D.XikmatullayevnaI. I.BobajanovB. (2024). The role of AI and web 2.0 tools in developing pragmatic competence of L2 English learners. SPAST Rep.1 (7), 10.69848/sreports.v1i7.5098

  • CrossRef
  • Google Scholar

• 39

  SeligmanM. E. P.CsikszentmihalyiM. (2000). Positive psychology: an introduction. Am. Psychol.55 (1), 5–14. 10.1037/0003-066X.55.1.5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 40

  Thi-Nhu NgoT.ChenH.-J.LaiK.-W. (2024). The effectiveness of automatic speech recognition in ESL/EFL pronunciation: a meta-analysis. ReCALL36 (1), 4–21. 10.1017/S0958344023000113

  • CrossRef
  • Google Scholar

• 41

  ZhaiC.WibowoS. (2023). A systematic review on artificial intelligence dialogue systems for enhancing English as foreign language students’ interactional competence in the university. Comput. Educ.: Artif. Intell.4, 100134. 10.1016/j.caeai.2023.100134

  • CrossRef
  • Google Scholar

• 42

  ZhengS. (2024). The effects of chatbot use on foreign language Reading anxiety and Reading performance among Chinese secondary school students. Comput. Educ.: Artif. Intell.7, 100271. 10.1016/j.caeai.2024.100271

  • CrossRef
  • Google Scholar