International Journal of Language Learning and
Teaching Research
IJLLTR
Volume 1 (2026)
https://doi.org/10.65553/IJLLTR.260105
Computerized Dynamic
Assessment and Its Effects on Second Language Production: A Meta-Analytic
Perspective
Han Shi Xiaolei Ma
Basic Education
College, National University of Defense Technology, 410000, China
Corresponding author: maxiaolei@nudt.edu.cn
Abstract
A
substantial body of research in the field of second language acquisition has
demonstrated that computerized dynamic assessment (C-DA) has positive effects
on learners’ second language (L2) proficiency. However, a comprehensive
research synthesis concerning the overall effects of C-DA on L2 production
remains limited. To address this gap, the present study employs meta-analytic
methods and incorporates data from L2 studies with 889 participants to quantify
its effects on L2 production. The main findings are as follows: (1) C-DA
enhances learners’ writing and speaking abilities, with a particularly
significant positive effect on oral production; (2) C-DA has more pronounced
effects on the fluency and accuracy of both written and oral output; and (3)
although intervention duration, learners’ L2 proficiency, and educational level
do not significantly moderate the overall effects, all subgroups show certain
positive intervention effects, with more pronounced effects observed for
short-term interventions, intermediate-to-upper-intermediate learners, and
university learners. Drawing on the synthesis of empirical research results,
this study intends intends to enrich the understanding of
C-DA and L2 acquisition, eventually offering implications for L2 writing and
speaking instruction.
Keywords:
computerized
dynamic assessment; second language production; L2 writing; L2 speaking;
meta-analysis
Dynamic Assessment (DA) originates from the concept of the Zone of Proximal Development (ZPD) in Vygotsky’s (1978) sociocultural theory. Unlike traditional assessment approaches that primarily focus on learners’ current performance, DA underscores the mediated interaction among the assessor, the learner, and the assessment task, integrating personalized intervention with assessment (Feuerstein et al., 1988). It intends to overcome the limitations of the separation between teaching and assessment, attempting to stimulate the learners’ potential. With the rapid development of information technology, the affordance of digital technology in second language (L2) education has profoundly reshaped traditional pedagogical practices. Therefore, integrating digital technologies with dynamic assessment has become an inevitable trend, leading to the emergence of Computerized Dynamic Assessment (C-DA). C-DA refers to a form of immediate dynamic assessment conducted through a computer platform, which employs algorithm-based models to produce immediate, dynamic, and individualized diagnostic feedback (Poehner & Lantolf, 2013). Furthermore, the defining feature of C-DA lies in the capacity to provide continuous mediation that responds to learners’ difficulties and is aligned with their ZPD. Through analyzing learners’ real-time online performance, C-DA flexibly adjusts the intervention strategies and enhances the standardization of mediation thereby probing L2 learner’s potential to surpass existing abilities (Lu et al., 2026).
While research on C-DA in L2
learning has expanded rapidly, many studies predominantly focus on its
effectiveness in fostering learners’ language proficiency. Early studies
examined not only broad language skills, including listening, speaking,
reading, writing, translation, and grammar (Mardian
& Nafissi, 2022; Meng et
al., 2024; Pileh
Roud et al., 2021; Yang
& Qian, 2020), but also more
specific areas, such as paraphrasing competence and the comprehension of
conversational implicatures (Geng et al., 2022;
Qian et al., 2023; Qin
& van Compernolle, 2021). Recent
studies have shifted their focus toward examining the intervention effects of
different forms of dynamic assessment, particularly comparisons between C-DA
and interactive dynamic assessment approaches, such as group and rater mediated
dynamic assessment (Ghahderijani et al., 2021; Sherkuziyeva et al., 2023). More importantly, a review of the C-DA literature reveals
numerous empirical investigations targeting productive skills, particularly
writing and speaking, with substantial variation among them (Ebadi & Vakili, 2019; Jiang, 2026; Li et
al., 2020; Sherkuziyeva
et al., 2023; Xiao, 2024). Meta-analysis, a quantitative research methodology,
enables researchers to synthesize
findings from multiple empirical studies, calculate overall effect sizes, and
identify potential moderators influencing research outcomes (Cai, 2012). Given that an existing
meta-analysis has already established the overall positive effects of C-DA on
broad measures of L2 performance (Lu et al., 2026), the present study specifically explores its effects on
the written and oral production of English as a foreign language (EFL) learners. By adopting a meta-analytic approach and
integrating effect sizes from previous studies, this study is expected to yield
empirical evidence that inspire educators in designing applications for C-DA
and even artificial intelligence-enhanced dynamic assessment in L2 education.
C-DA can be defined as an immediate assessment approach
that relies on digital platforms to evaluate learners’ current performance and
provide standardized mediation (Geng et al., 2022; Tai & Wang, 2022). C-DA was subsequently developed as a
technology-mediated extension of DA (Lu et al., 2026). DA is fundamentally grounded in Vygotsky’s (1978) sociocultural theory (SCT). The notion of ZPD under the
framework of SCT insists that learners’ emerging abilities can be identified
through their performance under appropriate assistance, combining educational
psychology theories and advanced information technology. C-DA delivers
mediation through computerized procedures (Poehner & Lantolf, 2013). It employs computer-supported adaptive feedback and
analytical techniques to overcome the limitations of traditional static
assessments, ultimately aiming to predict and promote learners’ potential
development. Feuerstein et al.’s (1988) concept of mediated learning experience (MLE) further
highlights the role of mediation in cognitive development. From the perspective
of C-DA, learner development depends not only on exposure to a task but also on
sufficient assistance. In a computerized environment, the digital platform is
expected to perform some functions conventionally undertaken by a teacher or
assessor, such as diagnosing errors promptly and offering adaptive
intervention. Nevertheless, computer delivery alone does not guarantee
meaningful mediation. Through such mediation, learners strive to overcome
individual cognitive constraints and improve their proficiency.
Dynamic assessment interventions
can generally be categorized into two major types: interactionist and
interventionist DA (Lantolf & Poehner, 2004). Existing research has primarily adopted the
interventionist approach, because pre-designed mediation is easier to
standardize and deliver to large groups of learners efficiently. Based on when
mediation is offered, Sternberg and Grigorenko (2002) proposed two general DA formats: the “cake” format and
the “sandwich” format. The sandwich format consists of three sequential stages:
pre-test, intervention, and post-test, requiring strict experimental control
and systematic intervention design. In the cake format, mediation is embedded
within the assessment sequence. This model was proposed by Guthke and Beckmann
(2000), which intended to simulate
children’s language development with computer assistance. It advocates
providing graduated prompts according to learners’ ongoing performance, marking
an initial attempt of technology-enhanced dynamic assessment.
Research on C-DA has gradually
shifted from cognitive and educational psychology to second language
acquisition. Correspondingly, C-DA research has progressed from theoretical
exploration to practical applications, with increasingly refined research objectives,
more targeted mediation procedures, and diversified outcome analyses. The majority of C-DA studies have focused on discrete
linguistic items (e.g., specific grammatical structures) or broad language
skills (e.g., listening and reading) (Jiang,
2026). Existing studies have examined
language proficiency development in specific linguistic dimensions, including
paraphrasing, grammatical accuracy, and discourse coherence (Mardian
& Nafissi, 2022; Qian et
al., 2023; Yang & Qian, 2020). Moreover, researchers have conducted comparative
studies by examining C-DA alongside other forms of dynamic assessment (e.g.
group dynamic assessment), highlighting the relative advantages of standardized
computer mediation and flexible interpersonal mediation. (Feng
& Zou, 2026; Ren,
2019; Sherkuziyeva et al., 2023). Furthermore, researchers have increasingly
incorporated multidimensional indicators to provide stronger evidence for the
effectiveness of C-DA, among which complexity, accuracy, and fluency (CAF)
measures and cognitive load assessments are particularly representative (Babamoradi et al., 2018;
Ghahderijani et al., 2021; Imsa-ard & Barrot, 2026; Jiang, 2026).
Research on C-DA in L2 writing and
speaking tasks has been relatively more extensive (Feng
& Zou, 2026; Ghahderijani
et al., 2021; Jalambo
et al., 2025; Li et al., 2020).
In terms of L2 writing, researchers have mainly focused on investigating the
effects of C-DA on EFL writing proficiency and different forms of feedback
provision (Xu, 2022; Yang
& Qian, 2020). Li (2015) conducted a mixed-methods study combining pre-test,
post-test, and control group designs to compare different English writing
feedback approaches based on dynamic assessment theory. He found that
computer-assisted monitoring and individualized feedback supported improvements
in L2 writing. This aligns with recent empirical evidence demonstrating that
tailoring written feedback to fit learners’ individual characteristics, such as
cognitive learning styles, significantly enhances EFL writing outcomes (Lu & Pan, 2026). With the advancement of research, scholars have
extended the application of C-DA to more specialized writing contexts,
including academic writing (Ebadi & Vakili, 2019) and email requests (Nicholas et al., 2026). Comparing teacher-student face-to-face dynamic
assessment with C-DA, Ebadi and Vakili (2019) concluded
that C-DA not only facilitated the identification of learners’ developmental
errors but also encouraged autonomous correction. These preceding findings
suggest that the intervention effectiveness of C-DA can be extended to various
functional text types. The emergence of generative artificial intelligence (AI)
has created new possibilities for more adaptive forms of computer mediation.
Shen et al. (2026) creatively
compared chatbot-assisted dynamic assessment (CA-DA) and chatbot-assisted
non-dynamic assessment (CA-NDA). This study emphasizes that the CA-DA group
achieved comparable writing performance to both the CA-NDA and the comparison
group. Nevertheless, it should be highlighted that the CA-DA approach yielded
additional benefits in helping learners generate their ideas, developing their “maturing”
ideas, and emulating an authentic communication environment. Hence, the
contribution of CA-DA may also become visible in developing learners’ writing
ability.
Meanwhile, the application of C-DA
in L2 speaking instruction has also become a prominent research focus in this
field. The real-time corrective feedback mechanism of C-DA can improve the
efficiency of traditional speaking instruction by rapidly and accurately
identifying pronunciation deviations, grammatical errors, and pragmatic
inadequacies in learners’ oral production. C-DA generates individualized
assessments and adaptive training programs tailored to learners’ specific
needs. It not only reduces learners’ anxiety during oral communication but also
provides targeted support for enhancing their oral comprehension and production
abilities (Ghahderijani et al., 2021). Sherkuziyeva et
al. (2023) conducted a strictly designed control experiment
comparing C-DA with traditional instructional assessment approaches to track
the distinguished improvements of learners’ speaking complexity, accuracy, and
fluency. The findings demonstrated that learners receiving C-DA intervention
achieved significant improvements in speaking fluency and logical coherence,
further highlighting the advantages of the well-designed application of
technology-enhanced dynamic assessment.
Existing research has consistently
demonstrated that C-DA contributes to influencing learners’ L2 proficiency.
However, findings remain inconsistent and fragmented in individual studies due
to heterogeneity in mediation design, assessment-related factors, learner
populations, and study quality. For instance, Ren (2019) found that C-DA significantly improved learners’
writing performance in terms of lexical density, lexical sophistication, and
accuracy, whereas no significant effects were observed for lexical diversity,
syntactic complexity, or fluency. In contrast, Sherkuziyeva
et al. (2023) reported
significant differences between the experimental and control groups in terms of
post-test complexity, with the experimental group outperforming the control
group in L2 production complexity. To synthesize existing empirical evidence,
the present study introduces meta-analysis as a statistical approach. Compared
with traditional narrative reviews, meta-analysis represents a quantitative
secondary analysis of complicated research findings (Shen & Zhang, 2020). Meta-analysis enables researchers to systematically
account for sampling errors and statistically non-significant findings (Plonsky
& Oswald, 2015), quantify
heterogeneity across studies, and identify potential sources contributing to
variations in effect sizes. In recent years, this methodology has been widely
adopted in the field of second language acquisition (Cai, 2012; Shen et al., 2025;
Wang et al., 2024). Lu et al. (2026) synthesized 35 studies using a three-level
meta-analytic model and reported large positive effects for both cake-format
and sandwich-format C-DA on L2 performance. However, their assessment context
covers comprehensive L2 performance, covering grammar, listening and
pragmatics, etc. The effectiveness of C-DA on productive skills such as
speaking and writing was not separately measured by employing the meta-analytic
approach.
This meta-analysis is a
fine-grained investigation on the effects of C-DA in promoting L2 learners’
production. Specifically, it aims to synthesize existing empirical research,
calculate the overall pooled effect size, conduct subgroup analyses and
moderator analyses, eventually providing a comprehensive evaluation and
offering practical guidance for future L2 teaching practices. The present study
addresses the following research questions:
RQ1. What is the overall effect of C-DA on L2 production?
RQ2. How do the effects of C-DA vary across production
types and measurement indicators?
RQ3. Do research duration, learners’ L2 proficiency
levels, and instructional contexts moderate the effects of C-DA on L2
production?
To ensure a comprehensive and precise
literature search, a combination of various methods was employed, including
systematic searches of authoritative databases, targeted manual searches of
core journals identified through expert consultation and their recurrent
citation in C-DA studies (Lu et al., 2026). English-language studies were retrieved from the Web
of Science, the Education Resources Information Center (ERIC), APA PsycInfo, and ProQuest Dissertations and Theses. The
Boolean search string was utilized: TS = ((“computerized dynamic assessment” OR
“computer-mediated dynamic assessment” OR “mobile-mediated dynamic assessment”)
AND (“second language production” OR performance OR “L2 writing” OR “EFL
writing” OR “L2 speaking” OR “EFL speaking” OR complexity OR accuracy OR
fluency)). Meanwhile, Chinese-language studies were retrieved from the China
National Knowledge Infrastructure (CNKI). The search terms included “计算机动态评估” (computerized dynamic assessment), “动态评估” (dynamic assessment), “二语产出”
(L2 production), “二语写作” (L2 writing), “英语写作” (EFL writing), “二语口语” (L2
speaking), and “英语口语” (EFL speaking). The reference
lists of relevant published studies were manually searched to ensure the
inclusion of pertinent and eligible studies that might have been missed in the
earlier database searches. A total of 855 records from academic databases were
identified.
The inclusion criteria for the meta-analysis
were as follows:
(1) Studies investigated the
effects of C-DA on L2 production.
(2) Studies were original
empirical research.
(3) Studies involved at least one
task modality, either speaking or writing.
(4) Studies reported at least one
dependent variable measuring EFL learners’ L2 production.
(5) Experimental design included a
control group.
(6) Studies provided sufficient
statistical data (e.g., sample size, mean, and standard deviation) to calculate
effect sizes.
(7) Studies were published between
2014 and 2026. The year 2014 marked a turning point in the rapid development of
educational technology (EdTech), providing the technical foundation for
empirical studies on C-DA.
Figure 1. PRISMA flow diagram
Figure 1 presents a flowchart
detailing the steps of literature search, screening and inclusion. The whole
coding procedure followed the PRISMA guidelines to ensure transparency.
Initially, a total of 855 records were identified through comprehensive and
thorough searching. Subsequently, all retrieved records were imported into
EndNote for duplicate removal and abstract screening, leaving 72 unique records
for further screening. Then, a detailed full-text review based on inclusion
criteria was conducted, leading to the exclusion of 21 records. After excluding
original studies that failed to meet these criteria, 12 primary studies were
ultimately selected as the final articles for this meta-analysis, involving 889
participants.
The coding scheme was developed based on our
research purpose and the meta-analysis guidelines for L2 research proposed by
Plonsky and Oswald (2015), which covered four main
categories: (1) bibliographic information: author, year of publication and
journal; (2) sample characteristics: sample size (for both experimental and
control groups) and participant demographics (e.g., age, academic status, native
language, and L2 proficiency); (3) methodological features: research design
(e.g., C-DA implementation methods, intervention duration, and control group
setup), task types (e.g., speaking or writing tests), and L2 production
measures (e.g., complexity, accuracy, fluency, and overall scores); and (4)
results: descriptive statistics and specific effect sizes.
Regarding the coding process, to
ensure accuracy, two researchers coded the data independently. A high level of
agreement was achieved between the independent coders (Cohen’s κ = 0.91). Any
discrepancies during the coding process were resolved and a final decision was reached through sufficient discussion.
This study conducted a meta-analysis using
Comprehensive Meta-Analysis 3.0 (CMA). First, effect sizes were extracted by
entering the sample sizes, means, standard deviations, and other relevant data
into CMA. Hedges’s g was chosen as the effect size measure due to the small
sample sizes present in this analysis (Lu et al., 2026). Following Cohen’s (1988) benchmarks, values of 0.2, 0.5, and 0.8 were respectively viewed as
the thresholds for small, medium, and large effects. The next step was to
calculate the combined effect size and its confidence interval, so as to estimate the overall effect of C-DA on L2
production. Publication bias was subsequently assessed by conducting a
comprehensive assessment using a funnel plot, the trim-and-fill procedure, Begg’s
rank correlation test, and Egger’s regression test. Heterogeneity was also
examined for the overall analysis and within subgroups. The results of
heterogeneity tests determined the selection of appropriate meta-analytic
models to compute pooled effect sizes. Finally, subgroup and moderator analyses
were conducted to explore potential sources of variation across studies.
This study combined the funnel
plot with statistical tests to examine the risk of publication bias from
multiple perspectives. Figure 2 shows that the distribution of effect sizes was
not perfectly symmetrical, suggesting that there is a risk of publication bias.
The funnel plot indicated a potential threat to the validity of meta-analytic
findings. Three complementary statistical procedures were also conducted. Duval-Tweedie
trim-and-fill iterative analysis demonstrated that the effect sizes before and
after adjustment were identical, so that no missing studies needed to be
imputed and no substantial publication bias was detected. Meanwhile, Begg’s
rank correlation test showed that Kendall’s tau without continuity correction
was 0.333 (p = 0.131). After continuity correction, the tau decreased to 0.318 (p = 0.150). Neither result
reached statistical significance (p > 0.05), which showed a relatively balanced
distribution of effect sizes and no significant association between effect size
and standard error. Accordingly, Begg’s test provided no clear evidence of
publication bias. In contrast, the result of Egger’s regression test (p = 0.009) indicated possible
publication bias.
Egger’s
regression test yielded the evidence of possible publication bias, whereas the
trim-and-fill procedure and Begg’s rank correlation test did not support this
finding. Rosenthal’s fail-safe N, as a robustness check, was conducted and
further showed that 358 additional studies with null results would be required
to render the overall effect statistically nonsignificant. This number
substantially exceeded the conventional threshold of 5k + 10 = 70 (k = 12),
suggesting that publication bias is unlikely to have had a substantial
influence on the meta-analysis. Taken together, these findings suggest that the
overall conclusions were relatively reliable and stable.
Figure 2. Funnel plot of Hedges’s g
The random-effects model was selected for the
meta-analysis. The random-effects model allows the true effects to vary across
studies and preserves more data in statistical estimation process, in contrast
to fixed-effects models that assume homogeneity across the selected studies and
samples (Borenstein et al., 2010). To evaluate the overall effect of
C-DA on L2 production and examine its heterogeneity, a forest plot and
effect-size analysis were generated. As shown in Figures 3 and 4, the pooled
effect size was Hedges’ g = 0.837 (95% CI [0.462, 1.212], p < 0.001), suggesting that
computerized dynamic assessment exerts a substantial positive effect on L2
production. The heterogeneity test reflects varying degrees of heterogeneity
among the independent samples (Shen et al., 2025). The analysis showed substantial
heterogeneity (I2 = 86.020%), reflecting that most of the
variability across effect estimates was due to between-study heterogeneity
rather than sampling error. In the context of this substantial heterogeneity,
the study continued to adopt random-effects models to conduct subsequent
subgroup analyses and moderator-effect tests for the sake of disclosing
potential sources of heterogeneity.
Figure 3. Forest plot of overall effect
size
Figure 4. Results of effect size analysis
Effect sizes among
different output types
This meta-analysis synthesized 12 empirical
studies, one of which analyzed the effects of C-DA on both written and oral
production. To compare the differential effects of C-DA on these two output
modes, this study conducted a subgroup analysis. As illustrated in Table 1,
C-DA had a positive effect on L2 writing production, with a pooled effect size
of Hedges’ g = 0.634 (95% CI [0.334, 0.935], p < 0.001), while the effect size
for oral output was larger (g = 2.257, 95% CI [0.752, 3.762], p = 0.003). However, the I2 values of
both subgroups exceeding 75% revealed the substantial within-group
heterogeneity. In addition, the confidence interval for the speaking subgroup
was considerably wider than that for the writing subgroup. This wider interval
represented greater uncertainty surrounding the estimated effect on speaking,
which may be attributed to the small sample sizes.
Table 1. Subgroup analysis
for different output types
|
L2
production |
k |
Sample |
g |
p |
95% CI |
Heterogeneity |
||
|
Lower |
Upper |
Q |
I² |
|||||
|
Writing |
10 |
764 |
0.634 |
<0.001 |
0.334 |
0.935 |
37.085 |
75.732 |
|
Speaking |
3 |
189 |
2.257 |
0.003 |
0.752 |
3.762 |
32.465 |
93.840 |
Effect sizes among
different measurement indicators
Among the empirical studies included in this
meta-analysis, three studies simultaneously employed multiple measurement
indicators in their research design. Table 2 indicates that the effects of all
four outcome measurement indicators were statistically significant (p < 0.05). Meanwhile, C-DA exerted
differential effects on different measurement indicators. Its positive effects
were more pronounced for fluency (g = 2.440, 95% CI [1.974, 2.905], p < 0.001) and accuracy (g =1.803,
95% CI [0.515, 3.091], p = 0.006), their effect sizes substantially
exceeding those for overall score (g = 0.983) and complexity (g = 1.402). The
overall score group (I² = 90.461%, Q = 94.346), the complexity group (I2 =
94.657%, Q = 56.145), and the accuracy group (I2 = 92.288%, Q =
25.935) all showed high I2 values. This high level of heterogeneity may arise
from pronounced differences in sample characteristics (e.g., L2 proficiency),
or in measurement instruments. By contrast, heterogeneity in the fluency group
was close to zero (I2 = 0.000%, Q = 0.473), implying a high degree
of consistency among the effect sizes. Nevertheless, this result warrants
cautious interpretation, given the limited sample size in this subgroup.
Table 2. Subgroup analysis
for different measurement indicators
|
Measurement
indicators |
k |
Sample |
g |
p |
95% CI |
Heterogeneity |
||
|
Lower |
Upper |
Q |
I² |
|||||
|
Overall score |
10 |
754 |
0.983 |
<0.001 |
0.482 |
1.484 |
94.346 |
90.461 |
|
Complexity |
4 |
298 |
1.402 |
0.015 |
0.272 |
2.532 |
56.145 |
94.657 |
|
Fluency |
2 |
124 |
2.440 |
<0.001 |
1.974 |
2.905 |
0.473 |
0.000 |
|
Accuracy |
3 |
177 |
1.803 |
0.006 |
0.515 |
3.091 |
25.935 |
92.288 |
Potential moderators were
identified by reviewing previous relevant meta-analyses and considering the
design and data characteristics of the included studies. Intervention duration,
learners’ L2 proficiency level, and instructional setting were selected as the
three main moderators and moderator analyses were conducted using a random-effects
model to evaluate the corresponding effect sizes. Table 3 presents the results.
Intervention
duration was the first moderator to be examined. Two categories under
Intervention duration were classified, including interventions lasting less
than 15 weeks and interventions lasting more than 15 weeks. Studies with
interventions lasting less than 15 weeks produced a statistically significant
positive pooled effect of C-DA on L2 production (g = 1.014, 95% CI [0.537,
1.492], p < 0.001). For
interventions lasting more than 15 weeks, the pooled effect was moderate and
did not reach statistical significance (g = 0.502, 95% CI [−0.162, 1.165], p = 0.138). The results demonstrated that short-term
interventions significantly promoted L2 production. However, the between-group
difference was not statistically significant (Qbetween
= 1.513, p = 0.219). The current
evidence is insufficient to support intervention duration as a significant
moderator of the effects of C-DA on L2 production.
Secondly,
differences in effect-size estimates were observed across L2 proficiency
levels. In the present study, participants’ L2 proficiency was coded according
to the Common European Framework of
Reference for Languages: Learning, Teaching, Assessment (CEFR) (Council of
Europe, 2020). As some Chinese studies employed locally defined proficiency
indicators in the Chinese educational context, this meta-analysis uniformly
mapped these Chinese academic indicators onto the CEFR framework for reporting
participants’ proficiency, ensuring data comparability. Two researchers
independently completed the coding with a high level of inter-rater agreement.
The statistical results illustrated that L2 learners at the B1 level (g =
0.911, 95% CI [0.343, 1.479], p = 0.002) and B2
level (g = 1.635, 95% CI [0.635, 2.636], p = 0.001), namely intermediate and upper-intermediate learners, had
significant improvement in L2 production. The positive effects observed for
these two proficiency levels were substantially larger than those for the A2
and C1 groups. Neither the A2 nor the C1 groups exhibited a statistically
significant pooled effect. Nevertheless, the between-group difference in effect
sizes across proficiency levels was not statistically significant (Qbetween = 4.328, p = 0.228). Hence, L2 proficiency can not be
viewed as a significant moderator.
Finally,
the last examined moderator in this meta-analysis was the instructional
setting. Both the university group (g = 0.909) and the English-major group (g =
0.869) showed significant positive effects on L2 production. Effect sizes were
largest for C-DA at the university level, whereas the intervention effect for
the high school group was not statistically significant (g = 0.632, 95% CI
[-0.371, 1.636], p = 0.217). Given that the test
of between-group differences was not significant, instructional level was also
not identified as a moderator of effect-size variation (Qbetween
= 0.214, p = 0.898 > 0.05).
Table 3.
Results of moderator analysis
|
Moderator |
Type |
k |
g |
p |
95% CI |
Qb |
p(Qb) |
|
|
Lower |
Upper |
|||||||
|
Intervention duration |
≤15weeks |
8 |
1.014 |
<0.001 |
0.537 |
1.492 |
1.513 |
0.219 |
|
>15weeks |
4 |
0.502 |
0.138 |
-0.162 |
1.165 |
|||
|
L2 proficiency level |
A2 |
1 |
0.329 |
0.635 |
-1.030 |
1.688 |
4.328 |
0.228 |
|
B1 |
6 |
0.911 |
0.002 |
0.343 |
1.479 |
|||
|
B2 |
2 |
1.635 |
0.001 |
0.635 |
2.636 |
|||
|
C1 |
3 |
0.377 |
0.354 |
-0.420 |
1.174 |
|||
|
Instructional setting |
High school |
2 |
0.632 |
0.217 |
-0.371 |
1.636 |
0.214 |
0.898 |
|
University |
5 |
0.909 |
0.006 |
0.261 |
1.556 |
|||
|
University |
5 |
0.869 |
0.009 |
0.214 |
1.523 |
|||
This meta-analysis synthesized effect sizes from 12
experimental studies. The results evidently reveal that C-DA significantly
influenced learners’ written and oral L2 production. It is also broadly
consistent with previous research, including Sherkuziyeva
et al. (2023). This finding supports the view that C-DA enhances
learners’ task performance through strategic support when learners encounter
difficulties, enabling them to apply strategies or knowledge that they could
not deploy independently (Lu et al., 2026).
Previous studies demonstrate that this approach may compensate for limitations
in teacher-led assessment, such as limited time and delayed feedback. By
providing structured assistance, C-DA, as a pedagogical strategy, can
effectively activate the ZPD. Specifically, learners overcome the constraints
of their current competence and gradually internalize the linguistic and
cognitive resources required for more demanding activities (Wang,
2000). However, the substantial
heterogeneity observed reflects that the estimated effects varied considerably.
Such high heterogeneity indicates that even though C-DA effectiveness was
generally robust across learners, the overall effects as a form of mediation in
cognitive development are likely to be influenced by specific assessment
requirements and instructional contexts.
The empirical studies coded in this meta-analysis employed diverse experimental
designs.
Researchers evaluated L2
production using multidimensional measures, including overall scores as well as complexity, accuracy, and fluency (CAF)
indices, across different production types, namely writing and speaking. To address whether the effects of C-DA varied across
these dimensions, two sets of subgroup analyses were conducted:
The
first subgroup is written and oral production. Under the random-effects model,
the speaking subgroup produced a larger pooled effect-size estimate than the
writing subgroup. However, this difference should be interpreted with caution
because the speaking subgroup contained relatively few studies. The automated
corrective feedback provided by C-DA is immediate and targeted (Li et al.,
2020), enabling L2 learners to draw on models and
independently address pronunciation problems, grammatical errors, and pragmatic
strategies within a relatively short period. These features may drive the
immediate improvements in oral performance. Written production, by contrast,
requires not only the correction and supplementation of linguistic knowledge,
but also sustained cognitive processing to develop more elaborate content and
rigorous logic. Xiao (2024) reported that computerized mediation was associated with reductions in
technical, lexical, and discourse-level errors, whereas improvements in overall
textual organization were less evident. Similarly, in Ren’s (2019) intervention study, C-DA produced no significant improvements in lexical
diversity, syntactic complexity, or fluency. Sherkuziyeva
et al. (2023) disagree with the above viewpoints. They reported a significant posttest
difference in writing complexity between the experimental and control groups,
with the experimental group performing better. Overall, the available evidence
suggests that C-DA does not affect all dimensions of written production
equally. Such variability may stem from the fact that writing tasks require the
engagement of deeper cognitive structures (Xue & Pang, 2000). Meanwhile, the influence of C-DA on higher-order cognitive functions may
require longer periods of mediation to make intervention effect become evident.
Consequently, the effects of C-DA on writing have not yet demonstrated
consistent patterns across studies, with the pooled effect remaining moderate.
The
second subgroup concerns measurement indicators. C-DA produced positive effects
across the examined measures of L2 production. The effects on fluency and
accuracy were numerically larger than those on overall performance scores and
complexity. As complexity, accuracy, and fluency interact dynamically and
follow different developmental trajectories (Norris & Ortega, 2009), these three measures are not necessarily enhanced to the same extent.
But accuracy and fluency can develop concurrently under some conditions (Yu, 2020). The data-analytic function of C-DA facilitates the statistical analysis
of linguistic features in discourse, detecting spelling errors, grammatical and
syntactic problems, and the frequency of cohesive devices. Its automated
corrective feedback can therefore provide immediate and clear evaluations that
lead learners to consolidate their linguistic knowledge and refine their
expressions. Thus, C-DA can exert strong positive effects on fluency and
accuracy. Overall scores and complexity are broader constructs influenced by
multiple factors. Consequently, the cumulative effects on these measures
require a longer period to emerge.
The between-group heterogeneity tests reveal that intervention duration, L2 proficiency level, and instructional setting did not
significantly moderate the effects of C-DA on L2 production. Although no significant between-group differences were
identified, the subgroup analyses still revealed meaningful intervention
effects within individual subgroups.
First,
studies with interventions lasting less than 15 weeks yielded a relatively large pooled effect size for C-DA, whereas the effects
reported in studies lasting more than 15 weeks did not reach statistical
significance. Effects of intervention duration on L2 production may be
nonlinear. Cognitive load theory provides a plausible explanation for this
finding. As a key component of instructional design, intervention duration, if
inappropriately determined, may impose excessive extraneous cognitive load on
learners (Paas et al., 2003).
Well-designed short-term interventions help minimize such cognitive load while
using the immediate feedback afforded by C-DA to increase learners’ interest
and self-efficacy.
Second,
the intervention effects were statistically significant among learners at the
B1-B2 levels (intermediate to upper-intermediate), with effect sizes exceeding
those observed among learners at the C1 (advanced) and A2 (elementary) levels.
This result is consistent with the meta-analytic findings of Shen et al. (2019). They reported that intermediate-level learners are more responsive to
instructional interventions.
Learners at the C1 level may already possess high L2 proficiency, so their room
for further improvement becomes limited, making it difficult for C-DA
interventions to overcome the “ceiling” of their language development. Learners
at the elementary level represent the other extreme. On
account of their inadequate linguistic knowledge, C-DA may
primarily address gaps in basic language knowledge rather than facilitate
deeper cognitive processing. Learners at the B1-B2 levels demonstrate greater
developmental plasticity and are better positioned to internalize the benefits
of the intervention.
Third,
the university group showed the largest pooled effect size, followed by the
university English-major group. The high school group also exhibited a positive
effect, although it did not reach statistical significance. This ordering
should be interpreted as a descriptive gradient based on effect sizes. The
effectiveness of C-DA may depend on learners’ autonomy and willingness to
participate in exploratory learning (Lu et al., 2026). University students generally possess higher levels of metacognitive
awareness and learner autonomy, which enables them to interpret and use the
graduated prompts and immediate feedback provided by C-DA more effectively. Indeed,
as Li and Mizumoto (2026) emphasize in the context of technology- and GenAI-mediated learning,
fostering autonomous and efficacious tool use, rather than passive reliance, is
critical for university EFL learners to maximize the developmental benefits of
automated feedback (see also Teng, 2026). Furthermore, unlike EFL teaching in high school, university courses may
have access to more favorable conditions for implementing C-DA, such as
technological facilities, online platforms, and instructional organization.
Sociocultural and pedagogical backgrounds need to be taken
into account in C-DA implementation (Lu et al., 2026). These abundant pedagogical resources and advanced teaching facilities in
the university allow their benefits to be fully leveraged. Previous studies
conducted in university settings have primarily shown that C-DA can enhance
learners’ L2 performance (Shen et al., 2026; Yang & Qian, 2020) and
may positively influence learning motivation, autonomous learning, and
self-regulation (Estrada-Araoz et al., 2023).
This meta-analysis synthesized
12 empirical studies from Chinese and international literature to investigate
the effects of C-DA on production and achieved three main findings. Firstly,
C-DA had positive effects on both written and oral production, with a
particularly large pooled effect for speaking.
Secondly, the positive effects of C-DA were more pronounced for fluency and
accuracy than other dimensions of measurement indicators. Thirdly, although
intervention duration, learners’ L2 proficiency, and educational level did not
exert significant moderating effects on L2 production, all moderator analyses
demonstrated that positive intervention effects of moderators varied in
degrees. Specifically, relatively larger effects were observed for shorter
interventions, intermediate and upper-intermediate learners, and university
students. The results highlighted that appropriately designed short-term C-DA
interventions may support the development of L2 production. Learners at
intermediate and upper-intermediate levels also showed larger effect-size
estimates than those at lower or advanced levels.
Our
findings provide reference and implications for the wider use of C-DA in L2
educational environments. They may help teachers optimize instructional design, select assessment platforms suited to their
pedagogical objectives, determine an appropriate intervention duration, and
provide mediation aligned with EFL learners’ ZPD. With the current trend toward
the increasing integration of digital technology and L2 teaching, this study is
also conducive to informing the future incorporation of AI-supported dynamic
assessment tools.
Despite
the comprehensive nature of this meta-analysis, several limitations should be
noted. First and foremost, due to the limited number of studies involved, the
educational backgrounds of learners were relatively restricted, which affects
the generalizability of the findings to other cultural and educational
contexts. In addition, some subgroups contained only a small number of studies
or effect sizes. This reduced the statistical power to detect between-group
differences and increased the uncertainty of the subgroup estimates. Therefore,
the nonsignificant moderator results should be interpreted with caution.
Finally, the analysis did not report the effects of C-DA on other L2 abilities
and did not consider other potentially important moderators.
Several possible directions
for future research can be drawn from the present study. Future meta-analyses
should include a larger number of studies and learners from more diverse
educational, linguistic, and cultural backgrounds, which contributes to increasing
the statistical power of moderator analyses and ensuring a broader
applicability of the results. Moreover, future research could also examine the
effects of C-DA on reading, listening, translation, and other dimensions of
language ability. Additionally, a wider range of potential moderators are
recommended to be considered in future meta-analyses. Such work would provide a
more comprehensive understanding of the conditions under which C-DA supports L2
development.
This research received no external funding.
Informed consent to participate was obtained from all
individual participants included in the study.
The datasets generated and analyzed during the current
study are available from the corresponding author on reasonable request.
The authors declare that they have no competing
interests.
Both authors listed in the study have materially
participated in the research and article preparation. Additionally, they have
approved the final article. SH conceived of the presented idea and carried out
the experiment. With support from MXL,
SH performed the analysis, drafted the manuscript and designed the figures.
Babamoradi, P., Nasiri, M., &
Mohammadi, E. (2018). Learners’ attitudes toward using dynamic assessment in
teaching and assessing IELTS Writing Task 1. International Journal of
Language Testing, 8(1), 1–11.
Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R.
(2010). A basic introduction to fixed-effect and random-effects models for
meta-analysis. Research Synthesis Methods, 1(2), 97–111. https://doi.org/10.1002/jrsm.12
Ebadi, S. & S. Vakili.
(2019). Exploring EFL learners’ developmental errors in academic writing
through face-to-face and computer-mediated dynamic assessment. Computer
Assisted Language Learning, 35(4). https://doi.org/10.1080/09588221.2019.1698616
Feng, S., & Zou, J. (2026). Group homogeneity enhances dynamic
assessment efficacy in L2 English genitive learning: Preliminary evidence from
integrating computerized and group DA. Language Testing in Asia, 16,
Article 27. https://doi.org/10.1186/s40468-026-00447-3
Geng, X., Meng, Y. R., & Du, W. B. (2022). Jisuanji dongtai
pinggu de eryu fazhan shanbian he yanjiu kuangjia [Research on
computerized L2 dynamic assessment: Development and framework]. Xiandai Waiyu [Modern Foreign
Languages], 45(1), 126–136.
Ghahderijani, B. H., Namaziandost,
E., Tavakoli, M., Kumar, T., & Magizov, R.
(2021). The comparative effect of group dynamic assessment (GDA) and
computerized dynamic assessment (C-DA) on Iranian upper-intermediate EFL
learners’ speaking complexity, accuracy, and fluency (CAF). Language Testing
in Asia, 11, Article 25. https://doi.org/10.1186/s40468-021-00144-3
Guthke, J., & Beckmann, J. F. (2000). The learning test concept and
its applications in practice. In C. S. Lidz & J. G. Elliott (Eds.),
Dynamic assessment: Prevailing models and applications (pp. 17–69). Elsevier Science.
Imsa-ard, P., & Barrot, J. S.
(2026). Combining multimodal technology-mediated and peer feedback: Effects on
second language (L2) learners’ complexity, accuracy, and fluency (CAF) in
writing. Computer Assisted Language Learning, 39(4), 1023–1061. https://doi.org/10.1080/09588221.2024.2443770
Jiang, Z. (2026). Scaffolding early-stage simultaneous interpreting
learning using computerized dynamic assessment. Language Learning &
Technology, 30(1), 1–25. https://doi.org/10.64152/10125/73701
Lantolf, J. P., & Poehner, M. E.
(2004). Dynamic assessment of L2 development: Bringing the past into the
future. Journal of Applied Linguistics, 1(1), 49–72. https://doi.org/10.1558/japl.1.1.49.55872
Li, O.,
& Mizumoto, A. (2026). Measuring autonomous and efficacious GenAI use and
GenAI reliance in EFL learning: Initial scale development and validation. Applied
Language Sciences, 2(1). https://doi.org/10.65553/ALS.260110
Lu, Q., Chen, M., Yang, L., Li, S., & Chen, Y. (2026). The
effectiveness of computerized dynamic assessment in improving L2 performance: A
three-level meta-analysis. ReCALL. Advance
online publication. https://doi.org/10.1017/S0958344026100561
Lu, X.,
& Pan, L. (2026). Tailoring written feedback to fit: The role of cognitive
learning styles in improving EFL writing with feedback. Applied Language
Sciences, 2(1). https://doi.org/10.65553/ALS.260112
Mardian, F., & Nafissi,
Z. (2022). Synchronous computer-mediated corrective feedback and EFL learners’ grammatical knowledge
development: A sociocultural perspective. Iranian Journal of Language
Teaching Research, 10(2), 115–136.
Meng, Y., Fu, H., & Wang, C. (2024). The effectiveness of computerized
listening dynamic assessment: Attribute-based mediation model. Language
Learning & Technology, 28(1), 1–28. https://doi.org/10.64152/10125/73580
Nicholas, A., Blake, J., Perkins, J., & Mozgovoy, M. (2026). Evaluating the
effectiveness of a computerised dynamic assessment of
L2 English email requests. Computer Assisted Language Learning, 39(1–2), 303–335. https://doi.org/10.1080/09588221.2024.2374775
Norris, J. M., & Ortega, L. (2009). Towards an organic approach to
investigating CAF in instructed SLA: The case of complexity. Applied
Linguistics, 30(4), 555–578. https://doi.org/10.1093/applin/amp044
Paas, F., Renkl, A., & Sweller, J. (2003). Cognitive load
theory and instructional design: Recent developments. Educational
Psychologist, 38(1), 1–4. https://doi.org/10.1207/S15326985EP3801_1
Pileh Roud, L. F., & Hidri, S.
(2021). Toward a sociocultural approach to computerized dynamic assessment of
the TOEFL iBT listening comprehension test. Education and Information
Technologies, 26(4), 4943–4968. https://doi.org/10.1007/s10639-021-10498-z
Plonsky, L., & Oswald, F. L. (2015). Meta-analyzing second language
research. In L. Plonsky (Ed.), Advancing quantitative methods in second
language research (pp. 106–128).
Routledge. https://doi.org/10.4324/9781315870908-6
Poehner, M. E., & Lantolf, J. P. (2013). Bringing the ZPD into the equation:
Capturing L2 development during computerized dynamic assessment (C-DA). Language
Teaching Research, 17(3), 323–342. https://doi.org/10.1177/1362168813482935
Qian, X., Meng, Y., & Yue, Z. (2023). Effects of
dynamic-assessment-informed blended instruction on the development of academic
English paraphrasing ability. Foreign Language Education, 44(2), 63–68. https://doi.org/10.16362/j.cnki.cn61-1023/h.2023.02.014
Qin, T., & van Compernolle, R. A. (2021). Computerized dynamic assessment of implicature
comprehension in L2 Chinese. Language Learning & Technology, 25(2),
55–74. https://doi.org/10.64152/10125/73433
Ren, K. (2019). Effects of using Juku Pigai in senior high school
English writing instruction [Master’s thesis, Tianjin University].
Shen, J., Hou, Z., & Min, S. (2026). Exploring the use of chatbots in
EFL argumentative writing: From the perspective of dynamic assessment. International
Journal of Applied Linguistics, 36(1), 725–736. https://doi.org/10.1111/ijal.12809
Shen, Y., Zhang, L. J., & Zhou, L. (2019). A meta-analysis of empirical
research on the effects of L2 listening strategy instruction. Foreign
Languages in China, 16(6), 57–66. https://doi.org/10.13564/j.cnki.issn.1672-9382.2019.06.009
Shen, Y., & Zhang, L. J. (2020). Recent advances in meta-analysis in
second language research in China and abroad, 2010–2019. Foreign Language
World, (3), 70–79.
Shen, Y., Zhang, L. J., & Sun, Q. (2025). The effects of task
complexity on L2 written production: A meta-analysis. Modern Foreign
Languages, 48(1), 77–89. https://doi.org/10.20071/j.cnki.xdwy.20241030.006
Sherkuziyeva, N., Gabidullina,
F. I., Ahmed Abdel-Al Ibrahim, K., & Bayat, S. (2023). The comparative
effect of computerized dynamic assessment and rater-mediated assessment on EFL
learners’ oral proficiency, writing performance, and test anxiety. Language
Testing in Asia, 13, Article 15. https://doi.org/10.1186/s40468-023-00227-3
Teng,
M. F. (2026). Metacognition in using GenAI for language teaching and learning. Emerging
Educational Technologies, 1(1). https://doi.org/10.65553/EET.260101
Vakili, S., & Ebadi, S. (2022). Exploring EFL
learners’ developmental errors in academic writing through face-to-face and
computer-mediated dynamic assessment. Computer Assisted Language Learning,
35(3), 345–380. https://doi.org/10.1080/09588221.2019.1698616
Vygotsky, L. S. (1978). Mind in society: The development of higher
psychological processes (M. Cole, V. John-Steiner, S. Scribner, & E. Souberman, Eds.). Harvard University
Press.
Wang, Q., Gao, Y., & Wang, X. (2024). The relationships between
burnout, job satisfaction, and emotion regulation among foreign language
teachers: A meta-analytic review. Acta Psychologica, 250, Article 104545. https://doi.org/10.1016/j.actpsy.2024.104545
Xiao, Y. (2024). Effects of dynamic assessment on the English writing
performance of English majors in a blended-learning context [Master’s thesis, Jiangxi Normal
University].
Yang, Y., & Qian, D. D. (2020). Promoting L2 English learners’ reading
proficiency through computerized dynamic assessment. Computer Assisted
Language Learning, 33(5–6), 628–652. https://doi.org/10.1080/09588221.2019.1585882
Han Shi is a doctoral student of Basic
Education College, National University of Defense Technology, China. Her areas
of interest are the development of EFL teachers, the integration of technology
in EFL education, and the synergistic intersection of positive psychology with
EFL education research. She focuses on frontier research in the aforementioned fields and actively participates in language
teaching seminars. Her theoretical accomplishments and research capacity have
been steadily improved.
Email: shihan@nudt.edu.cn
ORCID ID:
https://orcid.org/0009-0003-0223-1769
Xiaolei Ma works as a Ph.D. supervisor in Basic Education College at National
University of Defense Technology. He serves as an executive council member of
the Language Governance Research Society of the Chinese Society for Language
Modernization, a member of the Hunan Provincial Teaching Guidance Committee, a
member of the Discipline Appraisal Group of the Hunan Provincial Academic
Degrees Committee, and an executive council member of the Translators
Association of Hunan. His research interests lie in technology-assisted
language learning and foreign language education. His academic publications
have appeared in a number of leading Chinese core
journals (CSSCI indexed).
Email: maxiaolei@nudt.edu.cn
ORCID ID:
https://orcid.org/0009-0003-6060-0980