Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Humanities and Social Sciences Communications volume 12, Article number: 556 (2025)
23k
39
305
Metrics details
This study investigates the efficacy of an AI-enhanced reading platform in enhancing reading comprehension and influencing psychological variables such as motivation, anxiety, and cognitive load among second language (L2) learners. A mixed-methods approach was employed, combining quantitative analyses of reading comprehension scores and psychological measures with qualitative exploration through focus group discussions. A total of 300 participants were randomly assigned to either the experimental group, which used the AI platform, or the control group, which used traditional digital resources. Quantitative results revealed significant improvements in reading comprehension, motivation, anxiety reduction, and cognitive load management among participants using the AI platform compared to the control group. Effect sizes indicated large to moderate impacts of the intervention across all measured variables. Biometric data analysis, including heart rate variability (HRV) and eye-tracking metrics, provided objective evidence supporting the platform’s efficacy in reducing stress levels and promoting more efficient reading behaviors. Qualitative findings illuminated participants’ experiences, highlighting enhanced engagement, reduced anxiety, development of improved reading strategies, and appreciation for the user-friendly platform design. The study underscores the potential of AI-driven educational technologies to optimize language learning outcomes by providing personalized, adaptive, and supportive learning environments.
Reading comprehension is a critical skill for learners of a second language (L2), particularly for English as a Foreign Language (EFL) students. It involves complex cognitive processes that require the integration of vocabulary, grammar, and prior knowledge to construct meaning from text (Grabe & Stoller 2020; Jeon & Yamashita 2014). Despite its importance for academic success and everyday communication, many EFL learners struggle with developing this skill due to insufficient instructional support and the inherent challenges of navigating a second language. Vygotsky’s Zone of Proximal Development (ZPD) provides a theoretical framework for understanding how learners develop reading skills through guided interaction. ZPD refers to the gap between a learner’s current independent ability and their potential development level achievable through support from teachers or peers (Vygotsky 1978). It highlights the importance of scaffolding to bridge this gap (Dabarera et al. 2014; Mežek et al. 2022).
Artificial Intelligence (AI) offers transformative potential in language learning by providing personalized, adaptive instruction via tools like Intelligent Tutoring Systems and Natural Language Processing (Chen et al. 2020; Fathi et al. 2025; Lee & Hwang 2022). Research shows AI enhances vocabulary, grammar, and overall language skills (Kim 2019; Rad et al. 2023), while interactive features like chatbots and gamified platforms boost engagement and motivation (Divekar et al. 2021; Yin & Fathi 2025; Zou et al. 2023). Yet, challenges such as data privacy, infrastructure needs, and risks of over-reliance on AI persist, highlighting the importance of complementing human instruction (Pedró et al. 2019; Hooda et al. 2022; Wang et al. 2023).
This study addresses these gaps by examining the impact of an AI-enhanced reading platform on L2 reading comprehension among Chinese EFL learners. Specifically, it investigates how the integration of biometric feedback mechanisms, such as heart rate variability (HRV) sensors and eye-tracking technology, can further personalize and enhance the learning experience. The inclusion of biometric feedback is based on its ability to provide objective, real-time data on learners’ physiological and cognitive states. HRV, a well-established measure of autonomic nervous system activity, reflects stress levels and cognitive load, allowing the platform to detect when learners are experiencing excessive strain or disengagement. Similarly, eye-tracking technology offers granular insights into reading behavior, such as fixation patterns and regressions, which are indicative of comprehension challenges and cognitive processing demands. Together, these tools enable the platform to dynamically adjust instructional strategies and scaffold support, ensuring learners remain within their Zone of Proximal Development (ZPD). Previous research has demonstrated the effectiveness of biometric feedback in providing real-time insights into cognitive and emotional states, thereby informing adaptive instructional strategies (Shaffer & Ginsberg 2017; Rayner 1998). For instance, studies have shown that eye-tracking metrics, such as fixation duration and regressions, can accurately reflect levels of cognitive load during reading (Duchowski et al. 2018; Nasri et al. 2024). Similarly, research has demonstrated that HRV can be a reliable indicator of anxiety and stress in learning environments (Kim et al. 2018; Le et al. 2018).
The significance of this study lies in advancing understanding of how AI combined with biometric feedback can enhance L2 reading comprehension. By offering personalized learning that adapts to learners’ cognitive and emotional states, the platform aims to foster a more effective learning environment. This research contributes to the expanding literature on AI in education, offering practical insights into incorporating advanced technologies into language instruction. The novelty of the study is the real-time integration of biometric feedback with AI to personalize reading materials and instructional strategies, thereby providing a comprehensive approach to improving reading comprehension. Additionally, this research explores how these technologies affect learner motivation, anxiety, and cognitive load, clarifying both benefits and challenges.
The purpose of this study is to evaluate the efficacy of an AI-enhanced reading platform, equipped with biometric feedback mechanisms, in improving L2 reading comprehension among Chinese EFL learners. By integrating advanced AI technologies with biometric feedback, this study aims to fill the existing gap in personalized and adaptive learning experiences. The insights gained from this research will contribute to the development of effective instructional strategies for language learning and the practical application of AI and biometric technologies in educational settings.
Vygotsky’s Zone of Proximal Development (ZPD) underscores the role of targeted scaffolding in guiding learners from what they can do independently to higher levels of performance with assistance (Lantolf et al. 2014; Vygotsky 1978). In L2 reading, our AI platform operationalizes this principle by customizing the level of text difficulty and feedback in real time. For example, if eye-tracking data reveal repeated fixations, the platform may simplify passages or insert brief explanations, ensuring learners remain within their ZPD (Poehner & Lantolf 2010). Through adaptive scaffolding, learners receive support that fosters incremental skill gains without overwhelming them (Liu et al. 2024; Mežek et al. 2022; Dabarera et al. 2014).
Biometric data (e.g., HRV, eye-tracking) supplement ZPD-based support by indicating when a learner might be struggling or showing signs of cognitive overload (Kim et al. 2018; Duchowski et al. 2018). This allows the system to adjust reading tasks or provide motivational prompts as needed (Wang et al. 2023). Such adaptive responsiveness aligns with Vygotsky’s emphasis on continuous interaction between learner and instructor—here, translated into real-time AI assistance that guides comprehension and encourages self-regulation (Pan et al. 2024).
Cognitive Load Theory (CLT) further complements this approach by addressing how learners process information during reading (Sweller 1988; Paas et al. 2003). In our study, the AI platform helps manage intrinsic load by matching text complexity to individual proficiency and reduces extraneous load through clear explanations and minimal distractions. Feedback on reading progress is tailored to promote germane load, so learners can focus on building effective reading strategies (Feng 2025; Hudon et al. 2021). Biometric cues guide these adjustments, helping learners stay engaged without becoming overloaded. This alignment with CLT ensures sustained attention and efficient reading while also supporting incremental skill development (Sweller 1988). By integrating ZPD’s adaptive scaffolding with CLT’s cognitive load management, our study aims to enhance L2 reading comprehension while minimizing cognitive overload and anxiety, addressing key challenges in language learning.
Artificial Intelligence (AI) has introduced transformative advancements in language education, reshaping how learners engage with language skills and addressing diverse instructional, cognitive, and ethical dimensions (Liang et al. 2023; Wei 2023). By offering personalized learning opportunities, promoting conversational fluency, and supporting cognitive and emotional needs, AI tools are creating tailored and impactful educational experiences (Hsiao & Chang 2024; Xiu-Yi 2024; Xin & Derakhshan 2024).
One significant application of AI in language learning is through Intelligent Tutoring Systems (ITS), which use advanced algorithms to customize reading and writing exercises based on individual needs. These systems dynamically adjust difficulty levels and deliver instant feedback to ensure that learners receive guidance aligned with their specific abilities (Chen et al. 2020; Lee & Hwang 2022; Pan et al. 2024; Muthmainnah et al. 2024; Halkiopoulos & Gkintoni 2024). Incorporating NLP, ITS platforms go beyond identifying errors, providing detailed corrections and actionable suggestions to facilitate iterative learning (Shadiev & Feng 2023; Chen et al. 2021). Evidence underscores the effectiveness of these tools in enhancing vocabulary acquisition, grammatical accuracy, and overall fluency (Fathi & Rahimi 2024; Kim 2019; Rad et al. 2023; Chen et al. 2020). By responding to performance patterns and offering targeted scaffolding, these systems have demonstrated the ability to cater to diverse learner profiles and foster sustained improvements across skill areas.
In addition to personalized instruction, AI-powered conversational agents such as chatbots have emerged as versatile tools for improving speaking and listening skills (Fathi et al. 2024). These agents simulate realistic dialogs, enabling learners to practice in low-stress, interactive environments (Divekar et al. 2021; Zou et al. 2023). They dynamically interpret and respond to user input, thereby fostering communicative competence and enhancing confidence in oral language use (Rusmiyanto et al. 2023; Pokrivcakova 2019). Although learners frequently report gains in fluency, pronunciation, and self-assurance, the success of these systems depends on their alignment with user proficiency levels. Simplistic chatbots may fail to engage advanced learners, while overly complex interactions can overwhelm less proficient users, emphasizing the importance of designing systems that match learners’ needs (Xiu-Yi 2024; Wei 2023; Hsiao & Chang 2024).
AI tools also play a critical role in enhancing motivation and engagement by integrating adaptive content delivery, gamification elements, and real-time feedback mechanisms (Rahimi et al. 2024). Features such as progress tracking, point systems, and reward-based incentives create an engaging and goal-driven learning environment, encouraging active participation (Moybeka et al. 2023; Chiu et al. 2023). These tools not only maintain learner interest but also reduce anxiety by adjusting task difficulty to prevent feelings of overwhelm (Zheng 2024; Xin & Derakhshan 2024). However, systems with poor design or excessive automation can inadvertently cause frustration or disengagement, particularly when feedback is delayed or unclear (Xiu-Yi 2024). Addressing these limitations requires a focus on user-centered design and careful consideration of learners’ emotional and motivational needs to ensure a supportive and productive educational experience (Zheng, 2024; Moybeka et al. 2023; Chiu et al. 2023).
Cognitive engagement is another critical area where AI has demonstrated substantial benefits. By reducing extraneous cognitive load, well-designed AI platforms allow learners to focus on essential tasks, streamlining the learning process and enhancing comprehension (Sweller 1988; Klepsch et al. 2017; Wei 2023; Hsiao & Chang 2024). Beyond supporting foundational skills, AI tools show potential for fostering higher-order cognitive abilities such as critical thinking and problem-solving. Interactive and immersive environments that include scenario-based tasks or adaptive simulations encourage learners to analyze, synthesize, and evaluate information in meaningful ways (Xin & Derakhshan 2024; Liu & Fan 2025). These features highlight the capacity of AI to balance basic skill acquisition with the cultivation of complex cognitive processes, providing learners with opportunities for comprehensive intellectual growth (Liang et al. 2023; Zhai & Wibowo 2023; Law 2024).
Despite these advancements, the integration of AI into language education is accompanied by notable challenges. Issues such as data privacy, technological infrastructure, and the need for teacher training present significant barriers to effective implementation (Pedró et al. 2019; Hooda et al. 2022; Chen et al. 2020). Ensuring transparency and compliance with data protection standards is essential for building trust among users. Furthermore, ethical concerns surrounding plagiarism, misuse of AI tools, and inequities in access to technology underscore the importance of establishing clear guidelines and equitable practices (Dakakni & Safa 2023; Kartal & Yeşilyurt 2024; Huang et al. 2023). Over-reliance on automated tools could also hinder the development of self-regulated learning and meaningful teacher-learner interactions, highlighting the necessity of balancing technology with human facilitation (Wang et al. 2023; Valijärvi & Tarsoly 2019).
In summary, AI is revolutionizing language education by offering personalized learning experiences, supporting conversational fluency, and addressing cognitive and emotional dimensions of language learning. However, its successful adoption depends on strategic implementation, ethical oversight, and collaborative efforts among educators, researchers, and developers. Continued research is essential to refine AI tools and ensure their integration as effective, inclusive, and ethical complements to traditional language instruction (Zhai & Wibowo 2023; Law 2024; Dakakni & Safa 2023).
L2 reading comprehension is shaped by cognitive, linguistic, and instructional factors (Grabe & Stoller 2020; Jeon & Yamashita 2014). Among the cognitive factors, working memory (WM) is crucial, especially in processing language-based information. Research shows a positive correlation between WM capacity and reading proficiency (Jeon & Yamashita 2014; Linck et al. 2014), with verbal WM tasks more strongly linked to reading comprehension than non-verbal tasks (Shin 2020). This suggests that verbal tasks better reflect WM’s role in L2 reading (Linck et al. 2014). Vocabulary knowledge also plays a vital role, facilitating comprehension and inferencing (Gottardo & Mueller 2009; Grabe & Stoller 2020). A strong vocabulary allows learners to better understand texts and make inferences from context. Grammar further supports syntactic parsing and meaning construction, enhancing comprehension (Edele & Stanat 2016; Zhang & Zhang 2022). Together, vocabulary and grammar highlight the need for integrated language instruction that addresses both components in reading development (Droop & Verhoeven 2003).
Task-based instruction (TBI) has proven effective in improving L2 reading comprehension by engaging learners in meaningful tasks that promote communicative competence and reduce anxiety (Prabhu 1987). TBI encourages prior knowledge activation and collaborative learning, focusing on both meaning and form (Sidek 2012). The interdependence hypothesis (Cummins 1979, 1991) suggests that L1 proficiency, especially in phonological awareness and decoding, supports L2 reading development (Geva & Siegel 2000; Genesee & Geva 2006). This cross-linguistic transfer emphasizes the importance of bilingual literacy in L2 reading.
Technological innovations have enriched L2 reading instruction by offering tools that support both cognitive and affective aspects of learning. Mobile Assisted Language Learning (MALL) has been recognized as an effective way to enhance L2 reading comprehension, enabling more interactive and personalized learning (Gutiérrez-Colón et al. 2023). Similarly, computer-assisted language learning (CALL) glossing, where unfamiliar vocabulary is explained within the text, improves comprehension, especially when accompanied by visual aids (Taylor 2021). These tools promote comprehension and encourage independent reading (Liaw & English 2017). However, Peng et al. (2023) emphasize that the integration of technology in L2 reading requires careful attention to factors such as student autonomy, ICT competence, and effective teacher support.
Recent studies on AI in L2 reading show its potential to provide adaptive learning experiences. Zheng et al. (2024) explored an AI chatbot, “Reading Bot,” which reduced foreign language reading anxiety (FLRA) but did not significantly improve reading performance compared to traditional instruction. This suggests AI can address affective aspects of L2 reading, though its impact on cognitive skills requires further refinement. Similarly, Çelik et al. (2024) found that AI-driven simplification of authentic texts improved comprehension and inferencing but had no effect on reading anxiety. This highlights AI’s potential to enhance cognitive engagement with complex texts while underscoring the need for strategies to address the emotional aspects of reading. Broader reviews of AI in language learning further highlight its transformative potential. Jose and Jayaron Jose (2024) discuss how AI-supported tools like text simplifiers and real-time feedback mechanisms can make complex texts more accessible, promote student engagement, and improve reading outcomes across various contexts. While AI holds promise for enhancing L2 reading, further research is needed to optimize its use and address its limitations in both cognitive and affective domains.
In conclusion, the literature reveals the intersection of cognitive, linguistic, and technological factors in shaping L2 reading comprehension. While traditional instruction remains essential, technologies like AI, MALL, and CALL offer new opportunities for improving reading skills. Future research should focus on optimizing these technologies within pedagogical frameworks that address both cognitive and emotional aspects of learning. This will enable more effective and engaging L2 reading experiences.
Three hundred Chinese EFL learners were recruited from a university language program through flyers, online forum postings, and email invitations. The participants were all undergraduate students majoring in various fields, including computer science, international relations, and economics. They were predominantly homogeneous in terms of proficiency, classified as B1-B2 according to the Common European Framework of Reference for Languages (CEFR), to ensure a consistent baseline for evaluating the intervention effects. To confirm the B1-B2 proficiency levels and establish a baseline for measuring reading comprehension gains, all participants completed the reading component of the Preliminary English Test (PET), a standardized EFL proficiency test. This ensured that all participants met the inclusion criteria and that the experimental and control groups were comparable in their initial reading abilities. The age of the participants ranged from 18 to 35 years (M = 22.41, SD = 4.26), with the majority (78%) falling between the ages of 20 and 24. The sample consisted of 171 females (57%) and 129 males (43%).
Prior to the study, none of the participants had any previous experience with the Smart Sparrow platform. To ensure familiarity with the platform’s interface and functionalities, all participants underwent a standardized 30-min training session before commencing the intervention. This training covered basic navigation, accessing reading materials, and engaging with interactive exercises.
Participants were then randomly assigned to the experimental (n = 150) or control group (n = 150) using a computer-generated sequence, minimizing selection bias and enhancing internal validity. The experimental group used an AI-enhanced reading platform with advanced NLP algorithms and biometric feedback, while the control group engaged with traditional digital reading resources.
All participants provided written informed consent before the study, with a clear briefing on the study’s purpose, procedures, risks, and benefits. The consent process followed ethical guidelines and ensured participant rights, confidentiality, and the ability to withdraw at any time without consequence.
The experimental group used Smart Sparrow, an AI-driven platform chosen for its flexible course design and easy integration with biometric feedback tools. Unlike other reading platforms, Smart Sparrow offers fine-grained control over content and real-time adaptation based on user performance, making it well-suited for B1-B2 level Chinese EFL learners. Its compatibility with HRV sensors (Polar H10) and eye-tracking devices (Tobii Pro X3-120) allowed us to embed physiological data directly into adaptive reading tasks, enabling on-the-spot adjustments for text difficulty, additional hints, or stress-reduction prompts.
Key functionalities include advanced NLP for sentence complexity analysis, reinforcement learning algorithms to refine tasks based on learner interactions, and open API tools that simplify biometric data integration. This direct data flow between biometric inputs and the AI system was essential for examining how stress levels and reading behaviors influenced comprehension.
Before the intervention, participants took part in a 30-min training on basic platform navigation, device setup, and calibration for both HRV sensors and eye trackers. They practiced calibration under supervision to ensure accurate data collection and learned how to interpret basic feedback cues. This preparation minimized technical issues and helped participants feel comfortable with the platform and biometric devices throughout the study.
For the control group, traditional digital reading resources were used, consisting of fixed reading materials and comprehension exercises without adaptive features. To ensure comparability with the experimental group’s adaptive materials, we first reviewed B1-B2 level resources (e.g., textbooks, online libraries) to select texts similar in topic and genre to those in Smart Sparrow’s database. This ensured both groups encountered a range of narrative, informational, and opinion-based texts.
We then applied quantitative readability measures, including the Flesch-Kincaid Grade Level and Gunning Fog Index, to both the control group’s texts and a sample of texts from the experimental group, adjusting the control group materials to match the experimental group’s average difficulty. Comprehension exercises for the control group were designed to mirror those in the experimental group, using similar question formats, word counts, and estimated completion times.
A pilot study (n = 20) with participants from the target population tested both the control materials and a simulated Smart Sparrow experience, collecting feedback on perceived difficulty, engagement, and time-on-task. Additionally, we controlled the total word count of reading materials per session to ensure equivalence between groups and conducted lexical analysis to ensure comparable vocabulary overlap, minimizing potential confounds.
The use of biometric feedback mechanisms, such as HRV sensors and eye-tracking technology, was integral to this study. Ensuring the accuracy and reliability of these tools is crucial for the validity of the collected data and subsequent findings. The selection of the Polar H10 HRV sensor and the Tobii Pro X3-120 eye-tracker was based on their demonstrated accuracy and reliability in previous research. These devices have been widely used in studies examining cognitive workload, emotional responses, and reading behaviors (Nunan et al. 2010; Holmqvist et al. 2011).
The HRV sensors used in this study were the Polar H10, known for its accuracy and reliability in cognitive and emotional monitoring (Nunan et al. 2010; Laborde et al. 2017; Shaffer et al. 2014). To assess reliability, we conducted a test-retest study with 30 participants, who wore the Polar H10 sensors during two reading sessions one week apart, using identical materials. We calculated the RMSSD for each session, and the intraclass correlation coefficient (ICC) between the two sets of RMSSD values was 0.92 (95% CI: 0.85–0.96), indicating excellent reliability. Additionally, internal consistency within sessions was assessed using Cronbach’s alpha, which was 0.87, further supporting the reliability of the HRV data.
The eye-tracking technology used in this study was the Tobii Pro X3-120, known for its precision in tracking eye movements (Holmqvist et al. 2011; Rayner et al. 2009). Holmqvist et al. (2011) reported that the device measures fixation durations and saccades with an error rate of less than 0.5 degrees, ensuring high accuracy. Reliability was assessed through split-half reliability, yielding a coefficient of 0.85, indicating strong internal consistency. The average fixation duration was 250 ms (SD = 32 ms) for the experimental group and 285 ms (SD = 38 ms) for the control group. The average number of regressions per 100 words was 4.5 (SD = 1.2) for the experimental group and 6.5 (SD = 1.6) for the control group. These findings, along with the high reliability coefficient, affirm the consistency and validity of the eye-tracking data as an indicator of reading behavior and cognitive load.
Before the study began, both the HRV sensors and eye-tracking devices were calibrated according to manufacturer guidelines. For the Polar H10 HRV sensors, this involved ensuring proper electrode placement, moistening electrodes for optimal signal transmission, and adjusting strap tightness. The Tobii Pro X3-120 eye-tracking system was calibrated using a 9-point procedure, with participants fixating on specific screen points to map their eye movements.
To maintain calibration accuracy, periodic checks were conducted during each session. Prior to each of the three 30-min reading segments, both devices underwent brief recalibrations: a check of electrode placement and signal quality for the HRV sensors and a 5-point calibration for the eye-tracking system. These procedures ensured reliable biometric data throughout each session.
The control group used fixed digital reading resources with comprehension exercises, selected for their relevance and appropriateness for B1-B2 proficiency levels, ensuring content and complexity were comparable to those in the experimental group.
Reading comprehension was assessed using the reading component of the Preliminary English Test (PET), a standardized EFL proficiency test developed by Cambridge Assessment English. The PET is designed to evaluate the language ability of learners at the B1 level of the CEFR. The reading component includes five parts, which test various aspects of reading skills such as understanding main ideas, locating specific information, and identifying text structure. The test comprises multiple-choice questions, matching tasks, and gap-filling exercises.
To ensure the validity and reliability of the PET in this study, two parallel test forms were developed for the pre-test and post-test. These forms were carefully constructed to be equivalent in terms of content, format, and difficulty level, drawing upon items from the same test administration period. A pilot study was conducted with 30 participants from a comparable population to establish the equivalence of the two forms. The results indicated a high correlation between the scores on the two forms (r = 0.92, p < 0.001), supporting their comparability. Additionally, item difficulty indices and discrimination coefficients were analyzed, showing no significant differences between the forms (p > 0.05). The internal consistency of both PET reading component versions was high. For the pre-test version, Cronbach’s alpha was 0.82, and for the post-test version, it was 0.84. Construct validity was assessed through CFA, and the model demonstrated acceptable fit indices: χ2/df = 2.16, CFI = 0.95, TLI = 0.93, RMSEA = 0.06 (90% CI = 0.04–0.08), SRMR = 0.04.
Motivation was assessed using a modified version of the L2 Learning Motivation Scale (adapted from Ryan 2005). This 5-item self-report scale measured participants’ overall motivation towards learning English as a Second Language (L2). Participants rated their agreement with each statement on a 5-point Likert scale (1 = Strongly Disagree, 5 = Strongly Agree). The scale focused on aspects of instrumental motivation, such as the perceived importance and effort associated with learning English. The internal consistency of the motivation scale was examined at both pre- and post-test. Cronbach’s alpha was 0.87 for the pre-test and 0.89 for the post-test, indicating strong reliability. CFA results supported the scale’s construct validity: χ2/df = 2.03, CFI = 0.96, TLI = 0.94, RMSEA = 0.05 (90% CI = 0.03–0.07), SRMR = 0.03.
Anxiety levels were measured using the well-established Foreign Language Classroom Anxiety Scale (FLCAS) (Horwitz et al. 1986). This 33-item self-report scale assessed the specific anxiety associated with language learning environments. Participants rated the frequency of their experiences with anxiety-related statements (e.g., “I don’t feel pressure to prepare very well for language class”) on a Likert scale. The FLCAS has demonstrated high internal consistency (Cronbach’s alpha >0.80) and positive correlations with other measures of anxiety in L2 learners. In this study, the FLCAS demonstrated excellent reliability at both measurement points. Cronbach’s alpha was 0.89 for the pre-test and 0.91 for the post-test. CFA results indicated strong construct validity: χ2/df = 2.11, CFI = 0.94, TLI = 0.92, RMSEA = 0.06 (90% CI = 0.04–0.08), SRMR = 0.04.
Cognitive load was measured using the Differentiated Cognitive Load Questionnaire (Klepsch et al. 2017). This instrument included 2 items for Intrinsic Cognitive Load (ICL), 3 items for Extraneous Cognitive Load (ECL), and 3 items for Germane Cognitive Load (GCL). Participants rated these items on a 7-point Likert scale from 1 (not at all) to 7 (completely). To assess consistency, Cronbach’s alpha values were calculated separately for the pre-test and post-test. The overall internal consistency for the Cognitive Load Questionnaire was satisfactory, with α = 0.83 for the pre-test and α = 0.85 for the post-test. The subscales also demonstrated acceptable reliability for both measurement points: ICL (pre α = 0.78, post α = 0.80), ECL (pre α = 0.81, post α = 0.83), and GCL (pre α = 0.79, post α = 0.81). CFA results confirmed the construct validity of the questionnaire: χ2/df = 2.24, CFI = 0.93, TLI = 0.91, RMSEA = 0.07 (90% CI = 0.05–0.09), SRMR = 0.05.
To gain deeper insights into participants’ experiences with the AI-enhanced reading platform, focus group discussions were conducted following the intervention. Participants for the focus groups were selected using a stratified random sampling method to ensure a representative sample of the larger participant pool. This approach considered various factors such as age, gender, and initial proficiency level to capture diverse perspectives. Each focus group consisted of 8–10 participants, with a total of four groups formed for in-depth discussions. The focus group discussions were semi-structured, guided by a set of predetermined questions aimed at exploring learners’ engagement, motivation, anxiety, reading strategies, and overall user experience with the AI platform. Discussions were audio-recorded and transcribed verbatim for analysis. Thematic analysis was employed to identify key themes and patterns within the qualitative data, providing a comprehensive understanding of the platform’s impact beyond the quantitative measures.
The study was conducted over a 12-week period, with participants attending weekly 2-h sessions held in a computer lab equipped with the necessary hardware and software for the research. These sessions were meticulously designed to ensure a high degree of consistency between the experimental and control groups, with the exception of the independent variable—the use of the AI-enhanced reading platform. This rigorous approach aimed to strengthen the study’s internal validity and guarantee that any observed differences in outcomes could be confidently attributed to the AI intervention.
Each session adhered to a structured format, consisting of three key components:
At the outset of each session, participants completed a brief questionnaire to assess their current motivation and anxiety levels using the Language Learning Motivation Scale (LLMS) and the Foreign Language Classroom Anxiety Scale (FLCAS). These standardized measures provided valuable insights into participants’ psychological states. Concurrently, biometric devices were calibrated to ensure accurate data collection. For the experimental group, this involved proper positioning and functioning of wearable chest straps for heart rate variability (HRV) and desktop-mounted eye trackers. The control group underwent the same biometric setup to maintain consistency in data collection procedures and minimize potential bias.
The core component of each session was a 90-min reading period, divided into three manageable 30-min segments to promote focused learning with flexibility. During these segments, participants engaged with reading materials tailored to their B1-B2 CEFR proficiency level. The key distinction between the groups emerged here. The experimental group utilized Smart Sparrow, which dynamically adjusted the difficulty and type of reading materials based on continuous performance assessments and real-time biometric feedback. For instance, the platform could adapt by presenting easier texts or providing supplementary explanations if a participant’s eye-tracking data revealed prolonged fixations on specific words or sentences, indicating potential comprehension challenges. Similarly, HRV data could inform adjustments aimed at managing participants’ cognitive load and emotional states during reading tasks. In contrast, the control group worked with a fixed set of digital reading materials and comprehension exercises that were identical in content and difficulty level to those initially presented to the experimental group. However, these resources lacked adaptive features.
Following the reading session, participants completed a comprehension test designed to evaluate their understanding of the reading materials. These tests were comparable in format and difficulty across both groups, ensuring that any observed differences in performance could be attributed solely to the intervention. Participants also revisited the pre-session questionnaires to reassess their motivation and anxiety levels. Additionally, they reported their perceived cognitive load, providing subjective data to complement the objective biometric measures.
To minimize instructor bias and ensure consistency, both groups were supervised by the same experienced EFL instructor. The instructor underwent training to standardize interactions and maintain neutrality. This training included an overview of the study’s objectives, procedures, and the instructor’s roles, particularly regarding the experimental and control group conditions, permissible interactions, and the importance of impartiality. To ensure consistent communication, the instructor received scripted responses for common scenarios, such as answering questions or addressing technical issues. The instructor practiced these scripts in mock sessions with research assistants acting as participants from both groups, receiving feedback to refine their approach.
Throughout the study, the instructor’s interactions were monitored via audio recordings and observation, with regular feedback provided to ensure adherence to protocols. A subset of sessions was independently reviewed by multiple researchers, achieving a high inter-rater reliability (κ > 0.80), confirming consistency in evaluating the instructor’s adherence to the standardized protocols.
The reading materials for both groups were selected to match B1-B2 proficiency levels, using a diverse corpus from reputable educational publishers and online libraries. This included narrative texts, informational articles, and opinion pieces to expose participants to various writing styles and topics. Content was chosen with attention to cultural sensitivity. Topics deemed potentially controversial or sensitive in the Chinese context, such as politics and religion, were excluded. Selected texts covered subjects like travel, technology, health, environment, and everyday life to ensure variety and engagement.
Comprehension exercises assessed literal, inferential, and critical reading skills through multiple-choice, short-answer, and summary tasks. Most exercises were developed specifically for the study, aligned with the target proficiency level, while some were adapted from standardized assessments (e.g., Cambridge PET, IELTS) for reliability and validity.
The comprehension tests underwent rigorous validation, including expert reviews by EFL instructors to assess clarity and difficulty, and pilot testing with a small sample of learners to refine the tests before the main study.
The data analysis for this study employed a mixed-methods approach to comprehensively evaluate the impact of the AI-enhanced reading platform on various outcomes, including reading comprehension, motivation, anxiety, and cognitive load (Tabachnick & Fidell 2019). Quantitative data were analyzed using SPSS, beginning with the calculation of descriptive statistics (means and standard deviations) to provide an overview of the participants’ performance and psychological states. Subsequently, a Multivariate Analysis of Variance (MANOVA) was conducted to assess the overall effect of the intervention on the combined set of dependent variables, followed by univariate ANOVAs to examine the intervention’s effect on each dependent variable separately (Cohen 1988). To enhance internal validity, potential confounding variables, such as initial proficiency levels and baseline psychological measures, were incorporated as covariates in the analyses. Furthermore, to account for individual differences in learning trajectories, we employed mixed-effects modeling, which allowed us to examine both the overall effect of the intervention and individual variations in response to the AI platform, accommodating the nested structure of the data and providing a more nuanced understanding of the intervention’s effects (Bates et al. 2015).
To complement the quantitative findings, focus group discussions were conducted with participants from the experimental group at the conclusion of the 12-week intervention, allowing for a detailed exploration of their experiences while ensuring that key topics were covered (Krueger & Casey 2014). The focus group discussions were audio-recorded and transcribed verbatim. The transcribed data were then analyzed using thematic analysis, a well-established method for identifying, analyzing, and reporting patterns within qualitative data (Braun & Clarke 2006). Thematic analysis involved several steps: familiarization with the data, coding of the transcripts, searching for themes among the codes, reviewing themes, defining and naming themes, and producing the final report. To enhance the reliability of the analysis, initial coding was conducted independently by two researchers. Codes were then compared and reconciled, with discrepancies addressed through discussion. This collaborative coding process ensured a thorough and unbiased analysis of the data. Emerging themes were subsequently categorized into broader domains relevant to the research questions, such as usability of the platform, perceived effectiveness, engagement, and emotional responses.
The results of this study are presented in two main sections: quantitative results from the statistical analyses of the reading comprehension scores, motivation, anxiety, and cognitive load; and qualitative results from the thematic analysis of focus group discussions.
Descriptive statistics for the pre- and post-intervention measures of reading comprehension, motivation, anxiety, and cognitive load are presented in Table 1.
At baseline, both groups were comparable across all variables, indicating successful random assignment and suggesting that any post-intervention differences were due to the intervention. Pre-intervention means for reading comprehension were 65.4 (SD = 7.8) for the experimental group and 64.8 (SD = 8.0) for the control group. Motivation levels were 3.5 (SD = 0.6) for the experimental group and 3.6 (SD = 0.5) for the control group. Anxiety scores were 3.8 (SD = 0.7) and 3.7 (SD = 0.8) for the experimental and control groups, respectively. Cognitive load was 4.2 (SD = 0.6) for the experimental group and 4.1 (SD = 0.7) for the control group.
The MANOVA revealed a significant multivariate effect of the intervention, Wilks’ Lambda = 0.32, F(4, 295) = 34.65, p < 0.001, η2 = 0.68, indicating a substantial overall impact. In other words, learners in the experimental group showed significantly different outcomes across all dependent variables compared to the control group. To better understand these differences, we conducted follow-up univariate ANOVAs for each dependent variable.
For reading comprehension, the intervention had a significant effect, F(1, 298) = 118.34, p < 0.001, η2 = 0.28. The experimental group’s post-intervention mean (M = 82.6, SD = 6.2) was 12.4 points higher than the control group’s mean (M = 70.2, SD = 7.5), suggesting the platform effectively enhanced reading skills, likely through adaptive learning and real-time feedback.
In terms of motivation, the experimental group showed a significant increase, F(1, 298) = 45.78, p < 0.001, η2 = 0.13, with the post-intervention mean for the experimental group (M = 4.3, SD = 0.5) higher than the control group’s (M = 3.8, SD = 0.6). This improvement points to the platform’s engaging, personalized approach to content delivery.
Regarding anxiety, the experimental group experienced a significant reduction, F(1, 298) = 72.21, p < 0.001, η2 = 0.19. The mean anxiety score for the experimental group decreased to 2.5 (SD = 0.6) from 3.4 (SD = 0.7) in the control group, suggesting the AI platform alleviated anxiety by providing adaptive support and real-time adjustments that reduced cognitive strain.
Finally, cognitive load was significantly lower in the experimental group, F(1, 298) = 61.45, p < 0.001, η2 = 0.17. The experimental group’s post-intervention cognitive load mean (M = 3.1, SD = 0.5) was 0.9 points lower than the control group’s mean (M = 4.0, SD = 0.6), indicating the platform helped streamline the learning process by adjusting task difficulty and offering supportive feedback (Table 2).
These results highlight the significant positive effects of the AI-enhanced reading platform across all outcomes, with large effect sizes for reading comprehension and anxiety, and moderate effects for motivation and cognitive load. This suggests the platform’s potential to enhance L2 reading comprehension and improve psychological aspects of learning, making it a valuable tool for educational settings.
Biometric data analysis was conducted to objectively assess the physiological impact of the AI-enhanced reading platform and corroborate the self-reported measures of cognitive load and anxiety. This section explores two key biometric measures: heart rate variability (HRV) and eye-tracking metrics. These measures were chosen due to their established ability to provide real-time insights into cognitive and emotional states during learning tasks (Shaffer & Ginsberg 2017; Rayner 1998).
Although differences in HRV and eye-tracking metrics between the experimental and control groups might initially appear to result directly from the AI platform’s real-time feedback, the subsequent analysis serves several crucial purposes. By examining overall patterns in HRV and eye-tracking metrics, we gain a nuanced understanding of the platform’s impact on learners’ physiological and cognitive processes. This analysis goes beyond the immediate effects of real-time feedback, allowing us to explore the dynamic interaction between the learner, the AI platform, and the learning task. For example, HRV trends over time revealed distinct phases: an initial decrease followed by a gradual increase.
The initial decrease in HRV among the experimental group indicated heightened engagement and concentration as participants interacted with the AI platform. During cognitively demanding tasks, lower HRV is associated with increased mental effort and focus (Delliaux et al. 2019; Hansen et al. 2004; Laborde et al. 2017). This phase reflects the participants’ active engagement and adaptation to the adaptive, challenging environment provided by the platform, rather than increased stress. Over the course of the 12-week study, HRV levels in the experimental group gradually increased, suggesting a reduction in overall stress and a better balance between focus and relaxation. This upward trend indicates that as participants became more accustomed to the platform, their physiological responses shifted from high engagement to a more relaxed state, enhancing their learning efficiency.
Additionally, this analysis offers a valuable opportunity for triangulation by comparing objective physiological data with self-reported measures. This strengthens the validity of our conclusions and provides a more holistic picture of the learning experience. For instance, the initial decrease in HRV in the experimental group coincided with self-reported decreases in cognitive load and anxiety, indicating that the lower HRV was reflective of increased cognitive engagement rather than heightened stress. Subsequent increases in HRV corresponded with reports of reduced stress, reinforcing the dual role of HRV as an indicator of both engagement and stress reduction. Furthermore, analyzing the biometric data in detail enables us to identify subtle patterns and individual differences in learners’ responses to the AI platform, which could inform future research and the development of more personalized learning interventions. For example, examining individual variations in eye-tracking metrics—such as fixation durations and saccade lengths—offers insights into specific areas where learners may struggle, revealing aspects of their cognitive processing strategies.
HRV analysis focused on participants’ physiological stress responses. While higher HRV at rest is generally associated with a more relaxed state and greater autonomic flexibility, the relationship between HRV and cognitive performance during tasks is more complex (Laborde et al. 2017). In cognitively demanding contexts, lower HRV may reflect not only heightened cognitive engagement but also increased concentration and mental effort (Delliaux et al. 2019; Hansen et al. 2004). In our study, a statistically significant difference in HRV between the experimental and control groups was observed (t(298) = 11.28, p < 0.001), with the experimental group showing lower HRV (M = 55.6, SD = 5.3) compared to the control group (M = 62.4, SD = 6.1), as shown in Table 3.
This initial lower HRV in the experimental group suggests a state of concentrated engagement with the AI platform rather than increased stress. As participants adapted to the platform, HRV levels gradually increased, indicating a reduction in stress and a more balanced autonomic response. By the end of the intervention, the experimental group’s HRV surpassed that of the control group, demonstrating successful adaptation and decreased stress.
HRV is a complex measure that reflects the interaction between the sympathetic and parasympathetic branches of the autonomic nervous system. Although lower HRV can indicate stress during tasks, it can also signify increased cognitive engagement and focus, depending on the context (Laborde et al. 2017). The observed initial lower HRV in the experimental group likely reflects the platform’s ability to foster concentration, which may contribute to improved learning outcomes. Over the 12-week study, this initial drop in HRV was followed by a gradual increase, suggesting a decline in stress levels and an improvement in reading efficiency. Importantly, both the experimental and control groups exhibited HRV values within the normal range for healthy adults (Shaffer & Ginsberg 2017). The significant difference between the groups emphasizes HRV’s sensitivity to changes in cognitive and emotional states, highlighting the potential impact of the AI platform on physiological responses. These findings align with prior research suggesting that interventions that reduce cognitive load and stress can enhance HRV, improving autonomic regulation and cognitive performance (Shaffer & Ginsberg 2017). The AI platform’s ability to adapt reading materials and provide supportive feedback likely contributed to this positive outcome by reducing stress and supporting cognitive function.
To further illustrate these findings, we included a graph depicting the average HRV (root mean square of the successive differences, RMSSD) for both groups throughout the 12-week intervention period. This graph clearly shows two distinct phases for the experimental group: an initial decrease in HRV, followed by a gradual increase. The experimental group’s HRV, despite starting at a lower baseline, ultimately surpassed that of the control group, demonstrating successful adaptation and decreased stress.
As shown in Fig. 1, the experimental group exhibited a gradual increase in HRV after the initial decrease, indicating a consistent reduction in stress levels over time. In contrast, the control group’s HRV values fluctuated slightly but showed no significant upward trend. This divergence becomes more pronounced from the fourth week onward, suggesting that the AI platform’s adaptive features had a cumulative positive effect on reducing stress. The visual evidence, highlighting the two distinct phases of HRV change in the experimental group, complements the statistical findings, offering a comprehensive understanding of the AI platform’s efficacy in promoting both engagement and relaxation in the learning environment.
Average HRV (RMSSD) over 12 weeks for experimental and control groups.
Eye-tracking data provided further insights into learners’ cognitive processes during reading. Key metrics analyzed included fixation durations, regression counts, and saccade lengths. Fixation durations reflect the time spent processing individual words, with shorter durations indicating faster word recognition and comprehension (Rayner 1998). Regression counts, or backward eye movements in the text, signal comprehension difficulties, with fewer regressions indicating smoother, more confident reading. Saccade length, which measures the distance covered by eye movements between fixations, reflects the efficiency of visual scanning and the ability to process larger chunks of text. These metrics, when analyzed together, provide a holistic view of reading proficiency and cognitive load.
To assess the differences in saccade length between the groups, we conducted an independent samples t-test. The results revealed a significant difference (t(298) = 4.85, p < 0.001), with the experimental group exhibiting longer saccades (M = 8.2 characters, SD = 1.5) compared to the control group (M = 7.1 characters, SD = 1.8). This finding suggests that the experimental group, using the AI-enhanced platform, exhibited more efficient visual scanning and information processing during reading, as longer saccades indicate the ability to process larger units of text with each eye movement, a hallmark of skilled reading (Rayner 1998).
As shown in the scatterplot (Fig. 2), saccade length is represented by the size of each data point, offering an additional dimension of insight into reading efficiency. Participants in the experimental group (blue points) exhibited longer saccades, reinforcing their ability to process text more efficiently. This pattern is consistent with the observed shorter fixation durations and fewer regressions. Conversely, control group participants (red points) displayed shorter saccades alongside longer fixation durations and more regressions, indicative of less efficient reading strategies. Together, these metrics underscore the superior reading efficiency of the experimental group, highlighting the role of the platform in fostering more effective cognitive strategies.
Scatterplot of eye-tracking metrics: fixation duration vs. regression counts, with saccade length as point size.
Independent samples t-tests were conducted to compare each eye-tracking metric between the groups, revealing significant differences in both fixation durations and regressions (Table 4). The experimental group demonstrated shorter fixation durations (M = 225 ms, SD = 30 ms) and fewer regressions per 100 words (M = 4.2, SD = 1.1) compared to the control group (fixation duration: M = 290 ms, SD = 35 ms; regressions per 100 words: M = 6.8, SD = 1.5) (fixation duration: t(298) = 13.45, p < 0.001; regressions: t(298) = 14.78, p < 0.001). These results further support the experimental group’s enhanced ability to integrate and process text efficiently, as evidenced by their saccade patterns.
These significant differences suggest that the experimental group, utilizing the AI-enhanced platform, read more efficiently and fluently. The adaptive features likely facilitated this improvement by providing personalized learning materials and adjusting difficulty based on performance and real-time biometric feedback. The inclusion of saccade length analysis complements the findings from fixation durations and regressions, providing a comprehensive view of reading efficiency. The convergence of self-reported measures and biometric data further strengthens the study’s conclusions. Participants in the experimental group reported lower cognitive load and anxiety, which aligns with the observed reductions in HRV and improvements in eye-tracking metrics, providing objective support for the platform’s efficacy in enhancing reading performance.
To further investigate the impact of the AI-enhanced reading platform while accounting for individual differences in learning trajectories, we employed mixed-effects modeling. This statistical approach allows for the examination of both fixed effects, representing the overall influence of the intervention, and random effects, capturing variations across individual participants. Mixed-effects models are particularly well-suited for this study as they accommodate the nested structure of the data, with repeated measures for each participant, and provide a nuanced understanding of the intervention’s effects.
Linear mixed-effects models were constructed for each dependent variable: reading comprehension, motivation, anxiety, and cognitive load. Additionally, we included heart rate variability (HRV) and eye-tracking metrics as separate dependent variables in their own mixed-effects models to assess their specific responses to the intervention. Fixed effects included the intervention group (experimental vs. control), time (pre-test vs. post-test), and the interaction between group and time to assess differential effects of the intervention. Pre-test scores were included as covariates to control for baseline differences, ensuring that the observed effects were attributable to the intervention. Additionally, we included age (centered at the sample mean) and gender (coded as 0 = male, 1 = female) as fixed effects in the models to explore their potential influence on the outcomes. For the biometric measures, HRV and eye-tracking metrics were treated as additional fixed effects in their respective models to examine their relationship with the primary outcomes. To account for individual variability, random intercepts for participants were included in all models. For cognitive load, random slopes for time were also included to capture individual differences in learning trajectories.
Model selection followed a systematic approach, utilizing likelihood ratio tests and Akaike Information Criterion (AIC) to determine the best-fitting models for each outcome variable. All models were estimated using the lme4 package in R (Bates et al. 2015), and conditional R-squared (R2c) values were calculated to evaluate the proportion of variance explained by both fixed and random effects. The final models, including fixed effect estimates and significance levels, are presented in Table 5.
Random effects were included to account for variability across participants. Estimates for random intercepts were as follows: reading comprehension (σ = 2.85), motivation (σ = 0.32), anxiety (σ = 0.40), HRV (σ = 1.50), and eye-tracking metrics (σ = 0.50). For cognitive load, HRV, and eye-tracking metrics, random slopes for time were also included to capture individual differences in learning trajectories (σ_slope = 0.15 for cognitive load, σ_slope = 0.10 for HRV, and σ_slope = 0.07 for eye-tracking metrics). Likelihood ratio tests demonstrated that including random slopes for time significantly enhanced the model fit for cognitive load (χ2(1) = 7.85, p = 0.005) and eye-tracking metrics (χ2(1) = 5.60, p = 0.018), highlighting individual differences in how participants’ cognitive load and eye-tracking metrics evolved over the intervention.
To evaluate overall model fit, both conditional R2c and Bayesian Information Criterion (BIC) were calculated. These metrics provide complementary insights: R2c measures the proportion of variance explained by the model, including both fixed and random effects, with higher values indicating a better fit. BIC assesses the trade-off between model fit and complexity, where lower values suggest a more efficient model. The results are summarized in Table 6.
The R2c values, ranging from 0.65 to 0.83, show that the models account for a substantial portion of the variance in each outcome. For example, an R2c of 0.75 for reading comprehension means that 75% of the variability is explained by the model. The BIC values, such as 1850 for reading comprehension, allow comparison across models, with lower values indicating a better balance of fit and simplicity.
The mixed-effects models provided additional insights beyond the MANOVA/ANOVA analyses, emphasizing individual variability in baseline performance and learning trajectories. Specifically, the random slopes for cognitive load showed that the AI platform was especially beneficial for learners with higher initial cognitive load. Furthermore, while age and gender were included as fixed effects, their influence was not statistically significant (age: p = 0.060; gender: p = 0.153), suggesting that the AI platform’s effectiveness was largely consistent across these variables.
Additionally, biometric measures (HRV and eye-tracking metrics) were analyzed in separate mixed-effects models to assess their specific responses to the AI platform. These models revealed that the AI platform significantly influenced HRV and eye-tracking metrics, aligning with improvements in the primary outcomes. For HRV, the model showed a significant negative effect of the intervention (Group (Experimental): −0.85, p < 0.001), which could suggest increased stress since lower HRV is often linked to higher stress levels. However, in this study, the AI platform’s adaptive features may have increased cognitive engagement, temporarily reducing HRV. This is supported by the significant Group × Time interaction (Estimate = −0.20, p < 0.001), indicating that HRV differences between groups changed over time. Further analysis of HRV trends could clarify whether adaptation to the platform later reduced stress. For the primary outcomes, the Group × Time interaction terms showed that, compared to the control group, the experimental group had greater gains over time: reading comprehension increased by an additional 8 points (Estimate = 8.00, p < 0.001), motivation rose by an additional 0.65 points (Estimate = 0.65, p < 0.001), and anxiety decreased by an additional 0.72 points (Estimate = −0.72, p < 0.001). Participants with higher baseline cognitive load experienced more substantial reductions in cognitive load, reinforcing the platform’s adaptability to diverse learner needs. These results validate the platform’s effectiveness and highlight the value of addressing individual variability in educational interventions, with adaptive technologies fostering a responsive and inclusive learning environment.
The qualitative component of this study aimed to gain a deeper understanding of participants’ experiences with the AI-enhanced reading platform. Focus group discussions were conducted and analyzed using thematic analysis, allowing for an exploration of how the platform influenced learners’ engagement, motivation, anxiety, reading strategies, and overall user experience beyond the quantitative data. The analysis revealed four key themes that provide a rich tapestry of how the platform impacted learners on a personal level.
Enhanced Engagement and Motivation emerged as a prominent theme. Participants frequently mentioned the platform’s ability to personalize reading materials and provide targeted feedback, which made the learning experience more engaging. One participant remarked, “The AI made reading articles feel less like a chore and more like a conversation tailored to my interests.” Another noted, “The adjustments to difficulty level kept me challenged but not overwhelmed. It felt like the platform was adapting to my learning pace, which kept me motivated.” Interactive features such as built-in dictionaries and gamified elements significantly contributed to increased engagement. A participant shared, “I found myself actively looking up unfamiliar words because the dictionary was so easy to access. It made me feel more involved in the reading process.” The points and badges for completing tasks provided a fun and motivating way to stay engaged, making progress feel like a series of mini-achievements. Another participant mentioned, “The badges were a nice touch. They gave me a sense of accomplishment and kept me coming back for more.”
Reduced Anxiety and Stress was another evident theme. Participants reported a significant reduction in anxiety when using the AI platform, attributing this to the real-time adjustments in difficulty based on performance. One participant shared, “In the past, I would get discouraged by difficult reading materials. But with the AI platform, I knew it would adjust the difficulty if I got stuck, which helped me stay calm and focused.” Another mentioned, “The platform made me feel more confident in my reading abilities because it wasn’t constantly pushing me beyond my limits.” The supportive prompts and guidance provided by the platform, such as hints and suggestions, were also seen as helpful in mitigating stress. Participants felt less alone and more secure, knowing that the platform was there to guide them if needed, akin to having a tutor by their side. One participant expressed, “The hints were incredibly helpful. Whenever I felt stuck, I knew I could rely on them to get back on track.”
The third theme, Development of Improved Reading Strategies, highlighted how the platform contributed to better reading habits and strategic approaches. Participants appreciated the strategic feedback provided by the platform, which went beyond merely identifying errors. It offered specific suggestions for improvement, such as recommending strategies for vocabulary acquisition or comprehension techniques. One participant noted, “The AI platform didn’t just tell me I made a mistake; it also suggested ways to learn from it. This helped me develop better reading habits and become a more strategic reader.” Additionally, the platform’s integration with real-time biometric feedback fostered a sense of metacognitive awareness. Participants who opted for this feature could see their heart rate and eye-tracking data, helping them understand their own cognitive states during reading. This feature was particularly beneficial in helping learners realize the connection between their emotions and reading comprehension. As one participant commented, “Seeing my heart rate go down as I became more comfortable with the text was fascinating.” Another participant added, “Understanding how my stress levels affected my reading helped me develop strategies to stay calm and focused.”
Finally, the theme of User-Friendly Interface and Platform Design emerged as a crucial factor in the platform’s success. Participants consistently praised the intuitive and user-friendly interface, which made navigation and interaction with the platform effortless. One participant shared, “The platform was so easy to use, even for someone not very tech-savvy. I was able to find everything I needed without any difficulty.” The seamless integration of various features, such as the dictionary, note-taking tools, and progress trackers, was also seen as a significant positive aspect. Participants could easily access these tools without interrupting their reading flow, enhancing their overall learning experience. One participant remarked, “Having the dictionary built right in was amazing. I could look up unfamiliar words without leaving the text, which saved me time and kept me focused.” Another participant emphasized, “The note-taking feature was a game-changer. I could jot down thoughts and insights without disrupting my reading.”
Although the majority of feedback was positive, some participants did express challenges or neutral experiences with the platform. A few noted that the biometric feedback, although innovative, could sometimes be distracting. One participant shared, “I found myself paying more attention to my heart rate than to the reading material, which was a bit counterproductive.” Another participant raised concerns about data privacy, saying, “I’m not entirely comfortable with how my biometric data is being used and stored. It makes me a bit uneasy.” Additionally, while many appreciated the gamified elements, a small number felt indifferent. One participant commented, “The badges and points didn’t really motivate me. I prefer a more straightforward approach to learning.” These perspectives highlight the importance of considering individual differences and privacy concerns when implementing AI and biometric technologies in educational settings.
Overall, the qualitative results suggest that participants generally perceived the AI-enhanced reading platform as beneficial for their reading comprehension, motivation, anxiety reduction, and cognitive load management compared to traditional digital resources. The integration of real-time adaptive features and biometric feedback was seen as instrumental in creating a supportive and engaging learning environment, underscoring the potential of AI-driven educational technologies to enhance language learning outcomes. However, some participants also noted challenges, such as distractions from biometric feedback and concerns about data privacy, indicating areas for further refinement. These insights, enriched by the diverse voices of participants, highlight the platform’s multifaceted impact on learners’ experiences while emphasizing the need to address individual preferences and ethical considerations.
The findings of this study highlight the effectiveness of AI-enhanced reading platforms in improving L2 reading comprehension, motivation, anxiety, and cognitive load among Chinese EFL learners. These results are consistent with existing theoretical frameworks and empirical studies, underscoring the role of AI in enhancing language learning experiences.
The significant improvement in reading comprehension observed in the experimental group suggests that the AI platform effectively supports reading skill development. According to Vygotsky’s ZPD theory, learning is most effective when scaffolding and guided interaction are provided (Vygotsky 1978). The AI platform, by adapting reading materials to learners’ proficiency levels and offering real-time feedback, functions as a dynamic form of scaffolding, aiding learners as they engage with increasingly complex texts (Grabe & Stoller 2020; Hsiao & Chang 2024). This finding aligns with previous research (Mežek et al. 2022; Dabarera et al. 2014), which emphasized the importance of scaffolding and feedback in improving L2 reading comprehension and self-regulation.
The ability of the AI platform to tailor learning experiences also reflects Cummins’ (1979, 1991) concept of leveraging L1 proficiency to support L2 learning, drawing on shared cognitive and linguistic resources (Chen et al. 2021; Gutiérrez-Colón et al. 2023). Moreover, the significant random intercepts for reading comprehension observed in the mixed-effects models indicate that individual differences, such as learner characteristics and initial proficiency levels, influence the effectiveness of the intervention (Vygotsky 1978; Droop & Verhoeven 2003; Geva & Siegel 2000). This highlights the importance of adaptive learning technologies that cater to diverse learner needs (Gottardo & Mueller 2009; Edele & Stanat 2016).
The observed increase in motivation further supports the positive impact of the AI platform. Vygotsky’s ZPD suggests that social interaction with more knowledgeable others can enhance learner motivation (Vygotsky 1978; Chiu et al. 2023). Incorporating principles from SDT (Ryan & Deci 2000), the AI platform promoted autonomy and competence, both crucial for intrinsic motivation. Through personalized feedback and adaptive learning paths, the platform served as a knowledgeable guide, fostering learner engagement (Chen et al. 2020; Lee & Hwang 2022). This is consistent with previous studies (Chen et al. 2020; Lee & Hwang 2022) showing that AI-driven tools can significantly boost motivation and engagement. One participant noted, “The platform made learning fun and interactive, which kept me coming back to study more,” while another shared, “I felt like the platform understood my needs and adjusted the content accordingly, which made me feel supported and encouraged.”
The reduction in anxiety among the experimental group aligns with research suggesting that adaptive learning environments can alleviate stress and foster a more supportive atmosphere (Horwitz et al. 1986; Çelik et al. 2024). By adjusting reading material difficulty in real time, the AI platform prevented learners from feeling overwhelmed by challenging texts. This is consistent with anxiety-appraisal theories, which focus on the role of cognitive appraisals in anxiety (Lazarus & Folkman 1984; Spielberger 1972). The platform’s adaptive difficulty adjustment likely reduced learners’ anxiety by providing a sense of control, promoting more confident learning (Chiu et al. 2023). As one participant stated, “Knowing the platform would adjust if I struggled helped me stay calm and focused.” The interaction effects observed in the mixed-effects modeling support the role of adaptive features in reducing anxiety and are in line with Cognitive Load Theory, which emphasizes the importance of managing cognitive demands to reduce stress and improve learning efficiency (Sweller 1988; Dabarera et al. 2014). These findings also reflect Self-Determination Theory’s emphasis on learner autonomy, which has been shown to reduce anxiety (Ryan & Deci, 2000), as well as research on anxiety-appraisal theories, which highlight the role of perceived control in mitigating anxiety (Lazarus & Folkman 1984).
Regarding cognitive load, the experimental group’s reported decrease suggests that the AI platform helped manage the mental effort required during reading tasks. This aligns with Sweller’s Cognitive Load Theory (1988), which posits that learning is optimized when cognitive load is appropriately managed. Real-time feedback and adaptive support likely contributed to a more efficient learning process by aligning instruction with learners’ immediate needs (Chen et al. 2021; Grabe & Stoller 2020). One participant remarked, “The real-time feedback was like having a tutor guiding me through the difficult parts, making the whole process less stressful.” The significant random slope for cognitive load found in the mixed-effects analysis suggests that the AI platform’s adaptive feedback mechanism played a key role in managing cognitive load, reinforcing CLT’s emphasis on matching instructional complexity with learners’ cognitive capacities (Sweller 1988; Dabarera et al. 2014).
The qualitative data from the focus group discussions further corroborate these findings. Learners reported enhanced engagement, reduced anxiety, and improved reading strategies, as well as a user-friendly interface. These participants valued the platform’s personalized learning paths and interactive features, which made learning more enjoyable. Their experiences mirrored the quantitative findings and provided deeper insight into the platform’s effectiveness. One participant shared, “The platform’s built-in dictionary made looking up words so easy, I felt more involved in the text,” and another noted, “The gamified elements made learning feel like a game, which kept me motivated and eager to learn.” The development of improved reading strategies observed here aligns with previous research (Gutiérrez-Colón et al. 2023; Taylor 2021), which highlighted how AI can foster strategic reading behaviors and enhance comprehension. The platform’s ability to provide strategic feedback and promote metacognitive awareness supports Vygotsky’s scaffolding and self-regulated learning framework (Vygotsky 1978; Mežek et al. 2022). As one participant noted, “The AI platform didn’t just point out mistakes; it suggested strategies to improve, which really helped me develop better reading habits.”
Additionally, the intuitive and user-friendly design of the platform contributed to a positive learning experience. Learners appreciated the seamless integration of features like the dictionary and progress trackers, which made the platform easy to navigate. This is consistent with research on the importance of user-centered design in educational technology (Liaw & English 2017; Chen et al. 2021). One participant commented, “Navigating the platform was so intuitive; I could focus on learning without getting frustrated by the technology,” while another said, “Having all the tools I needed in one place made the whole learning experience smooth and efficient.”
Biometric data analysis further supported the self-reported measures of reduced anxiety and cognitive load. The lower HRV and more efficient eye movements in the experimental group suggest that the AI platform not only improved reading performance but also created a less stressful and more cognitively manageable learning environment. This finding is consistent with literature linking physiological measures to cognitive and emotional states during learning tasks (Shaffer & Ginsberg 2017; Rayner 1998; Holmqvist et al. 2011; Duchowski et al. 2018).
However, despite these positive outcomes, several challenges and limitations must be acknowledged when implementing AI in language education. Issues such as data privacy, the need for robust technological infrastructure, and the potential for over-reliance on AI tools must be addressed to ensure ethical and effective use of AI in educational settings (Pedró et al. 2019; Hooda et al. 2022; Dakakni & Safa 2023). It is crucial that AI applications comply with ethical standards and data protection regulations to maintain learner trust and safeguard personal information. Furthermore, while AI tools offer significant advantages, it is important to balance their use with human interaction. Insights from sociocultural theory emphasize the pivotal role of teachers in language learning, suggesting that AI should be viewed as a complementary tool rather than a replacement for human instruction (Hsiao & Chang 2024; Wang et al. 2023; Valijärvi & Tarsoly 2019). The successful implementation of AI in education requires collaboration between researchers, educators, and developers to create a future in which AI empowers both teachers and learners in the language classroom (Chen et al., 2020; Chiu et al., 2023).
Taken together, this study demonstrates the significant potential of AI-enhanced reading platforms to improve L2 reading comprehension, motivation, reduce anxiety, and manage cognitive load among Chinese EFL learners. These findings are in line with theoretical and empirical literature on scaffolding, cognitive load, and the use of AI in education. By leveraging adaptive technology and biometric feedback, AI-driven tools can create supportive and engaging learning environments that facilitate language acquisition. However, the challenges related to data privacy, technological infrastructure, and the balance between AI and human interaction need to be carefully considered to ensure the ethical and effective use of AI in education.
The present study investigated the impact of an AI-enhanced reading platform, integrated with biometric feedback mechanisms, on L2 reading comprehension among Chinese EFL learners. The findings reveal significant improvements in reading comprehension, motivation, anxiety reduction, and cognitive load management in the experimental group compared to the control group. These results highlight the potential of leveraging AI technology and biometric feedback to create personalized and adaptive learning environments that cater to individual learners’ needs and states. Specifically, the AI platform’s ability to provide real-time adjustments to reading material difficulty, based on performance and biometric data, significantly enhanced learners’ reading comprehension and motivation. Additionally, the reduction in anxiety and cognitive load observed in the experimental group underscores the importance of a supportive and adaptive learning environment in facilitating language acquisition. The integration of advanced AI technologies and biometric feedback not only supported learners in their reading tasks but also contributed to a more engaging and less stressful learning experience.
This study contributes significantly to the theoretical understanding of L2 reading comprehension by integrating Vygotsky’s concept of ZPD with modern AI technologies. By doing so, it demonstrates how AI-driven adaptive scaffolding can effectively bridge the gap between learners’ current abilities and their potential capabilities, thereby operationalizing ZPD within digital learning environments. The findings support the notion that adaptive scaffolding, provided by AI-driven platforms, can effectively bridge the gap between learners’ current abilities and their potential capabilities. By incorporating biometric feedback, the study extends the application of ZPD, demonstrating how real-time data on learners’ cognitive and emotional states can further personalize and enhance the learning process. This integration of biometric data with ZPD provides a more dynamic framework for understanding learner engagement and progression, offering insights into how physiological indicators can inform instructional adjustments. This dynamic and responsive approach to scaffolding, informed by real-time physiological data, represents a novel contribution to the understanding of ZPD in the context of AI-enhanced learning.
Moreover, this study goes beyond applying a single theoretical framework by integrating multiple perspectives, including SDT, Cognitive Load Theory (CLT), and Anxiety-Appraisal Theories, to provide a more holistic understanding of how AI can impact motivation, anxiety, and cognitive load in language learning. By synthesizing these theories, the research offers a comprehensive model that illustrates the interplay between motivational, cognitive, and emotional factors in AI-facilitated language acquisition. SDT (Deci & Ryan, 1985) highlights the importance of intrinsic motivation, autonomy, and competence in fostering effective learning, and the study shows how AI technologies can support these dimensions of motivation. Cognitive Load Theory (Sweller 1988) emphasizes the need to manage mental load to optimize learning, and the study demonstrates how AI can reduce extraneous cognitive load by providing adaptive feedback. Additionally, Anxiety-Appraisal Theory (Lazarus & Folkman 1984) is employed to understand how learners’ emotional states, such as anxiety, are influenced by AI interactions, with the study showing how real-time feedback can alleviate anxiety and improve learning outcomes. This multi-theoretical approach not only enhances our understanding of the mechanisms through which AI influences L2 reading comprehension but also provides a foundation for developing more effective AI-driven educational interventions. This multi-theoretical approach contributes to a more nuanced and comprehensive understanding of the complex interplay between technology, cognition, and affect in language acquisition.
The practical implications of this study are manifold, offering valuable guidance for educators, policymakers, and technology developers in implementing AI-enhanced learning tools in language education. Teachers can utilize AI-enhanced platforms to provide personalized reading instruction that adapts to the individual needs of students. The integration of biometric feedback helps educators better understand students’ cognitive and emotional states, allowing for more tailored and effective interventions. Educational policymakers should consider investing in AI technologies and supporting their integration into language learning curricula. Policies that promote the development and use of AI-enhanced educational tools can contribute to more effective and inclusive language education practices. Developers of educational technologies should focus on creating AI platforms that incorporate biometric feedback mechanisms. Such tools can provide more comprehensive and personalized learning experiences, enhancing the efficacy of language learning programs. By implementing these practical implications, stakeholders can enhance the quality of language education and support the development of more effective and engaging learning environments for EFL learners.
Despite the promising findings, this study has several limitations that need to be acknowledged. The study sample was limited to Chinese EFL learners at a specific proficiency level (B1-B2). Future research should include a more diverse sample, encompassing different proficiency levels and learners from various cultural and linguistic backgrounds, to enhance the generalizability of the findings. Additionally, the implementation of AI-enhanced platforms with biometric feedback requires substantial technological infrastructure, which may not be readily available in all educational settings. This limitation highlights the need for further research on the scalability and accessibility of such technologies in under-resourced regions.
Moreover, the 12-week duration of the study may not capture the long-term effects of using AI-enhanced reading platforms. Longitudinal studies are needed to investigate the sustained impact of AI and biometric feedback on L2 reading comprehension and other language skills. While the AI platform provided significant benefits, there is a potential risk of learners becoming overly reliant on AI tools. Future studies should explore strategies to balance the use of AI with the development of autonomous learning skills and human interaction in the educational process. Another limitation is the absence of an in-subject (cross-over) design, where the experimental group would switch to the control method and vice versa to observe within-subject changes. Incorporating such a design could provide deeper insights into the comparative effectiveness of the two teaching methods and strengthen the study’s internal validity. Future research could adopt this approach to further enhance the robustness of findings.
Additionally, factors such as working memory, vocabulary and grammar proficiency, and L1 language proficiency were not directly controlled in this study. While participants were classified under the same CEFR proficiency level (B1-B2) to ensure a consistent baseline, variations in these cognitive and linguistic factors could have influenced the outcomes. Future research should consider measuring and controlling for these variables to further isolate the effects of AI-enhanced reading platforms on reading comprehension and anxiety. Lastly, the use of biometric feedback raises concerns about data privacy and ethical issues. Ensuring that AI platforms comply with ethical standards and data protection regulations is crucial. Further research should explore best practices for safeguarding learner privacy and addressing ethical concerns in the use of AI and biometric technologies in education.
The datasets generated and analyzed during this study are not publicly accessible due to the potential risk of compromising participant privacy. However, these datasets can be obtained from the corresponding author upon reasonable request.
Bates D, Maechler M, Bolker B, Walker S (2015) Fitting linear mixed-effects models using lme4. J Stat Softw 67(1):1–48. https://doi.org/10.18637/jss.v067.i01
Article Google Scholar
Braun V, Clarke V (2006) Using thematic analysis in psychology. Qual Res Psychol 3(2):77–101. https://doi.org/10.1191/1478088706qp063oa
Article Google Scholar
Çelik F, Yangın Ersanlı C, Arslanbay G (2024) Does AI simplification of authentic blog texts improve reading comprehension, inferencing, and anxiety? A one-shot intervention in Turkish EFL context. Int Rev Res Open Distrib Learn 25(3):287–303. https://doi.org/10.19173/irrodl.v25i3.7779
Article Google Scholar
Chen X, Xie H, Hwang GJ (2020) A multi-perspective study on artificial intelligence in education: grants, conferences, journals, software tools, institutions, and researchers. Comput Educ Artif Intell 1:100005. https://doi.org/10.1016/j.caeai.2020.100005
Article Google Scholar
Chen X, Zou D, Xie H, Cheng G (2021) Twenty years of personalized language learning. Educ Technol Soc 24(1):205–222
Google Scholar
Chiu TK, Moorhouse BL, Chai CS, Ismailov M (2023) Teacher support and student motivation to learn with artificial intelligence (AI) based chatbot. Interact Learn Environ. https://doi.org/10.1080/10494820.2023.2253861
Cohen J (1988) Statistical power analysis for the behavioral sciences, 2nd edn. Hillsdale, NJ: Erlbaum
Cummins J (1979) Linguistic interdependence and the educational development of bilingual children. Rev Educ Res 49(2):222–251. https://doi.org/10.3102/00346543049002222
Article Google Scholar
Cummins J (1991) Interdependence of first- and second-language proficiency in bilingual children. In: Byialystok E (ed) Language processing in bilingual children. Cambridge University Press, 70–89. https://doi.org/10.1017/CBO9780511620652.006
Dabarera C, Renandya WA, Zhang LJ (2014) The impact of metacognitive scaffolding and monitoring on reading comprehension. System 42:462–473. https://doi.org/10.1016/j.system.2013.12.020
Article Google Scholar
Dakakni D, Safa N (2023) Artificial intelligence in the L2 classroom: Implications and challenges on ethics and equity in higher education: a 21st century Pandora’s box. Comput Educ Artif Intell 5:100179. https://doi.org/10.1016/j.caeai.2023.100179
Article Google Scholar
Deci EL, Ryan RM (1985) The general causality orientations scale: Self-determination in personality. J Res Personality 19(2):109–134. https://doi.org/10.1016/0092-6566(85)90023-6
Article Google Scholar
Delliaux S, Delaforge A, Deharo JC, Chaumet G (2019) Mental workload alters heart rate variability, lowering non-linear dynamics. Front Physiol 10:565. https://doi.org/10.3389/fphys.2019.00565
Article PubMed PubMed Central Google Scholar
Divekar RR, Lepp H, Chopade P, Albin A, Brenner D, Ramanarayanan V (2021, July) Conversational agents in language education: where they fit and their research challenges. In: International conference on human-computer interaction. Springer International Publishing, Cham, 272–279. https://doi.org/10.1007/978-3-030-90179-0_35
Droop M, Verhoeven L (2003) Language proficiency and reading ability in first- and second-language learners. Read Res Q 38(1):78–103. https://doi.org/10.1598/RRQ.38.1.4
Article Google Scholar
Duchowski AT, Biele C, Niedzielska A, Krejtz I (2018) Eye tracking cognitive load using pupil diameter and microsaccades with fixed gaze. PLoS ONE 13(9):e0203629. https://doi.org/10.1371/journal.pone.0203629
Article CAS PubMed PubMed Central Google Scholar
Edele A, Stanat P (2016) The role of first-language listening comprehension in second-language reading comprehension. J Educ Psychol 108(2):163–180. https://doi.org/10.1037/edu0000060
Article Google Scholar
Fathi J, Rahimi M, Derakhshan A (2024) Improving EFL learners’ speaking skills and willingness to communicate via artificial intelligence-mediated interactions. System 121:103254. https://doi.org/10.1016/j.system.2024.103254
Article Google Scholar
Fathi J, Rahimi M (2024) Utilising artificial intelligence-enhanced writing mediation to develop academic writing skills in EFL learners: a qualitative study. Comput Assisted Lang Learn. https://doi.org/10.1080/09588221.2024.2374772
Fathi J, Rahimi M, Teo T (2025) Applying intelligent personal assistants to develop fluency and comprehensibility, and reduce accentedness in EFL learners: an empirical study of Google Assistant. Lang Teach Res. https://doi.org/10.1177/13621688251317786
Feng L (2025) Investigating the effects of artificial intelligence-assisted language learning strategies on cognitive load and learning outcomes: a comparative study. J Educ Comput Res 62(8):1961–1994. https://doi.org/10.1177/07356331241268349
Article Google Scholar
Genesee F, Geva E (2006) Cross-linguistic relationships in working memory, phonological processes, and oral language. In D August & T Shanahan (eds). In Developing literacy in second-language learners: Report of the National Literacy Panel on language-minority children and youth. Mahwah, NJ: Lawrence Erlbaum Associates, 175–183
Geva E, Siegel LS (2000) Orthographic and cognitive factors in the concurrent development of basic reading skills in two languages. Read Writ: Interdiscip J 12(1/2):1–30. https://doi.org/10.1023/A:1008017710115
Article Google Scholar
Gottardo A, Mueller J (2009) Are first- and second-language factors related in predicting second-language reading comprehension? A study of Spanish-speaking children acquiring English as a second language from first to second grade. J Educ Psychol 101(2):330–344. https://doi.org/10.1037/a0014320
Article Google Scholar
Grabe W, Stoller FL (2020) Teaching and researching reading, 3rd edn. Routledge. https://doi.org/10.4324/9781315726274-2
Gutiérrez-Colón M, Frumuselu AD, Curell H (2023) Mobile-assisted language learning to enhance L2 reading comprehension: a selection of implementation studies between 2012–2017. Interact Learn Environ 31(2):854–862. https://doi.org/10.1080/10494820.2020.1813179
Article Google Scholar
Halkiopoulos C, Gkintoni E (2024) Leveraging AI in e-learning: personalized learning and adaptive assessment through cognitive neuropsychology—A systematic analysis. Electronics 13(18):3762. https://doi.org/10.3390/electronics13183762
Article Google Scholar
Hansen AL, Johnsen BH, Sollers JJ, Stenvik K, Thayer JF (2004) Heart rate variability and its relation to prefrontal cognitive function: the effects of training and detraining. Eur J Appl Physiol 93:263–272. https://doi.org/10.1007/s00421-004-1208-0
Article PubMed Google Scholar
Holmqvist K, Nyström M, Andersson R, Dewhurst R, Jarodzka H, Van de Weijer J (2011) Eye tracking: a comprehensive guide to methods and measures. Oxford: Oxford University Press
Hooda M, Rana C, Dahiya O, Rizwan A, Hossain MS (2022) Artificial intelligence for assessment and feedback to enhance student success in higher education. Math Probl Eng 2022:1–19. https://doi.org/10.1155/2022/7690103
Article Google Scholar
Horwitz EK, Horwitz MB, Cope J (1986) Foreign language classroom anxiety. Mod Lang J 70(2):125–132. https://doi.org/10.1111/j.1540-4781.1986.tb05256.x
Article Google Scholar
Hsiao JC, Chang JS (2024) Enhancing EFL reading and writing through AI-powered tools: design, implementation, and evaluation of an online course. Interact Learn Environ 32(9):4934–4949. https://doi.org/10.1080/10494820.2023.2207187
Article Google Scholar
Huang X, Zou D, Cheng G, Chen X, Xie H (2023) Trends, research issues and applications of artificial intelligence in language education. Educ Technol Soc 26(1):112–131
Google Scholar
Hudon A, Demazure T, Karran A, Léger PM, Sénécal S (2021) Explainable artificial intelligence (XAI): how the visualization of AI predictions affects user cognitive load and confidence. In: Information systems and neuroscience: NeuroIS Retreat 2021. Springer International Publishing, 237–246. https://doi.org/10.1007/978-3-030-88900-5_27
Jeon EH, Yamashita J (2014) L2 reading comprehension and its correlates: a meta-analysis. Lang Learn 64(1):160–212. https://doi.org/10.1111/lang.12034
Article Google Scholar
Jose J, Jayaron Jose B (2024) An overview of incorporating artificial intelligence tools for promoting learners’ reading skills. In: Innovative and intelligent digital technologies; towards an increased efficiency, vol 1. pp 603–613. https://doi.org/10.1007/978-3-031-70399-7_46
Kartal G, Yeşilyurt YE (2024) A bibliometric analysis of artificial intelligence in L2 teaching and applied linguistics between 1995 and 2022. ReCALL. https://doi.org/10.1017/S0958344024000077
Kim HG, Cheon EJ, Bai DS, Lee YH, Koo BH (2018) Stress and heart rate variability: a meta-analysis and review of the literature. Psychiatry Investig 15(3):235–245. https://doi.org/10.30773/pi.2017.08.17
Article PubMed PubMed Central Google Scholar
Kim NY (2019) A study on the use of artificial intelligence chatbots for improving English grammar skills. JDig Converg 17(8):37–46
Google Scholar
Klepsch M, Schmitz F, Seufert T (2017) Development and validation of two instruments measuring intrinsic, extraneous, and germane cognitive load. Front Psychol 8:294028. https://doi.org/10.3389/fpsyg.2017.01997
Article Google Scholar
Krueger RA, Casey MA (2014) Focus groups: a practical guide for applied research, 5th edn. Thousand Oaks, CA: SAGE Publications
Laborde S, Mosley E, Thayer JF (2017) Heart rate variability and cardiac vagal tone in psychophysiological research–recommendations for experiment planning, data analysis, and data reporting. Front Psychol 8:238557. https://doi.org/10.3389/fpsyg.2017.00213
Article Google Scholar
Lantolf JP, Thorne SL, Poehner ME (2014) Sociocultural theory and second language development. In: Theories in second language acquisition. New York, NY: Routledge, 221–240
Law, L (2024). Application of generative artificial intelligence (GenAI) in language teaching and learning: a scoping literature review. Comput Educ Open. https://doi.org/10.1016/j.caeo.2024.100174
Lazarus RS, Folkman S (1984) Stress, appraisal, and coping. Springer, New York
Google Scholar
Le Y, Liu J, Deng C, Dai DY (2018) Heart rate variability reflects the effects of emotional design principle on mental effort in multimedia learning. Comput Hum Behav 89:40–47. https://doi.org/10.1016/j.chb.2018.07.037
Article Google Scholar
Lee JY, Hwang Y (2022) A meta-analysis of the effects of using AI chatbot in Korean EFL education. Stud Engl Lang Lit 48(1):213–243. https://doi.org/10.21087/nsell.2022.11.83.213
Article Google Scholar
Liang JC, Hwang GJ, Chen MRA, Darmawansah D (2023) Roles and research foci of artificial intelligence in language education: an integrated bibliographic analysis and systematic review approach. Interact Learn Environ 31(7):4270–4296. https://doi.org/10.1080/10494820.2021.1958348
Article Google Scholar
Liaw, M-L, & English, K (2017). Technologies for teaching and learning L2 Reading. In: The handbook of technology and second language teaching and learning. Wiley Online Library, 62–76. https://doi.org/10.1002/9781118914069.ch5
Linck JA, Osthus P, Koeth JT, Bunting MF (2014) Working memory and second language comprehension and production: a meta-analysis. Psychon Bull Rev 21(4):861–883. https://doi.org/10.3758/s13423-013-0565-2
Article PubMed Google Scholar
Liu GZ, Fathi J, Rahimi M (2024) Using digital gamification to improve language achievement, foreign language enjoyment, and ideal L2 self: a case of English as a foreign language learners. J Comput Assist Learn 40(4):1347–1364. https://doi.org/10.1111/jcal.12954
Article Google Scholar
Liu H, Fan J (2025) AI‐mediated communication in EFL classrooms: the role of technical and pedagogical stimuli and the mediating effects of AI literacy and enjoyment. Eur J Educ 60(1):e12813. https://doi.org/10.1111/ejed.12813
Article Google Scholar
Mežek Š, McGrath L, Negretti R, Berggren J (2022) Scaffolding L2 academic reading and self‐regulation through task and feedback. TESOL Q 56(1):41–67. https://doi.org/10.1002/tesq.3018
Article Google Scholar
Moybeka AM, Syariatin N, Tatipang DP, Mushthoza DA, Dewi NPJL, Tineh S (2023) Artificial intelligence and English classroom: the implications of AI toward EFL students’ motivation. Edumaspul J Pendidik 7(2):2444–2454. https://doi.org/10.33487/edumaspul.v7i2.6669
Article Google Scholar
Muthmainnah M, Cardoso L, Alsbbagh YAMR, Al Yakin A, Apriani E (2024) Advancing sustainable learning by boosting student self-regulated learning and feedback through AI-driven personalized in EFL education. In: International conference on explainable artificial intelligence in the digital sustainability. Springer Nature, Cham, Switzerland, 36–54. https://doi.org/10.1007/978-3-031-63717-9_3
Nasri M, Kosa M, Chukoskie L, Moghaddam M, Harteveld C (2024, October) Exploring eye tracking to detect cognitive load in complex virtual reality training. In: 2024 IEEE international symposium on mixed and augmented reality adjunct (ISMAR-Adjunct). IEEE, 51–54. https://doi.org/10.1109/ISMAR-Adjunct64951.2024.00022
Nunan D, Sandercock GR, Brodie DA (2010) A quantitative systematic review of normal values for short‐term heart rate variability in healthy adults. Pacing Clin Electrophysiol 33(11):1407–1417. https://doi.org/10.1111/j.1540-8159.2010.02841.x
Article PubMed Google Scholar
Paas F, Renkl A, Sweller J (2003) Cognitive load theory and instructional design: Recent developments. Educ Psychol 38(1):1–4. https://doi.org/10.1207/S15326985EP3801_1
Article Google Scholar
Pan M, Guo K, Lai C (2024) Using artificial intelligence chatbots to support English-as-a-foreign language students’ self-regulated reading. RELC J. https://doi.org/10.1177/00336882241264030
Pedró F, Subosa M, Rivas A, Valverde P (2019) Artificial intelligence in education: challenges and opportunities for sustainable development. UNESCO, Paris
Google Scholar
Peng Y, Wang Y, Hu J (2023) Examining ICT attitudes, use and support in blended learning settings for students’ reading performance: Approaches of artificial intelligence and multilevel model. Comput Educ 203:104846. https://doi.org/10.1016/j.compedu.2023.104846
Article Google Scholar
Poehner ME, Lantolf JP (2010) Vygotsky’s teaching-assessment dialectic and L2 education: the case for dynamic assessment. Mind Cult Act 17(4):312–330. https://doi.org/10.1080/10749030903338509
Article Google Scholar
Pokrivcakova S (2019) Preparing teachers for the application of AI-powered technologies in foreign language education. J Lang Cult Educ 7(3):135–153. https://doi.org/10.2478/jolace-2019-0025
Article Google Scholar
Prabhu NS (1987) Second language pedagogy. Oxford: Oxford University Press
Rad HS, Alipour R, Jafarpour A (2023) Using artificial intelligence to foster students’ writing feedback literacy, engagement, and outcome: a case of Wordtune application. Interact Learn Environ. https://doi.org/10.1080/10494820.2023.2208170
Rahimi M, Fathi J, Zou, D (2024) Exploring the impact of automated written corrective feedback on the academic writing skills of EFL learners: an activity theory perspective. Educ Inf Technol. https://doi.org/10.1007/s10639-024-12896-5
Rayner K (1998) Eye movements in reading and information processing: 20 years of research. Psychol Bull 124(3):372–422. https://doi.org/10.1037/0033-2909.124.3.372
Article CAS PubMed Google Scholar
Rayner K, Castelhano MS, Yang J (2009) Eye movements and the perceptual span in older and younger readers. Psychol Aging 24(3):755. https://doi.org/10.1037/a0014300
Article PubMed Google Scholar
Rusmiyanto R, Huriati N, Fitriani N, Tyas NK, Rofi’i A, Sari MN (2023) The role of artificial intelligence (AI) in developing English language learner’s communication skills. J Educ 6(1):750–757. https://doi.org/10.31004/joe.v6i1.2990
Article Google Scholar
Ryan RM, Deci EL (2000) Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist 55(1):68. https://doi.org/10.1037/0003-066X.55.1.68
Article CAS PubMed Google Scholar
Ryan S (2005) Motivational factors questionnaire. School of English Studies, University of Nottingham, Nottingham
Google Scholar
Shadiev R, Feng Y (2023) Using automated corrective feedback tools in language learning: a review study. Interact Learn Environ. https://doi.org/10.1080/10494820.2022.2153145
Shaffer F, Ginsberg JP (2017) An overview of heart rate variability metrics and norms. Front Public Health 5:258. https://doi.org/10.3389/fpubh.2017.00258
Article PubMed PubMed Central Google Scholar
Shaffer F, McCraty R, Zerr CL (2014) A healthy heart is not a metronome: an integrative review of the heart’s anatomy and heart rate variability. Front Psychol 5:108292. https://doi.org/10.3389/fpsyg.2014.01040
Article Google Scholar
Shin J. (2020) A meta-analysis of the relationship between working memory and second language reading comprehension: Does task type matter? Applied Psycholinguistics 41(4):873–900. https://doi.org/10.1017/S0142716420000272
Article Google Scholar
Sidek HM (2012) EFL reading instruction: communicative task-based approach. Int J Instruct 5(2):109–128
ADS Google Scholar
Spielberger, CD (1972) Conceptual and methodological issues in anxiety research. Current trends in theory andresearch (pp. 481–493). Academic Press
Sweller J (1988) Cognitive load during problem solving: eeffects on learning. Cogn Sci 12(2):257–285. https://doi.org/10.1207/s15516709cog1202_4
Article Google Scholar
Tabachnick BG, Fidell LS (2019) Using multivariate statistics, 7th edn. New York, NY: Pearson
Taylor AM (2021) Technology and reading: the effects of CALL glossing. Psychol Rep 124(5):2092–2118. https://doi.org/10.1177/0033294120954139
Article PubMed Google Scholar
Valijärvi RL, Tarsoly E (2019) ‘Language students as critical users of Google Translate’: pitfalls and possibilities. Pract Res High Educ 12(1):61–74
Google Scholar
Vygotsky LS (1978) Mind in society: The development of higher psychological processes. Harvard University Press. https://www.hup.harvard.edu/books/9780674576292
Wang X, Liu Q, Pang H, Tan SC, Lei J, Wallace MP, Li L (2023) What matters in AI-supported learning: a study of human-AI interactions in language learning using cluster analysis and epistemic network analysis. Comput Educ 194:104703. https://doi.org/10.1016/j.compedu.2022.104703
Article Google Scholar
Wei L (2023) Artificial intelligence in language instruction: impact on English learning achievement, L2 motivation, and self-regulated learning. Front Psychol 14:1261955. https://doi.org/10.3389/fpsyg.2023.1261955
Article PubMed PubMed Central Google Scholar
Xin Z, Derakhshan A (2024) From excitement to anxiety: exploring English as a foreign language learners’ emotional experiences in the artificial intelligence‐powered classrooms. Eur J Educ. https://doi.org/10.1111/ejed.12845
Xiu-Yi WU (2024) AI in L2 learning: a meta-analysis of contextual, instructional, and social-emotional moderators. System. https://doi.org/10.1016/j.system.2024.103498
Yin L, Fathi J (2025) Exploring the motivational dynamics of Chinese learners on tandem and Hellotalk: a self-determination theory perspective. Learn Motiv 90:102113. https://doi.org/10.1016/j.lmot.2025.102113
Article Google Scholar
Zhai C, Wibowo S (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. https://doi.org/10.1016/j.caeai.2023.100134
Article Google Scholar
Zhang S, Zhang X (2022) The relationship between vocabulary knowledge and L2 reading/listening comprehension: a meta-analysis. Lang Teach Res 26(4):696–725. https://doi.org/10.1177/1362168820913998
Article ADS Google Scholar
Zheng S (2024) The effects of chatbot use on foreign language reading anxiety and reading performance amongChinese secondary school students Computers and Education: Artificial Intelligence 7:100271. https://doi.org/10.1016/j.caeai.2024.100271
Article Google Scholar
Zheng S (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. https://doi.org/10.1016/j.caeai.2024.100271
Article Google Scholar
Zou B, Guan X, Shao Y, Chen P (2023) Supporting speaking practice by social network-based interaction in artificial intelligence (AI)-assisted language learning. Sustainability 15(4):2872. https://doi.org/10.3390/su15042872
Article Google Scholar
Download references
School of Foreign Languages, Zhengzhou Shengda University of Economic Business & Management, Zhengzhou, China
Hang Yuan
PubMed Google Scholar
Hang Yuan designed the study, collected and analyzed the data, and wrote the manuscript as the sole contributor.
Correspondence to Hang Yuan.
The author declares no competing interests.
This study was approved by the School of Foreign Languages at Zhengzhou Shengda University of Economic Business & Management (Ethics Approval No. SFL-2023-047, granted on October 12, 2023). All procedures involving human participants followed the ethical standards of the institutional and national research committees and aligned with the 1964 Helsinki Declaration and its amendments.
All participants voluntarily joined this study, which posed no physical, emotional, or psychological risks. Written informed consent was obtained from participants and their legal guardians between January 8, 2024, and February 29, 2024, with a clear explanation of procedures, tasks, and their right to withdraw at any time without penalty.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Reprints and permissions
Yuan, H. Artificial intelligence in language learning: biometric feedback and adaptive reading for improved comprehension and reduced anxiety. Humanit Soc Sci Commun 12, 556 (2025). https://doi.org/10.1057/s41599-025-04878-w
Download citation
Received:
Accepted:
Published:
Version of record:
DOI: https://doi.org/10.1057/s41599-025-04878-w
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
Humanities and Social Sciences Communications (Humanit Soc Sci Commun)
ISSN 2662-9992 (online)
© 2026 Springer Nature Limited