The study enrolled students from several academic fields at Universidad Peruana Unión in the Puno Region. Participants were in their second through fifth years of university, and all were enrolled during the second academic semester of 2025. The sample included 137 female and 117 male students: 91 studied humanities and education, 78 engineering and architecture, 46 business sciences, 34 health sciences and five theology. English-language learning was a shared institutional requirement rather than a subject restricted to language majors, allowing the analysis to examine language strategies within a multidisciplinary student body. English proficiency was self-reported: 68.5% described their proficiency as basic, 28.7% as intermediate and 2.8% as advanced.

Researchers assessed AI self-efficacy with the six-item General Self-Efficacy Scale for the Use of Artificial Intelligence, adapted for Peruvian university students. Each item was rated from 1, “not at all true,” to 4, “exactly true.” Language-learning strategies were measured with an 18-item inventory covering memory, cognitive, compensation, metacognitive, affective and social strategies, with responses ranging from “never or almost never” to “always or almost always.” These dimensions capture practices such as retaining and processing language information, filling knowledge gaps, planning and monitoring learning, managing emotions and learning through interaction. Satisfaction with studies was measured with a three-item scale evaluating students’ overall academic experience. All data came from the same digital questionnaire, with no personally identifiable information collected.

Before estimating the proposed relationships, the researchers examined the instruments with confirmatory factor analysis. The three-factor measurement model showed acceptable robust fit, with a comparative fit index of .954, Tucker–Lewis index of .943, root mean square error of approximation of .067 and standardized root mean square residual of .055. All factor loadings were statistically significant, ranging from .613 to .767 for AI self-efficacy, .726 to .849 for the six strategy dimensions and .795 to .876 for satisfaction with studies. Heterotrait–monotrait ratios were below the .85 criterion used to assess construct distinctiveness, indicating that the measures represented related but distinguishable concepts. The structural model was estimated with robust maximum likelihood, and a complementary bootstrap analysis with 5,000 resamples examined uncertainty in the indirect effect.

In the structural model, AI self-efficacy was positively associated with language-learning strategies, with a standardized coefficient of .431 and p less than .001. This path accounted for 18.5% of the variance in the latent strategy construct. The descriptive analysis showed a corresponding positive correlation between the two measures, r = .37, p less than .001. The pattern means that students reporting greater confidence in their ability to use AI also tended to report more frequent or extensive use of strategies for managing English learning. It does not show that using AI caused students to adopt those strategies, nor does it mean that greater AI use automatically produces better strategic learning.

Language-learning strategies had the strongest standardized association with satisfaction with studies, with a coefficient of .523 and p less than .001. The strategies–satisfaction correlation was r = .51, p less than .001, compared with r = .25, p less than .001, between AI self-efficacy and satisfaction. Together, AI self-efficacy and language-learning strategies explained 30.9% of the variance in the modeled satisfaction variable. The association between strategies and satisfaction was consistent with the idea that students who plan, monitor, evaluate and adapt their learning, while managing cognitive, emotional and social demands, may also evaluate their wider university experience more favorably. The study did not determine whether students’ satisfaction with English learning specifically drove this result; satisfaction was assessed more broadly as an evaluation of their studies.

The direct path from AI self-efficacy to satisfaction was small and statistically nonsignificant after language-learning strategies were included in the model, with a standardized coefficient of .068 and p = .421. The standardized indirect effect was .225, p less than .001, and the total association was .294, p = .001. In the bootstrap analysis, the unstandardized indirect effect was B = .360, with a 95% confidence interval from .190 to .576; because the interval excluded zero, the result was consistent with the primary model. The findings therefore support the authors’ fourth hypothesis at a statistical level, but not a claim that language strategies transmit the influence of AI confidence in a proven causal chain. The ordering remains theoretically specified rather than temporally demonstrated.

The results offer a more qualified view of AI confidence than a simple technology-centered argument would suggest. AI self-efficacy reflects students’ perceived capability to use artificial intelligence effectively for academic tasks, whereas satisfaction with studies is a broader judgment about their educational experience. The model indicates that much of the observed relationship between these constructs was statistically accounted for by language-learning strategies. It also places particular emphasis on what students do with their perceived technological capability: organizing language learning, monitoring progress, responding to difficulties, regulating emotions and engaging socially may be more closely tied to academic satisfaction than confidence in operating AI alone. The study did not collect detailed information about how often students used AI, which applications they used or what purposes those applications served, so it could not determine whether patterns differed between occasional, intensive or purpose-specific users.

Several limitations constrain the findings. The convenience sample came from one university, limiting generalizability to Peruvian higher education as a whole and to institutions with different curricula, technology access or approaches to English instruction. All variables were self-reported at a single time point, creating the possibility of social desirability, subjective interpretation and common-method effects. The cross-sectional design cannot show whether changes in AI self-efficacy precede changes in strategy use, or whether strategy use precedes changes in satisfaction. In addition, the six strategy dimensions were modeled as indicators of an overall latent construct, so the analysis could not reveal whether particular forms of strategic learning, such as cognitive, metacognitive or affective strategies, were more strongly related to either AI confidence or satisfaction. The findings consequently identify a pattern of associations rather than a tested intervention, effect of AI use or proven educational mechanism.

For universities exploring how artificial intelligence fits into learning, the results point toward a student-centered interpretation of technological competence. The authors suggest that AI training could be considered alongside support for metacognitive planning and monitoring, information processing, emotional regulation and collaborative learning. Within English education, students might use AI to organize activities, seek feedback, practice skills, monitor progress and critically reflect on generated responses rather than treating it only as a tool for completing tasks. Those are educational directions suggested by the model, not practices tested in this study, and the researchers did not establish that any specific combination of AI training and strategy instruction improves satisfaction. Longitudinal, experimental and multicenter research using behavioral measures of AI use would be needed to determine whether the observed relationships change over time, respond to intervention or hold across different universities and student groups. The central result is that AI confidence alone was not closely linked to satisfaction in the model, whereas confidence and satisfaction were connected through the strategic ways students reported managing their language learning.

Subject of Research: Associations among AI self-efficacy, language-learning strategies and academic satisfaction in Peruvian undergraduates

Article Title: AI self-efficacy and satisfaction with studies among Peruvian university students: the indirect role of language learning strategies

Article References: Sacsi-Huamani, Y., Chusi-Pacco, R., Saire-Perez, Y. Z., Mendoza-Gutierrez, Y., Coaquira Tuco, C. M., & Huamanttupa Mar, W. (2026). AI self-efficacy and satisfaction with studies among Peruvian university students: the indirect role of language learning strategies. Frontiers in Education, 11, 1978846. https://doi.org/10.3389/feduc.2026.1978846

Image Credits: AI Generated

DOI: 10.3389/feduc.2026.1978846

Keywords: artificial intelligence, AI self-efficacy, language learning, study strategies, academic satisfaction, university students, Peru, structural equation modeling, higher education, self-efficacy, satisfaction, studies

The intervention combined a flipped sequence with gamified in-class activities. For the flipped component, the teacher uploaded PowerPoint slides, videos and audio recordings to the class WhatsApp group before each session, covering the concepts addressed in class. During the subsequent online lesson, students played digital games on those concepts using Kahoot and LearningApps. Activity formats included quizzes, cloze tests, matching and pairing games, sentence ordering, millionaire-style questions, image matching, audio or video prompts, number-line exercises and group assignments. The content addressed grammar, vocabulary, discourse conventions, picture description, résumés and cover letters, parts of speech, tense, modal verbs, conditionals, dangling modifiers, parallelism, instructions and technical proposals. The course ran from July to December 2020, with 90-minute laboratory sessions on weekdays except Wednesday.

All 28 students enrolled in the Foundation English course experienced the intervention, but only 10 volunteered for end-of-semester interviews and provided the data analysed in the study. The interviewees included three women and seven men and had a mean age of 18. Placement in the course was determined by an institutional English proficiency test, producing a cohort described as relatively homogeneous in language proficiency. Researchers selected a qualitative case-study design because the question concerned students’ own accounts of emotion, motivation, confidence and preference rather than responses on predefined psychological scales. Each semi-structured, one-to-one interview lasted about 20 minutes, was audio-recorded with consent and was transcribed verbatim by the first author. Each transcript was checked against its recording before analysis.

The transcripts were analysed in NVivo 14 using a codebook-based thematic approach. The first author conducted the initial coding, consolidated related codes into 10 thematic nodes under the parent construct of the affective domain and refined the structure against the full corpus. Two co-authors who had not taught the cohort independently reviewed all 10 coded transcripts against the node definitions; discrepancies were resolved through discussion, and no formal intercoder agreement statistic was calculated. No new thematic nodes appeared after the eighth transcript, while the final two interviews confirmed and enriched existing themes. Word-frequency queries, co-occurrence analysis, coding summaries and sentiment coding were used as descriptive tools. The researchers explicitly treated these outputs as aids to interpretation rather than statistical evidence of theme importance.

Motivation appeared in all 10 transcripts and accounted for 17.9% of coded transcript text, while the engaging nature of the games also appeared in all 10 and covered 16.3%. Language suitability was coded in nine interviews and covered 14.7%. Students described persisting with difficult Kahoot questions rather than abandoning them, sometimes playing beyond scheduled class time and describing picture-description tasks as imaginative and interesting. They connected the activities with vocabulary, grammar, speaking and writing, although the study did not test the perceived skill improvements. The percentages describe the share of transcript text assigned to each node, not effects in the wider student population. They can overlap because an excerpt could receive more than one code; across the 10 nodes, total coverage was 110% by design.

Collaboration was another prominent feature. A preference for playing together appeared in eight transcripts, and social presence appeared in seven. Learners reported choosing group play for unfamiliar or difficult topics and individual play when they felt confident with the material. Group activities created opportunities to ask questions, share responses and learn from peers, while individual games allowed students to test their ability with less immediate social pressure. One participant explained that when a topic was familiar, they played alone, whereas otherwise they learned from others. Some students also said they could support classmates as well as learn from them. The pattern was interpreted through social constructivist and self-efficacy theories, but those frameworks informed the study’s analysis rather than functioning as experimentally tested mechanisms.

Students did not regard the two platforms as interchangeable. Coded references to Kahoot were more often associated with motivation and excitement, consistent with its competitive, timed quiz format. LearningApps was more frequently associated with social presence and language suitability because its varied templates supported activities such as matching pairs, cloze exercises and vocabulary games. These were descriptive tendencies rather than numerical platform ratings: participants did not comment equally on both services, and references were unevenly distributed across interviews. Overall, the accounts suggest complementarity, with Kahoot providing competitive energy and LearningApps supporting focused, skill-specific practice, but the study did not experimentally compare the platforms or establish that either produced better learning outcomes.

One of the clearest findings concerned the timing of instruction. Despite the flipped design, students often preferred to receive conceptual explanation during class before being allowed to play. Their comments suggested that materials shared through WhatsApp were not consistently prepared, either because students failed to complete the preparation or because some concepts were difficult to understand independently. Continuous gameplay also produced boredom for at least one participant. These observations qualify the assumption that placing explanatory materials before class automatically creates a successful flipped sequence. The study did not test alternative classroom sequences or evaluate readiness checks. Instead, it identified timing as a design issue: learners in this particular case generally wanted brief in-class conceptual input before gameplay, while uninterrupted play was not always engaging.

The findings also showed that positive and mixed responses coexisted. Learners reported that choice over when, how and with whom to play could support autonomy; feedback, rewards and graduated difficulty could reinforce perceived competence; and collaborative formats could foster relatedness. Accounts of persistence, absorption and rapid feedback resembled the flow experience described in the study’s theoretical framework, while peer support resembled social scaffolding. At the same time, enthusiasm was accompanied by boredom, inconsistent preparation, platform preferences and variation in prior experience with educational games. Six of the participants reported that they had not played classroom games before, making initial unfamiliarity part of the experience rather than evidence of a longer-term adaptation pattern.

The evidence base remains narrow and should not be read as proof that gamification improves learning or motivation. The study relied on retrospective self-reports from volunteer interviewees in one pandemic-era course, and the instructor’s dual role as teacher and interviewer could have encouraged socially desirable responses even though final grades had been submitted before interviews. Coding was performed by one researcher with co-author review rather than independent double-coding. No pre- and post-intervention proficiency tests, performance scores, classroom observations or platform analytics were used, and the games primarily targeted language knowledge rather than systematically assessing listening, speaking, reading and writing. Future work could triangulate interviews with observations, analytics and proficiency measures, follow learners beyond the novelty phase and include more diverse academic programmes. Within these limits, the case provides detailed qualitative evidence about how game design, collaboration, challenge and the timing of conceptual support shaped students’ reported motivation, confidence and sense of connection in an online English class.

Subject of Research: Perceived affective experiences of undergraduate learners in a gamified flipped online English course

Article Title: Gamified flipped learning and the affective domain of language learning: a qualitative case study of online English classes

Article References: Sivakami, B., Jayasudha, M., & Sanmuganathan, K. (2026). Gamified flipped learning and the affective domain of language learning: a qualitative case study of online English classes. Frontiers in Education, 11, Article 1888127. https://doi.org/10.3389/feduc.2026.1888127

Image Credits: AI Generated

DOI: 10.3389/feduc.2026.1888127

Keywords: gamified learning, flipped classroom, online English, student motivation, social presence, qualitative case study, Kahoot, LearningApps, higher education, language learning, Gamified, flipped

Sleep has long been recognized as a critical partner in memory consolidation, the dynamic process through which newly acquired information is progressively stabilized and reorganized. During non-rapid eye movement (NREM) sleep, two hallmark brain events dominate the landscape: slow oscillations (SOs), large, slow waves that sweep across the cortex roughly once per second, and sleep spindles, brief bursts of rhythmic brain activity that last about one to three seconds. Prior work has consistently linked memory performance to the power and density of these events, and more specifically to the precise timing with which spindles nest within the SO cycle. Fast spindles, those in the 12 to 15 Hz range, tend to couple with the up-state of the SO, while slow spindles, in the 9 to 12 Hz range, tend to align with the down-state. This cross-frequency coupling is increasingly viewed as the mechanistically specific marker of sleep-dependent consolidation, more informative than the density of either event alone.

Yet the extent to which these neural markers relate to changes in speech-perceptual learning over time has remained under-explored. Speech-perceptual learning is thought to engage both declarative and procedural memory systems, leaving open questions about how sleep contributes to changes in speech perception. Some studies show that talker-specific perceptual learning remains stable over 24 hours, while others report improvements following daytime rest independent of sleep. In synthetic speech learning paradigms, sleep has been shown to restore performance declines observed during wake and protect newly learned mappings from forgetting. For nonnative speech contrasts such as the Hindi dental-retroflex distinction, overnight sleep, but not equivalent wake intervals, has been associated with stabilization against interference and improved identification of novel talkers. Notably, improvements in identification do not always correspond to gains in discrimination, because identification requires a listener to assign a single token to a learned category, whereas discrimination requires only a same-different judgement between two tokens and can in principle be performed on acoustic detail alone. These tasks index qualitatively different perceptual skills and may point to distinct effects of sleep on task performance.

To probe these questions, a team of researchers from the University of Delaware, Georgetown University, and Florida Atlantic University recruited 29 monolingual English-speaking adults aged 18 to 35, spanning a broad range of language ability from those with typical language histories to those with a history of DLD. DLD is a neurodevelopmental disorder that affects language acquisition in approximately 7 percent of children in the United States, and beyond language learning difficulties, individuals with DLD also experience deficits in speech perception that can further impact communication. Participants were classified as having DLD if they reported a history of receiving services and obtained scores that resulted in a positive index for DLD under methods devised by Fidler and colleagues, which utilize raw scores of a 15-word spelling and modified token tests. Participants were assigned to the typically developed group if they indicated no history of speech, language, or reading remediation and obtained a negative index. The researchers excluded individuals with current ADHD, since ADHD medication can alter sleep, and screened for dyslexia to ensure that performance on the DLD classification tasks could not be attributed to a disorder of reading.

The experimental task centered on a notoriously difficult nonnative contrast: the Hindi dental-retroflex distinction, contrasting the retroflex /ɖ/ and dental /d̪/ consonants in the context of minimally pairing words. Participants first completed a baseline discrimination task in which they heard two speech tokens and indicated whether the word onsets belonged to the same or different speech sound category. They were then exposed to a familiarization sequence pairing the auditory tokens with novel visual objects, followed by 200 trials of two-item forced-choice identification training with written feedback after each response. Perceptual ability was measured immediately after training, and again after a delay that included a polysomnography-monitored nap. To take advantage of the natural circadian rhythms of sleep propensity that peak in the early afternoon for this age group, all participants were scheduled from noon to 5 p.m. for a single session, and the laboratory was dimly lit with low-frequency red light to promote sleepiness.

During the nap, participants underwent a lights-out period of 2.5 hours while EEG, EOG, and EMG were recorded continuously at 500 Hz using a 32-channel electrode cap. The researchers used automated sleep scoring software to classify sleep into stage 1, stage 2, slow-wave sleep, and REM. They then identified slow oscillations by low-pass filtering the EEG signal at 3.5 Hz and detecting negative-to-positive zero-crossings within an interval of 0.8 to 2 seconds, requiring the negative peak amplitude to exceed 1.25 times the averaged negative peak. Spindles were detected by band-pass filtering the signal between 9 and 12 Hz for slow spindles and 13 to 16 Hz for fast spindles, computing the root mean square of the filtered signals, and flagging events where the RMS signal remained above 1.5 times the standard deviation for 0.5 to 3 seconds. Slow spindles were quantified over frontal channels F3 and F4, reflecting their predominantly anterior distribution, while fast spindles were quantified over central channels C3 and C4. SO-spindle coupling was estimated using the continuous synchronization index described by Ladenbauer and colleagues, with coupling estimates restricted to stage 2 within NREM sleep because not all participants entered slow-wave sleep during their rest period.

The behavioral results revealed a nuanced picture. Discrimination performance improved significantly from baseline to pre-nap, confirming that the training itself was effective, but there was no significant change in discrimination across the nap interval. In contrast, identification performance improved significantly across the nap interval, with a main effect of Time that was statistically significant. The magnitude of offline changes to perceptual behavior was not modulated by language status, which differs from a previous account that reported a significant difference between TD and DLD adults in perceptual performance following overnight sleep. One possibility is that a nap is too short a duration of sleep for such group-level observations to emerge. Another may be that the study was underpowered in its DLD sample to detect a significant interaction. Indeed, an exploratory paired t-test conducted on identification scores by group suggests that whereas the TD participants improved, the DLD participants did not.

The sleep physiology results were equally revealing. After controlling for baseline and pre-nap discrimination performance, greater slow spindle density significantly predicted better post-nap discrimination performance, whereas greater SO density significantly predicted poorer post-nap performance. Fast spindle density was not significantly associated with post-nap discrimination performance. Neither the association with slow spindle density nor the association with SO density were moderated by language status. The negative correlation between SO density and changes in perceptual discrimination performance is particularly intriguing. NREM sleep characterized by a high density of SOs is thought to support systems-level reorganization, abstraction, and integration. For speech categories, that process may preserve category-level information while reducing access to token-specific acoustic detail. Higher SO density could therefore accompany stronger abstraction but poorer performance on a task requiring fine-grained acoustic comparison. This is consistent with recent observations that slow-wave sleep promotes the preservation of category-level representations at the expense of item-level information.

Perhaps the most striking finding emerged from the coupling analyses. For slow spindles, the coupling strength of SO-slow spindle coupling was negatively associated with language difficulties in those with DLD, meaning that those with stronger SO-spindle coupling had better language skills within the DLD sample. This correlation, with a Spearman rank-order coefficient of -0.71 and a p-value of 0.047, raises the intriguing possibility that SO-spindle coupling may serve a compensatory or protective role in the context of language disorder. However, given the small number of participants with DLD, this finding should be interpreted cautiously. The researchers also observed a numerically lower slow spindle density in adults with DLD relative to TD, with a p-value of 0.067, and exploratory analyses suggested a potential difference in the scalp topography of fast spindle activity between the two groups. Interestingly, prior reports of spindle activity in populations with learning disabilities have suggested that children with dyslexia have higher sleep spindle density relative to TD children, though these earlier studies did not examine spindle topography.

The study has several important limitations that the authors acknowledge. The DLD group was substantially smaller than the TD group, limiting the ability to conduct meaningful between-group comparisons. Nap sleep may not provide sufficient time for a complete sleep cycle, including deeper stages of NREM sleep such as slow-wave sleep, and because approximately half of the participants did not reach this stage, the effects observed here may not fully reflect those associated with longer sleep periods. The study also did not include a wake control group, nor did it include baseline polysomnographic measures of sleep, meaning that although the post-nap improvements observed are consistent with sleep-dependent consolidation, the researchers cannot rule out the possibility that comparable gains would have occurred in the absence of sleep. Future studies should recruit larger and more balanced samples, examine the effects of a full night of sleep, incorporate measures of generalization alongside assessments of trained-item perception, and include a wake control group to allow for a more definitive assessment of the specific contributions of sleep to memory consolidation. Nevertheless, these findings point to a potential relationship between sleep physiology and perceptual-linguistic representation, and suggest that the timing of sleep spindles relative to slow oscillations may be a promising target for understanding individual differences in language ability.

Subject of Research: Sleep-dependent consolidation of speech-sound memory in adults with and without developmental language disorder

Article Title: Speech-sound memory consolidation over a daytime nap in adults with and without developmental language disorder

Article References: Solbi, A., Thimmugari, R. G., van Zelst, A. L., & Earle, F. S. (2026). Speech-sound memory consolidation over a daytime nap in adults with and without developmental language disorder. Frontiers in Behavioral Neuroscience, 20, Article 1903903. https://doi.org/10.3389/fnbeh.2026.1903903

Image Credits: AI Generated

DOI: 10.3389/fnbeh.2026.1903903

Keywords: developmental language disorder, sleep spindles, slow oscillations, memory consolidation, speech perception, NREM sleep, polysomnography, Hindi dental-retroflex contrast, SO-spindle coupling, perceptual learning, neurodevelopmental disorders, daytime nap

The study, published in Frontiers in Language Sciences, was led by Takara Kenza Allal-Sumoto and Hiroaki Mizuhara. It builds on their earlier behavioral and electrophysiological work, which showed that second language learners process spoken sentences faster when the sentences contain words they can productively use, not just words they understand. The new paper asks a deeper question: does vocabulary knowledge change not only how strongly individual brain regions respond, but how those regions talk to each other in real time during listening? To find out, the researchers reanalyzed an EEG dataset from their previous experiments, this time through the lens of network neuroscience.

The participants were native Japanese speakers with solid English proficiency, all scoring above thresholds corresponding to at least CEFR B1 level on standardized tests such as TOEIC or TOEFL. Before entering the EEG session, each participant completed a screening phase with a production task and a comprehension task. In the production task, they had to type a missing word from a memorized English phrase; in the comprehension task, they translated target phrases into Japanese. This allowed the researchers to classify every phrase, for every individual participant, into three categories: productive knowledge, meaning the learner could both understand and produce it; only comprehensive knowledge, meaning recognition without production; and non-comprehensive knowledge, meaning the phrase was essentially unknown.

This participant-specific classification is one of the study’s key methodological strengths. The target items were formulaic sequences, three- to four-word adverbial or adjectival phrases such as “just in case,” “in a row,” or “by all accounts,” drawn from large English corpora and embedded in eight- to ten-word sentences. The sentences were synthesized as audio at 150 words per minute, roughly the pace of natural native speech. During the EEG session, participants listened to sentences and rated their confidence in understanding each one, a manipulation designed mainly to keep attention sustained. Crucially, the same phrase could be productive for one participant and merely comprehensive for another, which means the brain differences observed reflect the learner’s knowledge state rather than the intrinsic properties of the words themselves.

The analytical approach was technically ambitious. The researchers computed the weighted phase lag index, or wPLI, a measure of phase synchronization between EEG channels that is relatively robust to volume conduction, the blurring artifact that occurs when electrical activity spreads across the scalp. Phase synchronization captures whether two brain regions oscillate in a coordinated rhythm, which many neuroscientists interpret as a signature of functional communication. The team estimated wPLI separately for theta, alpha, and beta frequency bands, across five consecutive 500-millisecond time windows surrounding the onset of the final word of each target phrase. They then used the Network-Based Statistic, a permutation-based method, to search for connected subnetworks of electrodes whose connectivity differed between vocabulary knowledge levels.

The results revealed a striking temporal split. Before the target word was even heard, alpha-band connectivity was stronger for sentences containing non-comprehensive phrases than for those containing known phrases, in three separate comparisons. In other words, when listeners were about to encounter vocabulary they did not know, their brains showed heightened alpha connectivity in advance. The researchers interpret this as proactive control: when lexical representations are weak or absent, listeners may ramp up top-down attention and inhibitory gating to cope with anticipated difficulty, an adaptive strategy consistent with previous work linking alpha oscillations to attentional suppression and speech intelligibility.

After the target word onset, the pattern flipped. Sentences containing productively known phrases showed stronger connectivity in theta, beta, and alpha bands across the 0 to 500 and 500 to 1,000 millisecond windows, in five of the tested contrasts. Theta connectivity is often associated with lexico-semantic binding, the process of linking word meanings into a coherent interpretation, while beta connectivity has been tied to syntactic unification, the assembly of words into phrase- and sentence-level structure. The authors suggest that productive vocabulary knowledge enables more efficient semantic binding and structural integration, allowing the network to coordinate more richly once the word arrives. In the later window, continued theta and alpha connectivity may reflect stabilization of the developing interpretation as competition among candidate meanings is resolved.

Graph-theoretic characterization of the significant subnetworks added another layer. The researchers examined betweenness centrality and the participation coefficient, measures that capture whether a node acts as an intermediary or bridge between different parts of the network. Elevated values for productive knowledge appeared most consistently at right-occipital, midline-frontal, and left fronto-central and temporo-frontal electrode sites. Notably, the effects showed up as localized increases in bridging roles rather than as broad boosts in within-module hubness, suggesting that knowing a word productively may change how information is routed across the network, not simply how much activity concentrates in any single hub.

Here, however, comes the crucial caveat, and the authors are admirably upfront about it. Because the study surveyed 45 combinations of frequency bands, time windows, and contrasts without pre-registered hypotheses, they applied a rigorous family of statistical corrections. Eight subnetworks survived the initial cluster-based correction, but none survived Benjamini-Hochberg false discovery rate correction across the full analysis family. The researchers therefore frame all of their findings as exploratory and hypothesis-generating rather than confirmatory. Community structure, tested against null models, held up in only one of the eight subnetworks, and theta-band networks were often too sparse and fragmented for reliable interpretation. Confidence ratings, which rose with knowledge level for nearly all participants, did not correlate with connectivity differences after correction, arguing against the alternative explanation that the effects simply reflect how sure listeners felt.

The study also carries practical implications that extend beyond the laboratory. The authors point out that real-life conversation in a foreign language is doubly demanding: listeners must comprehend what they hear while simultaneously preparing what to say next. If productive vocabulary knowledge genuinely supports more efficient neural coordination during comprehension, then pushing learners beyond passive recognition toward active production could reduce the cognitive load of real-time communication. Knowing what a word means appears to be the minimum requirement for understanding a sentence, but it is not enough for fast, effortless processing. As the authors conclude, vocabulary knowledge may shape comprehension not only through how much the brain activates, but through how information is routed across functional networks, with differences in knowledge level appearing as differences in route selection and role allocation. Confirmatory studies with pre-registered hypotheses targeting the alpha and theta windows identified here will be needed to establish whether these exploratory patterns replicate, but the study offers a compelling roadmap for connecting the depth of word knowledge to the choreography of brain networks in real time.

Subject of Research: Functional EEG connectivity and vocabulary knowledge level during second language listening comprehension

Article Title: Neural connectivity changes related to vocabulary knowledge effects during second language comprehension

Article References: Allal-Sumoto, T. K., & Mizuhara, H. (2026). Neural connectivity changes related to vocabulary knowledge effects during second language comprehension. Frontiers in Language Sciences, 5, Article 1798627. https://doi.org/10.3389/flang.2026.1798627

Image Credits: AI Generated

DOI: 10.3389/flang.2026.1798627

Keywords: EEG, functional connectivity, second language comprehension, vocabulary knowledge, weighted phase lag index, network-based statistic, theta oscillations, alpha oscillations, graph theory, listening comprehension, language learning, cognitive neuroscience

The study, conducted by Meysam Muhammadpour of the University of Guilan and published in BMC Psychology, set out to answer a question that has long divided researchers in second language acquisition: is listening comprehension limited primarily by cognitive capacity, the sheer amount of information a learner can hold and manipulate in mind, or by metacognitive regulation, the learner’s ability to plan, monitor, and evaluate their own comprehension? Rather than treating these as rival explanations, the research adopted an integrative framework that views them as complementary, one governing resource availability and the other governing resource regulation.

To test this idea, the researcher recruited 103 male learners aged 15 to 18 from a private language institute in Babol, Iran, all classified as intermediate English as a Foreign Language students. After eligibility screening, participants were randomly assigned to one of four conditions: a working memory training group, a metacognitive intervention group, a combined group receiving working memory training embedded in metacognitive instruction, and a control group that continued with regular instruction. Following attrition-related exclusions, 83 learners completed every phase of the experiment, leaving final group sizes of 19, 21, 20, and 23 respectively.

The design was a randomized pretest-posttest controlled experiment, a gold standard structure for isolating causal effects in educational research. Before the interventions began, all participants completed measures of listening comprehension and working memory capacity, including a Digit Span Test, a classic assessment in which participants must recall increasingly long sequences of digits, either forward or backward. The same battery was administered after the training period, allowing the researcher to compare gains across the four groups while statistically adjusting for initial differences.

The working memory training condition targeted the cognitive machinery underlying listening. Working memory capacity is conventionally understood as a limited pool of attentional resources that determines how much linguistic material a learner can activate, store, and integrate in real time. For a language learner, this capacity constrains everything from holding a clause in mind while parsing its grammar to retaining a speaker’s earlier point long enough to connect it with what follows. Training tasks of the kind used in this study are designed to stretch that capacity through repeated, adaptive practice on demanding memory and attention exercises.

The metacognitive intervention, by contrast, did not aim to enlarge any mental muscle. Instead, it taught learners to regulate the resources they already possess. Metacognition in listening research typically encompasses planning before listening, such as activating background knowledge and setting a purpose; monitoring during listening, such as noticing when comprehension breaks down and deploying repair strategies; and evaluating afterward, such as judging how well the message was understood and what might be done differently next time. Learners in this condition were guided through explicit instruction and practice in these self-regulatory processes, transforming listening from a passive act of reception into an active, strategic performance.

The combined condition integrated the two approaches, embedding working memory training within the metacognitive intervention. The theoretical logic is elegant: a larger cognitive reservoir should make strategic regulation more effective, while better regulation should allow learners to deploy their available capacity more efficiently. If the resource availability and resource regulation frameworks truly capture distinct contributions to listening, then addressing both simultaneously should yield the strongest outcomes of all.

The results were revealing, and in some respects more nuanced than the integrative hypothesis predicted. All three active treatments produced improvements, confirming that both cognitive and metacognitive training are worthwhile. For listening comprehension specifically, the combined condition and the metacognitive intervention alone were the standout performers, both significantly enhancing learners’ ability to understand spoken English. For working memory capacity, the combined condition and working memory training alone were comparably effective, with no statistically significant difference between them. Intriguingly, in both outcome measures the combined group posted a descriptively higher adjusted mean than either single treatment, but that descriptive advantage fell short of statistical significance when compared with metacognitive intervention for listening or working memory training for memory capacity.

These findings carry a double message. On one hand, they validate the integrative framework: learners who received both forms of training never did worse than those who received either alone, and they tended to score highest on both outcomes. On the other hand, they temper expectations about additive benefits. The metacognitive intervention proved nearly as powerful as the full package for listening, and working memory training alone matched the package for building memory capacity. The practical implication is that educators seeking efficient interventions may achieve most of the benefit by choosing the training matched to their goal, while the combined approach remains a theoretically attractive option whose extra value awaits stronger empirical confirmation.

For the millions of English learners worldwide who struggle with listening, the study offers a hopeful reframing. The skills that determine whether a fast-moving conversation in a second language becomes comprehensible are not fixed traits but trainable systems. Teaching learners to plan, monitor, and evaluate their listening appears to be among the most reliable ways to improve comprehension, while targeted cognitive exercise can measurably expand the memory capacity that listening depends on. The next generation of research, the author notes, must continue to probe whether the combined approach delivers benefits beyond its components, but the evidence already suggests that listening instruction should look beyond the ear and toward the mind that interprets what the ear receives.

Subject of Research: The combined effects of working memory training and metacognitive intervention on English listening comprehension and working memory capacity in adolescent EFL learners

Article Title: Beyond hearing: investigating the combined impact of working memory and metacognitive training on EFL listening comprehension and working memory capacity

Article References: Muhammadpour, M. (2026). Beyond hearing: investigating the combined impact of working memory and metacognitive training on EFL listening comprehension and working memory capacity. BMC Psychology. https://doi.org/10.1186/s40359-026-05746-1

Image Credits: AI Generated

DOI: 10.1186/s40359-026-05746-1

Keywords: working memory, metacognition, listening comprehension, EFL, second language acquisition, cognitive training, metacognitive intervention, working memory capacity, randomized controlled trial, language learning, psycholinguistics, BMC Psychology

The research, led by Abdo Hasan AL-Qadri of Xi’an Eurasia University together with colleagues at institutions in Saudi Arabia and Algeria, set out to build and test a structural model of student engagement in university-level online EFL learning. Rather than asking a single question about whether students like online classes, the team mapped the relationships among three psychological constructs: self-regulation, which encompasses the ability to plan, monitor, and evaluate one’s own learning; online learning pathways, which describe how students participate across different components of online instruction; and student engagement itself, the behavioral and emotional investment learners make in their studies. The goal was to determine not just whether these factors correlate, but how they connect in a coherent causal architecture.

To do this, the researchers collected data from 625 undergraduate EFL students enrolled at three public universities in Algeria. The choice of setting is significant. Algerian higher education operates under substantial institutional and infrastructural constraints, and the country’s universities were pushed rapidly into digital delivery during the COVID-19 pandemic, a transition that previous scholarship has described as emergency remote teaching rather than carefully designed online learning. Studying engagement in such a context provides a demanding test of the model: if self-regulation predicts engagement where connectivity is patchy and digital pedagogy is still maturing, the relationship is unlikely to be an artifact of privileged, well-resourced learning environments.

Methodologically, the study followed a rigorous two-stage analytical sequence. First, the team administered validated questionnaires measuring self-regulation, online learning pathways, and student engagement, then subjected the responses to exploratory and confirmatory factor analyses. Exploratory factor analysis allows the underlying structure of a questionnaire to emerge from the data itself, revealing which items cluster together as dimensions of a single construct. Confirmatory factor analysis then tests whether that structure holds when imposed on the data as a fixed hypothesis, a crucial safeguard against mistaking statistical noise for genuine psychological dimensions. Only after the measurement structure was established did the researchers proceed to structural equation modelling, the technique that allows simultaneous testing of direct and indirect pathways among multiple constructs.

The results paint a clear picture. Self-regulation was positively associated with student engagement through both direct and indirect routes. In other words, students who are better at setting goals, tracking their progress, and reflecting on their strategies invest more effort and attention in their online English courses, even before any other factor is considered. But the indirect pathway is where the model becomes genuinely interesting. Self-regulation was also strongly related to online learning pathways, the patterns of participation students adopt across the various components of their digital courses, and those pathways were in turn positively associated with engagement outcomes. Statistical testing confirmed a significant indirect association: self-regulated learners navigate online learning more effectively, and that effective navigation feeds engagement.

Think of it as a chain of psychological transmission. A student who plans a study schedule, monitors comprehension while watching a recorded lecture, and evaluates performance after an online quiz is not merely accumulating good habits in isolation. That self-regulatory machinery shapes how the student engages with the architecture of online learning itself, which platforms to use, which activities to prioritize, how to distribute effort across asynchronous and interactive components. The model suggests that this structured participation is a conduit through which self-regulation translates into sustained engagement. Students who lack that machinery, by implication, may flounder not because they cannot learn English online, but because they never establish the participation patterns that make online learning cohere.

The explanatory power of the model adds weight to these conclusions. The structural equation model accounted for a substantial proportion of the variance in both online learning pathways and student engagement, indicating that the constructs the researchers chose capture much of what actually matters in these settings. In educational research, where dozens of competing variables, from family background to institutional quality, can dilute any single model’s reach, explaining a large share of engagement through self-regulation and participation pathways is a notable result. It suggests that in constrained online EFL environments, the learner’s internal regulatory capacity and the participation patterns it produces are among the dominant determinants of engagement.

These findings align with a growing body of international evidence. Systematic reviews of self-regulated learning strategies in blended and online higher education have repeatedly found that students who deploy planning, monitoring, and evaluation strategies achieve better outcomes and persist longer. Research on massive open online courses has shown that self-regulation predicts retention, and studies of EFL learners in contexts from Iran to China have linked self-regulatory skill to motivation, engagement with feedback, and language performance. What the Algerian study adds is an integrated structural model that positions online learning pathways as a mediating mechanism, giving researchers and instructors a more precise account of how self-regulation exerts its influence rather than simply documenting that it does.

The practical implications are considerable. If self-regulation drives engagement partly through the pathways students take through online learning, then interventions should target both. Universities operating under infrastructural constraints, and indeed any institution delivering online instruction, could incorporate explicit training in self-regulatory strategies, goal setting, self-monitoring, and reflective evaluation, into language curricula rather than assuming students arrive with these skills. Course designers, meanwhile, could structure online platforms so that productive participation pathways are visible and easy to follow, reducing the navigational burden on students whose regulatory capacities are still developing. The study’s authors suggest that learners’ capacity to plan, monitor, and evaluate their learning is closely linked to engagement in online EFL contexts, particularly participation across different components of online learning, a formulation that points directly at course architecture as a lever for improvement.

There are also broader lessons for the post-pandemic university. The mass shift to online instruction did not merely relocate classrooms; it redistributed cognitive labor from instructors to learners. Students in Algiers, like students everywhere, were suddenly responsible for managing their own academic activity in environments that offered fewer external supports. This study provides empirical evidence that in such systems, the psychological resources students bring to that task matter enormously, and that the way those resources shape digital participation is measurable, modelable, and potentially trainable. As online and blended learning becomes a permanent fixture of higher education rather than an emergency measure, understanding the mechanics of engagement, who stays invested, who drifts away, and why, may prove to be one of the most consequential questions educational science can answer. This research offers a rigorous first draft of that answer, drawn from a setting where the stakes for getting it right are especially high.

Subject of Research: A structural equation model of self-regulation, online learning pathways, and student engagement in online EFL learning among Algerian university students

Article Title: A structural model of student engagement in online EFL learning: self-regulation and online learning pathways

Article References: AL-Qadri, A. H., Al-khresheh, M. H., Abdelhadi, A., Saraa, N., Zhang, J., Alruwaili, S. F., & Almayez, M. (2026). A structural model of student engagement in online EFL learning: self-regulation and online learning pathways. SN Social Sciences, 6(10), Article 529. https://doi.org/10.1007/s43545-026-01794-x

Image Credits: AI Generated

DOI: 10.1007/s43545-026-01794-x

Keywords: self-regulation, student engagement, online learning, EFL, higher education, structural equation modelling, Algeria, online learning pathways, language learning, educational psychology, structural, model

The system targets a skill that decades of second-language acquisition research have identified as foundational: morphological awareness, the ability to slice unfamiliar words into their meaningful building blocks. A learner who recognizes that “international” decomposes into the prefix inter-, the root nat, and the suffixes -ion and -al can decode whole families of words instead of memorizing them one by one. Large language models, the authors note, are remarkably good at performing this decomposition in real time, yet the conversational interfaces through which we access them leave the resulting knowledge trapped in unstructured chat logs, fragmented and impossible to consolidate systematically.

Morphowiki’s architecture is deliberately simple and rests on three layers. A generation layer orchestrates seven Agent Skills, small callable command modules that direct the language model through specific tasks. A data layer stores everything as plain-text Markdown files organized into directories for words, prefixes, roots, and suffixes. A review layer runs a lightweight local Node.js server with a single-page web application and a JSON datastore that tracks spaced-repetition scheduling. Because the entire knowledge base lives in Markdown and JSON rather than a proprietary database, it can be read by humans, parsed by programs, diffed by version-control systems, and shared freely, in sharp contrast to the closed data silos of commercial study platforms.

The workflow follows an “initialize, ingest, expand, review” cycle. After a one-command setup that builds the directory structure and installs dependencies, the learner feeds in an unfamiliar word or an entire sentence. The system tokenizes the input, filters out stopwords, both a default list of roughly 150 English function words and a personalized exclusion list the learner can extend, and then applies what the authors call an architecturally bounded recursive decomposition. Given “international,” the pipeline isolates the root nation (itself nat plus -ion), expands to national, and resolves the target word with the prefix inter-, creating a separate Markdown page for each layer and wiring every page bidirectionally to its constituent morphemes. A visited-set mechanism prevents the recursion from looping endlessly.

Once a word is archived, the /flashcard command enrolls it into the review queue after a strict three-part check: the word’s Markdown page must exist, the word must not be on the stopword list, and no duplicate entry may exist in the JSON datastore. Scheduling follows the classic SuperMemo-2 algorithm, initialized with an easiness factor of 2.5 and a zero interval. When the learner runs the review command, a local server identifies due cards, parses the relevant Markdown on the fly, and presents a four-button evaluation, Again, Hard, Good, or Easy. The authors condensed the traditional six-grade SM-2 scale with a mathematical justification: all failed retrievals reset the interval and repetition counters without touching the easiness factor, so they can safely be merged into a single “Again” button.

Two further commands push the system beyond passive flashcards. The /new-word command scans the learner’s accumulated morpheme pool and synthesizes derivatives that are not yet recorded, for example proposing “nationalize” once the root nat has been mastered, then automatically running the ingestion and linking pipeline. Because language models can hallucinate plausible-sounding pseudowords, the skill specification embeds a validation step requiring the model to confirm that each candidate is an attested modern English word. A /lint command then audits the knowledge base in two modes: a deterministic, token-free structural check of link symmetry and schema completeness, and a model-assisted semantic inspection of individual entries. Crucially, lint never edits anything on its own; it only raises flags, and every state change requires explicit human authorization.

What elevates the paper beyond a software description is its quantitative evaluation of the decomposition step. The authors assembled 950 test items, including a main set of 600 words stratified by vocabulary level and morpheme count, drawn from the intersection of a Taiwanese senior-high-school word list and the SIGMORPHON 2022 morpheme segmentation shared task. Against the unsupervised baseline BPE, which scored an F1 of 29.13, and Morfessor 2.0, which reached 47.26, Morphowiki paired with Claude Opus 5 achieved an F1 of 83.06 and an exact-match rate of 74.83 percent. On a subset of 200 words containing four or more morphemes, the baselines managed essentially nothing, 0.00 and 3.00 percent exact match, while Morphowiki still resolved 46.50 percent correctly.

The most intellectually interesting result comes from a two-step ablation that separates the contribution of the underlying model from that of the Agent Skills scaffolding. With the model held fixed, removing the skill file and then the repository’s project instructions revealed that the two components do very different work. The project instructions, which describe the project in etymological terms but prescribe no segmentation rules, actually degraded every measure on the main set and nearly tripled over-segmentation of simple words, from 6.7 to 23.3 percent. The skill file, by contrast, added more than 52 points of exact-match accuracy on words recorded in canonical form and corrected 17 of the over-segmentation errors the instructions induced. The authors conclude that the skill does not teach the model how to segment strings, a capability it already possesses, but rather imposes canonical discipline: knowing when to restore a base to its dictionary form, when to record an assimilated prefix canonically, and where recursion should stop.

The team is candid about the system’s limits. Decomposition quality remains tethered to the reliability of the underlying model, and the deployed configuration still over-segments monomorphemic words at 12.0 percent, a trade-off traced to the etymological priming of the project instructions rather than the skill itself. The reference standard used for scoring is itself noisy, with coverage of the most elementary vocabulary at only 43.1 percent, so the reported accuracies likely underestimate true performance. Cross-model comparison covered only two models from a single provider, the pipeline is calibrated exclusively for English derivational morphology, and cost is nontrivial: roughly 9 to 23 US dollars per thousand words processed, which positions the framework for curated vocabulary construction rather than bulk corpus processing. Pedagogical efficacy, finally, has not yet been tested in classrooms.

Even with those caveats, Morphowiki speaks to a tension now rippling through education research. Recent studies on cognitive offloading and “metacognitive laziness” suggest that instantaneous AI answers can undermine the deep semantic engagement that durable memory requires, a concern rooted in the classic levels-of-processing work of Craik and Lockhart. By forcing learners through active retrieval, spaced review, and human-approved curation, Morphowiki is designed to make the AI a builder of infrastructure rather than a dispenser of answers. The authors frame it as a reproducible testbed on which researchers in computer-assisted language learning can systematically compare models, prompting strategies, and scheduling algorithms on a standardized, transparent data format. If the broader lesson holds, the future of AI-assisted learning may depend less on how cleverly a chatbot explains a word, and more on whether the explanation leaves behind something the learner can keep.

Subject of Research: An open-source Agent Skills framework that uses large language models to decompose English vocabulary into morphemes, build bidirectionally linked knowledge graphs, and schedule spaced-repetition review.

Article Title: Morphowiki: reusable agent-skill infrastructure for morphology-centered vocabulary knowledge construction

Article References: Chen, Y.-T., Yao, K.-C., Chang, S.-H., & Lin, Y.-C. (2026). Morphowiki: reusable agent-skill infrastructure for morphology-centered vocabulary knowledge construction. SoftwareX, 36, Article 103114. https://doi.org/10.1016/j.softx.2026.103114

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103114

Keywords: Morphowiki, large language models, morphology, vocabulary acquisition, spaced repetition, Agent Skills, computer-assisted language learning, open-source software, second language acquisition, cognitive offloading, Markdown knowledge graph, SM-2 algorithm

The research, published in Discover Education, was conducted by Yang Zhang and Ying Zhou of Kaifeng Vocational College of Culture and Arts in Henan Province. The pair set out to test whether AI-assisted speaking practice could improve oral English among first-year students in higher vocational education, a sector in China where English instruction has traditionally emphasized reading, writing, and examination performance over genuine communicative ability. Their quasi-experimental study involved 92 students aged 18 to 20, all enrolled in a compulsory College English course and all with at least six years of prior English study behind them.

The design was straightforward but carefully controlled. Two intact classes of 46 students each were compared. Both groups received the standard four hours per week of conventional College English instruction from the same teacher, ensuring that any difference in outcomes could not be attributed to differences in classroom teaching. The experimental group, however, added three thirty-minute sessions per week of AI-assisted speaking practice using iFLYTEK Spark, version 3.5, a generative AI platform developed by one of China’s leading speech technology companies. The control group spent the equivalent additional time on conventional pair-work and group speaking activities covering the same textbook topics at comparable task complexity.

The technology at the heart of the intervention is more sophisticated than a simple chatbot. iFLYTEK Spark integrates automatic speech recognition, natural language processing, and generative dialogue capabilities. For speaking practice, it offers phoneme-level pronunciation assessment with visual feedback on individual sounds, intonation, and stress; real-time scoring aligned with IELTS-style criteria on a zero-to-nine scale covering pronunciation, vocabulary, grammar, fluency, and coherence; interactive AI-generated dialogue for open-ended conversation, role-play, and topic-based discussion; and detailed feedback reports pinpointing specific areas for improvement. Each practice session followed a three-phase cycle: a five-minute warm-up dialogue with the AI on topics related to the current unit, a fifteen-minute block of guided role-play and one-minute monologues with immediate feedback, and a ten-minute self-reflection phase in which students reviewed their AI-generated reports and re-recorded weak segments.

At the end of the twelve weeks, students’ speaking was assessed through a standardized task consisting of a two-minute monologue on a randomly assigned topic and a one-minute interactive exchange with an examiner. Three trained English teachers, each with at least five years of experience, rated the audio recordings independently and blind to group assignment, using an analytic rubric covering pronunciation, language use, fluency and coherence, and interaction. Inter-rater reliability, measured with intraclass correlation coefficients, ranged from 0.810 to 0.926, indicating good to excellent agreement. The results were striking: the experimental group achieved a significantly higher mean overall posttest score of 3.69 compared with 3.19 for the control group, a difference that was statistically significant with a medium-to-large effect size of Hedges’ g equal to 0.63.

Perhaps the most intriguing finding lay in the pattern across subdimensions. The largest effect appeared in interaction, the ability to engage in meaningful conversational exchange, where the effect size reached 0.75 and remained significant even after conservative Bonferroni correction for multiple comparisons. Differences in language use and in fluency and coherence showed medium effect sizes of 0.49 and 0.50 respectively, though these did not survive the strictest statistical threshold and should be treated as suggestive. Pronunciation, by contrast, showed only a smaller and non-significant difference of 0.38. The authors suggest that segmental pronunciation accuracy may require more specialized and sustained phonological training than general-purpose AI conversation tools typically provide, or that twelve weeks was simply too short for such gains to emerge, or that students prioritized communicative fluency over phonological precision during practice.

The theoretical scaffolding behind the study helps explain why the intervention worked. Swain’s Output Hypothesis holds that language production, not merely comprehension, drives second language acquisition by forcing learners to process language deeply and notice gaps in their own competence. The AI sessions demanded substantial pushed output across varied tasks. Krashen’s Affective Filter Hypothesis proposes that anxiety can block acquisition, and the AI’s non-judgmental, private practice environment appears to have lowered that barrier. Long’s Interaction Hypothesis emphasizes negotiation of meaning, a function the AI’s conversational role-play partially fulfilled by prompting students to clarify, rephrase, and respond in real time. The pronounced interaction advantage observed in the data aligns closely with this third mechanism.

Interviews conducted after the intervention with one teacher and three students added a human dimension to the statistics. All three students reported that the AI dramatically increased their opportunities for spoken practice, offering unlimited attempts without the time constraints and turn-taking limits of classroom pair work. Two students said they felt less embarrassed making mistakes in front of the machine because there was no fear of social judgment, with one describing the AI as a patient practice partner that never showed frustration. The teacher observed that students who were typically reluctant to speak in class became more willing participants after gaining confidence through AI sessions. Yet the participants also flagged limitations: the AI’s conversational responses sometimes felt mechanical or repetitive, one student preferred combining AI practice with human interaction, and the teacher stressed that continued teacher involvement remained essential for guiding reflection and correcting persistent errors the AI failed to address.

The authors are candid about the study’s constraints. Because intact classes rather than randomly assigned individuals were used, definitive causal claims are not possible. Baseline comparability relied on a coarse institutional proficiency classification rather than a dedicated standardized speaking pretest, so pre-existing differences in oral skill cannot be fully ruled out. The sample came from a single college in Henan Province, the qualitative sample was small, and a novelty effect from using exciting new technology cannot be excluded. The posttest-only design means the findings should be read as group differences under approximate baseline comparability rather than demonstrated learning gains.

Even with those caveats, the study offers a replicable template for educators: a structured cycle of warm-up dialogue, guided practice, and self-reflection, layered on top of conventional instruction rather than replacing it. As generative AI tools with speech recognition spread through education systems worldwide, the evidence suggests their greatest value may lie not in replacing teachers but in multiplying opportunities for the one activity that language learning cannot do without: speaking, freely and often, without fear.

Subject of Research: Effects of AI-assisted speaking practice on spoken English performance in higher vocational education

Article Title: Effects of AI-assisted speaking practice on spoken English performance in higher vocational education

Article References: Zhang, Y., & Zhou, Y. (2026). Effects of AI-assisted speaking practice on spoken English performance in higher vocational education. Discover Education, 5(1), Article 1158. https://doi.org/10.1007/s44217-026-02260-9

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02260-9

Keywords: artificial intelligence, spoken English, higher vocational education, iFLYTEK Spark, generative AI, speech recognition, language learning, quasi-experimental study, speaking anxiety, second language acquisition, China, educational technology

The stakes are high in a country where spoken English opens doors to social, academic, and professional opportunity, yet where proficiency remains stubbornly low. Pakistan ranked 67th among 116 countries in the Education First English Proficiency Report, and earlier research by the same team found that 88 percent of secondary school students performed at only the most basic speaking levels, classified as A1 or A2 on the Common European Framework of Reference (CEFR). Just 12 percent reached the pre-intermediate B1 level, and not a single student achieved advanced proficiency. The reason, the researchers argue, is structural: Pakistani public schools emphasize rote memorization and written exercises, leaving students with years of English instruction but almost no meaningful oral practice.

The study was grounded in Vygotsky’s Sociocultural Theory, which holds that learners advance through interaction with a More Knowledgeable Other within their Zone of Proximal Development, the space between what a learner can do alone and what they can achieve with guidance. In this framing, ChatGPT served as a scaffold: an endlessly patient conversational partner that adjusted its questions to the student’s level, prompted elaboration, and corrected errors in real time. After each individual practice session, students joined collaborative classroom activities such as role plays, pair work, and group discussions, allowing them to consolidate what the AI practice had unlocked.

The intervention itself was built with the ADDIE instructional design model and aligned with the Grade 9 English syllabus. Each week began with a five-minute teacher briefing, followed by sixty minutes of ChatGPT-assisted speaking practice in the school’s IT laboratory, and closed with twenty-five minutes of collaborative learning. Students used standardized prompts to steer the conversation. In one documented exchange on the topic of child labor, a student first asked ChatGPT to act as an English tutor and provide a word bank of useful vocabulary, then asked factual questions about the topic, and finally instructed the AI to act as a speaking partner, asking questions one at a time and giving brief feedback after each answer. Crucially, students kept the same conversation thread throughout the twelve weeks, so the AI retained the history of prior interactions and could build on them.

Measuring the outcome required a rigorous assessment framework. The researchers developed a Spoken English Proficiency Test modeled loosely on the IELTS structure, comprising a warm-up conversation, a four-to-five-minute monologue, and a five-to-six-minute dialogue. Each task was scored against CEFR descriptors across five dimensions: range, the breadth of vocabulary and expression; accuracy, grammatical correctness; fluency, the smoothness of delivery; interaction, the ability to engage a conversational partner; and coherence, the logical organization of speech. Each component was scored from 1 to 6, corresponding to CEFR levels A1 through C2, for a maximum total of 60 points. The instrument showed exceptional internal consistency, with a Cronbach’s alpha of 0.964.

In a novel twist, the assessment itself was conducted by ChatGPT. The researchers uploaded the CEFR scale, the marking scheme, and the test materials, then prompted the AI to act as a CEFR English-speaking examiner. Students spoke directly to ChatGPT in Voice Mode while the researcher supervised, and the AI generated real-time transcripts and scored the speaking components. Human language experts, blinded to group allocation, oversaw quality control and independently scored the interaction component of the monologue task, since ChatGPT could not evaluate non-verbal engagement such as facial expressions and gestures. This human-AI collaborative approach aimed to combine standardized, consistent scoring with expert judgment.

The results were unambiguous. Before the intervention, the two groups were statistically equivalent, with all students clustered in the basic A1-A2 range. After twelve weeks, the experimental group’s mean overall score reached 36.56, compared with 17.60 for the control group, a difference the Mann-Whitney U test confirmed as highly significant, with U = 1.50, z = -6.08, and p < .001. The effect size of r = .86 is unusually large for an educational intervention. Perhaps more telling is the raw gain: the ChatGPT group improved by an average of 20.44 points, while the control group gained just 1.04. On the CEFR scale, experimental students progressed from Basic User to Independent User levels, with one learner reaching C1, the threshold of Proficient User.

The improvements held across every measured dimension and both task types. In monologue performance, where students spoke independently on topics such as their dream job, the experimental group significantly outperformed the control group on range, accuracy, fluency, interaction, and coherence, all at p < .001. The same pattern emerged in dialogue tasks, where students conversed on everyday topics such as weekend plans. This across-the-board pattern matters because previous studies of ChatGPT in language education tended to focus on motivation, confidence, or overall proficiency; this study is among the first to isolate and document gains in the specific qualitative dimensions that define spoken competence under an internationally recognized framework.

The findings align with a growing body of international evidence. Studies in other contexts have reported that interactions with ChatGPT improve oral speaking performance, that consistent engagement with AI-assisted speaking activities produces greater gains than sporadic use, and that twelve-week chatbot interventions help learners outperform traditionally taught peers. Researchers have also documented improvements in vocabulary use, grammatical accuracy, and fluency from repeated AI-supported practice. Yet the literature carries caveats: systematic reviews warn about ethical concerns, uneven quality of AI-generated feedback, and the fact that chatbots cannot fully replicate human interaction. Effectiveness, the evidence suggests, depends less on the technology itself than on pedagogical integration, teacher guidance, and instructional design, all of which were central to the Pakistani module.

The authors are careful about the limits of their claims. The sample comprised fifty female students from a single urban public school, a homogeneous group in age, socio-economic background, and educational history, which may have reduced variability and inflated effect sizes. The twelve-week duration cannot speak to long-term retention, and the researchers recommend delayed post-tests, larger and more diverse samples, and mixed-methods designs to capture how the learning actually happens. The theoretical interpretation, that ChatGPT provided scaffolding within students’ zones of proximal development, remains a plausible explanation rather than a directly demonstrated mechanism. Still, the practical implications are hard to ignore. For classrooms where one teacher faces dozens of students and individual speaking practice is a logistical impossibility, an AI conversational partner offers a scalable supplement, and the study suggests that policymakers weighing digital infrastructure and teacher training for AI-supported language education now have experimental evidence to inform those decisions.

Subject of Research: Effectiveness of a ChatGPT-assisted learning module for improving English-speaking proficiency among secondary school students

Article Title: Integration of ChatGPT to enhance English speaking skills among secondary school students in Pakistan

Article References: Nageen, S., Khawaja, H. U. H., Sarwar, M., Moin, M., Imran, Z., & Jabeen, S. (2026). Integration of ChatGPT to enhance English speaking skills among secondary school students in Pakistan. Discover Education, 5(1), Article 1137. https://doi.org/10.1007/s44217-026-02278-z

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02278-z

Keywords: ChatGPT, artificial intelligence, English speaking proficiency, CEFR, Pakistan, secondary education, language learning, experimental study, educational technology, Vygotsky, scaffolding, randomized controlled trial

The context is stark. Global forced displacement rose from roughly 40 million people in 2012 to 108.4 million in 2022, and Sweden’s foreign-born population now stands at around two million residents, approximately 20 percent of the population. Yet integration into the Swedish labour market remains slow. The average refugee waits about three years for a first job, and after fifteen years in the country only 59 percent of refugee women are established in working life, compared with 82 percent of native-born women. Women with refugee backgrounds face the longest periods of inactivity, and once employed they risk double discrimination based on both ethnicity and gender, which can mean exclusion from workplace networks, blocked promotions, and lower pay.

Previous research has catalogued the barriers and facilitators of labour market entry: higher education speeds the transition, language learning and vocational training help, and strategies such as volunteering, using social networks, and accepting low-quality survival jobs are common. But the Swedish team wanted to move beyond this structural accounting. Drawing on the sociologist Vanessa May’s conceptualisation of belonging and on Yuval-Davis’s politics of belonging, they reframed the question. Belonging, they argue, is not a fixed end-state achieved once someone is employed. It is a dynamic, continually negotiated process shaped by interactions, emotions, and power dynamics, and it captures dimensions of integration that employment statistics cannot see.

The researchers recruited eleven women aged 22 to 44, all from non-European countries, all with work experience in Sweden and all able to speak Swedish. They had lived in Sweden between three and ten years and were recruited through a local church, a Swedish for Immigrants (SFI) school, and Women On Wednesdays, a networking organisation for foreign women. Each participant took part in a life-story interview lasting between 60 and 130 minutes, with a median of 100 minutes, and face-to-face sessions included a deliberate coffee break to ease conversation. The transcripts were analysed using reflexive thematic analysis in an abductive logic, moving back and forth between the women’s narratives and the theoretical concepts of belonging until four robust themes emerged.

The first theme, navigating language and everyday communication, revealed how deeply linguistic hierarchies structure workplace life. The women described Swedish as the currency of participation, the language that opened doors to recognition and trust. One participant initially refused to learn the language because she did not see her future in Sweden, but a difficult first job changed her mind. If you cannot speak Swedish, she explained, someone else can go to the boss faster and tell what happened, and that is unfair. The study found that formal language instruction was insufficient on its own; workplaces became crucial arenas where women gained confidence by being surrounded by Swedish-speaking colleagues and given opportunities to speak and be heard.

The second theme concerned the embodied negotiation of cultural norms. Women described adjusting dress, gesture, tone, and even humour as they learned Swedish workplace codes. One participant explained that her hijab was part of her identity but not always a religious statement; at work it signalled her boundaries, such as not hugging colleagues, and when those boundaries were respected she felt safe and accepted. She also noted that some immigrant women removed the hijab to avoid discrimination or missed opportunities. Others described learning to interpret jokes and indirect communication over months of listening and observation. The researchers interpret these accounts as a gradual, embodied learning process in which women remained authentic to themselves while navigating expectations implicitly coded as Swedish, exposing the conditional nature of inclusion.

Relational belonging emerged as perhaps the most powerful force in the women’s trajectories. Informal mentors, Swedish families who extended invitations, internship supervisors who helped write CVs and served as references, and community organisations all filled gaps that formal integration programmes could not. Crucially, the quality of relational encounters mattered more than mere access to networks. Trust, moral affirmation, and sustained interpersonal investment enabled women to move from isolation toward participation, and some described these relationships as forming a new family. Many women then turned around and supported others, organising study groups, book circles, and introductions to networks, describing the satisfaction of giving back once they had received help themselves.

The sensory and temporal dimensions of belonging proved equally striking. Several women described visceral, physical reactions at moments of recognition: one trained nurse from Iran cried the first time she took a blood sample during her internship, shattering an internalised label that she could not handle anything. Others described shaking with nerves before colleagues began to trust them, and then trusting themselves more in turn. These affective moments, the researchers argue, show that belonging is not merely cognitive but corporeal, and that affirmation in hierarchical workplaces is often sporadic and must be repeatedly earned. Temporally, the women’s narratives traced a journey from dependency to independence, from waiting to strategic action, with civil society organisations acting as anchors that helped re-establish daily rhythms and future-oriented aspirations.

The study’s implications are pointed, particularly as Sweden’s government shifts toward stricter migration and integration requirements emphasising language, self-sufficiency, and reduced asylum-related immigration. The authors argue that policies equating labour market participation with integration miss the point: workplaces must actively cultivate cultural, relational, sensory, and temporal belonging through mentorship programmes, early access to meaningful work, and managers who understand that inclusion depends on respecting difference rather than demanding assimilation. Belonging, the findings insist, is an evolving process rather than an achieved state, one that shapes identity, agency, health, and the capacity to imagine a future. When a woman can say, as one participant did after many rejected applications, that she declared she would work at a certain company someday and last week signed the contract, the pathway to working life has done more than deliver a paycheque. It has delivered a self.

Subject of Research: Immigrant women's experiences of belonging during their pathways into working life in Sweden

Article Title: Belonging as a part of the pathway to working life: immigrant women’s experiences in Sweden

Article References: Lydell, M., Garell, C., Lindgren, E.-C., Zimmerman Nilsson, M.-H., & Hertting, K. (2026). Belonging as a part of the pathway to working life: immigrant women’s experiences in Sweden. BMC Public Health, 26(1), Article 2808. https://doi.org/10.1186/s12889-026-29688-w

Image Credits: AI Generated

DOI: 10.1186/s12889-026-29688-w

Keywords: belonging, integration, immigrant women, Sweden, labour market, refugees, qualitative research, thematic analysis, workplace inclusion, language learning, social networks, public health