pith. sign in

arxiv: 2604.27493 · v1 · submitted 2026-04-30 · 💻 cs.HC

Why Learners Drift In and Out: Examining Intermittent Discontinuance in AI-Mediated Informal Digital English Learning (AI-IDLE) Using SEM and fsQCA

Yiran Du , Huimin He This is my paper

Pith reviewed 2026-05-07 09:24 UTC · model grok-4.3

classification 💻 cs.HC
keywords AI-IDLEintermittent discontinuanceinformal learningEnglish as a foreign languagestructural equation modelingfsQCAenjoymentboredom
0
0 comments X

The pith

Perceptions of AI intelligence and interactivity cut intermittent discontinuance in English learning by boosting enjoyment

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper explores why learners temporarily stop using AI tools for informal English practice outside class. It applies a cognition-affect-conation model to survey data from 632 Chinese university EFL students and analyzes the results with structural equation modeling plus fuzzy-set qualitative comparative analysis. The modeling shows that seeing AI as intelligent, interactive, and personalized lowers discontinuance indirectly by raising enjoyment, while seeing it as ineffective, uncontrollable, or complex raises discontinuance indirectly by raising boredom. The configurational analysis identifies four distinct combinations of these factors that each produce the same outcome of pausing use. The work shifts attention from getting learners to start with AI tools to keeping them engaged over repeated sessions.

Core claim

Intermittent discontinuance in AI-mediated informal digital English learning (AI-IDLE) is indirectly reduced by perceived intelligence, interactivity, and personalisation through enjoyment and increased by perceived ineffectiveness, uncontrollability, and complexity through boredom, as shown by SEM results on 632 learners. FsQCA identifies four configurational pathways to discontinuance arising from varied mixes of cognitive barriers and affective disengagement. These results extend prior work on AI-IDLE from adoption to post-adoption behavior and indicate the value of designing enjoyable, controllable, and manageable AI experiences for language learners.

What carries the argument

Cognition-affect-conation framework applied via structural equation modelling and fuzzy-set qualitative comparative analysis to map how AI tool perceptions influence emotional states that drive discontinuance behavior.

If this is right

  • Designers of AI English learning tools should enhance perceived intelligence, interactivity, and personalisation to promote enjoyment and reduce pauses in use.
  • Tools should minimize perceptions of ineffectiveness, uncontrollability, and complexity to avoid boredom-induced discontinuance.
  • Multiple configurational pathways mean that different learners may require tailored approaches to prevent temporary withdrawal.
  • Research on AI-mediated learning should expand beyond initial adoption to address sustaining long-term informal use.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • These mechanisms could generalize to AI tools for learning other subjects in informal settings.
  • Tracking actual usage data over time rather than relying on single surveys could test the predictive power of these perceptions.
  • Interventions focused on improving controllability and reducing complexity might have wide effects on retention in digital education platforms.

Load-bearing premise

The self-reported perceptions of AI tool characteristics and associated enjoyment or boredom levels validly and causally account for intermittent discontinuance among the surveyed Chinese EFL learners.

What would settle it

An experiment that manipulates the perceived intelligence or complexity of an AI English tool and finds no corresponding change in enjoyment, boredom, or actual rates of intermittent discontinuance in a comparable group of learners.

Figures

Figures reproduced from arXiv: 2604.27493 by Huimin He, Yiran Du.

Figure 1
Figure 1. Figure 1: The Conceptual Model 3.2 Enjoyment and Boredom Enjoyment and boredom are included in the conceptual model because they represent two central affective experiences that may explain learners’ voluntary engagement or withdrawal in AI-mediated informal digital learning of English (AI-IDLE) (Liu, Zou, et al., 2025). As AI-IDLE typically occurs outside formal instruction and is largely self-regulated, learners’ … view at source ↗
Figure 2
Figure 2. Figure 2: Structural Model Results Note. *** p < 0.001; ** p < 0.01 view at source ↗
read the original abstract

This study examined intermittent discontinuance in AI-mediated informal digital learning of English (AI-IDLE) through the cognition-affect-conation framework. Survey data were collected from 632 Chinese university EFL learners with prior AI-IDLE experience and analysed using structural equation modelling and fuzzy-set qualitative comparative analysis. The SEM results showed that perceived intelligence, perceived interactivity, and perceived personalisation reduced AI-IDLE intermittent discontinuance indirectly through enjoyment, whereas perceived ineffectiveness, perceived uncontrollability, and perceived complexity increased discontinuance indirectly through boredom. The fsQCA results further identified four configurational pathways leading to intermittent discontinuance, indicating that learners' temporary withdrawal from AI-IDLE can result from different combinations of cognitive barriers and affective disengagement. These findings extend AI-IDLE research from adoption and continuance to post-adoption discontinuance and highlight the need to design AI-supported English learning experiences that are enjoyable, personalised, controllable, and cognitively manageable.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript examines intermittent discontinuance in AI-mediated informal digital English learning (AI-IDLE) among Chinese university EFL learners through the cognition-affect-conation framework. It reports SEM results from a survey of 632 participants showing that perceived intelligence, interactivity, and personalization reduce discontinuance indirectly via enjoyment, while perceived ineffectiveness, uncontrollability, and complexity increase it via boredom; fsQCA identifies four configurational pathways to discontinuance.

Significance. If the measurement and analytical results prove robust, the work usefully extends AI-IDLE research from adoption/continuance to post-adoption discontinuance and supplies design implications for enjoyable, controllable AI tools. The dual use of SEM (for net indirect effects) and fsQCA (for configurations) is a methodological asset that can reveal both average relations and equifinal pathways. The sample size provides reasonable power for the SEM component.

major comments (3)
  1. Abstract and Results section: The language that positive perceptions 'reduced AI-IDLE intermittent discontinuance indirectly through enjoyment' and negative perceptions 'increased discontinuance indirectly through boredom' asserts directional causal mediation. The single-timepoint cross-sectional survey cannot establish temporal precedence, rule out reverse causation (e.g., prior discontinuance shaping current perceptions), or exclude unmeasured confounders. This causal framing is load-bearing for the central claim and must be revised to associational language or supported by additional evidence such as longitudinal data.
  2. Methodology and Results sections: The manuscript must report explicit tests for common-method bias (e.g., Harman's single-factor test, full collinearity assessment, or marker-variable approach) and full measurement-model diagnostics (factor loadings, AVE, CR, HTMT). Without these, the validity of the eight self-reported constructs and the indirect-effect estimates cannot be confirmed.
  3. fsQCA analysis (presumably §4 or §5): The calibration thresholds for fuzzy-set membership (e.g., for enjoyment, boredom, and discontinuance) should be justified as pre-specified on substantive grounds rather than chosen post-hoc to maximize coverage or consistency. Post-hoc calibration risks capitalizing on sample idiosyncrasies and undermines the configurational claims.
minor comments (2)
  1. Abstract: The response rate, exact sampling frame, and any exclusion criteria for the 632 respondents should be stated to allow assessment of selection bias.
  2. Discussion: The four fsQCA pathways should be compared explicitly with the SEM indirect effects to clarify where the two methods converge or diverge.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback on our manuscript. We have carefully considered each comment and revised the paper to address the concerns raised. Below, we provide point-by-point responses.

read point-by-point responses

  1. Referee: Abstract and Results section: The language that positive perceptions 'reduced AI-IDLE intermittent discontinuance indirectly through enjoyment' and negative perceptions 'increased discontinuance indirectly through boredom' asserts directional causal mediation. The single-timepoint cross-sectional survey cannot establish temporal precedence, rule out reverse causation (e.g., prior discontinuance shaping current perceptions), or exclude unmeasured confounders. This causal framing is load-bearing for the central claim and must be revised to associational language or supported by additional evidence such as longitudinal data.

    Authors: We fully acknowledge the limitations of our cross-sectional survey design in establishing causal relationships. Accordingly, we have revised the abstract and the Results section to employ associational rather than causal language. The revised text now indicates that positive perceptions 'were associated with lower AI-IDLE intermittent discontinuance indirectly through enjoyment' and that negative perceptions 'were associated with higher discontinuance indirectly through boredom'. This adjustment accurately reflects the correlational nature of the findings while preserving the reported indirect associations. We do not have longitudinal data available to support stronger causal claims at this stage. revision: yes

  2. Referee: Methodology and Results sections: The manuscript must report explicit tests for common-method bias (e.g., Harman's single-factor test, full collinearity assessment, or marker-variable approach) and full measurement-model diagnostics (factor loadings, AVE, CR, HTMT). Without these, the validity of the eight self-reported constructs and the indirect-effect estimates cannot be confirmed.

    Authors: We appreciate the referee's emphasis on methodological rigor. In the revised version of the manuscript, we have added a dedicated subsection in the Methodology reporting the results of common method bias tests, including Harman's single-factor test (showing the first factor accounted for less than 50% of variance) and a full collinearity assessment using variance inflation factors (all VIFs below 3.3). Furthermore, we now present complete measurement model diagnostics, encompassing standardized factor loadings (all > 0.7), AVE values (all > 0.5), CR values (all > 0.7), and HTMT ratios (all < 0.85), thereby confirming the convergent and discriminant validity of the constructs. revision: yes

  3. Referee: fsQCA analysis (presumably §4 or §5): The calibration thresholds for fuzzy-set membership (e.g., for enjoyment, boredom, and discontinuance) should be justified as pre-specified on substantive grounds rather than chosen post-hoc to maximize coverage or consistency. Post-hoc calibration risks capitalizing on sample idiosyncrasies and undermines the configurational claims.

    Authors: We agree that transparent justification of calibration is essential. The thresholds were pre-specified based on substantive theoretical considerations and established fsQCA guidelines prior to conducting the analysis. We used the 5th, 50th, and 95th percentiles of the empirical distributions as anchors for full non-membership, crossover, and full membership, respectively, consistent with recommendations in the fsQCA literature for continuous variables and aligned with the theoretical ranges of the constructs in educational psychology. We have now elaborated on this rationale in the Methods section, including citations to relevant methodological sources, to demonstrate that the choices were not data-driven post-hoc adjustments. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical statistical analysis is self-contained

full rationale

The paper applies standard SEM and fsQCA to cross-sectional survey data from 632 participants to estimate associations and configurations among measured constructs. Reported indirect effects and pathways are outputs of statistical estimation on observed variables rather than any self-referential definitions, fitted parameters renamed as predictions, or load-bearing self-citation chains that reduce the central claims to prior inputs by construction. The cognition-affect-conation framework is invoked as an established lens without uniqueness theorems or ansatzes smuggled from the authors' own prior work. The derivation chain consists of data collection, measurement validation, and model fitting against external benchmarks of survey-based mediation and configurational analysis, with no steps that collapse the reported results to tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard social-science statistical assumptions and an established theoretical framework; no new free parameters, invented entities, or ad-hoc axioms beyond those implicit in SEM and fsQCA are introduced.

axioms (2)
  • domain assumption Standard assumptions of structural equation modeling (linearity, multivariate normality, no omitted variable bias)
    Implicitly required by the reported SEM analysis of indirect effects.
  • domain assumption The cognition-affect-conation framework provides an appropriate causal structure for AI-IDLE behavior
    The study is explicitly framed through this framework.

pith-pipeline@v0.9.0 · 5472 in / 1507 out tokens · 89586 ms · 2026-05-07T09:24:01.406581+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages

  1. [1]

    Zhang & Liu, 2025)

    Introduction Informal digital learning of English (IDLE) refers to learners’ self-directed English learning through digital technologies beyond formal instruction (Y . Zhang & Liu, 2025). With recent advances in artificial intelligence, AI-mediated informal digital learning of English (AI-IDLE) has emerged, enabling learners to use tools such as ChatGPT, ...

    work page 2025

  2. [2]

    Zhang & Liu, 2025)

    Literature Review 2.1 AI-Mediated Informal Digital English Learning (AI-IDLE) Informal digital learning of English (IDLE) has been conceptualised as learners’ self-directed and out-of-class engagement with digital technologies and online resources for English learning purposes, including activities such as watching videos, interacting on social media, gam...

    work page 2025

  3. [3]

    Theoretical Framework and Hypothesis Development 3.1 Cognition–Affect–Conation Framework The cognition–affect–conation framework provides a suitable theoretical basis for this study because it explains behaviour as a sequential process in which individuals’ beliefs shape their emotional responses, which in turn influence behavioural tendencies (Zhou & Zha...

    work page 2024

  4. [4]

    The Conceptual Model 3.2 Enjoyment and Boredom Enjoyment and boredom are included in the conceptual model because they represent two central affective experiences that may explain learners’ voluntary engagement or withdrawal in AI-mediated informal digital learning of English (AI-IDLE) (Liu, Zou, et al., 2025). As AI-IDLE typically occurs outside formal i...

    work page 2025

  5. [5]

    Methodology 4.1 Research Design This study adopted a quantitative cross-sectional design to examine learners’ intermittent discontinuance in AI-mediated informal digital English learning (AI-IDLE). Guided by the cognition–affect–conation framework, it investigated how cognitive evaluations of AI tools influenced enjoyment and boredom, which in turn predic...

    work page 2023

  6. [6]

    Characteristic Category n % Gender Male 287 45.4 Female 345 54.6 Age 18–20 218 34.5 21–23 301 47.6 24 or above 113 17.9 Study level Undergraduate 468 74.1 Postgraduate 164 25.9 Academic discipline STEM 296 46.8 Non-STEM 336 53.2 4.3 Measurement All constructs, perceived intelligence (Balakrishnan et al., 2022), perceived interactivity (F. Wang et al., 202...

    work page 2022

  7. [7]

    All items were rated on a five-point Likert scale ranging from 1 = strongly disagree to 5 = strongly agree

    (see Table 2). All items were rated on a five-point Likert scale ranging from 1 = strongly disagree to 5 = strongly agree. The English items were translated into Chinese and checked through back-translation to ensure semantic equivalence (Klotz et al., 2023). A pilot test was conducted with 30 university students who had experience using AI tools for info...

    work page 2023

  8. [8]

    Specifically, enjoyment was negatively associated with intermittent discontinuance, whereas boredom was positively associated with intermittent discontinuance

    Discussion 6.1 Net Effects of Cognitive and Affective Factors on AI-IDLE Intermittent Discontinuance The SEM results provide empirical support for the cognition–affect–conation framework, indicating that learners’ cognitive evaluations of AI tools shape affective experiences, which subsequently influence AI-IDLE intermittent discontinuance. Specifically, ...

    work page 2024

  9. [9]

    by showing that these negative evaluations operate through affective disengagement. Overall, the net-effect results indicate that AI-IDLE intermittent discontinuance is not merely a behavioural response to technological limitations, but an affectively mediated outcome: positive perceptions reduce discontinuance by enhancing enjoyment, whereas negative per...

    work page 2012

  10. [10]

    The remaining configurations highlight a control–complexity mechanism

    and that boredom is a salient negative emotion in language learning that weakens persistence (Li et al., 2023; Zhang & Li, 2025). The remaining configurations highlight a control–complexity mechanism. Paths 3 and 4 show that perceived uncontrollability and perceived complexity can jointly lead to intermittent discontinuance, especially when enjoyment is a...

    work page 2023

  11. [11]

    Conclusion This study examined AI-IDLE intermittent discontinuance through the cognition–affect–conation framework by integrating SEM and fsQCA. The SEM results showed that positive cognitive evaluations, including perceived intelligence, interactivity, and personalisation, reduced intermittent discontinuance indirectly through enjoyment, whereas negative...

    work page doi:10.1007/s10639-023-11635-6 2023

  12. [12]

    https://doi.org/10.1016/j.ijer.2026.102963 Wang, C., Du, Y., & Zou, B

    International Journal of Educational Research, 137, 102963. https://doi.org/10.1016/j.ijer.2026.102963 Wang, C., Du, Y., & Zou, B. (2026). Learners’ acceptance and use of multimodal artificial intelligence (AI)‐generated content in AI‐mediated informal digital learning of English. International Journal of Applied Linguistics, 36(1), 927–940. https://doi.o...

    work page doi:10.1016/j.ijer.2026.102963 2026

  13. [13]

    https://doi.org/10.3390/healthcare11020161 Zadorozhnyy, A., & Lee, J. S. (2025). Informal Digital Learning of English and willingness to communicate in a second language: Self-efficacy beliefs as a mediator. Computer Assisted Language Learning, 38(4), 669–689. https://doi.org/10.1080/09588221.2023.2215279 Zhang, J., & Li, R. (2025). Mapping the research o...

    work page doi:10.3390/healthcare11020161 2025