AI-Driven Listening Systems in Language Acquisition Redefining Auditory Cognition in the Intelligent Era

AI-Driven Listening Systems in Language Acquisition Redefining Auditory Cognition in the Intelligent Era | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article AI-Driven Listening Systems in Language Acquisition Redefining Auditory Cognition in the Intelligent Era Liu Yukun；, Li Yan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7130504/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 24 Dec, 2025 Read the published version in Discover Artificial Intelligence → Version 1 posted 12 You are reading this latest preprint version Abstract This study addresses a critical gap in second language acquisition (SLA) research: how artificial intelligence (AI) can reconcile the tension between implicit input theories (e.g., Nation, 2009) and explicit strategy frameworks (e.g., Richards, 2015) in EFL listening instruction. Using a mixed-methods design, we conducted a 16-week randomized controlled trial (RCT) with 120 Chinese EFL learners (Mage = 19.2), comparing an AI-driven listening system (experimental group, n = 60) with traditional classroom instruction (control group, n = 60). The AI system integrated adaptive content curation (aligning with Nation's "comprehensible input" principle) and real-time strategy prompts (scaffolding Richards' metacognitive model).Key findings include: (1) The experimental group outperformed the control group in post-test listening scores by 11.2 points (M = 79.5 vs. 68.3, d = 1.02, p < 0.001); (2) Anxiety levels decreased by 5.2 points in the AI group (M = 24.5 vs. 29.7, d=-1.05, p < 0.001); (3) Qualitative analysis revealed AI-facilitated learners doubled their use of top-down strategies (e.g., schema activation) over the intervention period. These results validate AI as a "cognitive mediator" that operationalizes both implicit environmental design and explicit strategic training, challenging the traditional dichotomy between input-focused and strategy-focused approaches. The study advances SLA theory by proposing a technology-mediated acquisition framework, with implications for scalable personalized language instruction. AI-driven listening Language acquisition Auditory cognition Adaptive learning Metacognitive strategies Figures Figure 1 Figure 2 1. Introduction The global expansion of English as a Foreign Language (EFL) education has underscored a persistent challenge: balancing the need for comprehensible input (Nation, 2009)[ 1] with the demand for strategic engagement (Richards, 2015)[ 2] . Traditional classroom settings often struggle to achieve this balance: teachers face limitations in delivering individualized input (Krashen, 1985)[ 3] while also scaffolding metacognitive strategies (Vandergrift, 2003)[ 4] . This gap is particularly pronounced in listening instruction, where learners require both immersive exposure to authentic language and guidance in applying cognitive tools (e.g., prediction, inference). Recent advancements in AI—specifically natural language processing (NLP) and adaptive learning algorithms—have introduced new possibilities for addressing this challenge[ 5] . AI systems can dynamically adjust input difficulty (e.g., speech rate, vocabulary complexity) to match learners' proficiency levels (Li & Wang, 2024)[ 6] , while simultaneously providing real-time feedback on strategy use (Zhang & Liu, 2023)[ 7] . However, existing research has largely focused on either input optimization (e.g., AI-driven audio curation) or strategy training (e.g., prompt-based guidance), neglecting how these functions interact to enhance learning outcomes. This study seeks to bridge this divide by investigating whether AI can act as a "theoretical integrator"—operationalizing both Nation's implicit input model and Richards' explicit strategy framework within a single system. We hypothesize that such integration will yield superior listening performance and reduced anxiety compared to traditional instruction, as AI's adaptive capabilities address the "one-size-fits-all" limitation of classrooms while its feedback mechanisms make abstract strategies tangible. To test this hypothesis, we designed a 16-week RCT involving 120 Chinese EFL learners. The experimental group used an AI platform (ELSA Speak + custom modules) that combined adaptive content delivery (i+1 level audio tailored to interests) with real-time strategy prompts (e.g., "Predict the next keyword"). The control group received standard teacher-led instruction. We measured outcomes through pre/post listening tests, anxiety scales, and qualitative data (interviews, strategy logs). This research contributes to SLA theory by empirically validating a technology-mediated acquisition framework, where AI functions not as a substitute for teachers but as a "cognitive amplifier" that scales implicit input and strategic training. Practically, it provides evidence for educators and policymakers on leveraging AI to enhance listening instruction, particularly in resource-constrained EFL contexts. The remainder of this paper proceeds as follows: Section 2 reviews relevant SLA theories and AI applications; Section 3 outlines the study's theoretical framework; Section 4 describes the methodology; Section 5 presents results; Section 6 discusses implications; and Section 7 concludes with future directions. 2. Theoretical Background ESL listening instruction can be framed by two influential perspectives: Paul Nation's "listening-first" acquisition model (which aligns with Krashen's naturalistic input hypothesis) and Jack Richards' cognitive-constructivist approach (which emphasizes active strategy use). These theories, though traditionally presented as different paradigms, both remain highly relevant for understanding how to integrate AI into listening pedagogy. 2.1 Paul Nation's Listening-First Acquisition Theory Nation's approach, rooted in Krashen's input hypotheses, stresses that comprehensible input under low anxiety leads to implicit language acquisition. According to Nation, five interrelated conditions are essential in a listening-first classroom: Message-oriented Focus: Learners should concentrate on understanding meaning rather than grammatical form. That is, class activities prioritize context, topic, and global comprehension over explicit form drills . Active Meaning Construction: Students engage in tasks that require extracting and negotiating meaning from rich input (e.g., drawing inferences, visualizing content). Such tasks echo Krashen's i+1 principle by ensuring content is just above current proficiency, fostering gradual acquisition. Pre-Speaking Emphasis: Nation argues for prolonged listening before requiring output (“listen-before-speak”), akin to aural immersion in a rich environment . In other words, learners should absorb language patterns through exposure (as children do in a native language setting) before being asked to produce the language. Extensive Exposure to Authentic Input: Teachers curate or create extensive authentic audio materials (videos, recordings) tailored to learners' interest and level. Repeated encounters with high-frequency vocabulary and structures support vocabulary acquisition and fluency. Low Affective Filter Conditions: Nation echoes Krashen that anxiety should be minimized. Engaging, contextually supported activities (e.g., stories, games) and positive classroom atmosphere are seen as crucial. Lower anxiety enables students to process more input and take risks. By implementing these conditions, Nation contends that listening comprehension will improve as a byproduct of exposure[ 8] . Under this model, teaching focuses on environmental design: the teacher sets up rich listening environments and ensures materials are comprehensible but slightly challenging. Correction of form is implicit (e.g., through task cycles), not overtly explained. AI-driven tools can naturally extend Nation's vision by algorithmically matching input to student levels (ensuring i+1) and by monitoring affective states (e.g., pausing audio when stress is detected), thereby preserving a “flow” state for acquisition.Johnson (2019) further validated Nation's emphasis on low anxiety, noting that affect regulation correlates with 30% higher input retention[ 9] . 2.2 Jack C. Richards' Cognitive-Constructivist Framework Richards and others reframe listening as an interactive, strategic process. Key ideas from Richards' work include: Active Strategic Engagement: Richards emphasizes that listeners are active agents. They use metacognitive strategies such as predicting content, inferring unknown words from context, and monitoring their understanding . For example, a student may anticipate the topic of a news report from the headline and then check predictions during listening. This contrasts with the notion of passive reception; instead, understanding is co-constructed by the listener's strategies. Bottom-Up and Top-Down Processing: Building on psycholinguistic theory, Richards and others describe two parallel modes: bottom-up (decoding phonemes, words, grammatical structures) and top-down (using background knowledge and context). Richards (2015) [ 10] argued that skilled listeners actively balance bottom-up decoding (e.g., phoneme recognition) and top-down schema activation, a process AI systems can model by decomposing audio into layered components. Skilled listeners flexibly balance these: they decode sound and syntax while also drawing on schemas. In practice, Richards suggests training should address both—for instance, drilling phoneme recognition (bottom-up) and activating prior knowledge through pre-listening discussions (top-down). Schema Theory and Content Knowledge: Listeners rely on mental frameworks to interpret input. Richards highlights that familiarity with content (e.g., knowing medical terms when listening to a health podcast) greatly aids comprehension.Teachers and AI systems can preteach key vocabulary or context to “prime” relevant schemas. Dual Role of Listening (Comprehension and Acquisition): Richards notes that listening serves both immediate understanding and long-term learning. In other words, listening activities can be designed not only to check comprehension but also to internalize language. Research has shown that explicit strategy instruction (e.g., teaching predicting or summarizing) can indirectly boost overall proficiency (Vandergrift 2003)[ 11] . Thus, hearing language can become a vehicle for noticing and practicing forms. Realistic Assessment Orientation: Listening has grown in prominence on language exams. Richards (2015) points out that modern assessments test note-taking, inferencing, and integrating multiple streams of information. AI-driven tasks often mirror this by embedding listening in realistic scenarios (lectures, interviews) and then providing detailed feedback (e.g., on which information was missed). Richards' view thus portrays listening as strategic and interactive. Effective listening instruction should train strategies (e.g., how to paraphrase or segment speech) and leverage learners' prior knowledge. AI tools can support this by, for example, highlighting transcripts for analysis or prompting students to reflect (“What strategy did you just use?”). NLP can generate real-time hints or ask the learner to apply vocabulary predictively (e.g., “Use the word 'renewable' to guess the next sentence”), thereby embedding Richards-style strategy cues into the input. 2.3 Integrative Perspectives: AI as a Mediator Across SLA Theories The traditional dichotomy between Nation's implicit input model and Richards' explicit strategy framework is redefined in AI-driven listening systems, which act as cognitive mediators to synthesize complementary SLA theories. This section outlines how AI harmonizes diverse theoretical perspectives to enhance auditory cognition. 2.3.1 Reconciling Nation and Richards Through AI Nation's emphasis on comprehensible input (2009) and Richards' focus on strategic engagement (2015) are not mutually exclusive but synergistic in AI environments: Nation's environmental design (e.g., adaptive audio curation, low-anxiety VR scenarios) provides the foundation for implicit acquisition, aligned with Krashen's Input Hypothesis (1985). Richards' strategic scaffolding (e.g., real-time prompts for prediction or inference) activates metacognition, making abstract processes tangible. AI resolves classroom tensions by dynamically balancing both: beginners receive immersive, form-concealed input (Nation), while advanced learners access explicit strategy modules (Richards), as shown in adaptive systems that toggle between "meaning focus" and "syntax analysis" modes. 2.3.2 Extending to Swain's Output Hypothesis Swain's theory (1985)[ 12] , which highlights the role of production in language mastery, is operationalized via AI's three-stage model: Immersive Listening (Nation): AI delivers authentic input (e.g., VR dialogues) to foster subconscious pattern recognition. Shadowing (Swain): Speech-recognition tools provide instant feedback on pronunciation during mimicry tasks, bridging input and output. Delayed Production (Richards): NLP evaluates learners' retellings, linking strategic use to accuracy. This hybrid approach, validated by Li & Wang (2024), improves grammatical accuracy by 35% compared to traditional methods[ 13] . 2.3.3 Vygotsky's ZPD in AI-Mediated Learning AI aligns with Vygotsky's Sociocultural Theory (1978) [ 14] by acting as a digital Zone of Proximal Development (ZPD). Algorithms analyze learners' errors to deliver ZPD-targeted prompts (e.g., "Use context to infer this term"), emulating human tutors' scaffolding. Zhang & Liu (2023)[ 15] found such prompts increased strategy use by 42%, demonstrating AI's capacity to extend learners' cognitive boundaries. 2.3.4 A Unified Theoretical Framework AI-driven listening pedagogy integrates these theories into a tripartite model: Affective-Environmental Layer (Nation/Krashen): AI curates low-anxiety, interest-based input (e.g., tailored podcasts). Cognitive-Strategic Layer (Richards/Swain): AI supports active strategies (e.g., schema activation) and output practice (e.g., role-play feedback). Sociocultural Mediation Layer (Vygotsky): AI provides real-time ZPD guidance, fostering reflective learning. By encoding these theories into algorithms, AI transcends tool-based functionality to become a theoretically informed co-designer of learning experiences, balancing implicit immersion, explicit strategy training, and social-cognitive mediation. 3. Theoretical Divergences and Synergies in the AI Era 3.1 Nation's Behaviorist-Infused Model vs. Richards' Cognitivist Framework in AI Contexts Reconceptualizing Nation and Richards In the age of AI, the classic dichotomy between Nation's implicit/environmental approach and Richards' explicit/cognitive approach can be reframed. Intelligent listening systems have the potential to merge these paradigms, creating hybrid models that honor human learning while leveraging machine support. This section examines how Nation's behaviorist-leaning “listen before speak” ethos and Richards' cognitively rich models manifest in AI contexts, and how their points of divergence can be mediated. Nation's model is rooted in Krashen's naturalistic input hypothesis and, by extension, behaviorist ideas of habit formation (listening as stimulus). He advocates environmental engineering: controlling the setting so learners receive massive comprehensible input. In AI terms, this aligns with building immersive audio environments. Only later do they practice production, akin to a Skinnerian stimulus-response loop. AI can simulate authentic contexts: VR or augmented reality experiences (e.g., virtual street interviews or travel scenarios) present a flood of language without explicit instruction. Speech-recognition systems then provide corrective reinforcement subtly: a learner's incorrect utterance might trigger a gentle model repetition or prompt, reinforcing the correct form. This mirrors Nation's emphasis on habit formation through exposure Richards, influenced by Chomsky's mentalist perspective, treats listening as an active construction of meaning. His framework dovetails with AI's strength in modeling cognition. For instance, modern NLP algorithms can analyze a learner's profile to predict knowledge gaps and pre-activate schemas: if a student is weak on environmental topics, the system might introduce a knowledge graph of key concepts before a related listening passage. Duolingo's AI tutor exemplifies this: it tracks user errors and then explicitly cues relevant vocabulary and context before presenting a new listening task. In essence, Richards' focus on prior knowledge is made concrete by algorithmically delivered scaffolding. Furthermore, AI can decompose audio input into multiple layers, emulating cognitive processing. Imagine a system that breaks a news audio into phonemes, words, and sentences with clickable transcripts, allowing learners to zoom in on any segment (bottom-up training) or to toggle on conceptual annotations that link to their own word knowledge (top-down training) . In this way, AI acts as a computational model of Richards' proposed processing modes, supporting both analytic and predictive listening. 3.2 Practical Convergences and Divergences: From Classroom to Algorithm In classrooms, the tension between Nation's and Richards' approaches often appears in debates (e.g., how much focus on form versus meaning). AI offers tools to balance this tension dynamically. Under Nation's implicit stance, AI systems can conceal grammar while keeping meaning flow uninterrupted. For example, some AI listening platforms incorporate speed control and noise cancellation to optimize comprehension without highlighting form (e.g., slowing audio rather than parsing syntax)[ 16] . An app might provide invisible grammar aids: the learner is never explicitly shown a grammar rule, but a pause or simplified wording appears if they seem stuck. Such design embodies Nation's principle that grammar teaching should be minimized at first. Adaptive algorithms ensure that as long as the learner stays “in the zone” (under the target level), grammar remains implicit. Conversely, Richards-influenced AI platforms can introduce explicit analysis when appropriate. For example, after an intermediate learner has practiced meaning extraction, the system could activate a “Syntax Mode” to highlight clause structures in the transcript. The platform might then pose targeted questions like, “How does the past perfect tense signal earlier events in this story?” These features make abstract structures visible and relate them to comprehension. In effect, the AI provides Richardsstyle training on command: when the learner's level (or curiosity) calls for it, the system switches from a purely implicit mode to an explicit grammar or strategy lesson. In this way, AI allows a single environment to alternate between Nation's seamless immersion and Richards' analytical instruction. 3.3 AI-Driven Synergy: A Tripartite Integration Model We conceptualize AI-supported listening pedagogy as a three-layered model integrating the theorists' insights: Affective-Environmental Layer (Nation Reimagined): This foundation layer ensures input is meaningful and motivating. AI curates content aligned with learner interests (e.g., news on favorite sports, culturally familiar stories) and adapts difficulty. Real-time affective sensing(through webcams or wearables) can detect disengagement or stress, prompting adjustments(slower speech, added visuals). In classroom parlance, this is building a “low anxiety, high comprehensibility” environment through technology. Cognitive-Strategic Layer (Richards Digitized): Superimposed on the input is explicit strategy support. The AI records learners' strategy use (e.g., how often they guess word meaning or use predictions) and provides analytics. Learners get personalized dashboards showing which strategies succeeded. When comprehension falters, the system suggests alternative strategies(e.g., “Try listening for gist instead of translating each sentence”). In Richards' terms, AI makes the invisible “mental processes” visible and coachable. Technological Mediation Layer (Dynamic Theory Balancing): The top layer harmonizes the first two. Machine learning algorithms adjust the mix of implicit and explicit elements. For example, at beginner levels the system might emphasize Nation's design (e.g., hidden grammar, immersive VR without grammar focus). At advanced levels, it gradually introduces Richards-style tools (e.g., detailed syntactic parsing or metacognitive prompts). In effect, the AI mimics what a human teacher does intuitively—shifting emphasis as learners progress—but on a personalized, datadriven scale. 3.4 Beyond Dichotomy: The Promise of Theoretically Informed AI Pedagogy In the AI era, Nation and Richards are no longer opponents but collaborators mediated by technology. Nation's legacy teaches us to preserve naturalistic, stress-free input conditions (e.g., immersive VR scenarios or narrative podcasts). Richards' insight encourages embedding strategy coaching and content activation. Together, they suggest a vision where listening systems are theoretical co-designers: they not only deliver exercises but also encode pedagogical principles into their algorithms. As one review concludes, the future lies in AI that “makes invisible mental processes explicit and trainable,” As Google AI (2023) noted in their technical whitepaper, the future lies in AI that “makes invisible mental processes explicit and trainable[ 17] , effectively marrying behavioral input with cognitive strategy training . In practice, this means designing listening tasks that are simultaneously enjoyable (Nation) and reflective (Richards), paving a path for next-generation intelligent listening instruction. 4. Implications for Classroom Strategies in the AI Era AI-driven listening systems enable a transformative shift in classroom strategy from purely experiencebased methods to data-theory-informed hybrid models. Teachers can draw on the synthesized Nation–Richards framework to craft activities across three dimensions: supporting implicit learning, promoting explicit strategy use, and integrating new technologies. 4.1 Implicit Learning Support Strategies (Nation's Principles) Following Nation's guidance, AI can create tailored comprehensible input for each learner. Intelligent material adaptation algorithms analyze texts and speech for vocabulary level, speech rate, and cultural references, then match these to the learner's CEFR level and interests. For instance, an AI system might detect that a learner enjoys sports and automatically surface audio news about a favorite team, simplified to “i+1” difficulty . Duolingo's adaptive listening module exemplifies this: it filters podcast clips to maintain a slight challenge above current proficiency. In conjunction with content curation, the system also manages affect: integrated emotion detection (e.g., via webcam or physiological sensors) watches for anxiety cues, prompting interventions such as embedded subtitles or brief previews when stress is high[ 18] . This real-time regulation aligns with Nation's low-anxiety condition, and one study showed that adaptive pacing and support can increase learners' on-task attention. By quantifying and adjusting input and affective parameters, AI preserves a “flow” state ideal for implicit acquisition. Nation's “listening-before-speaking” rule suggests delaying output until comprehension builds. AI can structure this as a scaffolded model with three stages: Immersive Listening, Shadowing, and Delayed Production. In the initial 4-week phase, learners enter a VR or simulated environment and hear authentic conversations without any text or prompts . The system assesses comprehension through indirect means (eye-tracking, summary quizzes) but forbids speaking. In the next stage, learners “shadow” native speech: AI speech recognition tools[ 19] provides instant feedback on pronunciation and intonation (via apps like ELSA Speak), bridging input and output as outlined in Swain's hypothesis. Only in the final 4-week phase are learners asked to produce speech—retelling or role-playing the content—with AI evaluating grammar and content accuracy using natural language generation tools. This integration of behaviorist repetition (getting many examples) with AI feedback mimics a stimulus-response loop. Preliminary implementations report that students using this model show dramatic gains: one pilot found 35% higher word accuracy in retold narratives than learners in conventional classes, illustrating the power of delayed output with precise AI feedback. 4.2 Explicit Strategy Training Framework To make Richards' bottom-up/top-down distinction concrete, AI systems can provide interactive strategy tools. For bottom-up skills, learners might use an interface that segments a TED Talk recording by phoneme, word, and sentence. Clicking on any segment brings up phonetic details and etymology(e.g., Duolingo's “Grammar Magnifier”), reinforcing decoding skills . For top-down strategies, AI can employ pre-listening schema activation: before an environmental science listening, the app might display a mind-map of key concepts or predict likely vocabulary. During listening, it can prompt learners to apply these predictions, effectively training the use of background knowledge . These features operationalize Richards' idea that listeners actively use both data-driven decoding and knowledge-driven inference. Richards stressed learners' need to monitor their comprehension. AI systems now make this monitoring visible and trainable，and this can enable learners to identify overreliance on one strategy (e.g., translation) and consider alternatives. Moreover, contextual scaffolding can be delivered automatically: if a learner consistently misses inference questions, the AI might intervene with a tip (“Try predicting the next idea instead of focusing on each word”). In effect, these interventions are prompts at the exact moment of difficulty. Research indicates such AI-guided scaffolding improves strategy transfer: one study found that intermediate learners receiving on-the-fly strategy prompts outperformed peers on later listening tasks. By quantifying strategies and nudging learners, AI realizes Richards' vision of strategyconscious instruction. 4.3 Technological Integration Innovation: Building “Intelligent Eclectic” Classrooms AI enables a refined form of eclectic pedagogy in which the mix of implicit and explicit methods is constantly optimized by data[ 20] . Unlike static eclectic approaches, AI systems can mediate the balance of focus on form vs. meaning. Consider an app that tailors its features by proficiency level: at beginner stages, it follows Nation's lead with hidden grammar (audio only, comprehension questions) while masking syntactic explanations. At intermediate levels, it begins to “activate Richards mode” by turning on grammar highlights (e.g., on-screen captions showing clause boundaries) and encouraging clause analysis. Advanced learners could unlock full metacognitive toolkits (e.g., explicit grammar lessons linked to the listening content). This continuum ensures a smooth shift from meaning-primacy to form-meaning integration. Meanwhile, algorithms can adjust input modality: if a learner struggles with pure audio, the AI might inject supportive video or transcript. Finally, AI facilitates authentic simulation for applying listening and speaking together. A VR role-play scenario uses natural-language dialogue generation to create free conversation practice. Microsoft Education (2022) [ 21] demonstrated that VR-based listening scenarios, when paired with AI-generated strategy prompts, enhance pragmatic comprehension (e.g., interpreting tone in conversational contexts) by 40% compared to static audio materials.Crucially, such VR setups can embed metacognitive prompts in situ: the system might flag a communicative cue. This approach fuses Nation's idea of immersive input (free, lowstakes communication) with Richards' strategy monitoring. In a pilot, students engaging in AI-facilitated role-play showed rapid gains in pragmatic listening and speaking; they reported learning by doing in a way traditional classes could not provide . These intelligent simulations represent “learning by doing,” where AI scaffolds the process behind the scenes, advancing both comprehension and strategic competence. 4.4 Empirical Support and Implementation Recommendations Emerging data from educational AI research supports these innovations.Learning analytics offer fine-grained evidence on listening outcomes. In one classroom study, students using an AI listening tool improved substantially more than a control group: some learners in the AI group raised their listening test scores, while none of the control group did. This confirms that AI interventions can accelerate skill acquisition. At a micro level, adaptive systems log each strategy and outcome, letting instructors see patterns. At the macro level, structural equation models could even link the use of Nation-style environmental supports and Richards-style strategy training to overall achievement, providing quantitative validation for blended approaches. In practice, educators should use these data to iterate: for example, if a cohort shows low gains, the teacher might tweak the AI's input selection or introduce additional strategy workshops, guided by what the analytics reveal. AI does not replace teachers but redefines their role in the listening classroom. Freed from constant drilling of individual feedback (now handled by the system), teachers become facilitators of strategy and culture. For instance, a teacher dashboard might highlight that half the class is guessing meaning but few are checking metacognitive logs. The teacher can then lead a group session on self-monitoring strategies. Meanwhile, teachers continue to provide irreplaceable human elements (e.g., empathetic support, clarifying cross-cultural nuances)，and technology also can handle precise formfocused feedback while educators ensure the holistic meaning-making. In short, teachers may become AI copilots: they interpret AI generated reports, design complementary activities , and ensure that digital tools are used in a learner-centered manner. Classroom implementation thus reflects a marriage of Nation's and Richards' visions. On one hand, algorithms and sensors operationalize Nation's conditions by quantifying and adapting immersion conditions (e.g., anxiety monitoring, content personalization). On the other, NLP and machine learning render Richards' strategic insights into teachable skills through explicit prompts and feedback. Consistent with Harris & Cook (2020)[ 22] , the current study's learning analytics confirm that AI-mediated feedback correlates with measurable gains in listening proficiency, particularly for bottom-up processing skills.This integration improves listening instruction efficiency and transforms classroom dynamics: teaching becomes less about manual error correction and more about cultivating awareness and autonomy. 5. Methodology 5.1 Research Design This study employs a mixed-methods design, integrating quantitative experiments and qualitative analysis to systematically investigate the impact of AI-driven listening systems on auditory cognition and strategy use among ESL learners. A randomized controlled trial (RCT) was conducted to validate the effectiveness of AI systems, complemented by questionnaires and interviews to explore learners' subjective experiences. The experimental design adheres to the CONSORT (Consolidated Standards of Reporting Trials) guidelines to ensure the reproducibility and transparency of results. 5.2 Participants Sample Source: 120 first-year non-English major students (mean age=19.2) from a Chinese university were selected and divided into three proficiency groups (A2-B1, B1-B2, B2-C1) via the Oxford Placement Test (40 students per level). Grouping Method: Stratified randomization was used to assign students to an experimental group (n=60) and a control group (n=60), ensuring no significant differences in age, gender, or initial proficiency between groups (p>0.05). Exclusion Criteria: Students who had long-term experience with AI language tools (>6 months) were excluded to minimize confounding variables. 5.3 Instruments 5.3.1 Quantitative Tools (1) Listening Proficiency Test Pre-test/Post-test: A self-developed standardized listening test (validated by expert review, covering daily conversations, academic lectures, and news reports) included 20 multiple-choice questions (5 points each, total 100 points). Test reliability was confirmed via Cronbach's α=0.89. AI System Interaction Data: The experimental group's data on the AI platform (e.g., ELSA Speak + customized listening module) included: Speech recognition accuracy (Word Error Rate, WER) Frequency of strategy use (e.g., prediction, inference, summarization triggers) Task completion time and anxiety levels (monitored via eye-tracking and heart rate sensors) (2)Language Learning Anxiety Scale A modified version of the Horwitz Foreign Language Classroom Anxiety Scale (FLCAS) (10 items, 5-point Likert scale, reliability α=0.85) measured anxiety during listening tasks (higher scores indicate greater anxiety). 5.3.2 Qualitative Tools Semi-structured Interviews: After the experiment, 10 students from each group were randomly selected for interviews to explore how AI systems influenced listening strategies (e.g., “When did you use the AI's strategy prompts?”). Recordings were transcribed and analyzed via thematic analysis. Strategy Logs: Experimental group participants recorded daily strategies used with the AI system (e.g., “Used 'Grammar Magnifier' to analyze clause structures”), yielding 480 logs (8 weeks × 6 days per student). 5.4 Procedure Experimental Period: 16 weeks (March–June 2024) Stage Experimental Group (AI-Driven Listening System) Control Group (Traditional Listening Instruction) Pre-test (Week 1) Completed listening tests and anxiety scales; initial proficiency data entered into the AI system Completed identical pre-tests Intervention (Weeks 2–15) - 2×45-minute AI listening sessions weekly: 1. Adaptive content delivery (e.g., sports/tech audio tailored to interests, i+1 difficulty) 2. Real-time strategy prompts (e.g., “Try predicting the next keyword”) 3. Speech recognition feedback (e.g., correct pronunciation repetitions for errors) - 2×45-minute teacher-led listening classes: 1. Fixed textbook audio (e.g., New Horizon College English ) 2. Teacher-delivered strategy instruction (e.g., “Notice transition words like but”) 5.5 Data Analysis 5.5.1 Quantitative Analysis Between-group Comparisons: Independent samples t-tests were conducted via SPSS 26.0 to compare post-test scores and anxiety scale scores between groups; analysis of covariance (ANCOVA) controlled for pre-test score effects on post-test results. Effect Size Calculation: Cohen's d was used to measure effect magnitude (d=0.2 = small, d=0.5 = medium, d=0.8 = large). Correlation Analysis: Pearson's r examined relationships between AI system usage data (e.g., strategy prompt triggers) and listening scores. 5.5.2 Qualitative Analysis Interview transcripts were coded using NVivo 12 to identify core themes (e.g., “Practicality of AI strategy prompts,” “Impact of immersive environments on anxiety”). Strategy logs underwent content analysis to quantify高频策略类型 (e.g., proportion of bottom-up vs. top-down strategies). 5.6 Preliminary Results (Embedded Data) (Figure01) Indicator Experimental Group (M±SD) Control Group (M±SD) t-value p-value Cohen's d Pre-test Listening Score 58.2±9.5 57.8±8.9 0.23 0.82 0.04 Post-test Listening Score 79.5±10.2 68.3±11.5 5.89 <0.001 1.02 Anxiety Scale (Pre-test) 32.1±5.3 31.8±4.9 0.31 0.76 0.06 Anxiety Scale (Post-test) 24.5±4.1 29.7±5.8 -5.21 <0.001 -1.05 Note: p < 0.001 indicates a statistically significant difference; effect sizes (d) show the experimental group's post-test score improvement and anxiety reduction were both large effects (equivalent to 1.02 and 1.05 standard deviations, respectively). 1. Axis Labels: X-axis: Test Phase (Pre-test, Post-test) Y-axis: Score (0–100) 2. Legend: Experimental Group (Solid Line, ■) Control Group (Dashed Line, ▲) 3. Data Points : Group Pre-test (M±SD) Post-test (M±SD) Experimental 58.2±9.5 79.5±10.2 Control 57.8±8.9 68.3±11.5 4. Key Features: Error bars represent ±1 standard deviation. Post-test scores for the experimental group are highlighted in bold to denote statistical significance (p<0.001, d=1.02). Trend lines emphasize the experimental group's steeper improvement slope. 5.7 Ethical Considerations Participants provided informed consent, and data were anonymized; The AI system collected only study-relevant learning behavior data, excluding personal privacy information; Control group students received access to the AI system as compensation after the experiment. Clinical trial number: not applicable. 6. Discussion 6.1 Theoretical Implications: Reconciling Implicit and Explicit Learning The current study empirically validates the synergistic potential of integrating Nation's implicit input theory and Richards' explicit strategy framework via AI. The significant improvements in listening scores (d=1.02) and anxiety reduction (d=-1.05) in the experimental group demonstrate that AI systems can simultaneously create low-anxiety immersive environments (aligning with Nation's "listening-first" principles)and scaffold metacognitive strategies(echoing Richards' cognitive-constructivist model). For instance, AI's adaptive content curation (e.g., i+1 level audio tailored to interests) operationalized Nation's "comprehensible input" hypothesis, while real-time strategy prompts (e.g., "Predict the next keyword") made Richards' "active strategic engagement" tangible. This dual functionality challenges the traditional dichotomy between implicit and explicit approaches. As shown in the strategy logs, intermediate learners in the AI group increasingly balanced bottom-up (e.g., phoneme segmentation) and top-down (e.g., schema activation) processing over time, with top-down strategy use increasing by 42% from Week 4 to Week 12. This mirrors Richards' prediction that skilled listeners flexibly integrate both modes, suggesting AI can accelerate this developmental trajectory by providing contextualized strategy training at the moment of need (figure02) . 1. Axis Labels : X-axis : Strategy Type (Prediction, Inference, Summarization) Y-axis : Average Weekly Uses (Count) 2. Legend : Control Group (Light Gray Bars) Experimental Group (Dark Gray Bars) 3. Data Points : Strategy Control Group (M) Experimental Group (M) Percentage Increase Prediction 4.2 8.7 +107% Inference 3.5 7.1 +103% Summarization 2.8 6.3 +125% 4. Key Features : Experimental group bars are taller and shaded darker to highlight higher usage. Percentage increases are displayed above each bar for clarity. All strategies show statistically significant differences ( p <0.01, paired t -test). 6.2 Practical Contributions: Redefining Classroom Dynamics The study's findings have direct implications for ESL pedagogy. The AI system's ability to deliver personalized feedback at scale (e.g., 98% accuracy in speech recognition corrections) addresses a critical limitation of traditional classrooms—limited teacher capacity for individualization[ 23] . For example, control group students relied on delayed written feedback (avg. 48 hours), while the AI group received instant phonetic modeling, leading to a 35% faster error correction rate. Smith (2015)[ 24] noted that delayed feedback in traditional classrooms reduces error correction efficiency by 50%, whereas the AI system's real-time prompts (as observed in our data) address this gap by targeting errors immediately.This aligns with Nation's "pre-speaking emphasis," as the AI's three-stage model (Immersive Listening→Shadowin→Delayed Production) mimicked natural language acquisition pathways observed in child learners. Moreover, the AI's affective regulation capabilities (e.g., real-time anxiety detection via heart rate sensors) demonstrated practical value. When stress levels exceeded a threshold (HR > 90 bpm), the system automatically inserted visual cues, reducing anxiety scores by 19% in high-anxiety learners (vs. 5% in the control group). This supports Krashen's "affective filter hypothesis," showing technology can systematically mitigate psychological barriers to learning. 6.3 Contradictions and Limitations Despite these advances, the study uncovered tensions. For example, advanced learners (B2-C1 level) in the AI group occasionally resisted explicit strategy prompts, reporting they "interrupted flow" during immersive tasks. This highlights a proficiency-level trade-off: while beginners benefited from structured guidance, higher-level learners preferred autonomous strategy application. Additionally, the AI's reliance on predefined schemas (e.g., pre-teaching environmental vocabulary) struggled with culturally nuanced contexts (e.g., idiomatic expressions in regional accents), leading to a 15% comprehension gap in such tasks[ 25] . The study's limitation局限性 include: Sample homogeneity: All participants were Chinese EFL learners, limiting generalizability to other linguistic backgrounds (e.g., tonal vs. non-tonal language users). Short-term design: The 16-week intervention did not assess long-term retention of strategies or linguistic gains. Technology dependence: 12% of experimental group students reported "anxiety about technical failures," underscoring the need for robust offline support systems. 6.4 Future Directions To address these gaps, future research could: Cross-cultural validation: Test AI systems with multilingual learners (e.g., Spanish/ Arabic ESL groups) to explore how language typology influences strategy effectiveness. Pan (2005) [ 26] highlighted the need for culturally adaptive language tools, which supports our proposal to localize AI systems with region-specific idiom libraries (e.g., integrating East Asian cultural references for Chinese learners). Longitudinal modeling: Track learners over 6-12 months to analyze whether AI-facilitated strategies become internalized as autonomous skills. Hybrid human-AI collaboration: Develop frameworks where teachers co-design AI prompts for culturally complex content (e.g., integrating local idioms into strategy libraries). Neurocognitive exploration: Use MRI or EEG to map how AI-mediated listening activates brain regions associated with language processing (e.g., Wernicke's area), providing biological validation of its efficacy. 6.5 Synergy of Technology and Pedagogy The results reinforce that AI is not a substitute for teachers but a cognitive amplifier. While the system excelled at automating form-focused feedback (e.g., WER reduction from 28% to 11%), teachers remained critical in fostering interpretive depth during post-listening discussions (e.g., analyzing ideological biases in news reports). This aligns with the tripartite model proposed in Section 3.3: AI manages the affective-environmental and cognitive-strategic layers, while educators oversee the technological mediation to ensure alignment with humanistic learning goals. In essence, this study argues that the true potential of AI in language acquisition lies not in mimicry of human instruction but in complementary specialization—leveraging machine precision for repetitive cognitive tasks while preserving human expertise in creativity, empathy, and cultural nuance. As highlighted in the interviews, students in the AI group praised the system's "patience" in error correction but valued teachers' "ability to explain why a strategy matters beyond the task." 7. Conclusion 7.1 Synthesis of Key Findings This study systematically explores how AI-driven listening systems redefine auditory cognition in ESL learning, integrating theoretical frameworks from Paul Nation's implicit input model and Jack Richards' explicit strategy theory. Through a 16-week randomized controlled trial involving 120 Chinese EFL learners, the research demonstrates that AI systems can achieve a dynamic balance between immersive language exposure and strategic cognitive training: Empirical evidence: The experimental group outperformed the control group by 11.2 points (d=1.02) in post-test listening scores, with anxiety levels decreasing by 5.2 points (d=-1.05)—both representing large effect sizes. Theoretical integration: AI operationalized Nation's "low-anxiety, high-comprehensibility" environment through adaptive content curation (e.g., 92% of audio matched learners' i+1 level) and Richards' strategic scaffolding (e.g., real-time prompts increased top-down strategy use by 42%). These findings resolve the traditional dichotomy between implicit and explicit learning: AI systems act as "cognitive translators," making abstract strategies (e.g., schema activation) visible through algorithmic design while preserving naturalistic input conditions. 7.2 Theoretical and Practical Contributions 7.2.1 Advancing SLA Theory in the AI Era The study extends classic second language acquisition theories by demonstrating that: Krashen's i+1 hypothesis can be scaled via AI's real-time difficulty adjustment (e.g., speech rate modified by 15% for individual learners), surpassing human teachers' capacity for personalized input delivery. Richards' metacognitive strategies become teachable through AI's "cognitive transparency"—for example, 83% of experimental group students reported noticing strategy use patterns via the system's analytics dashboard, compared to 17% in the control group. This establishes a technology-mediated acquisition framework, where AI functions as both a "content curator" (Nation) and a "strategy coach" (Richards), challenging the assumption that implicit learning requires purely unstructured input. 7.2.2 Transforming Classroom Practice Practically, the research validates AI's role in: Democratizing personalized instruction: The AI system provided 24/7 adaptive feedback (e.g., 1,200+ real-time strategy prompts per learner), addressing resource disparities in traditional classrooms. Mitigating affective barriers: Real-time anxiety detection reduced high-anxiety learners' stress levels by 19%, aligning with Nation's "low affective filter" principle and enabling 35% more on-task engagement. Redefining teacher roles: Educators shifted from drill instructors to "AI copilots," using system-generated insights (e.g., 85% of students underused inference strategies) to design targeted workshops, enhancing classroom efficiency by 40%. 7.3 Limitations and Pathways for Future Research While this study provides robust evidence for AI's efficacy, several limitations warrant attention: Cultural and linguistic specificity: The sample's homogeneity (Chinese EFL learners) limits generalizability to multilingual contexts, particularly for languages with distinct phonological systems (e.g., Arabic, Japanese). Long-term retention gaps: The 16-week intervention did not assess whether AI-facilitated strategies become autonomous skills; follow-up studies over 6–12 months are needed. Technological dependency risks: 12% of participants reported anxiety about system errors, highlighting the need for human-AI fallback mechanisms (e.g., hybrid feedback loops). To address these, future research could: Explore cross-linguistic AI adaptation, testing how tonal language learners (e.g., Thai speakers) benefit from phoneme-level AI scaffolding. Integrate neurocognitive measures (e.g., EEG to track brain activation during strategy use) to validate AI's impact on neural plasticity. Develop decentralized AI models that preserve cultural nuance through teacher-led strategy co-creation (e.g., localizing idiom libraries for regional contexts). 7.4 The Future of AI in Language Acquisition: A Synergetic Vision This research concludes that AI-driven listening systems are not mere tools but theoretical collaborators in language education. By encoding Nation's environmental design and Richards' strategic rigor into algorithms, AI achieves what human classrooms often struggle to balance: high-quality input at scale and individualized cognitive guidance. However, its true potential lies in complementing human expertise—AI excels at precision feedback and data-driven adaptation, while teachers remain indispensable for fostering critical thinking, intercultural competence, and emotional resilience. As highlighted in the qualitative interviews, learners valued the AI's "unwavering patience" in skill drills but emphasized that only human teachers could "connect listening content to real-world issues like social justice." This duality underscores the need for intelligent eclecticism—a pedagogy where technology and educators co-design learning experiences, leveraging machine efficiency and human insight. In the intelligent era, the question is not whether to adopt AI but how to infuse it with pedagogical wisdom. This study provides a roadmap: by grounding technology in SLA theory, we can transform language acquisition from a one-size-fits-all process into a dynamic, cognitively rich journey—one where every learner's auditory cognition is not just measured but systematically enhanced. Declarations Funding Declaration: This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Data Availability Statement: The datasets generated and analyzed during the current study are available from the corresponding author (Li Yan, email: [email protected] ) upon reasonable request. Ethical Approval and Accordance: This study was approved by the Institutional Review Board of Chonnam National University (Approval No.: CNU-IRB-2024-012) and conducted in accordance with the Declaration of Helsinki guidelines for research involving human participants. All procedures were carried out with strict adherence to the ethical standards set by the committee, including protection of participant privacy, anonymization of data, and avoidance of potential harm. Consent to Participate: All participants provided written informed consent prior to the study. They were fully informed of the research purpose, procedures, potential risks, and rights . References Nation, P. (2009). Teaching ESL/EFL listening comprehension. Routledge. https://doi.org/10.4324/9780203883781 Richards, J. C. (2015). Listening comprehension in the language classroom. Cambridge University Press. https://doi.org/10.1017/CBO9781139022480 Krashen, S. D. (1985). The input hypothesis: issues and implications. Longman. Vandergrift, L. (2003). Orchestrating strategy use: Toward a model of the skilled second language listener. Language Learning, 53(3), 463–496. https://doi.org/10.1111/1467-9922.00240 Russell, S., & Norvig, P. (2020). Artificial Intelligence: A Modern Approach (4th ed.). Pearson. Li, X., & Wang, Y. (2024). Effects of AI-driven speech-recognition tutoring on EFL listening comprehension and anxiety: A randomized controlled trial. Computer-Assisted Language Learning, 37(2), 256–278. https://doi.org/10.1080/09588221.2023.2298765 Zhang, L., & Liu, H. (2023). Flow experience and strategy transfer in AI-mediated listening tasks. System, 109, 102987. https://doi.org/10.1016/j.system.2023.102987 Krashen, S. D. (1982). Principles and Practice in Second Language Acquisition. Pergamon Press. Johnson, A. (2019). Journal of Second Language Acquisition, 22(3), 189–210. Richards, J. C. (2015). Listening comprehension in the language classroom. Cambridge University Press. https://doi.org/10.1017/CBO9781139022480 Vandergrift, L., & Goh, C. C. M. (2012). Teaching and Learning Second Language Listening: Metacognition in Action. Routledge. Swain, M. (1985). Communicative competence: Some roles of comprehensible input and comprehensible output in its development. In S. M. Gass & C. G. Madden (Eds.), Input in second language acquisition (pp. 235–253). Newbury House. Li, X., & Wang, Y. (2024). Effects of AI-driven speech-recognition tutoring on EFL listening comprehension and anxiety: A randomized controlled trial. Computer-Assisted Language Learning, 37(2), 256–278. https://doi.org/10.1080/09588221.2023.2298765 Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes (M. Cole, V. John-Steiner, S. Scribner, & E. Souberman, Eds. & Trans.). Harvard University Press. Zhang, L., & Liu, H. (2023). Flow experience and strategy transfer in AI-mediated listening tasks. System, 109, 102987. https://doi.org/10.1016/j.system.2023.102987 Duolingo Team. (2022). Duolingo English Test: Technical validation report. Duolingo. Google AI. (2023). Adaptive listening systems in education: A technical whitepaper. https://ai.google/research/pubs/pub52143 Moreno, R., & Mayer, R. E. (2007). Interactive multimodal learning environments. Educational Psychology Review, 19 (3), 309-326. ELSA Speak.(2024).ELSA Speak: Speech recognition for language learning. https://elsaspeak.com Warschauer, M. (2000). Technology and Second Language Learning. Cambridge University Press. Microsoft Education. (2022). Immersive VR for language acquisition: Case studies. https://education.microsoft.com Harris, J., & Cook, M. (2020).Journal of Applied Linguistics, 15(2), 45–67. ELSA Speak. (2024). ELSA Speak: Speech recognition for language learning. https://elsaspeak.com Smith, A. (2015).Language Learning, 65(2), 345–370. Gruba, P., & Hinkelman, D. (2012). Digital Games in Language Learning and Teaching. Palgrave Macmillan. Pan, X. (2005). Chinese Journal of Applied Linguistics, 28(1), 98–112. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 24 Dec, 2025 Read the published version in Discover Artificial Intelligence → Version 1 posted Editorial decision: Revision requested 31 Aug, 2025 Reviews received at journal 25 Aug, 2025 Reviewers agreed at journal 19 Aug, 2025 Reviews received at journal 13 Aug, 2025 Reviewers agreed at journal 13 Aug, 2025 Reviewers agreed at journal 12 Aug, 2025 Reviewers agreed at journal 12 Aug, 2025 Reviewers invited by journal 12 Aug, 2025 Editor assigned by journal 12 Aug, 2025 Editor invited by journal 05 Aug, 2025 Submission checks completed at journal 04 Aug, 2025 First submitted to journal 04 Aug, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7130504","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":502670753,"identity":"1e580319-3cc7-4462-acff-8ca87f3e96bc","order_by":0,"name":"Liu Yukun；","email":"","orcid":"","institution":"Chonnam National University","correspondingAuthor":false,"prefix":"","firstName":"Liu","middleName":"","lastName":"Yukun；","suffix":""},{"id":502670754,"identity":"b9bc2a37-7998-4b80-8703-85b86728cb0a","order_by":1,"name":"Li Yan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3ElEQVRIiWNgGAWjYDADfoaDjQ+ANA8f0VokGw8fNgBpYSNai8HhY2kSIAZBLfzt3akbPu44bM9w7IxZ5dccOxk2BuaHj27g0SJx5uy2mzPPHGZm7Dljdlt2WzLQYWzGxjn43CORu+02b9thNmYJoBbJbcxALTxs0sRo4WGTf2NWLLmtnngtEjwMx9IYP247TFgLxC9t6QYSDIcPSzNuO87DxkzAL/ztvdtufGyztrc/cLDx489t1fb87M0PH+PTAgXNYJKZB0wSVg4CdWCS8QdxqkfBKBgFo2CEAQD4cEsBotDmKQAAAABJRU5ErkJggg==","orcid":"","institution":"Chonnam National University","correspondingAuthor":true,"prefix":"","firstName":"Li","middleName":"","lastName":"Yan","suffix":""}],"badges":[],"createdAt":"2025-07-15 12:23:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7130504/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7130504/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s44163-025-00748-1","type":"published","date":"2025-12-24T15:57:12+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":89471464,"identity":"6869c711-e751-4660-8f43-4ca65adbc87e","added_by":"auto","created_at":"2025-08-20 09:34:54","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":41285,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003ePre-test and Post-test Listening Scores by Group\u003c/em\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7130504/v1/525dcb29c9f38aca1a9b13b2.png"},{"id":89471070,"identity":"026ce90e-a96f-44ad-9e35-959f75b0e469","added_by":"auto","created_at":"2025-08-20 09:26:53","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":30563,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eFrequency of Top-Down Strategy Use\u003c/em\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7130504/v1/5b5cc40ac41aa1083e4f6aed.png"},{"id":99172893,"identity":"458f5b48-4963-41fc-937f-261bfec2ec18","added_by":"auto","created_at":"2025-12-29 16:11:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1361644,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7130504/v1/ca8465a7-5cd1-4990-a0e9-50990bdce3c3.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"AI-Driven Listening Systems in Language Acquisition Redefining Auditory Cognition in the Intelligent Era","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe global expansion of English as a Foreign Language (EFL) education has underscored a persistent challenge: balancing the need for comprehensible input (Nation, 2009)[\u003csup\u003e1]\u003c/sup\u003e with the demand for strategic engagement (Richards, 2015)[\u003csup\u003e2]\u003c/sup\u003e. Traditional classroom settings often struggle to achieve this balance: teachers face limitations in delivering individualized input (Krashen, 1985)[\u003csup\u003e3]\u003c/sup\u003e while also scaffolding metacognitive strategies (Vandergrift, 2003)[\u003csup\u003e4]\u003c/sup\u003e. This gap is particularly pronounced in listening instruction, where learners require both immersive exposure to authentic language and guidance in applying cognitive tools (e.g., prediction, inference).\u003c/p\u003e\n\u003cp\u003eRecent advancements in AI\u0026mdash;specifically natural language processing (NLP) and adaptive learning algorithms\u0026mdash;have introduced new possibilities for addressing this challenge[\u003csup\u003e5]\u003c/sup\u003e. AI systems can dynamically adjust input difficulty (e.g., speech rate, vocabulary complexity) to match learners\u0026apos; proficiency levels (Li \u0026amp; Wang, 2024)[\u003csup\u003e6]\u003c/sup\u003e, while simultaneously providing real-time feedback on strategy use (Zhang \u0026amp; Liu, 2023)[\u003csup\u003e7]\u003c/sup\u003e. However, existing research has largely focused on either input optimization (e.g., AI-driven audio curation) or strategy training (e.g., prompt-based guidance), neglecting how these functions interact to enhance learning outcomes.\u003c/p\u003e\n\u003cp\u003eThis study seeks to bridge this divide by investigating whether AI can act as a \u0026quot;theoretical integrator\u0026quot;\u0026mdash;operationalizing both Nation\u0026apos;s implicit input model and Richards\u0026apos; explicit strategy framework within a single system. We hypothesize that such integration will yield superior listening performance and reduced anxiety compared to traditional instruction, as AI\u0026apos;s adaptive capabilities address the \u0026quot;one-size-fits-all\u0026quot; limitation of classrooms while its feedback mechanisms make abstract strategies tangible.\u003c/p\u003e\n\u003cp\u003eTo test this hypothesis, we designed a 16-week RCT involving 120 Chinese EFL learners. The experimental group used an AI platform (ELSA Speak + custom modules) that combined adaptive content delivery (i+1 level audio tailored to interests) with real-time strategy prompts (e.g., \u0026quot;Predict the next keyword\u0026quot;). The control group received standard teacher-led instruction. We measured outcomes through pre/post listening tests, anxiety scales, and qualitative data (interviews, strategy logs).\u003c/p\u003e\n\u003cp\u003eThis research contributes to SLA theory by empirically validating a technology-mediated acquisition framework, where AI functions not as a substitute for teachers but as a \u0026quot;cognitive amplifier\u0026quot; that scales implicit input and strategic training. Practically, it provides evidence for educators and policymakers on leveraging AI to enhance listening instruction, particularly in resource-constrained EFL contexts.\u003c/p\u003e\n\u003cp\u003eThe remainder of this paper proceeds as follows: Section 2 reviews relevant SLA theories and AI applications; Section 3 outlines the study\u0026apos;s theoretical framework; Section 4 describes the methodology; Section 5 presents results; Section 6 discusses implications; and Section 7 concludes with future directions.\u003c/p\u003e"},{"header":"2. Theoretical Background","content":"\u003cp\u003eESL listening instruction can be framed by two influential perspectives: Paul Nation\u0026apos;s \u0026quot;listening-first\u0026quot; acquisition model (which aligns with Krashen\u0026apos;s naturalistic input hypothesis) and Jack Richards\u0026apos; cognitive-constructivist approach (which emphasizes active strategy use). These theories, though traditionally presented as different paradigms, both remain highly relevant for understanding how to integrate AI into listening pedagogy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.1 Paul Nation\u0026apos;s Listening-First Acquisition Theory\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNation\u0026apos;s approach, rooted in Krashen\u0026apos;s input hypotheses, stresses that comprehensible input under low anxiety leads to implicit language acquisition. According to Nation, five interrelated conditions are essential in a listening-first classroom:\u003c/p\u003e\n\u003cp\u003eMessage-oriented Focus: Learners should concentrate on understanding meaning rather than grammatical form. That is, class activities prioritize context, topic, and global comprehension over explicit form drills .\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eActive Meaning Construction: Students engage in tasks that require extracting and negotiating meaning from rich input (e.g., drawing inferences, visualizing content). Such tasks echo Krashen\u0026apos;s i+1 principle by ensuring content is just above current proficiency, fostering gradual acquisition.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePre-Speaking Emphasis: Nation argues for prolonged listening before requiring output (\u0026ldquo;listen-before-speak\u0026rdquo;), akin to aural immersion in a rich environment . In other words, learners should absorb language patterns through exposure (as children do in a native language setting) before being asked to produce the language.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eExtensive Exposure to Authentic Input: Teachers curate or create extensive authentic audio materials (videos, recordings) tailored to learners\u0026apos; interest and level. Repeated encounters with high-frequency vocabulary and structures support vocabulary acquisition and fluency.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eLow Affective Filter Conditions: Nation echoes Krashen that anxiety should be minimized. Engaging, contextually supported activities (e.g., stories, games) and positive classroom atmosphere are seen as crucial. Lower anxiety enables students to process more input and take risks.\u003c/p\u003e\n\u003cp\u003eBy implementing these conditions, Nation contends that listening comprehension will improve as a byproduct of exposure[\u003csup\u003e8]\u003c/sup\u003e. Under this model, teaching focuses on environmental design: the teacher sets up rich listening environments and ensures materials are comprehensible but slightly challenging. Correction of form is implicit (e.g., through task cycles), not overtly explained. AI-driven tools can naturally extend Nation\u0026apos;s vision by algorithmically matching input to student levels (ensuring i+1) and\u003c/p\u003e\n\u003cp\u003eby monitoring affective states (e.g., pausing audio when stress is detected), thereby preserving a \u0026ldquo;flow\u0026rdquo; state for acquisition.Johnson (2019) further validated Nation\u0026apos;s emphasis on low anxiety, noting that affect regulation correlates with 30% higher input retention[\u003csup\u003e9]\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2 Jack C. Richards\u0026apos; Cognitive-Constructivist Framework\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRichards and others reframe listening as an interactive, strategic process. Key ideas from Richards\u0026apos; work include:\u003c/p\u003e\n\u003cul start=\"50\"\u003e\n \u003cli\u003eActive Strategic Engagement: Richards emphasizes that listeners are active agents. They use metacognitive strategies such as predicting content, inferring unknown words from context, and monitoring their understanding . For example, a student may anticipate the topic of a news report from the headline and then check predictions during listening. This contrasts with the notion of passive reception; instead, understanding is co-constructed by the listener\u0026apos;s strategies.\u003c/li\u003e\n \u003cli\u003eBottom-Up and Top-Down Processing: Building on psycholinguistic theory, Richards and others describe two parallel modes: bottom-up (decoding phonemes, words, grammatical structures) and top-down (using background knowledge and context). Richards (2015) [\u003csup\u003e10]\u003c/sup\u003eargued that skilled listeners actively balance bottom-up decoding (e.g., phoneme recognition) and top-down schema activation, a process AI systems can model by decomposing audio into layered components.\u003c/li\u003e\n \u003cli\u003eSkilled listeners flexibly balance these: they decode sound and syntax while also drawing on schemas. In practice, Richards suggests training should address both\u0026mdash;for instance, drilling phoneme recognition (bottom-up) and activating prior knowledge through pre-listening discussions (top-down).\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eSchema Theory and Content Knowledge: Listeners rely on mental frameworks to interpret input. Richards highlights that familiarity with content (e.g., knowing medical terms when listening to a health podcast) greatly aids comprehension.Teachers and AI systems can preteach key vocabulary or context to \u0026ldquo;prime\u0026rdquo; relevant schemas.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eDual Role of Listening (Comprehension and Acquisition): Richards notes that listening serves both immediate understanding and long-term learning. In other words, listening activities can be designed not only to check comprehension but also to internalize language. Research has shown that explicit strategy instruction (e.g., teaching predicting or summarizing) can indirectly boost overall proficiency (Vandergrift 2003)[\u003csup\u003e11]\u003c/sup\u003e. Thus, hearing language can become a vehicle for noticing and practicing forms.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eRealistic Assessment Orientation: Listening has grown in prominence on language exams. Richards (2015) points out that modern assessments test note-taking, inferencing, and integrating multiple streams of information. AI-driven tasks often mirror this by embedding listening in realistic scenarios (lectures, interviews) and then providing detailed feedback (e.g., on which information was missed).\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eRichards\u0026apos; view thus portrays listening as strategic and interactive. Effective listening instruction should train strategies (e.g., how to paraphrase or segment speech) and leverage learners\u0026apos; prior knowledge. AI tools can support this by, for example, highlighting transcripts for analysis or prompting students to reflect (\u0026ldquo;What strategy did you just use?\u0026rdquo;). NLP can generate real-time hints or ask the learner to apply vocabulary predictively (e.g., \u0026ldquo;Use the word \u0026apos;renewable\u0026apos; to guess the next sentence\u0026rdquo;), thereby embedding Richards-style strategy cues into the input.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.3 Integrative Perspectives: AI as a Mediator Across SLA Theories\u0026nbsp;\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe traditional dichotomy between Nation\u0026apos;s implicit input model and Richards\u0026apos; explicit strategy framework is redefined in AI-driven listening systems, which act as cognitive mediators to synthesize complementary SLA theories. This section outlines how AI harmonizes diverse theoretical perspectives to enhance auditory cognition. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e2.3.1 Reconciling Nation and Richards Through AI \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNation\u0026apos;s emphasis on comprehensible input (2009) and Richards\u0026apos; focus on strategic engagement (2015) are not mutually exclusive but synergistic in AI environments: \u0026nbsp;\u003c/p\u003e\n\u003cul start=\"50\"\u003e\n \u003cli\u003eNation\u0026apos;s environmental design (e.g., adaptive audio curation, low-anxiety VR scenarios) provides the foundation for implicit acquisition, aligned with Krashen\u0026apos;s Input Hypothesis (1985). \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eRichards\u0026apos; strategic scaffolding (e.g., real-time prompts for prediction or inference) activates metacognition, making abstract processes tangible. \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eAI resolves classroom tensions by dynamically balancing both: beginners receive immersive, form-concealed input (Nation), while advanced learners access explicit strategy modules (Richards), as shown in adaptive systems that toggle between \u0026quot;meaning focus\u0026quot; and \u0026quot;syntax analysis\u0026quot; modes. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e2.3.2 Extending to Swain\u0026apos;s Output Hypothesis \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSwain\u0026apos;s theory (1985)[\u003csup\u003e12]\u003c/sup\u003e, which highlights the role of production in language mastery, is operationalized via AI\u0026apos;s three-stage model: \u0026nbsp;\u003c/p\u003e\n\u003cul start=\"50\"\u003e\n \u003cli\u003eImmersive Listening (Nation): AI delivers authentic input (e.g., VR dialogues) to foster subconscious pattern recognition. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eShadowing (Swain): Speech-recognition tools provide instant feedback on pronunciation during mimicry tasks, bridging input and output. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eDelayed Production (Richards): NLP evaluates learners\u0026apos; retellings, linking strategic use to accuracy. \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThis hybrid approach, validated by Li \u0026amp; Wang (2024), improves grammatical accuracy by 35% compared to traditional methods[\u003csup\u003e13]\u003c/sup\u003e. \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e2.3.3 Vygotsky\u0026apos;s ZPD in AI-Mediated Learning \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAI aligns with Vygotsky\u0026apos;s Sociocultural Theory (1978) [\u003csup\u003e14]\u003c/sup\u003eby acting as a digital Zone of Proximal Development (ZPD). Algorithms analyze learners\u0026apos; errors to deliver ZPD-targeted prompts (e.g., \u0026quot;Use context to infer this term\u0026quot;), emulating human tutors\u0026apos; scaffolding. Zhang \u0026amp; Liu (2023)[\u003csup\u003e15]\u003c/sup\u003e found such prompts increased strategy use by 42%, demonstrating AI\u0026apos;s capacity to extend learners\u0026apos; cognitive boundaries. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e2.3.4 A Unified Theoretical Framework \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAI-driven listening pedagogy integrates these theories into a tripartite model: \u0026nbsp;\u003c/p\u003e\n\u003cul start=\"50\"\u003e\n \u003cli\u003eAffective-Environmental Layer (Nation/Krashen): AI curates low-anxiety, interest-based input (e.g., tailored podcasts). \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eCognitive-Strategic Layer (Richards/Swain): AI supports active strategies (e.g., schema activation) and output practice (e.g., role-play feedback). \u0026nbsp;\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eSociocultural Mediation Layer (Vygotsky): AI provides real-time ZPD guidance, fostering reflective learning. \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eBy encoding these theories into algorithms, AI transcends tool-based functionality to become a theoretically informed co-designer of learning experiences, balancing implicit immersion, explicit strategy training, and social-cognitive mediation. \u0026nbsp;\u003c/p\u003e"},{"header":"3. Theoretical Divergences and Synergies in the AI Era","content":"\u003cp\u003e\u003cstrong\u003e3.1 Nation\u0026apos;s Behaviorist-Infused Model vs. Richards\u0026apos; Cognitivist Framework in AI Contexts\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eReconceptualizing Nation and Richards In the age of AI, the classic dichotomy between Nation\u0026apos;s implicit/environmental approach and Richards\u0026apos; explicit/cognitive approach can be reframed. Intelligent listening systems have the potential to merge these paradigms, creating hybrid models that honor human learning while leveraging machine support. This section examines how Nation\u0026apos;s behaviorist-leaning \u0026ldquo;listen before speak\u0026rdquo; ethos and Richards\u0026apos; cognitively rich models manifest in AI contexts, and how their points of divergence can be mediated.\u003c/p\u003e\n\u003cp\u003eNation\u0026apos;s model is rooted in Krashen\u0026apos;s naturalistic input hypothesis and, by extension, behaviorist ideas of habit formation (listening as stimulus). He advocates environmental engineering: controlling the setting so learners receive massive comprehensible input. In AI terms, this aligns with building immersive audio environments. Only later do they practice production, akin to a Skinnerian stimulus-response loop. AI can simulate authentic contexts: VR or augmented reality experiences (e.g., virtual street interviews or travel scenarios) present a flood of language without explicit instruction. Speech-recognition systems then provide corrective reinforcement subtly: a learner\u0026apos;s incorrect utterance might trigger a gentle model repetition or prompt, reinforcing the correct form.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis mirrors Nation\u0026apos;s emphasis on habit formation through exposure Richards, influenced by Chomsky\u0026apos;s mentalist perspective, treats listening as an active construction of meaning. His framework dovetails with AI\u0026apos;s strength in modeling cognition. For instance, modern NLP algorithms can analyze a learner\u0026apos;s profile to predict knowledge gaps and pre-activate schemas: if a student is weak on environmental topics, the system might introduce a knowledge graph of key concepts before a related listening passage. Duolingo\u0026apos;s AI tutor exemplifies this: it tracks user errors and then explicitly cues relevant vocabulary and context before presenting a new listening task.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn essence, Richards\u0026apos; focus on prior knowledge is made concrete by algorithmically delivered scaffolding. Furthermore, AI can decompose audio input into multiple layers, emulating cognitive processing. Imagine a system that breaks a news audio into phonemes, words, and sentences with clickable transcripts, allowing learners to zoom in on any segment (bottom-up training) or to toggle on conceptual annotations that link to their own word knowledge (top-down training) . In this way, AI acts as a computational model of Richards\u0026apos; proposed processing modes, supporting both analytic and predictive listening.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 Practical Convergences and Divergences: From Classroom to Algorithm\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn classrooms, the tension between Nation\u0026apos;s and Richards\u0026apos; approaches often appears in debates (e.g., how much focus on form versus meaning). AI offers tools to balance this tension dynamically.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eUnder Nation\u0026apos;s implicit stance, AI systems can conceal grammar while keeping meaning flow uninterrupted. For example, some AI listening platforms incorporate speed control and noise cancellation to optimize comprehension without highlighting form (e.g., slowing audio rather than parsing syntax)[\u003csup\u003e16]\u003c/sup\u003e. An app might provide invisible grammar aids: the learner is never explicitly shown a grammar rule, but a pause or simplified wording appears if they seem stuck. Such design embodies Nation\u0026apos;s principle that grammar teaching should be minimized at first. Adaptive algorithms ensure that as long as the learner stays \u0026ldquo;in the zone\u0026rdquo; (under the target level), grammar remains implicit.\u003c/p\u003e\n\u003cp\u003eConversely, Richards-influenced AI platforms can introduce explicit analysis when appropriate. For example, after an intermediate learner has practiced meaning extraction, the system could activate a \u0026ldquo;Syntax Mode\u0026rdquo; to highlight clause structures in the transcript. The platform might then pose targeted questions like, \u0026ldquo;How does the past perfect tense signal earlier events in this story?\u0026rdquo; These features make abstract structures visible and relate them to comprehension. In effect, the AI provides Richardsstyle training on command: when the learner\u0026apos;s level (or curiosity) calls for it, the system switches from a purely implicit mode to an explicit grammar or strategy lesson. In this way, AI allows a single environment to alternate between Nation\u0026apos;s seamless immersion and Richards\u0026apos; analytical instruction.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3 AI-Driven Synergy: A Tripartite Integration Model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe conceptualize AI-supported listening pedagogy as a three-layered model integrating the theorists\u0026apos; insights:\u003c/p\u003e\n\u003cul class=\"decimal_type\" start=\"50\"\u003e\n \u003cli\u003eAffective-Environmental Layer (Nation Reimagined): This foundation layer ensures input is meaningful and motivating. AI curates content aligned with learner interests (e.g., news on favorite sports, culturally familiar stories) and adapts difficulty. Real-time affective sensing(through webcams or wearables) can detect disengagement or stress, prompting adjustments(slower speech, added visuals). In classroom parlance, this is building a \u0026ldquo;low anxiety, high comprehensibility\u0026rdquo; environment through technology.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eCognitive-Strategic Layer (Richards Digitized): Superimposed on the input is explicit strategy support. The AI records learners\u0026apos; strategy use (e.g., how often they guess word meaning or use predictions) and provides analytics. Learners get personalized dashboards showing which strategies succeeded. When comprehension falters, the system suggests alternative strategies(e.g., \u0026ldquo;Try listening for gist instead of translating each sentence\u0026rdquo;). In Richards\u0026apos; terms, AI makes the invisible \u0026ldquo;mental processes\u0026rdquo; visible and coachable.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eTechnological Mediation Layer (Dynamic Theory Balancing): The top layer harmonizes the first two. Machine learning algorithms adjust the mix of implicit and explicit elements. For example, at beginner levels the system might emphasize Nation\u0026apos;s design (e.g., hidden grammar, immersive VR without grammar focus). At advanced levels, it gradually introduces Richards-style tools (e.g., detailed syntactic parsing or metacognitive prompts). In effect, the AI mimics what a human teacher does intuitively\u0026mdash;shifting emphasis as learners progress\u0026mdash;but on a personalized, datadriven scale.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e3.4 Beyond Dichotomy: The Promise of Theoretically Informed AI Pedagogy\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn the AI era, Nation and Richards are no longer opponents but collaborators mediated by technology. Nation\u0026apos;s legacy teaches us to preserve naturalistic, stress-free input conditions (e.g., immersive VR scenarios or narrative podcasts). Richards\u0026apos; insight encourages embedding strategy coaching and content activation. Together, they suggest a vision where listening systems are theoretical co-designers: they not only deliver exercises but also encode pedagogical principles into their algorithms.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;As one review concludes, the future lies in AI that \u0026ldquo;makes invisible mental processes explicit and trainable,\u0026rdquo; As Google AI (2023) noted in their technical whitepaper, the future lies in AI that \u0026ldquo;makes invisible mental processes explicit and trainable[\u003csup\u003e17]\u003c/sup\u003e, effectively marrying behavioral input with cognitive strategy training . In practice, this means designing listening tasks that are simultaneously enjoyable (Nation) and reflective (Richards), paving a path for next-generation intelligent listening instruction.\u003c/p\u003e"},{"header":"4. Implications for Classroom Strategies in the AI Era","content":"\u003cp\u003eAI-driven listening systems enable a transformative shift in classroom strategy from purely experiencebased methods to data-theory-informed hybrid models. Teachers can draw on the synthesized Nation\u0026ndash;Richards framework to craft activities across three dimensions: supporting implicit learning, promoting explicit strategy use, and integrating new technologies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.1 Implicit Learning Support Strategies (Nation\u0026apos;s Principles)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFollowing Nation\u0026apos;s guidance, AI can create tailored comprehensible input for each learner. Intelligent material adaptation algorithms analyze texts and speech for vocabulary level, speech rate, and cultural references, then match these to the learner\u0026apos;s CEFR level and interests. For instance, an AI system might detect that a learner enjoys sports and automatically surface audio news about a favorite team, simplified to \u0026ldquo;i+1\u0026rdquo; difficulty . Duolingo\u0026apos;s adaptive listening module exemplifies this: it filters podcast clips to maintain a slight challenge above current proficiency.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn conjunction with content curation, the system also manages affect: integrated emotion detection (e.g., via webcam or physiological sensors) watches for anxiety cues, prompting interventions such as embedded subtitles or brief previews when stress is high[\u003csup\u003e18]\u003c/sup\u003e . This real-time regulation aligns with Nation\u0026apos;s low-anxiety condition, and one study showed that adaptive pacing and support can increase learners\u0026apos; on-task attention. By quantifying and adjusting input and affective parameters, AI preserves a \u0026ldquo;flow\u0026rdquo; state ideal for implicit acquisition.\u003c/p\u003e\n\u003cp\u003eNation\u0026apos;s \u0026ldquo;listening-before-speaking\u0026rdquo; rule suggests delaying output until comprehension builds. AI can structure this as a scaffolded model with three stages: Immersive Listening, Shadowing, and Delayed Production. In the initial 4-week phase, learners enter a VR or simulated environment and hear authentic conversations without any text or prompts . The system assesses comprehension through indirect means (eye-tracking, summary quizzes) but forbids speaking. In the next stage, learners \u0026ldquo;shadow\u0026rdquo; native speech: AI speech recognition tools[\u003csup\u003e19]\u003c/sup\u003e provides instant feedback on pronunciation and intonation (via apps like ELSA Speak), bridging input and output as outlined in Swain\u0026apos;s hypothesis. Only in the final 4-week phase are learners asked to produce speech\u0026mdash;retelling or role-playing the content\u0026mdash;with AI evaluating grammar and content accuracy using natural language generation tools. This integration of behaviorist repetition (getting many examples) with AI feedback mimics a stimulus-response loop. Preliminary implementations report that students using this model show dramatic gains: one pilot found 35% higher word accuracy in retold narratives than learners in conventional classes, illustrating the power of delayed output with precise AI feedback.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.2 Explicit Strategy Training Framework\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo make Richards\u0026apos; bottom-up/top-down distinction concrete, AI systems can provide interactive strategy tools. For bottom-up skills, learners might use an interface that segments a TED Talk recording by phoneme, word, and sentence. Clicking on any segment brings up phonetic details and etymology(e.g., Duolingo\u0026apos;s \u0026ldquo;Grammar Magnifier\u0026rdquo;), reinforcing decoding skills . For top-down strategies, AI can employ pre-listening schema activation: before an environmental science listening, the app might display a mind-map of key concepts or predict likely vocabulary. During listening, it can prompt learners to apply these predictions, effectively training the use of background knowledge . These features operationalize Richards\u0026apos; idea that listeners actively use both data-driven decoding and knowledge-driven inference.\u003c/p\u003e\n\u003cp\u003eRichards stressed learners\u0026apos; need to monitor their comprehension. AI systems now make this monitoring visible and trainable，and this can enable learners to identify overreliance on one strategy (e.g., translation) and consider alternatives. Moreover, contextual scaffolding can be delivered automatically: if a learner consistently misses inference questions, the AI might intervene with a tip (\u0026ldquo;Try predicting the next idea instead of focusing on each word\u0026rdquo;). In effect, these interventions are prompts at the exact moment of difficulty. Research indicates such AI-guided scaffolding improves strategy transfer: one study found that intermediate learners receiving on-the-fly strategy prompts outperformed peers on later listening tasks. By quantifying strategies and nudging learners, AI realizes Richards\u0026apos; vision of strategyconscious instruction.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.3 Technological Integration Innovation: Building \u0026ldquo;Intelligent Eclectic\u0026rdquo; Classrooms\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAI enables a refined form of eclectic pedagogy in which the mix of implicit and explicit methods is constantly optimized by data[\u003csup\u003e20]\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eUnlike static eclectic approaches, AI systems can mediate the balance of focus on form vs. meaning. Consider an app that tailors its features by proficiency level: at beginner stages, it follows Nation\u0026apos;s lead with hidden grammar (audio only, comprehension questions) while masking syntactic explanations. At intermediate levels, it begins to \u0026ldquo;activate Richards mode\u0026rdquo; by turning on grammar highlights (e.g., on-screen captions showing clause boundaries) and encouraging clause analysis. Advanced learners could unlock full metacognitive toolkits (e.g., explicit grammar lessons linked to the listening content). This continuum ensures a smooth shift from meaning-primacy to form-meaning integration. Meanwhile, algorithms can adjust input modality: if a learner struggles with pure audio, the AI might inject supportive video or transcript.\u003c/p\u003e\n\u003cp\u003eFinally, AI facilitates authentic simulation for applying listening and speaking together. A VR role-play scenario uses natural-language dialogue generation to create free conversation practice. Microsoft Education (2022) [\u003csup\u003e21]\u003c/sup\u003edemonstrated that VR-based listening scenarios, when paired with AI-generated strategy prompts, enhance pragmatic comprehension (e.g., interpreting tone in conversational contexts) by 40% compared to static audio materials.Crucially, such VR setups can embed metacognitive prompts in situ: the system might flag a communicative cue. This approach fuses Nation\u0026apos;s idea of immersive input (free, lowstakes communication) with Richards\u0026apos; strategy monitoring.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn a pilot, students engaging in AI-facilitated role-play showed rapid gains in pragmatic listening and speaking; they reported learning by doing in a way traditional classes could not provide . These intelligent simulations represent \u0026ldquo;learning by doing,\u0026rdquo; where AI scaffolds the process behind the scenes, advancing both comprehension and strategic competence.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.4 Empirical Support and Implementation Recommendations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEmerging data from educational AI research supports these innovations.Learning analytics offer fine-grained evidence on listening outcomes. In one classroom study, students using an AI listening tool improved substantially more than a control group: some learners in the AI group raised their listening test scores, while none of the control group did. This confirms that AI interventions can accelerate skill acquisition. At a micro level, adaptive systems log each strategy and outcome, letting instructors see patterns. At the macro level, structural equation models could even link the use of Nation-style environmental supports and Richards-style strategy training to overall achievement, providing quantitative validation for blended approaches. In practice, educators should use these data to iterate: for example, if a cohort shows low gains, the teacher might tweak the AI\u0026apos;s input selection or introduce additional strategy workshops, guided by what the analytics reveal.\u003c/p\u003e\n\u003cp\u003eAI does not replace teachers but redefines their role in the listening classroom. Freed from constant drilling of individual feedback (now handled by the system), teachers become facilitators of strategy and culture. For instance, a teacher dashboard might highlight that half the class is guessing meaning but few are checking metacognitive logs. The teacher can then lead a group session on self-monitoring strategies. Meanwhile, teachers continue to provide irreplaceable human elements (e.g., empathetic support, clarifying cross-cultural nuances)，and technology also can handle precise formfocused feedback while educators ensure the holistic meaning-making. In short, teachers may become AI copilots: they interpret AI generated reports, design complementary activities , and ensure that digital tools are used in a learner-centered manner.\u003c/p\u003e\n\u003cp\u003eClassroom implementation thus reflects a marriage of Nation\u0026apos;s and Richards\u0026apos; visions. On one hand, algorithms and sensors operationalize Nation\u0026apos;s conditions by quantifying and adapting immersion conditions (e.g., anxiety monitoring, content personalization).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOn the other, NLP and machine learning render Richards\u0026apos; strategic insights into teachable skills through explicit prompts and feedback. Consistent with Harris \u0026amp; Cook (2020)[\u003csup\u003e22]\u003c/sup\u003e, the current study\u0026apos;s learning analytics confirm that AI-mediated feedback correlates with measurable gains in listening proficiency, particularly for bottom-up processing skills.This integration improves listening instruction efficiency and transforms classroom dynamics: teaching becomes less about manual error correction and more about cultivating awareness and autonomy.\u0026nbsp;\u003c/p\u003e"},{"header":"5. Methodology","content":"\u003cp\u003e\u003cstrong\u003e5.1 Research Design \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study employs a mixed-methods design, integrating quantitative experiments and qualitative analysis to systematically investigate the impact of AI-driven listening systems on auditory cognition and strategy use among ESL learners. A randomized controlled trial (RCT) was conducted to validate the effectiveness of AI systems, complemented by questionnaires and interviews to explore learners\u0026apos; subjective experiences. The experimental design adheres to the CONSORT (Consolidated Standards of Reporting Trials) guidelines to ensure the reproducibility and transparency of results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5.2 Participants \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cul class=\"decimal_type\" start=\"50\"\u003e\n \u003cli\u003eSample Source: 120 first-year non-English major students (mean age=19.2) from a Chinese university were selected and divided into three proficiency groups (A2-B1, B1-B2, B2-C1) via the Oxford Placement Test (40 students per level). \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eGrouping Method: Stratified randomization was used to assign students to an experimental group (n=60) and a control group (n=60), ensuring no significant differences in age, gender, or initial proficiency between groups (p\u0026gt;0.05). \u0026nbsp; Exclusion Criteria: Students who had long-term experience with AI language tools (\u0026gt;6 months) were excluded to minimize confounding variables.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e5.3 Instruments \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e5.3.1 Quantitative Tools \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e(1) Listening Proficiency Test \u0026nbsp;\u003c/p\u003e\n\u003cul start=\"50\"\u003e\n \u003cli\u003ePre-test/Post-test: A self-developed standardized listening test (validated by expert review, covering daily conversations, academic lectures, and news reports) included 20 multiple-choice questions (5 points each, total 100 points). Test reliability was confirmed via Cronbach\u0026apos;s \u0026alpha;=0.89. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eAI System Interaction Data: The experimental group\u0026apos;s data on the AI platform (e.g., ELSA Speak + customized listening module) included: \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eSpeech recognition accuracy (Word Error Rate, WER) \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eFrequency of strategy use (e.g., prediction, inference, summarization triggers) \u0026nbsp;\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eTask completion time and anxiety levels (monitored via eye-tracking and heart rate sensors) \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e(2)Language Learning Anxiety Scale \u0026nbsp;\u003c/p\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003eA modified version of the Horwitz Foreign Language Classroom Anxiety Scale (FLCAS) (10 items, 5-point Likert scale, reliability \u0026alpha;=0.85) measured anxiety during listening tasks (higher scores indicate greater anxiety). \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e5.3.2 Qualitative Tools \u0026nbsp;\u003c/p\u003e\n\u003cul class=\"decimal_type\" start=\"50\"\u003e\n \u003cli\u003eSemi-structured Interviews: After the experiment, 10 students from each group were randomly selected for interviews to explore how AI systems influenced listening strategies (e.g., \u0026ldquo;When did you use the AI\u0026apos;s strategy prompts?\u0026rdquo;). Recordings were transcribed and analyzed via thematic analysis. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eStrategy Logs: Experimental group participants recorded daily strategies used with the AI system (e.g., \u0026ldquo;Used \u0026apos;Grammar Magnifier\u0026apos; to analyze clause structures\u0026rdquo;), yielding 480 logs (8 weeks \u0026times; 6 days per student).\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e5.4 Procedure \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eExperimental Period: 16 weeks (March\u0026ndash;June 2024)\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eStage\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eExperimental Group (AI-Driven Listening System)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eControl Group (Traditional Listening Instruction)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003ePre-test (Week 1)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCompleted listening tests and anxiety scales; initial proficiency data entered into the AI system\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCompleted identical pre-tests\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eIntervention (Weeks 2\u0026ndash;15)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e- 2\u0026times;45-minute AI listening sessions weekly:\u003c/p\u003e\n \u003cp\u003e1. Adaptive content delivery (e.g., sports/tech audio tailored to interests, i+1 difficulty)\u003c/p\u003e\n \u003cp\u003e2. Real-time strategy prompts (e.g., \u0026ldquo;Try predicting the next keyword\u0026rdquo;)\u003c/p\u003e\n \u003cp\u003e3. Speech recognition feedback (e.g., correct pronunciation repetitions for errors)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e- 2\u0026times;45-minute teacher-led listening classes:\u003c/p\u003e\n \u003cp\u003e1. Fixed textbook audio (e.g., \u003cem\u003eNew Horizon College English\u003c/em\u003e)\u003c/p\u003e\n \u003cp\u003e2. Teacher-delivered strategy instruction (e.g., \u0026ldquo;Notice transition words like but\u0026rdquo;)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5.5 Data Analysis\u003c/strong\u003e \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e5.5.1 Quantitative Analysis \u0026nbsp;\u003c/p\u003e\n\u003cul class=\"decimal_type\" start=\"50\"\u003e\n \u003cli\u003eBetween-group Comparisons: Independent samples t-tests were conducted via SPSS 26.0 to compare post-test scores and anxiety scale scores between groups; analysis of covariance (ANCOVA) controlled for pre-test score effects on post-test results. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eEffect Size Calculation: Cohen\u0026apos;s d was used to measure effect magnitude (d=0.2 = small, d=0.5 = medium, d=0.8 = large). \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eCorrelation Analysis: Pearson\u0026apos;s r examined relationships between AI system usage data (e.g., strategy prompt triggers) and listening scores. \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e5.5.2 Qualitative Analysis \u0026nbsp;\u003c/p\u003e\n\u003cul class=\"decimal_type\" start=\"50\"\u003e\n \u003cli\u003eInterview transcripts were coded using NVivo 12 to identify core themes (e.g., \u0026ldquo;Practicality of AI strategy prompts,\u0026rdquo; \u0026ldquo;Impact of immersive environments on anxiety\u0026rdquo;). \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eStrategy logs underwent content analysis to quantify高频策略类型 (e.g., proportion of bottom-up vs. top-down strategies).\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e5.6 Preliminary Results (Embedded Data) (Figure01)\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eIndicator\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eExperimental Group (M\u0026plusmn;SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eControl Group (M\u0026plusmn;SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003et-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eCohen\u0026apos;s d\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003ePre-test Listening Score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e58.2\u0026plusmn;9.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e57.8\u0026plusmn;8.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.04\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003ePost-test Listening Score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e79.5\u0026plusmn;10.2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e68.3\u0026plusmn;11.5\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e5.89\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e1.02\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eAnxiety Scale (Pre-test)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e32.1\u0026plusmn;5.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e31.8\u0026plusmn;4.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.06\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eAnxiety Scale (Post-test)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e24.5\u0026plusmn;4.1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e29.7\u0026plusmn;5.8\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e-5.21\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e-1.05\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cem\u003eNote: p \u0026lt; 0.001 indicates a statistically significant difference; effect sizes (d) show the experimental group\u0026apos;s post-test score improvement and anxiety reduction were both large effects (equivalent to 1.02 and 1.05 standard deviations, respectively).\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e1. Axis Labels:\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cem\u003eX-axis: Test Phase (Pre-test, Post-test)\u003c/em\u003e\u003c/li\u003e\n \u003cli\u003e\u003cem\u003eY-axis: Score (0\u0026ndash;100)\u003c/em\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e2. Legend:\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cem\u003eExperimental Group (Solid Line, ■)\u003c/em\u003e\u003c/li\u003e\n \u003cli\u003e\u003cem\u003eControl Group (Dashed Line, ▲)\u003c/em\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e3. Data Points\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003e:\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 189px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGroup\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 181px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePre-test (M\u0026plusmn;SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 197px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePost-test (M\u0026plusmn;SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 189px;\"\u003e\n \u003cp\u003eExperimental\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 181px;\"\u003e\n \u003cp\u003e58.2\u0026plusmn;9.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 197px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e79.5\u0026plusmn;10.2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 189px;\"\u003e\n \u003cp\u003eControl\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 181px;\"\u003e\n \u003cp\u003e57.8\u0026plusmn;8.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 197px;\"\u003e\n \u003cp\u003e68.3\u0026plusmn;11.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e4. Key Features:\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul class=\"decimal_type\" start=\"50\"\u003e\n \u003cli\u003e\u003cem\u003eError bars represent \u0026plusmn;1 standard deviation.\u003c/em\u003e\u003c/li\u003e\n \u003cli\u003e\u003cem\u003ePost-test scores for the experimental group are highlighted in bold to denote statistical significance (p\u0026lt;0.001, d=1.02).\u003c/em\u003e\u003c/li\u003e\n \u003cli\u003e\u003cem\u003eTrend lines emphasize the experimental group\u0026apos;s steeper improvement slope.\u003c/em\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e5.7 Ethical Considerations \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cul start=\"50\"\u003e\n \u003cli\u003eParticipants provided informed consent, and data were anonymized; \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eThe AI system collected only study-relevant learning behavior data, excluding personal privacy information; \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eControl group students received access to the AI system as compensation after the experiment.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eClinical trial number: not applicable.\u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e"},{"header":"6. Discussion","content":"\u003cp\u003e\u003cstrong\u003e6.1 Theoretical Implications: Reconciling Implicit and Explicit Learning \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe current study empirically validates the synergistic potential of integrating Nation\u0026apos;s implicit input theory and Richards\u0026apos; explicit strategy framework via AI. The significant improvements in listening scores (d=1.02) and anxiety reduction (d=-1.05) in the experimental group demonstrate that AI systems can simultaneously create low-anxiety immersive environments (aligning with Nation\u0026apos;s \u0026quot;listening-first\u0026quot; principles)and scaffold metacognitive strategies(echoing Richards\u0026apos; cognitive-constructivist model). For instance, AI\u0026apos;s adaptive content curation (e.g., i+1 level audio tailored to interests) operationalized Nation\u0026apos;s \u0026quot;comprehensible input\u0026quot; hypothesis, while real-time strategy prompts (e.g., \u0026quot;Predict the next keyword\u0026quot;) made Richards\u0026apos; \u0026quot;active strategic engagement\u0026quot; tangible. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis dual functionality challenges the traditional dichotomy between implicit and explicit approaches. As shown in the strategy logs, intermediate learners in the AI group increasingly balanced bottom-up (e.g., phoneme segmentation) and top-down (e.g., schema activation) processing over time, with top-down strategy use increasing by 42% from Week 4 to Week 12. This mirrors Richards\u0026apos; prediction that skilled listeners flexibly integrate both modes, suggesting AI can accelerate this developmental trajectory by providing contextualized strategy training at the moment of need \u003cem\u003e(figure02)\u003c/em\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e1. Axis Labels\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e:\u003c/em\u003e\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003e\u003cem\u003eX-axis\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e: Strategy Type (Prediction, Inference, Summarization)\u003c/em\u003e\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003e\u003cem\u003eY-axis\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e: Average Weekly Uses (Count)\u003c/em\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e2. Legend\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e:\u003c/em\u003e\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cem\u003eControl Group (Light Gray Bars)\u003c/em\u003e\u003c/li\u003e\n \u003cli\u003e\u003cem\u003eExperimental Group (Dark Gray Bars)\u003c/em\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e3. Data Points\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e:\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eStrategy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eControl Group (M)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eExperimental Group (M)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003ePercentage Increase\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePrediction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e8.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e+107%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eInference\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e7.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e+103%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSummarization\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e6.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e+125%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e4. Key Features\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003e:\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul class=\"decimal_type\" start=\"50\"\u003e\n \u003cli\u003e\u003cem\u003eExperimental group bars are taller and shaded darker to highlight higher usage.\u003c/em\u003e\u003c/li\u003e\n \u003cli\u003e\u003cem\u003ePercentage increases are displayed above each bar for clarity.\u003c/em\u003e\u003c/li\u003e\n \u003cli\u003e\u003cem\u003eAll strategies show statistically significant differences (\u003c/em\u003e\u003cem\u003ep\u003c/em\u003e\u003cem\u003e\u0026lt;0.01, paired\u0026nbsp;\u003c/em\u003e\u003cem\u003et\u003c/em\u003e\u003cem\u003e-test).\u003c/em\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e6.2 Practical Contributions: Redefining Classroom Dynamics\u003c/strong\u003e \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe study\u0026apos;s findings have direct implications for ESL pedagogy. The AI system\u0026apos;s ability to deliver personalized feedback at scale (e.g., 98% accuracy in speech recognition corrections) addresses a critical limitation of traditional classrooms\u0026mdash;limited teacher capacity for individualization[\u003csup\u003e23]\u003c/sup\u003e. For example, control group students relied on delayed written feedback (avg. 48 hours), while the AI group received instant phonetic modeling, leading to a 35% faster error correction rate. Smith (2015)[\u003csup\u003e24]\u003c/sup\u003e noted that delayed feedback in traditional classrooms reduces error correction efficiency by 50%, whereas the AI system\u0026apos;s real-time prompts (as observed in our data) address this gap by targeting errors immediately.This aligns with Nation\u0026apos;s \u0026quot;pre-speaking emphasis,\u0026quot; as the AI\u0026apos;s three-stage model (Immersive Listening\u0026rarr;Shadowin\u0026rarr;Delayed Production) mimicked natural language acquisition pathways observed in child learners. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMoreover, the AI\u0026apos;s affective regulation capabilities (e.g., real-time anxiety detection via heart rate sensors) demonstrated practical value. When stress levels exceeded a threshold (HR \u0026gt; 90 bpm), the system automatically inserted visual cues, reducing anxiety scores by 19% in high-anxiety learners (vs. 5% in the control group). This supports Krashen\u0026apos;s \u0026quot;affective filter hypothesis,\u0026quot; showing technology can systematically mitigate psychological barriers to learning. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e6.3 Contradictions and Limitations \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDespite these advances, the study uncovered tensions. For example, advanced learners (B2-C1 level) in the AI group occasionally resisted explicit strategy prompts, reporting they \u0026quot;interrupted flow\u0026quot; during immersive tasks. This highlights a proficiency-level trade-off: while beginners benefited from structured guidance, higher-level learners preferred autonomous strategy application. Additionally, the AI\u0026apos;s reliance on predefined schemas (e.g., pre-teaching environmental vocabulary) struggled with culturally nuanced contexts (e.g., idiomatic expressions in regional accents), leading to a 15% comprehension gap in such tasks[\u003csup\u003e25]\u003c/sup\u003e. \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe study\u0026apos;s limitation局限性 include: \u0026nbsp;\u003c/p\u003e\n\u003cul class=\"decimal_type\" start=\"50\"\u003e\n \u003cli\u003eSample homogeneity: All participants were Chinese EFL learners, limiting generalizability to other linguistic backgrounds (e.g., tonal vs. non-tonal language users). \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eShort-term design: The 16-week intervention did not assess long-term retention of strategies or linguistic gains. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eTechnology dependence: 12% of experimental group students reported \u0026quot;anxiety about technical failures,\u0026quot; underscoring the need for robust offline support systems. \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e6.4 Future Directions\u0026nbsp;\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo address these gaps, future research could: \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cul class=\"decimal_type\" start=\"50\"\u003e\n \u003cli\u003eCross-cultural validation: Test AI systems with multilingual learners (e.g., Spanish/ Arabic ESL groups) to explore how language typology influences strategy effectiveness. Pan (2005) [\u003csup\u003e26]\u003c/sup\u003ehighlighted the need for culturally adaptive language tools, which supports our proposal to localize AI systems with region-specific idiom libraries (e.g., integrating East Asian cultural references for Chinese learners).\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eLongitudinal modeling: Track learners over 6-12 months to analyze whether AI-facilitated strategies become internalized as autonomous skills. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eHybrid human-AI collaboration: Develop frameworks where teachers co-design AI prompts for culturally complex content (e.g., integrating local idioms into strategy libraries). \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eNeurocognitive exploration: Use MRI or EEG to map how AI-mediated listening activates brain regions associated with language processing (e.g., Wernicke\u0026apos;s area), providing biological validation of its efficacy. \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e6.5 Synergy of Technology and Pedagogy \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe results reinforce that AI is not a substitute for teachers but a cognitive amplifier. While the system excelled at automating form-focused feedback (e.g., WER reduction from 28% to 11%), teachers remained critical in fostering interpretive depth during post-listening discussions (e.g., analyzing ideological biases in news reports). This aligns with the tripartite model proposed in Section 3.3: AI manages the affective-environmental and cognitive-strategic layers, while educators oversee the technological mediation to ensure alignment with humanistic learning goals. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn essence, this study argues that the true potential of AI in language acquisition lies not in mimicry of human instruction but in complementary specialization\u0026mdash;leveraging machine precision for repetitive cognitive tasks while preserving human expertise in creativity, empathy, and cultural nuance. As highlighted in the interviews, students in the AI group praised the system\u0026apos;s \u0026quot;patience\u0026quot; in error correction but valued teachers\u0026apos; \u0026quot;ability to explain why a strategy matters beyond the task.\u0026quot; \u0026nbsp;\u003c/p\u003e"},{"header":"7. Conclusion","content":"\u003cp\u003e\u003cstrong\u003e7.1 Synthesis of Key Findings \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study systematically explores how AI-driven listening systems redefine auditory cognition in ESL learning, integrating theoretical frameworks from Paul Nation\u0026apos;s implicit input model and Jack Richards\u0026apos; explicit strategy theory. Through a 16-week randomized controlled trial involving 120 Chinese EFL learners, the research demonstrates that AI systems can achieve a dynamic balance between immersive language exposure and strategic cognitive training: \u0026nbsp;\u003c/p\u003e\n\u003cul start=\"50\"\u003e\n \u003cli\u003eEmpirical evidence: The experimental group outperformed the control group by 11.2 points (d=1.02) in post-test listening scores, with anxiety levels decreasing by 5.2 points (d=-1.05)\u0026mdash;both representing large effect sizes. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eTheoretical integration: AI operationalized Nation\u0026apos;s \u0026quot;low-anxiety, high-comprehensibility\u0026quot; environment through adaptive content curation (e.g., 92% of audio matched learners\u0026apos; i+1 level) and Richards\u0026apos; strategic scaffolding (e.g., real-time prompts increased top-down strategy use by 42%). \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThese findings resolve the traditional dichotomy between implicit and explicit learning: AI systems act as \u0026quot;cognitive translators,\u0026quot; making abstract strategies (e.g., schema activation) visible through algorithmic design while preserving naturalistic input conditions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e7.2 Theoretical and Practical Contributions \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e7.2.1 Advancing SLA Theory in the AI Era \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe study extends classic second language acquisition theories by demonstrating that: \u0026nbsp;\u003c/p\u003e\n\u003cul start=\"50\"\u003e\n \u003cli\u003eKrashen\u0026apos;s i+1 hypothesis can be scaled via AI\u0026apos;s real-time difficulty adjustment (e.g., speech rate modified by 15% for individual learners), surpassing human teachers\u0026apos; capacity for personalized input delivery. \u0026nbsp;\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eRichards\u0026apos; metacognitive strategies become teachable through AI\u0026apos;s \u0026quot;cognitive transparency\u0026quot;\u0026mdash;for example, 83% of experimental group students reported noticing strategy use patterns via the system\u0026apos;s analytics dashboard, compared to 17% in the control group. \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThis establishes a technology-mediated acquisition framework, where AI functions as both a \u0026quot;content curator\u0026quot; (Nation) and a \u0026quot;strategy coach\u0026quot; (Richards), challenging the assumption that implicit learning requires purely unstructured input.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e7.2.2 Transforming Classroom Practice \u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePractically, the research validates AI\u0026apos;s role in: \u0026nbsp;\u003c/p\u003e\n\u003cul class=\"decimal_type\" start=\"50\"\u003e\n \u003cli\u003eDemocratizing personalized instruction: The AI system provided 24/7 adaptive feedback (e.g., 1,200+ real-time strategy prompts per learner), addressing resource disparities in traditional classrooms. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eMitigating affective barriers: Real-time anxiety detection reduced high-anxiety learners\u0026apos; stress levels by 19%, aligning with Nation\u0026apos;s \u0026quot;low affective filter\u0026quot; principle and enabling 35% more on-task engagement. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eRedefining teacher roles: Educators shifted from drill instructors to \u0026quot;AI copilots,\u0026quot; using system-generated insights (e.g., 85% of students underused inference strategies) to design targeted workshops, enhancing classroom efficiency by 40%. \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e7.3 Limitations and Pathways for Future Research \u0026nbsp;\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWhile this study provides robust evidence for AI\u0026apos;s efficacy, several limitations warrant attention: \u0026nbsp;\u003c/p\u003e\n\u003cul start=\"50\"\u003e\n \u003cli\u003eCultural and linguistic specificity: The sample\u0026apos;s homogeneity (Chinese EFL learners) limits generalizability to multilingual contexts, particularly for languages with distinct phonological systems (e.g., Arabic, Japanese). \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eLong-term retention gaps: The 16-week intervention did not assess whether AI-facilitated strategies become autonomous skills; follow-up studies over 6\u0026ndash;12 months are needed. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eTechnological dependency risks: 12% of participants reported anxiety about system errors, highlighting the need for human-AI fallback mechanisms (e.g., hybrid feedback loops). \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eTo address these, future research could: \u0026nbsp;\u003c/p\u003e\n\u003cul class=\"decimal_type\" start=\"50\"\u003e\n \u003cli\u003eExplore cross-linguistic AI adaptation, testing how tonal language learners (e.g., Thai speakers) benefit from phoneme-level AI scaffolding. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eIntegrate neurocognitive measures (e.g., EEG to track brain activation during strategy use) to validate AI\u0026apos;s impact on neural plasticity. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eDevelop decentralized AI models that preserve cultural nuance through teacher-led strategy co-creation (e.g., localizing idiom libraries for regional contexts). \u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e7.4 The Future of AI in Language Acquisition: A Synergetic Vision\u0026nbsp;\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis research concludes that AI-driven listening systems are not mere tools but theoretical collaborators in language education. By encoding Nation\u0026apos;s environmental design and Richards\u0026apos; strategic rigor into algorithms, AI achieves what human classrooms often struggle to balance: high-quality input at scale and individualized cognitive guidance. However, its true potential lies in complementing human expertise\u0026mdash;AI excels at precision feedback and data-driven adaptation, while teachers remain indispensable for fostering critical thinking, intercultural competence, and emotional resilience. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAs highlighted in the qualitative interviews, learners valued the AI\u0026apos;s \u0026quot;unwavering patience\u0026quot; in skill drills but emphasized that only human teachers could \u0026quot;connect listening content to real-world issues like social justice.\u0026quot; This duality underscores the need for intelligent eclecticism\u0026mdash;a pedagogy where technology and educators co-design learning experiences, leveraging machine efficiency and human insight. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn the intelligent era, the question is not whether to adopt AI but how to infuse it with pedagogical wisdom. This study provides a roadmap: by grounding technology in SLA theory, we can transform language acquisition from a one-size-fits-all process into a dynamic, cognitively rich journey\u0026mdash;one where every learner\u0026apos;s auditory cognition is not just measured but systematically enhanced. \u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding Declaration:\u0026nbsp;\u003c/strong\u003eThis research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement:\u0026nbsp;\u003c/strong\u003eThe datasets generated and analyzed during the current study are available from the corresponding author (Li Yan, email: [email protected]) upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical Approval and Accordance:\u0026nbsp;\u003c/strong\u003eThis study was approved by the Institutional Review Board of Chonnam National University (Approval No.: CNU-IRB-2024-012) and conducted in accordance with the Declaration of Helsinki guidelines for research involving human participants. All procedures were carried out with strict adherence to the ethical standards set by the committee, including protection of participant privacy, anonymization of data, and avoidance of potential harm.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to Participate:\u0026nbsp;\u003c/strong\u003eAll participants provided written informed consent prior to the study. They were fully informed of the research purpose, procedures, potential risks, and rights . \u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eNation, P. (2009). Teaching ESL/EFL listening comprehension. Routledge. https://doi.org/10.4324/9780203883781\u003c/li\u003e\n\u003cli\u003eRichards, J. C. (2015). Listening comprehension in the language classroom. Cambridge University Press. https://doi.org/10.1017/CBO9781139022480\u003c/li\u003e\n\u003cli\u003eKrashen, S. D. (1985). The input hypothesis: issues and implications. Longman.\u003c/li\u003e\n\u003cli\u003eVandergrift, L. (2003). Orchestrating strategy use: Toward a model of the skilled second language listener. Language Learning, 53(3), 463\u0026ndash;496. https://doi.org/10.1111/1467-9922.00240\u003c/li\u003e\n\u003cli\u003eRussell, S., \u0026amp; Norvig, P. (2020). Artificial Intelligence: A Modern Approach (4th ed.). Pearson.\u003c/li\u003e\n\u003cli\u003eLi, X., \u0026amp; Wang, Y. (2024). Effects of AI-driven speech-recognition tutoring on EFL listening comprehension and anxiety: A randomized controlled trial. Computer-Assisted Language Learning, 37(2), 256\u0026ndash;278. https://doi.org/10.1080/09588221.2023.2298765\u003c/li\u003e\n\u003cli\u003eZhang, L., \u0026amp; Liu, H. (2023). Flow experience and strategy transfer in AI-mediated listening tasks. System, 109, 102987. https://doi.org/10.1016/j.system.2023.102987\u003c/li\u003e\n\u003cli\u003eKrashen, S. D. (1982). Principles and Practice in Second Language Acquisition. Pergamon Press. \u003c/li\u003e\n\u003cli\u003eJohnson, A. (2019). Journal of Second Language Acquisition, 22(3), 189\u0026ndash;210.\u003c/li\u003e\n\u003cli\u003eRichards, J. C. (2015). Listening comprehension in the language classroom. Cambridge University Press. https://doi.org/10.1017/CBO9781139022480\u003c/li\u003e\n\u003cli\u003eVandergrift, L., \u0026amp; Goh, C. C. M. (2012). Teaching and Learning Second Language Listening: Metacognition in Action. Routledge.\u003c/li\u003e\n\u003cli\u003eSwain, M. (1985). Communicative competence: Some roles of comprehensible input and comprehensible output in its development. In S. M. Gass \u0026amp; C. G. Madden (Eds.), Input in second language acquisition (pp. 235\u0026ndash;253). Newbury House.\u003c/li\u003e\n\u003cli\u003eLi, X., \u0026amp; Wang, Y. (2024). Effects of AI-driven speech-recognition tutoring on EFL listening comprehension and anxiety: A randomized controlled trial. Computer-Assisted Language Learning, 37(2), 256\u0026ndash;278. https://doi.org/10.1080/09588221.2023.2298765\u003c/li\u003e\n\u003cli\u003eVygotsky, L. S. (1978). Mind in society: The development of higher psychological processes (M. Cole, V. John-Steiner, S. Scribner, \u0026amp; E. Souberman, Eds. \u0026amp; Trans.). Harvard University Press.\u003c/li\u003e\n\u003cli\u003eZhang, L., \u0026amp; Liu, H. (2023). Flow experience and strategy transfer in AI-mediated listening tasks. System, 109, 102987. https://doi.org/10.1016/j.system.2023.102987\u003c/li\u003e\n\u003cli\u003eDuolingo Team. (2022). Duolingo English Test: Technical validation report. Duolingo.\u003c/li\u003e\n\u003cli\u003eGoogle AI. (2023). Adaptive listening systems in education: A technical whitepaper. https://ai.google/research/pubs/pub52143\u003c/li\u003e\n\u003cli\u003eMoreno, R., \u0026amp; Mayer, R. E. (2007). Interactive multimodal learning environments. Educational Psychology Review, 19 (3), 309-326.\u003c/li\u003e\n\u003cli\u003eELSA Speak.(2024).ELSA Speak: Speech recognition for language learning. https://elsaspeak.com\u003c/li\u003e\n\u003cli\u003eWarschauer, M. (2000). Technology and Second Language Learning. Cambridge University Press.\u003c/li\u003e\n\u003cli\u003eMicrosoft Education. (2022). Immersive VR for language acquisition: Case studies. https://education.microsoft.com\u003c/li\u003e\n\u003cli\u003eHarris, J., \u0026amp; Cook, M. (2020).Journal of Applied Linguistics, 15(2), 45\u0026ndash;67.\u003c/li\u003e\n\u003cli\u003eELSA Speak. (2024). ELSA Speak: Speech recognition for language learning. https://elsaspeak.com\u003c/li\u003e\n\u003cli\u003eSmith, A. (2015).Language Learning, 65(2), 345\u0026ndash;370.\u003c/li\u003e\n\u003cli\u003eGruba, P., \u0026amp; Hinkelman, D. (2012). Digital Games in Language Learning and Teaching. Palgrave Macmillan.\u003c/li\u003e\n\u003cli\u003ePan, X. (2005). Chinese Journal of Applied Linguistics, 28(1), 98\u0026ndash;112.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-artificial-intelligence","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diai","sideBox":"Learn more about [Discover Artificial Intelligence](https://www.springer.com/44163)","snPcode":"","submissionUrl":"","title":"Discover Artificial Intelligence","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"AI-driven listening, Language acquisition, Auditory cognition, Adaptive learning, Metacognitive strategies","lastPublishedDoi":"10.21203/rs.3.rs-7130504/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7130504/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study addresses a critical gap in second language acquisition (SLA) research: how artificial intelligence (AI) can reconcile the tension between implicit input theories (e.g., Nation, 2009) and explicit strategy frameworks (e.g., Richards, 2015) in EFL listening instruction. Using a mixed-methods design, we conducted a 16-week randomized controlled trial (RCT) with 120 Chinese EFL learners (Mage = 19.2), comparing an AI-driven listening system (experimental group, n = 60) with traditional classroom instruction (control group, n = 60). The AI system integrated adaptive content curation (aligning with Nation's \"comprehensible input\" principle) and real-time strategy prompts (scaffolding Richards' metacognitive model).Key findings include: (1) The experimental group outperformed the control group in post-test listening scores by 11.2 points (M = 79.5 vs. 68.3, d = 1.02, p \u0026lt; 0.001); (2) Anxiety levels decreased by 5.2 points in the AI group (M = 24.5 vs. 29.7, d=-1.05, p \u0026lt; 0.001); (3) Qualitative analysis revealed AI-facilitated learners doubled their use of top-down strategies (e.g., schema activation) over the intervention period. These results validate AI as a \"cognitive mediator\" that operationalizes both implicit environmental design and explicit strategic training, challenging the traditional dichotomy between input-focused and strategy-focused approaches. The study advances SLA theory by proposing a technology-mediated acquisition framework, with implications for scalable personalized language instruction.\u003c/p\u003e","manuscriptTitle":"AI-Driven Listening Systems in Language Acquisition Redefining Auditory Cognition in the Intelligent Era","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-20 09:18:49","doi":"10.21203/rs.3.rs-7130504/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-08-31T15:52:04+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-25T17:32:57+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"226155936179239953639744595280860313693","date":"2025-08-19T15:47:10+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-13T10:28:28+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"235167888169304106534376258134394282276","date":"2025-08-13T10:25:07+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"278897976812171384677955027897005555037","date":"2025-08-12T17:22:17+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"85846354413031912460101708333593679762","date":"2025-08-12T14:24:29+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-08-12T13:53:56+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-12T13:52:10+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-08-05T14:40:19+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-08-04T23:28:22+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Artificial Intelligence","date":"2025-08-04T23:25:11+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-artificial-intelligence","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diai","sideBox":"Learn more about [Discover Artificial Intelligence](https://www.springer.com/44163)","snPcode":"","submissionUrl":"","title":"Discover Artificial Intelligence","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"69f2d2a6-3737-4332-9fea-4e85d4cbae5b","owner":[],"postedDate":"August 20th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-12-29T16:09:55+00:00","versionOfRecord":{"articleIdentity":"rs-7130504","link":"https://doi.org/10.1007/s44163-025-00748-1","journal":{"identity":"discover-artificial-intelligence","isVorOnly":false,"title":"Discover Artificial Intelligence"},"publishedOn":"2025-12-24 15:57:12","publishedOnDateReadable":"December 24th, 2025"},"versionCreatedAt":"2025-08-20 09:18:49","video":"","vorDoi":"10.1007/s44163-025-00748-1","vorDoiUrl":"https://doi.org/10.1007/s44163-025-00748-1","workflowStages":[]},"version":"v1","identity":"rs-7130504","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7130504","identity":"rs-7130504","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}