International Journal of Language Learning and
Teaching Research
IJLLTR
Volume 1 (2026)
https://doi.org/10.65553/IJLLTR.260106
Integrating Perception and
Production through Technology-Supported Mandarin Pronunciation Instruction: A
Quasi-Experimental Classroom Study
Liu Shi
School of Foreign
Studies, North China University of Water Resources and Electric Power, Zhengzhou,
China
Yanping Yan
School of Foreign
Studies, North China University of Water Resources and Electric Power, Zhengzhou,
China
Jingjing Xing
The Ontario
Institute for Studies in Education, University of Toronto, Toronto, Canada
Jie Zhang
School of Foreign
Studies, North China University of Water Resources and Electric Power, Zhengzhou,
China
Corresponding author: Zhangjie1@ncwu.edu.cn
Digital speech technologies can
extend pronunciation practice beyond limited class time, but their pedagogical
value depends on principled integration with instruction and feedback. Grounded
in the revised Speech Learning Model (SLM-r), this study developed and piloted
a four-phase perception–production model for Mandarin pronunciation
instruction. The model was implemented through Hanyin
Yusheng, a WeChat mini-program combining auditory discrimination, recorded
production, automated speech evaluation, practice resources, and
teacher-accessible learning analytics. The project comprised two connected
studies. First, 40 intermediate learners completed parallel perception and
production tasks targeting Mandarin lexical tones, the unaspirated–aspirated
onset contrast represented by Pinyin d–t (/t/–/tʰ/),
and the finals /i/, /a/, and /u/. Perception and
production scores were strongly correlated (Spearman’s ρ = .755, p
< .001). Second, 32 learners in intact classes received either
technology-enhanced perception–production instruction (n = 16) or conventional
teacher-led instruction (n = 16). Across a four-week, eight-lesson
intervention, both groups completed a pretest, an immediate posttest, and a
seven-day follow-up test. A 2 × 3 mixed analysis showed significant main
effects of group, F(1, 30) = 21.821, p
< .001, ηp² = .421, and test occasion, F(2,
60) = 160.883, p < .001, ηp² = .843, but no significant group-by-time
interaction, F(2, 60) = 1.555, p = .222,
ηp² = .049. The findings provide preliminary evidence for the feasibility of
integrating perception, production, automated feedback, and spaced practice.
Because the interaction was nonsignificant, however, they do not warrant a
strong causal claim that the technology-enhanced condition produced greater
gains. Implications are discussed for technology-enhanced Mandarin
pronunciation pedagogy and more rigorous future trials.
Keywords:
Mandarin
pronunciation; second language speech learning; automated speech evaluation;
perception–production relationship; technology-enhanced language learning;
classroom intervention
Pronunciation is central to communicative success in an additional language because inaccurate segmental and suprasegmental patterns can impede intelligibility, lexical access, and interaction. Mandarin pronunciation is especially demanding. In addition to consonantal and vowel contrasts, learners must attend to lexical tone, where changes in fundamental-frequency height and contour distinguish word meanings. Learners therefore need both perceptual categories for relevant contrasts and articulatory and laryngeal routines for producing them. Pronunciation teaching, however, is often constrained by limited contact time, large classes, delayed feedback, and an emphasis on imitation without systematic perceptual learning.
Digital speech technologies may
alleviate some of these constraints. Mobile platforms can distribute short
listening and speaking tasks, provide repeated exposure to models, collect
spoken responses, and return immediate automated feedback. Automatic speech
recognition (ASR) and automated speech evaluation can increase practice
frequency and privacy, potentially supporting learner
autonomy while reducing the cost of individualized feedback. Evidence from
computer-assisted pronunciation training is broadly positive but heterogeneous.
A meta-analysis of 15 ASR pronunciation studies reported a medium overall
effect and identified feedback type, treatment duration, linguistic target, and
peer involvement as moderators (Ngo et al., 2024). Individual studies have likewise reported benefits
from ASR-supported speaking practice (Bashori et al., 2024; Sun, 2023),
mobile feedback (Dai & Wu, 2023),
and automated speech evaluation (Zou et al., 2023a). Technology, however, is not itself a pedagogy: opaque
scores, accent-related recognition errors, limited diagnosis, and
decontextualized repetition may fail to establish robust perceptual categories
(Wu et al., 2025).
Recent syntheses reinforce the
importance of aligning technological affordances with pronunciation pedagogy
and assessment. In a systematic review of computer-assisted pronunciation
training, Amrate and Tsai (2025) documented variation in instructional targets and
assessment practices and noted that relatively few studies used broader
measures such as intelligibility and comprehensibility. This observation
highlights the need to specify both the phonetic targets and the scope of the
outcomes when evaluating technology-supported pronunciation instruction.
This limitation is consequential
in Mandarin. Tone realization varies with prosodic and phonological context,
and perception and production may share some mechanisms while remaining partly
skill-specific (Hao, 2018; Li & DeKeyser, 2017). Visual-acoustic feedback can facilitate both Mandarin
tone perception and production (Chen, 2024), while perception–production relationships appear
stronger for critical contour cues than for every acoustic dimension (Leung
& Wang, 2020). Effective instruction
should therefore coordinate discrimination, guided production, interpretable
multimodal feedback, and repeated opportunities to revisit difficult
categories.
To address this need, the present
study developed a technology-supported perception–production model for teaching
Mandarin tones and selected segmental contrasts. The model was grounded in
SLM-r (Flege & Bohn, 2021), which conceptualizes second-language speech learning
as a dynamic, input-driven process shaped by the quantity and quality of
experience and by individual differences. It was implemented through Hanyin Yusheng (literally, Chinese Phonetics and Speech), a
project-developed WeChat mini-program connecting diagnostic assessment,
adaptive lesson planning, multimodal classroom work, automated feedback, and a
seven-day review cycle.
The study makes three
contributions. First, it extends technology-enhanced pronunciation research
beyond its predominantly English-learning contexts by examining Mandarin as an
additional language, where learners must coordinate lexical tones and segmental
contrasts. Second, it extends the application of the revised Speech Learning
Model (SLM-r) by considering technology-mediated learning environments as a new
context for phonetic experience, in which diagnostic input, repeated production
opportunities, and feedback cycles may shape the development of perceptual and
productive categories. Third, it treats perception and production as
coordinated but separately trainable objectives rather than assuming
that improvement in one automatically transfers to the other.
The original Speech Learning Model proposed that
second-language production accuracy is constrained by how learners perceive L2
sounds and classify them in relation to established first-language categories (Flege, 1995). SLM-r retains the
interaction between first- and additional-language phonetic systems but gives
greater attention to dynamic change across the lifespan, input distributions,
and individual differences (Flege & Bohn, 2021). Importantly, the revised account does not require a
simple perception-first sequence. Perceptual and production learning may
develop in parallel, at different rates, and under different task demands.
Empirical work supports both
association and relative independence. In Mandarin tone learning, overall
identification and read-aloud accuracy may correlate even when error patterns
differ by tone context and task (Hao, 2018). Leung and Wang (2020) found production–perception correspondence for critical
F0 contour cues but not for lower-weighted height cues. Training can also be
channel-specific: learners tend to perform best in the modality they practiced,
with incomplete transfer to the other modality (Li & DeKeyser, 2017). A correlation therefore neither makes perception and
production identical constructs nor implies that training only one will
necessarily improve the other.
Instruction should accordingly
include high-quality models, contrastive listening and categorization, explicit
articulatory explanation, controlled production, and feedback directing
attention to relevant acoustic or articulatory properties. Learners should move
iteratively between input and output: perception provides targets and error
awareness, whereas production sharpens attention to contrasts and tests whether
emerging categories can be used in speech.
While SLM-r provides a theoretical
account of how perceptual and productive phonological categories develop
through experience, the present study extends this perspective by examining how
technology-mediated environments may increase opportunities for noticing,
feedback, practice, and revision during pronunciation learning. Thus,
technology is conceptualized not as a replacement for phonological learning
mechanisms but as an infrastructure that facilitates interaction between
perception, production, and instructional mediation.
Pronunciation instruction is
effective overall, although effects depend strongly on what is taught and how
outcomes are measured. Lee et al. (2015) reported a large overall effect across 86 reports, with stronger effects
for longer treatments, fuller feedback, and controlled measures. Saito and
Plonsky (2019) showed that conclusions vary
according to whether outcomes target specific versus global pronunciation, use
human or acoustic ratings, and elicit controlled or spontaneous speech. Gains
on narrow word-reading tests should therefore not be equated with gains in
spontaneous-speech comprehensibility.
Computer-assisted pronunciation
training enables repetition and immediate response. Successful systems must
align technical functions with learning principles rather than foreground
novelty (Neri et al., 2002).
ASR can support learner autonomy by allowing learners to monitor recognition
results and revise their output (McCrocklin, 2016; McCrocklin & Levis, 2025), but feedback design matters (Wang & Young, 2015), and automated feedback is not inherently superior to
teacher or peer feedback (Dai & Wu, 2023).
Ngo et al. (2024) similarly
found larger ASR effects with explicit feedback, moderate or longer treatments,
and peer involvement. Recent systematic evidence also emphasizes the need to
align computer-assisted pronunciation training with explicit pedagogical and
assessment principles (Amrate & Tsai, 2025). Automated evaluation should thus function as one
source of formative evidence in a teacher-guided cycle, not as an unquestioned
replacement for professional judgment.
Recent empirical studies have
begun to examine AI-supported pronunciation practice in more targeted
instructional settings. Aly and Alsehibany (2025) investigated the integration
of ChatGPT and SpeechAce in EFL pronunciation
instruction and reported improvements in selected pronunciation and
motivational outcomes. Xodabande et al. (2025) likewise found that a ChatGPT-supported intervention
was associated with sustained pronunciation gains among Iranian EFL learners.
More directly relevant to Mandarin, Cheng et al. (2025) developed an AI-enhanced multimodal tone-training
program combining automatic speech recognition, dynamic time warping, and
visual feedback; their findings indicated improvements in both tone production
and perception among learners from non-tonal language backgrounds. Liu (2026) examined voice-based ChatGPT for CSL learners and
reported improvement in comprehensibility and tone, although gains in initials
and finals were comparatively smaller. These studies suggest growing potential
for AI-mediated pronunciation practice while also showing that outcomes may
vary by linguistic target, learner background, and feedback design.
Recent work also highlights the
motivational and affective conditions of technology use. Wang and Xue’s (2024) intervention found that
AI-driven chatbots enhanced Chinese EFL learners’ academic engagement,
suggesting that interactive affordances can affect participation as well as
linguistic practice. Liu et al. (2025) identified distinct emotional and autonomous-learning
profiles in AI-enhanced data-driven language learning, underscoring the
heterogeneity concealed by group averages. In a robot-assisted environment for
students preparing to teach Chinese internationally, Zhang and Yan (2025) found that grit and enjoyment contributed to
engagement, with enjoyment emerging as the stronger predictor. Together, these
findings imply that the effectiveness of digital pronunciation work may depend
on how technology structures agency, emotion, and sustained participation, not
merely on scoring accuracy.
Learner responses to automated
feedback are also receiving closer attention. Huang and Pan (2026), using a longitudinal
design, found that learners’ perceptions of the diagnostic value of AI
pronunciation feedback were associated with subsequent trust, pronunciation
self-efficacy, and pronunciation performance. In addition, Zou et al. (2025) examined learners’ acceptance of an AI
speech-evaluation program, underscoring the importance of user acceptance in
technology-supported speaking practice. These findings reinforce the view that
the pedagogical value of automated pronunciation tools depends not only on technical
feedback delivery but also on whether learners regard the feedback as useful,
trustworthy, and actionable.
Speech learning is inherently
multimodal: learners coordinate auditory input, visual representations of
articulatory movement or pitch, kinesthetic experience, orthographic
representations, and feedback on their own speech. In Mandarin tone teaching, a
four-week study found larger perceptual and production gains from Praat-generated audiovisual feedback than from researcher
feedback alone (Chen, 2024). Multimedia can support attention and retention when learner proficiency
and cognitive resources are considered (Teng, 2022, 2023), but poorly aligned displays
may increase processing demands. Visualizations and videos should direct
attention to a linguistic target rather than decorate a lesson.
Digital platforms can also create
a learning-analytics feedback loop. Pre-class responses reveal class-wide and
individual difficulties and help teachers allocate time and select examples;
post-class and follow-up responses indicate change and retention. This is
pedagogically different from merely digitizing worksheets because learner
evidence informs subsequent instruction. Human interpretation remains
essential. Qin and Dong (2025), for example, showed that learners’
lived affective interpretations of GenAI-mediated feedback did not map
automatically onto their actual revision behavior. Although their study
concerned writing, its sociocultural implication transfers to pronunciation:
automated output becomes developmentally useful only when learners interpret
it, act on it, and receive appropriate mediation. Accordingly, the present
model combines automated evaluation with teacher explanation in a predictable
four-phase routine.
Recent research on multimodal
generative AI in Chinese-language learning further suggests that engagement and
phonological development should not be treated as interchangeable outcomes.
Wang (2026), for
example, reported that multimodal GenAI supported task preparation and learner
participation, while improvements in tonal accuracy and delivery confidence
remained limited. This finding is relevant to the present study because it
cautions against assuming that greater access to multimodal technology or
stronger engagement necessarily translates into equivalent gains across
pronunciation dimensions.
Three gaps motivated the
project. First, technology-enhanced pronunciation research has concentrated on
English, with comparatively little classroom evidence on Mandarin tones and
segmental contrasts. Second, many interventions foreground production and automated
scoring without a theoretically explicit perceptual component. Third, some
studies infer effectiveness from experimental-group pre–post gains even when
the relevant group-by-time comparison is absent or nonsignificant. A persuasive
evaluation must consider change over time, between-group differences, and their
interaction.
The study addressed the following
questions:
To what extent are
Mandarin pronunciation perception and production scores associated among
intermediate learners completing parallel tasks?
How do pronunciation
scores vary by instructional group and test occasion during a four-week
classroom intervention?
What pedagogical
conclusions about the technology-supported perception–production model are
warranted by the observed main and interaction effects?
Study 1 employed a cross-sectional perception–production
design to examine the relationship between Mandarin pronunciation perception
and production among intermediate learners and to provide empirical evidence
for integrating these two dimensions in subsequent technology-supported
instruction.
Forty intermediate learners of Mandarin
participated in Study 1, including 20 women and 20 men, with ages ranging from
20 to 25 years. All participants had achieved intermediate HSK proficiency or
above and had studied Mandarin in China for approximately 2.5–3 years. At the
time of data collection, they were engaged in formal Mandarin learning programs
and had received systematic instruction in Mandarin pronunciation and
phonological features.
The
participants were selected because intermediate learners have acquired basic
Mandarin phonological knowledge while still experiencing challenges in
accurately perceiving and producing non-native phonological contrasts. This
proficiency level was therefore considered appropriate for examining the
relationship between pronunciation perception and production, as learners at
this stage possess sufficient linguistic experience to complete controlled
perception and production tasks while maintaining observable variability in
pronunciation performance.
All
participants voluntarily participated in the study and completed both
perception and production tasks. They were informed of the research procedures
and provided consent before data collection. Their participation did not affect
their course grades or academic evaluation. The participants were international
learners from five Southeast Asian countries: Vietnam, Laos, Cambodia,
Thailand, and Indonesia. Their diverse linguistic backgrounds provided a
cross-national context for examining Mandarin pronunciation perception and
production. The distribution of participants across the five countries is
reported separately for each study.
Study 1 employed parallel perception and production
tasks designed to assess learners’ ability to perceive and produce the same
Mandarin phonological contrasts. The tasks targeted three major pronunciation
domains: (a) the four Mandarin lexical tones, (b) the alveolar stop aspiration
contrast represented by Pinyin d and t (approximately /t/ and /tʰ/), and (c) the finals i, a,
and u. These targets were selected because they represent fundamental yet
challenging features of Mandarin pronunciation for second-language learners,
involving both suprasegmental and segmental contrasts.
The
test materials consisted of 24 matched syllable items. Specifically, the
materials included 16 tone-related items representing the four lexical tones, 4
aspiration-contrast items targeting the d–t distinction, and 4 final-vowel
items involving i, a, and u. This balanced item
structure enabled direct comparison between learners’ perceptual and productive
performance across the same phonological targets. The stimuli included
combinations such as dī, dí,
dǐ, dì, tī,
tí, tǐ, tì,
dā, dá, dǎ,
dà, dū, dú,
dǔ, and dù. The
matched-item design enabled direct comparison between learners’ perceptual and
productive performance on the same phonological targets.
The
perception task consisted of 24 forced-choice identification items. For each
item, participants listened to a recorded Mandarin syllable and selected the
corresponding option from the provided choices. The task was designed to
evaluate learners’ ability to distinguish the target phonological contrasts,
including tonal differences, aspiration contrasts, and final-vowel
distinctions.
The
production task consisted of 24 read-aloud items corresponding to the
perception stimuli. Participants were required to produce the same target
syllables orally, allowing their production accuracy to be compared with their
perceptual performance. By using parallel task designs, the study examined
whether learners who demonstrated stronger perceptual sensitivity to Mandarin
phonological contrasts also tended to achieve higher production accuracy.
The perception task, delivered through Wenjuanxing, contained 24 forced-choice identification
items recorded at a natural pace by a provincial broadcaster certified at
Putonghua Proficiency Test Level 1-B. Items were separated by 10 seconds;
replay, skipping, and repetition were disabled, and administration was
synchronously proctored via Zoom. The production task comprised 24 read-aloud
items based on the same targets. Participants read each item twice and
submitted recordings within a fixed period.
The
production recordings were segmented and denoised in Audacity before scoring.
Two trained Mandarin pronunciation instructors independently evaluated all
recordings based on a predefined scoring rubric. The rubric focused on three
target dimensions: lexical tone accuracy, aspiration contrast accuracy, and
final-vowel accuracy. Each item was coded as either accurate production (1) or
inaccurate production (0). The production score was calculated as the
percentage of correctly produced items within each domain and as an overall
accuracy score across all target items.
To
ensure the reliability of the scoring procedure, interrater agreement between
the two raters was examined using the intraclass correlation coefficient (ICC).
The results indicated excellent agreement between the raters (ICC = .91, p <
.001), suggesting high consistency in production scoring. Any disagreements
were discussed and resolved through joint review of the recordings before the
final dataset was established.
For
statistical analysis, accuracy scores were first summarized by target domain.
Because the score distributions were treated as nonparametric, Spearman’s rank
correlation analysis was conducted to examine the association between overall
perception and production performance.
To address Research Question 1, Study 1 first
examined learners’ performance across the three target pronunciation domains
and then investigated the association between pronunciation perception and
production. Descriptive accuracy patterns were summarized by target domain,
followed by a nonparametric correlation analysis to determine whether learners’
perceptual and productive abilities were systematically related.
Table 1
presents descriptive statistics for learners’ perception and production
accuracy across the three target domains. Overall, production accuracy was
descriptively higher than perception accuracy in all domains. The largest
difference was observed for finals, with production accuracy exceeding perception
accuracy by 18.5 percentage points, followed by lexical tones (13.6 percentage
points) and onset contrasts (3.5 percentage points). These descriptive patterns
suggest that learners demonstrated relatively stronger performance in producing
the target sounds than in identifying them under the current task conditions.
However, these differences should not be interpreted as evidence that
production is inherently easier than perception, because the two tasks involved
different response demands and were not designed for direct within-domain
comparison.
Table 1. Perception and
production accuracy by target domain in Study 1 (n = 40)
|
Target domain |
Perception |
Production |
||
|
|
M |
SD |
M |
SD |
|
Mandarin lexical tones |
57.0 |
18.6 |
70.6 |
16.4 |
|
Pinyin d–t onsets |
67.5 |
15.2 |
71.0 |
14.8 |
|
Finals i, a, and u |
64.0 |
17.1 |
82.5 |
13.5 |
|
Overall
accuracy |
62.8 |
13.7 |
74.7 |
12.9 |
Note. Accuracy scores
represent the percentage of correctly identified or produced target items.
Beyond
these descriptive differences, the study further examined whether learners’
perceptual and productive performances were statistically associated. Because
the score distributions were treated as nonparametric, Spearman’s rank
correlation analysis was conducted between overall perception and production
scores.
As
shown in Table 2, learners achieved a mean perception accuracy of 62.8% (SD =
13.7) and a mean production accuracy of 74.7% (SD = 12.9). Spearman’s rank
correlation analysis revealed a strong positive association between perception
and production performance (ρ = .755, p < .001, 95% CI [.580, .863]). This
finding indicates that learners who demonstrated higher sensitivity to Mandarin
phonological contrasts also tended to produce these contrasts more accurately.
Table 2. Spearman
correlation between overall perception and production scores in Study 1 (n =
40)
|
Variables |
M |
SD |
Spearman’s ρ |
|
Perception accuracy |
62.8 |
13.7 |
- |
|
Production accuracy |
74.7 |
12.9 |
.755*** |
Note. Spearman’s rho
was used because score distributions were treated as nonparametric. ***p <
.001.
The
findings provide empirical support for a close relationship between Mandarin
pronunciation perception and production. However, the correlation should be
interpreted cautiously. The observed association does not establish the causal
direction between the two abilities, nor does it indicate that improvement in
one modality will automatically result in improvement in the other. Rather, the
findings suggest that perception and production are interconnected but
partially distinct dimensions of second-language speech learning. This result
provides an empirical rationale for integrating perceptual training and
productive practice within the subsequent technology-supported
perception–production instructional model.
The two
studies were designed to address complementary questions rather than to compare
learners at the same proficiency level. Study 1 examined whether perception and
production were sufficiently associated to justify their joint inclusion in an
instructional model. The intermediate learners were able to complete parallel
perception and production tasks while still showing meaningful individual
variation. The observed association therefore provided a rationale for
coordinating the two skill dimensions, but it did not establish a
perception-to-production transfer effect or identify an optimal instructional
sequence. Study 2 extended this rationale to a less experienced learner group,
for whom explicit support in establishing perceptual categories and
articulatory routines may be particularly relevant. The different proficiency
profiles should consequently be understood as serving distinct purposes within
the two-study design, not as evidence that Study 1 directly predicts the
intervention outcomes in Study 2.
Study 2 employed a four-week
quasi-experimental classroom design to examine the implementation and
preliminary outcomes of the technology-supported perception-production
instructional model developed based on the findings of Study 1. Specifically,
the study compared learners receiving technology-enhanced perception-production
instruction with those receiving conventional teacher-led pronunciation
instruction. Because participants were assigned according to pre-existing
intact classes and no individual random assignment was conducted, the study was
classified as quasi-experimental rather than a randomized controlled experiment
(Ballance, 2024). This
design reflects authentic classroom conditions while allowing the researchers
to examine changes in pronunciation performance across different instructional
conditions. However, because participants were assigned by pre-existing intact
classes rather than through individual random assignment, observed differences
should be interpreted cautiously and should not be taken as definitive evidence
of causal effects.
A 2
(instructional group: technology-enhanced perception–production instruction vs.
conventional instruction) × 3 (test occasion: pretest, immediate posttest, and
seven-day follow-up test) mixed design was employed. The pretest was
administered before the intervention to establish baseline pronunciation
performance, the immediate posttest was conducted after the four-week
intervention to examine short-term learning gains, and the follow-up test was
administered seven days later to evaluate the retention of pronunciation
improvement.
Thirty-two international
learners of Mandarin participated in Study 2, including 16 women and 16 men,
with ages ranging from 22 to 25 years. Compared with the intermediate learners
recruited in Study 1 (N = 40), the participants in Study 2 had relatively
limited Mandarin learning experience, with approximately one year of exposure
to Mandarin. Before the intervention, none of the participants had received
systematic instruction specifically targeting Mandarin pronunciation. At the
time of data collection, they were enrolled in formal Mandarin language courses
and were receiving regular classroom-based Mandarin instruction.
The
participants were selected because learners with limited Mandarin learning experience
often encounter difficulties in establishing accurate perceptual categories and
articulatory routines for Mandarin phonological contrasts. This learner profile
was therefore considered appropriate for examining the feasibility and
pedagogical potential of the technology-supported perception–production
instructional model, as these learners had sufficient Mandarin learning
experience to participate in classroom pronunciation activities while still
demonstrating considerable potential for pronunciation development.
Participants
were assigned according to their pre-existing intact classes rather than
individual random assignment. One intact class consisting of 16 learners (8
women and 8 men) received technology-enhanced perception–production
instruction, whereas the other intact class consisting of 16 learners (8 women
and 8 men) received conventional teacher-led pronunciation instruction. Both
groups followed the same instructional schedule and targeted the same Mandarin
pronunciation features during the four-week intervention.
All
participants voluntarily participated in the study and completed the pretest,
immediate posttest, and seven-day follow-up test. They were informed of the
research procedures and provided informed consent before data collection.
Participation was voluntary and had no influence on their course grades or
academic evaluation.
Findings from Study 1,
together with the theoretical assumptions of SLM-r and previous pronunciation
training research, motivated the development of a four-phase, eight-step
technology-supported perception–production instructional
model. The model was designed to address production-heavy pronunciation
instruction in which learners repeatedly imitate sounds without sufficient
diagnostic perception training or systematic follow-up review.
Phase 1—Pre-class diagnosis. Learners (a)
watched short pronunciation videos and completed perception discrimination and
read-aloud tasks through the mini-program; the teacher then (b) reviewed the
diagnostic results and adjusted instructional sequence, practice duration, and
target difficulty accordingly.
Phase 2—Classroom instruction and practice. The teacher (c)
introduced pronunciation targets using class-level performance data, learner
recordings, and acoustic/pitch visualizations, explaining phonological
contrasts through auditory, visual, and articulatory cues. Learners then (d)
engaged in balanced perception and production activities, including
identification, discrimination, listen-and-act tasks, modeling, imitation, and
controlled production practice.
Phase 3—Immediate reassessment and feedback. Learners (e)
completed parallel post-lesson perception and production tasks; the platform
and teacher subsequently (f) provided item-level performance results,
corrective feedback, and individualized practice guidance.
Phase 4—Spaced review and follow-up. Learners (g)
completed brief perception and production activities over the following seven
days and (h) completed a follow-up test covering the same pronunciation targets
in a reshuffled order.
Figure 1. The Four-Phase, Eight-Step
Technology-Supported Perception–Production Instructional Model.
The
model conceptualizes digital tools as an infrastructure for diagnostic
assessment, formative feedback, adaptive practice, and follow-up learning
rather than as substitutes for teacher explanation or pedagogical mediation.
The same learner evidence that supports individual learning also informs
teachers’ decisions regarding subsequent instructional adjustments.
In
Study 2, Hanyin Yusheng served as the technological
infrastructure for the intervention group by supporting diagnostic assessment,
multimodal pronunciation practice, automated speech evaluation, learning
records, and follow-up review. The digital functions complemented rather than
replaced teacher explanation, feedback, and classroom interaction. Importantly,
automated speech evaluation scores were used as formative feedback indicators
rather than as independent outcome measures. The automated evaluation function
was intended to help learners identify possible pronunciation deviations and
support subsequent practice, while instructional decisions remained guided by
teacher interpretation and classroom evidence.
The same teacher taught both
groups through eight scheduled lessons over a four-week intervention period
(360 minutes in total). The instructional content, learning objectives, and
planned teaching time were held constant across groups. Target pronunciation
features included the four Mandarin lexical tones, the Pinyin d–t aspiration
contrast, the finals i, a, and u, and syllable
blending. Both groups were expected to develop declarative knowledge of the
target features and improve their accuracy in identifying and producing the
target items.
The
technology-enhanced group followed the four-phase perception–production
instructional model. Before class, learners watched pronunciation videos and
completed perception and production diagnostic tasks through the mini-program. The teacher used diagnostic results to
identify learner difficulties and adapt instructional sequencing, practice
duration, and target difficulty. Classroom activities included pronunciation
modeling, discrimination tasks, appropriate acoustic and pitch visualizations,
guided imitation, and balanced perception–production practice. After each
lesson, learners completed parallel post-lesson tasks, received item-level feedback,
and engaged in seven days of brief follow-up practice before completing the
delayed follow-up test.
The
control group received the same pronunciation targets, learning objectives,
total instructional time, and teacher. Conventional teacher-led instruction
included explicit explanation, gesture-based demonstration, exaggerated
modeling, contrastive examples, physical teaching aids, imitation, repetition,
and corrective feedback. Compared with the technology-enhanced condition, the
control condition did not involve mini-program-based diagnostics, automated
speech evaluation, or the seven-day platform-guided practice cycle.
All
participants completed the same pronunciation assessments at three time points:
pretest, immediate posttest, and seven-day follow-up test. The same target
items were used across the three assessments, with item order reshuffled in the
follow-up test to reduce potential memory effects.
A two-way mixed-design ANOVA
was conducted to examine the effects of instructional condition and test
occasion on learners’ Mandarin pronunciation performance. Instructional group
(technology-enhanced perception–production instruction vs. conventional
instruction) was treated as a between-subject factor, whereas test occasion
(pretest, immediate posttest, and seven-day follow-up test) was treated as a
within-subject factor.
Before
the analysis, descriptive statistics were calculated to summarize learners’
pronunciation accuracy across groups and test occasions. The ANOVA examined the
main effects of instructional group and test occasion, as well as the
interaction effect between group and test occasion. The interaction effect was
of particular interest because it indicated whether learners receiving
technology-enhanced perception–production instruction demonstrated a different
pattern of pronunciation development compared with those receiving conventional
instruction.
Effect
sizes were reported using partial eta squared (ηp2). When
significant effects were identified, pairwise comparisons of marginal means
were conducted to further examine differences across instructional groups and
test occasions.
To address Research Questions
2 and 3, Study 2 examined changes in learners’ composite pronunciation scores
across instructional conditions and test occasions. Descriptive statistics were
first calculated to summarize learners’ pronunciation performance across the
three assessment points, followed by a mixed-design ANOVA to examine the
effects of instructional group, test occasion, and their interaction.
As
shown in Table 3, both groups demonstrated continuous improvement across the
three assessment occasions. The technology-enhanced perception–production group
achieved higher mean scores than the control group at pretest, immediate
posttest, and follow-up. Specifically, the technology-enhanced group increased
from 2.13 (SD = 0.62) at pretest to 4.88 (SD = 0.89) at immediate posttest and
further to 5.69 (SD = 0.48) at follow-up. The control group increased from 1.56
(SD = 0.89) at pretest to 3.63 (SD = 1.26) at immediate posttest and 4.81 (SD =
0.66) at follow-up.
The
technology-enhanced group showed a gain of 2.75 points from pretest to
immediate posttest and 3.56 points from pretest to follow-up, whereas the
corresponding gains for the control group were 2.07 and 3.25 points,
respectively. Although the technology-enhanced group demonstrated higher
descriptive gains, the differences in improvement patterns were relatively
modest.
Table 3. Composite
pronunciation scores by group and test occasion
|
Group |
n |
Pretest M (SD) |
Immediate posttest M (SD) |
Follow-up M (SD) |
|
Technology-enhanced |
16 |
2.13 (0.62) |
4.88 (0.89) |
5.69 (0.48) |
|
Control |
16 |
1.56 (0.89) |
3.63 (1.26) |
4.81 (0.66) |
Note. Values are means and standard deviations of
the composite pronunciation score.
Figure 2. Mean composite pronunciation scores by
instructional group and test occasion. Error bars are omitted because standard
errors for model-estimated means were not retained in the source file.
Mixed -design ANOVA
A two-way mixed-design ANOVA
was conducted to examine whether pronunciation development differed according
to instructional condition and test occasion. Instructional group was treated
as a between-subject factor, whereas test occasion was treated as a within-subject
factor.
The
analysis revealed a significant main effect of instructional group, F(1, 30) = 21.821, p < .001, ηp2 = .421. Averaged across
the three assessment occasions, learners receiving technology-enhanced
perception–production instruction achieved significantly higher pronunciation
scores than those receiving conventional instruction.
A
significant main effect of test occasion was also observed, F(2, 60) = 160.883, p < .001, ηp² = .843, indicating that learners’
pronunciation performance improved significantly across the intervention period
and follow-up assessment.
However,
the interaction effect between instructional group and test occasion was not
statistically significant, F(2, 60) = 1.555, p = .222, ηp2 = .049.
This result indicates that although the technology-enhanced group achieved
higher overall scores, the two groups did not demonstrate significantly
different patterns of change across the three assessment occasions.
Table 4. Mixed-design
ANOVA of composite pronunciation scores
|
Effect |
df |
F |
p |
ηp² |
|
Group |
1, 30 |
21.821 |
< .001 |
.421 |
|
Test
occasion |
2, 60 |
160.883 |
< .001 |
.843 |
|
Group
× Test occasion |
2, 60 |
1.555 |
.222 |
.049 |
Given
the intact-class assignment and descriptive baseline differences, the
significant group main effect should be interpreted cautiously. Future research
should employ randomized assignment or baseline-adjusted models to provide
stronger evidence for comparative effectiveness.
Pairwise
Comparisons
Pairwise comparisons of
marginal means were conducted to further examine differences associated with
instructional group and test occasion. As shown in Table 5, the
technology-enhanced group scored significantly higher than the control group
when averaged across test occasions (mean difference = 0.896, p < .001).
Regarding
test occasion, pronunciation scores significantly increased from pretest to
immediate posttest (mean difference = −2.406, p < .001), from pretest to follow-up (mean difference = −3.406, p < .001), and from immediate posttest to
follow-up (mean difference = −1.000, p < .001). These results indicate sustained improvement across the
intervention and follow-up period.
However,
because the group-by-test occasion interaction was not significant, these
pairwise comparisons should be interpreted as evidence of overall group
differences and temporal improvement rather than evidence that the
technology-enhanced condition produced a significantly greater rate of
improvement than conventional instruction.
Table 5. Pairwise
comparisons of marginal group and test-occasion effects
|
Contrast |
Mean difference |
SE |
t |
p |
Cohen’s d |
|
Technology-enhanced
– control |
0.896 |
0.171 |
5.243 |
< .001 |
1.070 |
|
Pretest
– immediate posttest |
−2.406 |
0.209 |
−11.498 |
< .001 |
−2.875 |
|
Pretest
– follow-up |
−3.406 |
0.209 |
−16.277 |
< .001 |
−4.069 |
|
Immediate
posttest – follow-up |
−1.000 |
0.209 |
−4.779 |
< .001 |
−1.195 |
Overall,
Study 2 demonstrated that learners’ pronunciation performance improved
significantly over time and that the technology-enhanced perception–production
group achieved higher overall performance than the conventional instruction
group. Nevertheless, the nonsignificant group-by-test occasion interaction
suggests that the present findings provide evidence for the feasibility and
pedagogical potential of the technology-supported perception–production model
rather than definitive evidence of its causal superiority over conventional
instruction.
The present study provides empirical support for a close
relationship between Mandarin pronunciation perception and production. The
strong positive association identified in Study 1 is consistent with the
revised Speech Learning Model (SLM-r), which conceptualizes additional-language
phonetic development as a dynamic process shaped by experience, input
distribution, and interactions between perceptual and productive systems (Flege & Bohn, 2021). The
finding also accords with previous Mandarin pronunciation research showing that
overall identification accuracy and production performance can be associated,
although the strength of correspondence may vary across phonological features
and task conditions (Hao, 2018; Leung
& Wang, 2020). Rather than
suggesting that perception and production represent identical abilities, the
present findings indicate that they are interconnected but partially distinct
dimensions of second-language speech learning.
At the
same time, the results caution against a simple perception-first account in
which accurate perception necessarily precedes successful production. In Study
1, production accuracy was descriptively higher than perception accuracy across
all target domains. Such differences may reflect task format, response demands,
rating sensitivity, or the possibility that controlled production tasks capture
emerging articulatory routines before stable perceptual representations are
fully established. This interpretation is consistent with previous training
research demonstrating that pronunciation learning can be modality-specific and
that improvement in one channel does not automatically transfer to the other (Li &
DeKeyser, 2017). Therefore, effective
pronunciation instruction should not treat perception and production as
separate stages but should instead facilitate iterative movement between the
two: learners identify a contrast, attempt production, receive feedback, revise
their output, and return to perception with increased awareness.
The
four-phase instructional model developed in this study operationalizes such an
iterative perception–production cycle. Pre-class diagnosis makes individual
difficulties visible before instruction; classroom activities involving
discrimination, articulatory explanation, visualization, and controlled
production direct attention toward relevant phonetic features; immediate
reassessment creates a short feedback loop; and spaced follow-up practice
provides repeated exposure to the same targets. From a theoretical perspective,
the model extends SLM-r-informed pedagogy by increasing opportunities for
meaningful input, improving the quality of phonetic feedback, and allowing
learner performance patterns to inform subsequent instructional decisions.
Study 2 demonstrated that both instructional conditions
were associated with substantial improvement in Mandarin pronunciation
performance. This pattern is understandable given the focused four-week
intervention period, during which both groups received systematic instruction,
explicit modeling, repeated practice, and corrective feedback from the same
teacher. The significant main effect of test occasion therefore indicates that
learners in both instructional conditions improved over the study period and is
consistent with broader evidence that structured pronunciation instruction can
support second-language speech development and is consistent with previous
meta-analytic evidence showing that explicit pronunciation teaching can
facilitate second-language speech development (Amrate
& Tsai, 2025;
Lee et al., 2015; Saito
& Plonsky, 2019).
The
technology-enhanced group achieved a higher overall mean score across the three
assessment occasions. This pattern is consistent with previous findings
suggesting that visual feedback, automated speech evaluation, and
technology-supported practice may create additional opportunities for
pronunciation learning (Chen, 2024; Cheng et al., 2025; Ngo et
al., 2024; Xodabande et al., 2025; Zou et al., 2023a).
Several mechanisms may explain this pattern. First, diagnostic analytics may
help teachers identify learner difficulties and allocate instructional time
more efficiently. Second, the platform extends opportunities for individual
practice and feedback beyond classroom constraints. Third, the integration of
listening and speaking activities may help learners notice discrepancies
between intended phonological categories and their own productions. Finally,
predictable follow-up practice may support sustained engagement. Previous
research showing that AI-supported interaction can enhance learner engagement (Wang &
Xue, 2024), that enjoyment contributes to engagement in
robot-assisted Chinese-language learning contexts (Zhang & Yan, 2025), and that multimodal GenAI can shape engagement in Chinese-language
learning (Wang, 2026)
provides plausible motivational context, although these mechanisms were not
directly examined in the present study.
However,
the nonsignificant group-by-time interaction is a critical finding that
requires cautious interpretation. Although the technology-enhanced group
achieved higher mean scores at all three assessment occasions, the two groups
did not demonstrate statistically different developmental trajectories.
Furthermore, the technology-enhanced group showed a higher descriptive pretest
mean than the control group (2.13 vs. 1.56), suggesting that initial group
differences may have contributed to the significant main effect of
instructional group. Therefore, neither the significant group effect nor the
significant improvement over time alone can be interpreted as evidence that the
instructional model produced superior pronunciation gains.
The
most appropriate interpretation is that the technology-supported perception–production
model was feasible and pedagogically promising, and that learners using the
model achieved higher overall performance within this pilot context. However,
the present quasi-experimental design does not establish incremental causal
superiority over high-quality conventional instruction. The lack of a
significant interaction effect may be related to several factors, including the
strength of the comparison condition, limited intervention duration, relatively
small sample size, baseline differences between groups, or the sensitivity of
the composite pronunciation measure.
This
cautious conclusion remains theoretically meaningful. Technology-enhanced
pronunciation instruction may not necessarily outperform well-designed
teacher-led instruction within a short intervention period, particularly when
both conditions provide intensive practice and feedback. Previous research has
similarly shown that the effectiveness of automated speech feedback depends on
treatment duration, feedback characteristics, and learner involvement (Ngo et
al., 2024). Moreover, automated feedback should be viewed as
complementary to, rather than a replacement for, teacher and peer mediation (Dai &
Wu, 2023). In the present study, the control condition was
an active instructional condition involving modeling, gesture-based
explanation, contrastive demonstration, repetition, and correction. The
relatively similar developmental trajectories may therefore reflect the
effectiveness of both instructional approaches.
The value of the model does
not primarily come from the platform itself. Rather, it comes from how learner
evidence, automated feedback, and teacher decisions are connected during
pronunciation instruction. Hanyin Yusheng connects
multiple sources of learning evidence, including learners’ perceptual
responses, recorded productions, automated speech evaluation, and teacher
observation. These sources provide complementary rather than interchangeable
information. Automated evaluation can increase feedback frequency and practice
opportunities, but it cannot independently explain the phonetic basis of an
error or determine the most appropriate pedagogical response. Teachers remain
essential in interpreting learner performance, connecting feedback to
articulatory and acoustic features, and deciding when individual or class-level
intervention is required.
This
perspective moves beyond viewing technology as merely a delivery channel and
instead conceptualizes digital tools as mediational systems embedded within
instructional decision-making. It aligns with recent discussions of actionable
automated pronunciation feedback and learner-mediated feedback processing (Huang
& Pan, 2026; McCrocklin
& Levis, 2025), evidence that AI-mediated
interaction can support speaking development when it creates meaningful
opportunities for practice (Fathi et al., 2024), and early computer-assisted pronunciation research emphasizing the
importance of integrating technological affordances with pedagogical principles
(Neri et al., 2002).
This
perspective moves beyond viewing technology as merely a delivery channel and
instead conceptualizes digital tools as mediational systems embedded within
instructional decision-making. It aligns with previous calls for richer and
more actionable feedback mechanisms in AI-based pronunciation learning
environments (Zou et al., 2023a; Wang et al., 2025),
evidence that AI-mediated interaction can support speaking development when it
creates meaningful opportunities for practice (Fathi et al., 2024), and early computer-assisted pronunciation research emphasizing the
importance of integrating technological affordances with pedagogical principles
(Neri et al., 2002).
The
feedback ecology perspective also highlights learner variability. Recent
research indicates that learners may experience AI-enhanced learning
environments differently depending on their emotional responses, autonomy, and
engagement profiles (Liu et al., 2025). Similarly, Qin and Dong (2025) demonstrated that learners’ perceptions of AI-mediated feedback may not
always correspond directly with their actual behavioral responses. These
findings suggest that adaptive pronunciation instruction should not only
monitor whether feedback is delivered but also examine how learners interpret
and act upon feedback. In the present model, the most important affordance of
technology may therefore be its capacity to make learning patterns visible over
time and transform these patterns into actionable instructional decisions.
Multimedia
components should be understood within the same framework. Short videos, pitch
visualizations, auditory models, and learner recordings can enhance attention
to pronunciation targets only when they are pedagogically aligned. Previous
research has shown that audiovisual feedback can facilitate Mandarin tone
learning (Chen, 2024),
while learner characteristics influence how multimedia information is processed
and retained (Teng, 2022, 2023). Therefore, multimodal
resources should not be incorporated simply to increase technological
complexity; rather, each representation should serve a clear learning purpose,
such as linking pitch contours to perceptual judgments or connecting acoustic
information to production attempts.
The present study contributes
to digital pronunciation research by extending technology-supported
pronunciation instruction to Mandarin as an additional language. Much previous
computer-assisted pronunciation research has focused on English segmental features
or broad speaking outcomes. Mandarin pronunciation requires simultaneous
attention to lexical tone, aspiration contrasts, vowel quality, and
syllable-level coordination. By integrating lexical tones, the Pinyin d–t
aspiration contrast, and three finals within a unified instructional cycle, the
present study demonstrates a practical approach for coordinating segmental and
suprasegmental pronunciation development.
The
study also provides a transferable pedagogical routine for technology-enhanced
pronunciation teaching. The innovation does not depend on the specific software
architecture of Hanyin Yusheng. Instead, the
underlying principle is the alignment of diagnostic assessment, classroom
instruction, formative feedback, and delayed review. Instructors using
different digital environments, including learning management systems, mobile
applications, or institutionally approved speech tools, may adopt similar
routines by collecting brief learner responses, identifying emerging
difficulties, supporting balanced perception–production activities, and implementing follow-up practice.
More
broadly, this study suggests that the educational value of AI-supported
pronunciation learning lies not simply in automated scoring accuracy but in how
technology reorganizes relationships among learners, teachers, and feedback. By
positioning digital tools as supportive infrastructures rather than
replacements for teachers, the present model offers a theoretically grounded
and practically adaptable pathway for integrating artificial intelligence into
Mandarin pronunciation pedagogy.
First, pronunciation courses
should diagnose perception and production jointly. A learner who can imitate an
item may still fail to identify it in unfamiliar speech; one who categorizes
accurately may lack stable articulatory control. Parallel tasks help teachers
distinguish these profiles.
Second,
automated feedback should be specific, revisable, and interpreted with teacher
involvement. Learners need to know which feature deserves attention and what to
do next. A useful sequence is to identify the likely contrast, replay or
visualize a model, attempt production, inspect automated output, and seek
teacher or peer confirmation when results remain unstable.
Third,
analytics should lead to instructional decisions. Class-level error frequencies
can determine sequencing, time allocation, and grouping, while individual
records can trigger targeted rather than uniform repetition. Such
evidence-based use respects the individual variation foregrounded by SLM-r and
the heterogeneous AI-learning profiles reported by Liu et al. (2025).
Fourth,
follow-up measurement should be built into instruction. Immediate posttests may
capture short-term familiarity. Delayed or spaced assessment better indicates
whether categories remain available after a lesson. Future studies should
include novel and untrained items to test generalization rather than merely
recycling trained syllables.
This study developed and
examined a technology-supported perception–production instructional model for Mandarin pronunciation through two
connected investigations. Study 1 identified a strong positive association
between pronunciation perception and production among intermediate Mandarin
learners, providing empirical support for coordinating these two dimensions
while recognizing their partially independent developmental characteristics.
Study 2 further evaluated the classroom implementation of the model through a
quasi-experimental pilot. Learners’ pronunciation performance improved
significantly over time, and the technology-enhanced group achieved higher
overall scores than the conventional instruction group. However, the
nonsignificant group-by-time interaction indicates that the present findings do
not provide sufficient evidence that the model produced significantly greater
developmental gains than high-quality conventional instruction.
The
study contributes to technology-enhanced pronunciation research by
demonstrating how perception and production can be integrated within an
SLM-r-informed instructional framework. It further illustrates how diagnostic
listening, guided production, automated feedback, teacher interpretation,
learning analytics, and spaced review can be organized into a coherent
pronunciation learning cycle. Rather than positioning technology as a
replacement for teachers, the study highlights the value of digital tools as
mediational infrastructures that support evidence-informed instructional
decisions and sustained learner engagement.
Several limitations should be
acknowledged. First, Study 2 was based on a small intact-group
quasi-experimental design, with one intact class assigned to each condition.
Consequently, class-level characteristics were confounded with instructional
condition and cannot be separated statistically in the present design. Future
research should employ larger samples with multiple classes per condition,
randomized or matched-group designs, and baseline-adjusted analyses to
strengthen causal inference. Second, the technology-enhanced group completed
additional platform-based activities outside class, whereas the control group
did not follow an equivalent seven-day platform-guided practice cycle. Unless
out-of-class practice time was matched and recorded, the present comparison
should be understood as one between instructional packages that differed in
both technological support and opportunities for additional practice; the
contribution of the platform itself cannot therefore be isolated. Third, the
present study primarily examined trained pronunciation targets; future work
should include novel-item transfer, connected speech, and human-rated
communicative outcomes to evaluate broader learning effects. Fourth, because
the intervention combined multiple components, future studies should further
disentangle the contributions of automated feedback, adaptive instructional
planning, and spaced practice. Finally, future research should incorporate
learner affective and behavioral variables, such as engagement, anxiety, motivation,
and perceptions of AI-mediated feedback, to better understand how and for whom
technology-supported pronunciation instruction produces meaningful benefits.
The study reported in this paper represents a
contribution to “2027 Special Research Project for Building a Strong Education
Province in Henan Province: A Study on the Assessment and Promotion Strategies
of Cultural Consciousness Literacy Among Primary and Secondary School Students
in Henan Province” and “2027 Special Research Project for Building a Strong
Education Province in Henan Province: Research on Digital Collection and Living
Inheritance Mechanism of Henan Dialect Speech Resources” sponsored by Henan
Provincial Department of Education.
Informed consent to participate was obtained from all
individual participants included in the study.
The datasets generated and analysed
during the current study are available from the corresponding author on
reasonable request.
The authors declare that they have no competing
interests.
This work was supported by
North China University of Water Resources and Electric Power. The university
had no role in the design or implementation of this study. The authors are
grateful for the insightful comments offered by the editor and the anonymous
reviewers.
All listed authors materially
participated in the research and manuscript preparation and approved the final
version of the article.
Ahn, T.
Y., & Lee, S. M. (2016). User experience of a mobile speaking application
with automatic speech recognition for EFL learning. British Journal of
Educational Technology, 47(4), 778–786. https://doi.org/10.1111/bjet.12354
Aly, S.
M. A., & Alsehibany, R. A. (2025). Integrating AI-powered tools in EFL
pronunciation instruction: Effects on accuracy and L2 motivation. Computer
Assisted Language Learning. Advance online publication. https://doi.org/10.1080/09588221.2025.2534015
Amrate, M., & Tsai, P.-H. (2025). Computer-assisted
pronunciation training: A systematic review. ReCALL,
37(1), 22–42. https://doi.org/10.1017/S0958344024000181
Ballance,
O. J. (2024). Sampling and randomisation in
experimental and quasi-experimental CALL studies: Issues and recommendations
for design, reporting, review, and interpretation. ReCALL,
36(1), 58–71. https://doi.org/10.1017/S0958344023000162
Bashori, M., van Hout, R., Strik, H.,
& Cucchiarini, C. (2024). “Look, I can speak
correctly”: Learning vocabulary and pronunciation through websites equipped
with automatic speech recognition technology. Computer Assisted Language
Learning, 37(5–6), 1335–1363. https://doi.org/10.1080/09588221.2022.2080230
Chen,
M. (2024). Computer-aided feedback on the pronunciation of Mandarin Chinese
tones: Using Praat to promote multimedia foreign language learning. Computer
Assisted Language Learning, 37(3), 363–388. https://doi.org/10.1080/09588221.2022.2037652
Cheng, B., Liao, K., Xiang, Y., Zou, Y., Zhang, X., &
Zhang, Y. (2025). Development and validation of an AI-enhanced multimodal
training program: Evidence from non-native Mandarin tone learning. Computer
Assisted Language Learning, 1-25. https://doi.org/10.1080/09588221.2025.2571696
Dai, Y., & Wu, Z. (2023). Mobile-assisted pronunciation
learning with feedback from peers and/or automatic speech recognition: A
mixed-methods study. Computer Assisted Language Learning, 36(5–6),
861–884. https://doi.org/10.1080/09588221.2021.1952272
Fathi,
J., Rahimi, M., & Derakhshan, A. (2024). Improving EFL learners’ speaking
skills and willingness to communicate via artificial intelligence-mediated
interactions. System, 121, 103254. https://doi.org/10.1016/j.system.2024.103254
Flege,
J. E., & Bohn, O.-S. (2021). The revised Speech Learning Model (SLM-r). In
R. Wayland (Ed.), Second language speech learning: Theoretical and empirical
progress (pp. 3–83). Cambridge University Press. https://doi.org/10.1017/9781108886901.002
Hao,
Y.-C. (2018). Contextual effect in second language perception and production of
Mandarin tones. Speech Communication, 97, 32–42. https://doi.org/10.1016/j.specom.2017.12.015
Huang,
X., & Pan, P. (2026). From feedback appraisal to pronunciation performance:
A longitudinal study of learner responses to artificial intelligence-based
pronunciation feedback. International Journal of Applied Linguistics.
Advance online publication. https://doi.org/10.1111/ijal.70296
Lee,
J., Jang, J., & Plonsky, L. (2015). The effectiveness of second language
pronunciation instruction: A meta-analysis. Applied Linguistics, 36(3),
345–366. https://doi.org/10.1093/applin/amu040
Leung,
K. K. W., & Wang, Y. (2020). Production–perception relationship of Mandarin
tones as revealed by critical perceptual cues. The Journal of the Acoustical
Society of America, 147(4), EL301–EL306. https://doi.org/10.1121/10.0000963
Li, M.,
& DeKeyser, R. (2017). Perception practice, production practice, and
musical ability in L2 Mandarin tone-word learning. Studies in Second
Language Acquisition, 39(4), 593–620. https://doi.org/10.1017/S0272263116000358
Liu, S.
H. J. (2026). Effects of voice-based ChatGPT on Chinese-as-a-second-language
learners’ Mandarin pronunciation. International Journal of Applied
Linguistics. Advance online publication. https://doi.org/10.1111/ijal.70213
Liu,
H., Fan, J., & Xia, M. (2025). Exploring individual’s emotional and
autonomous learning profiles in AI-enhanced data-driven language learning: An
expanded SOR perspective. Learning and Individual Differences, 123,
102753. https://doi.org/10.1016/j.lindif.2025.102753
McCrocklin,
S., & Levis, J. (2025). Automatic speech recognition and pronunciation
learning. Language Teaching, 1–17. https://doi.org/10.1017/S0261444825100852
McCrocklin,
S. M. (2016). Pronunciation learner autonomy: The potential of automatic speech
recognition. System, 57, 25–42. https://doi.org/10.1016/j.system.2015.12.013
Neri,
A., Cucchiarini, C., Strik, H., & Boves, L.
(2002). The pedagogy–technology interface in computer assisted pronunciation
training. Computer Assisted Language Learning, 15(5), 441–467. https://doi.org/10.1076/call.15.5.441.13473
Ngo, T.
T.-N., Chen, H. H.-J., & Lai, K. K.-W. (2024). The effectiveness of
automatic speech recognition in ESL/EFL pronunciation: A meta-analysis. ReCALL, 36(1), 4–21. https://doi.org/10.1017/S0958344023000113
Qin,
L., & Dong, J. (2025). EFL learners’ perceptual perezhivaniya and actual writing revision behaviors mediated by GenAI: A
sociocultural theory perspective. System, 133, 103769.
https://doi.org/10.1016/j.system.2025.103769
Saito,
K., & Plonsky, L. (2019). Effects of second language pronunciation teaching
revisited: A proposed measurement framework and meta-analysis. Language
Learning, 69(3), 652–708. https://doi.org/10.1111/lang.12345
Sun, W.
(2023). The impact of automatic speech recognition technology on second
language pronunciation and speaking skills of EFL learners: A mixed methods
investigation. Frontiers in Psychology, 14, 1210187. https://doi.org/10.3389/fpsyg.2023.1210187
Teng,
M. F. (2022). Incidental L2 vocabulary learning from viewing captioned videos:
Effects of learner-related factors. System, 105, 102736. https://doi.org/10.1016/j.system.2022.102736
Teng,
M. F. (2023). The effectiveness of multimedia input on vocabulary learning and
retention. Innovation in Language Learning and Teaching, 17(3), 738–754.
https://doi.org/10.1080/17501229.2022.2131791
Wang,
Y. (2026). Exploring student engagement with multimodal generative AI in
task-based Chinese language learning. Computer Assisted Language Learning. Advance
online publication. https://doi.org/10.1080/09588221.2026.2645428
Wang, Q., & Wang, Y. (2026). Predictors of
Self‐Regulated Learning in AI‐Supported Language Education: A Social Cognitive
Theory Perspective. European Journal of Education, 61(3), e70824.
https://doi.org/10.1111/ejed.70824
Wang, Y., Derakhshan, A., & Ghasvand, F. (2025). EFL teachers’
generative artificial intelligence (GenAI) literacy: A scale development and
validation study. System, 133, 103791. https://doi.org/10.1016/j.system.2025.103791
Wang, Y., & Wang, Y. (2026b). Explaining
learning engagement in AI-assisted learning through teacher support and
achievement goals: Insights from goal-setting theory. Humanities and Social
Sciences Communications, 13(1), 1215. https://doi.org/10.1057/s41599‑026‑02872‑8
Wang,
Y., & Xue, L. (2024). Using AI-driven chatbots to foster Chinese EFL
students’ academic engagement: An intervention study. Computers in Human
Behavior, 159, 108353. https://doi.org/10.1016/j.chb.2024.108353
Wang,
Y.-H., & Young, S. S.-C. (2015). Effectiveness of feedback for enhancing
English pronunciation in an ASR-based CALL system. Journal of Computer
Assisted Learning, 31(6), 493–504. https://doi.org/10.1111/jcal.12079
Wu, J. G., Lee, S.-M.,
Zou, D., & Xian, J. (2025). A decade of innovation in language education:
Insights from eight leading CALL-related journals. Applied Language Sciences,
1(1). https://doi.org/10.65553/ALS.250102
Xodabande, I., Shiri, S., & Zohrabi, M. (2025). Exploring
the impacts of an AI-driven instructional intervention on Iranian EFL learners’
pronunciation skill development. Discover Education, 4(1), 307. https://doi.org/10.1007/s44217-025-00782-2
Zhang,
J., & Yan, Y. (2025). The power of grit and enjoyment on engagement of
students majoring in teaching Chinese to speakers of other languages (TCSOL) in
a robot-assisted language learning environment: A self-determination theory
perspective. Learning and Motivation, 92, 102185. https://doi.org/10.1016/j.lmot.2025.102185
Zou,
B., Du, Y., Wang, Z., Chen, J., & Zhang, W. (2023a). An investigation into
artificial intelligence speech evaluation programs with automatic feedback for
developing EFL learners’ speaking skills. SAGE Open, 13(3), 1–14. https://doi.org/10.1177/21582440231193818
Zou,
B., Guan, X., Shao, Y., & Chen, P. (2023b). Supporting speaking practice by
social network-based interaction in artificial intelligence-assisted language
learning. Sustainability, 15(4), 2872. https://doi.org/10.3390/su15042872
Zou,
B., Li, H., & Li, J. (2018). Exploring a curriculum app and a social
communication app for EFL learning. Computer Assisted Language Learning, 31(7),
694–713. https://doi.org/10.1080/09588221.2018.1438474
Zou,
B., Lyu, Q., Han, Y., Li, Z., & Zhang, W. (2025). Exploring students’
acceptance of an artificial intelligence speech evaluation program for EFL
speaking practice: An application of the Integrated Model of Technology
Acceptance. Computer Assisted Language Learning, 38(5–6), 1366–1391. https://doi.org/10.1080/09588221.2023.2278608
Liu Shi, Ph.D., is a Master’s
supervisor, Deputy Director and Lecturer in the Department of International
Chinese Education, School of Foreign Studies, North China University of Water
Resources and Electric Power, Zhengzhou, China. Her research interests include
second language acquisition, international Chinese education, and digital
technology-assisted language learning. Her recent works have appeared in
SSCI-indexed journals. She has led a number of
provincial- and ministerial-level research projects, including grants from the
Center for Language Education and Cooperation of the Ministry of Education and
the Henan Provincial Philosophy and Social Science Planning Project. She works
as an anonymous reviewer for several international journals, such as
Learning and Motivation, SAGE Open, and European
Journal of Education.
E-mail: Shiliu910803@ncwu.cn
ORCID: 0009-0008-4361-0198
Yaping Yan, Ph.D., is an Associate Professor of Applied
Linguistics. She received her master’s and doctoral degrees from the Department
of Chinese Language and Literature, Fudan University, Shanghai, China. Her
research interests include modern Chinese grammar, rhetoric, and international
Chinese education. She has published more than 60 papers in CSSCI source
journals, including Journal of World Chinese Language Teaching,
Language Teaching and Linguistic Studies, Chinese Journal, Studies in the
Chinese Language, Contemporary Rhetoric, Language Research, and Chinese Language
Learning, alongside numerous
university journals. In recent years, she has published three academic
monographs and co-authored two books. She has presided over several
national-level research projects in China and received multiple national
academic achievement awards.
E-mail:
yanyaping@ncwu.edu.cn
ORCID: 0009-0005-5207-5252
Jingjing Xing, Ph.D. candidate and Research Assistant in the
Department of Curriculum, Teaching and Learning, University of Toronto, Canada.
She holds an MA in Language and Literacy Education, an MSc in Intercultural
Communication from the University of Warwick, and a BA (Hons) in International
Communication Studies from the University of Nottingham Ningbo China. Her
research focuses on critical multilingual pedagogy and social justice in ESL
education. She has published in peer-reviewed journals and presented at
conferences including the Conference on College Composition and Communication.
Prior to her doctoral studies, she worked as a university lecturer in China,
where she led provincial-level research projects and received excellence awards
for teaching and research.
E-mail:
jingjing.xing@mail.utoronto.ca
ORCID: 0009-0009-0414-2974
Jie Zhang, Ph.D., is a senior lecturer at the School of
Foreign Studies, North China University of Water Resources and Electric Power,
Zhengzhou, China. She has long been devoted to teaching Chinese as a foreign
language and has undertaken Chinese language teaching assignments in India,
Thailand and South Korea. Currently, she serves as Head of the Department of
Chinese as a Foreign Language and has attained fruitful achievements in the
field of teaching Chinese as a foreign language. Her publications on language teaching
pedagogy, literary criticism, community education and other related themes have
appeared in Chinese core journals. She has presided over several
provincial-level research projects. Her research interests include
international Chinese language teaching, language teaching pedagogy, teacher
education, cross-cultural interpersonal factors in educational psychology, and
intercultural communication.
E-mail:
Zhangjie1@ncwu.edu.cn
ORCID: 0009-0002-6915-4924