Sage Journals: Discover world-class research

Abstract

The use of technology in language learning is an emerging research field, especially in terms of how learners’ attitudes and acceptance of technology affect the process and experiences of language acquisition. This study seeks to evaluate the role of Generative AI (GenAI) tools in enhancing foreign language proficiency and to explore the motivational factors influencing their adoption and success. Utilizing the Technology Acceptance Model (TAM), this study explores the role of GenAI in foreign language learning among Chinese students. A cross-sectional survey design was utilized to collect data from 548 Chinese university students who use GenAI tools for foreign language learning in and outside the English as a Foreign Language (EFL) classroom. Questionnaires were collected on three respective scales: the Foreign Language Motivation Scale (FLMS), the Foreign Language Proficiency Scale (FLPS), and the AI Technology Acceptance for Foreign Language Learning Scale (ATA-FLLS). Data were analyzed through descriptive statistics, correlation analysis, path analysis, and structural equation modeling (SEM) to explore the relationships between ATA-FLL, FLP, and FLM. The findings demonstrated that positive relationships between perceived usefulness (PU) and perceived ease of use (PEOU) have a significant impact on Foreign Language Proficiency (FLP). Also, results reveal that self-efficacy (SE) mediated the relationship between PEOU and FLP, which indicates the subtle interplay between the factors of the Technology Acceptance Model (TAM) and psychological mediators; those factors collaboratively contribute to students’ foreign language achievement.

Plain Language Summary

This research investigates the impact of Chinese university students’ utilization of generative artificial intelligence (GenAI) learning tools tailored for foreign language learning on their motivation and proficiency in foreign language acquisition. Findings indicate that a positive disposition towards GenAI enhances motivation, particularly self-efficacy, which in turn leads to improvements in the four language skills of listening, speaking, reading, and writing. Motivation acts as a mediating factor between the acceptance of GenAI and language proficiency advancement. The study builds on previous research but focuses more on GenAI tools tailored for foreign language learning and the university student population.

Keywords

Gen AI foreign language proficiency motivation technology acceptance model

Introduction

AI has profoundly transformed second language acquisition by reshaping how learners engage with foreign language proficiency, motivation, and learning outcomes (Almelhes 2023; Wu et al., 2024). Positive psychology, which seeks not only to understand learners’ processes better but also to aid in their holistic development and positive learning experiences, has also resulted in a “positive turn” in English as Foreign Language (EFL) acquisition research (F. Huang et al., 2024). This change has a major effect on the psychological moods and personal well-being of students, which is why it is particularly relevant in AI-integrated and GenAI learning environments (Wang et al., 2024).

Two main reasons for implementing GenAI in foreign language instruction are particularly significant and relevant at this point: First, GenAI technologies are especially suited for learning foreign languages since their primary technology is designed to comprehend and produce human language (OpenAI, 2023), which closely aligns with the goals of language learning. Learners can instantly apply vocabulary and grammatical structures in a dialogic setting with GenAI tools, which offer interactive conversational practice that is essential for fluency development (Kasneci et al., 2023). Language learning, by nature, is based on experience and, thus, requires ongoing interaction and engagement with the medium of language (Li & Jeong, 2020). With English considered a global lingua franca (Zheng et al., 2024), however, there is an ever-growing demand for creativity and efficiency in ways to promote awareness of the English language. Practicing this way requires learners to internalize linguistic features and subtle interactions, which differ from experiential disciplines. Language researchers (Jansen et al., 2024) observed that rapid feedback from GenAI platforms can mimic the social interaction-driven process of natural language acquisition, enabling a more organic language skill development that is challenging to duplicate in other academic fields. GenAI systems are beneficial for language learners by providing speech recognition, interactive dialogues, and other advantages like being accessible 24/7 and rapid feedback (Adeshola & Adepoju, 2024). The personalized and interactive nature of GenAI makes it even more attractive. But both teachers and students must recognize its limitations and unpredictability (Kasneci et al., 2023). Although there are pedagogical justifications for GenAI, whether technology is effective in education is contingent upon the knowledge and acceptance of students. Using the Technology Acceptance Model (TAM), this study investigates the interactions of major TAM constructs in predicting Foreign Language Proficiency (FLP): Perceived Usefulness (PU), Perceived Ease of Use (PEOU), Attitude toward Use (ATU), and Behavioral Intention (BI) with motivational factors (Learning Needs, Self-Efficacy, and Achievement Motivation). Specifically, this study examines the mediating role of motivation in the relationship between GenAI acceptance and FLP in Chinese EFL university students. This study utilizes motivation to address the gap in understanding the role of psychological factors in GenAI acceptance and learning performance.

Literature Review

Technology Acceptance Model

Based on the Theory of Reasoned Action (TRA), Davis (1989) developed the Technology Acceptance Model (TAM). The model’s main thesis posits that behavioral intention to use technology is shaped by beliefs about its usefulness and ease of use rather than general attitudes. By following a simple methodology, TAM sought to provide a framework for investigating a broad range of behaviors displayed by technology users (Davis, 1989). TAM’s primary objective was to shed light on the mechanisms influencing technology adoption to predict user behavior and offer a theoretical framework for successful technology deployment.

The model shows four main characteristics: Perceived Usefulness (PU), Perceived Ease of Use (PEOU), Attitude Toward Using (ATU), and Behavioral Intention (BI). PU reflects the degree to which a person believes that using a certain technology enhances performance, while PEOU refers to the extent to which users believe that interacting with a system is free of effort. These factors further exert an impact on users’ attitude toward using technology (Attitude Toward Using (ATU), behavioral intention (BI), and actual usage behavior (Davis, 1989). This construct was conceptualized based on Bandura’s idea of outcome judgment, which is the belief that behavior results in a favorable outcome (Bandura, 1982). Based on data supporting the relationship between system performance expectancy and system utilization, perceived usefulness was operationalized (Marikyan & Papagiannidis, 2024).

When users perceive that technology may improve their academic or career outcomes, their intention to use it increases. According to Nilashi and Abumalloh (2025), TAM is a popular framework for forecasting and elucidating user acceptance of a variety of technologies in information systems, human-computer communication, and educational technology. Therefore, this study uses a technological approach to understand how PU, PEOU, ATU, and BI of students toward GenAI influence their foreign language proficiency (FLP).

Generative AI Acceptance in Foreign Language Learning

Advanced AI systems that use machine learning techniques to create original text, image, and audio output are referred to as generative AI (GenAI) (Baidoo-Anu & Ansah, 2023). The advancement of neural network topologies and deep learning methods has enabled GenAI to produce highly contextualized and human-like responses (Pack & Maloney, 2023). By offering interactive and personalized language practice, GenAI has created new opportunities and challenges in foreign language (FL) education (Gao et al., 2024). By creating a realistic conversational experience suited to the needs of each learner and providing immediate feedback that is contextually relevant, it enhances language learning effectiveness and learner engagement (Almelhes, 2023). The incorporation of chatbots, adaptive learning platforms, intelligent tutoring systems, and other GenAI technologies is transforming traditional pedagogy practices (Law, 2024). Hsiao and Chang (2023), in a study of 43 senior high school English as a foreign language (EFL) students in Taiwan, demonstrated that AI-enhanced tools significantly improved students’ performance in online writing and reading courses. By offering customized practice sessions and adaptive learning paths that address the student’s individual needs, GenAI applications strengthen second language competence by addressing specific learning gaps and accelerating progress (Wu et al., 2021).

Role of Students’ Motivation in Foreign Language Learning

In language learning environments, it is imperative that the affective characteristics of each learner (learning engagement, well-being, and motivation) interact dynamically with GenAI-based tools (Creely, 2024). When GenAI is recognized as a friendly and non-judgmental facilitator (Lievertz, 2019), which invites active participation in language learning processes and increases self-efficacy (Kim, 2020), levels of motivation and engagement among learners typically increase. According to Deci and Ryan’s Self-Determination Theory (SDT), individuals possess an inherent capacity to become intrinsically motivated and satisfy the three basic psychological needs of autonomy, competence, and relatedness. SDT offers a strong foundation for understanding why language learners during their studies choose artificial intelligence technology (Zheng et al., 2024). Research has shown that generative AI (GenAI) developments can enhance human-computer interaction, productivity, and student motivation. By allowing students to see their progress and achievement in their learning activities, GenAI boosts learning motivation and strengthens self-confidence. By enabling students to select and pursue lots of learning content independently, GenAI encourages autonomy, proficiency, and a sense of purpose. It also reinforces positive learning behaviors and sustained motivations.

Within the framework provided by the Self-Determination Theory, external factors such as perceived language ability, autonomy in technology use, and chances for social interaction play key roles in shaping motivation. Wang and Xue (2024) pointed out that in such a setting, integrated in EFL classrooms, students gain confidence from their language abilities, feel more control over how they learn, and interact more with each other. This contributes to motivating teachers and students in language learning efforts to keep pushing on.

For example, Wei (2023) compared the language learning experience of 60 university students using AI-mediated and traditional approaches. Students who utilized an AI-based language approach demonstrated higher confidence, improved English learning performance, stronger learning motivation, and greater self-regulated learning compared to those in the traditional approach. Similarly, studies on the effect of AI chatbots indicate that learners who felt supported by AI tools reported fewer negative learning experiences and greater satisfaction with their learning process (Linh et al., 2022).

UNIPUS AI: Foreign Language Learning Tool

UNIPUS is an online platform that supports independent English learning among university students in China. It works in tandem with the New Horizon College English Audio-Visual and Speaking Course and the New Horizon College English Reading and Writing Course. The platform includes digital courses, teaching management, interactive modules, a resource center, and other integrated tools (Wang & Xue, 2024). With the development of AI technology, UNIPUS has integrated AI into its system, transforming it into UNIPUS AI. Students can engage in conversations with the AI virtual assistant ZIYAN, inquiring about text themes, literary styles, structures, or other lesson content. Additionally, they can select lesson text directly, copy it with a single click, and swiftly edit their queries. After submitting their answers, students can click on the question number to view detailed answer analyses, in which ZIYAN provides problem-solving strategies and targeted feedback. Moreover, ZIYAN facilitates oral dialogue practice and provides evaluations that help students understand their strengths and weaknesses in spoken language. In this study, UNIPUS AI serves as the GenAI platform through which students’ AI-TAM acceptance, FLM, and FLP are examined.

Rationale

Learning a second language is quite a complex process, and researchers have used various models to explain the interaction of learning a second language and learner motivation. However, the rapid rise of GenAI tools has fundamentally altered the learning environment, and research has not yet clarified how such technologies influence learner motivation. The TAM has elucidated new technology adoption and successful implementations (J. Huang & Mizumoto, 2024). Therefore, this study extends TAM by integrating motivational constructs derived from SDT to examine how GenAI acceptance associates with FLP. While previous research has examined motivation as a moderator (Zheng et al., 2024), this study examines its mediating role by whether learning needs (LN), self-efficacy (SE), achievement motivation (AM), identified regulation (IR), negative motivation (NM), and situational motivation (SM) explain the relationship between GenAI acceptance and FLP. Figure 1 shows the mediation conceptual model of this study, which depicts the relationship among TAM constructs (PU, PEOU, ATU, and BI), FLP (Listening, Speaking, Reading, and Writing), and FLM (Learning Needs (LN), Self-efficacy (SE), Achievement Motivation (AM), Identified Regulation (IR), Negative Motivation (NM), and Situational Motivation (SM)).

Figure 1.

Conceptual framework.

Hypotheses

Based on the given objectives, we propose the following hypotheses:

H1: Based on the Technology Acceptance Model (TAM), Chinese EFL university students’ usage and attitudes towards GenAI positively predicts their foreign language proficiency (FLP).

H2: There is a significant positive correlation between the Technology Acceptance Model (TAM) of GenAI and foreign language motivation (FLM) among Chinese EFL university students.

H3: The foreign language motivation (FLM) mediates the relationship between the Technology Acceptance Model (TAM) of GenAI and foreign language proficiency (FLP) among Chinese EFL university students.

Methods

Research Design

This study focuses on the role of generative AI acceptance, foreign language proficiency, and foreign language motivation among Chinese university students and explores their relationships through a quantitative research design. Data were collected from university students learning English as a Foreign Language (EFL) by using a structured questionnaire. The study aims to explore the impact of generative AI acceptance and foreign language motivation (FLM) on university students’ foreign language proficiency (FLP), and to examine the mediating role of foreign language motivation (FLM) between generative AI acceptance and foreign language proficiency (FLP).

Sampling and Data Collection

The data for the study were collected by the WJX online platform in five universities in two eastern Chinese cities. The target group consists of students who study English as a foreign language (EFL) in universities and use the UNIPUS AI or other GenAI tools simultaneously. The questionnaire was conducted through the WJX online platform, and rewards were set up to achieve the highest response rate (100%) (no missing data) and obtain high-quality questionnaires. The research collected 194 questionnaires as pilot samples, which were used to detect the reliability and validity of the questionnaires for adjusting the questionnaire structure, and all scales were administered in Chinese after forward-backward translation validation. After the questionnaire structure was adjusted, 548 valid questionnaires were collected for formal sample analysis.

Measurement

The Foreign Language Motivation Scale (FLMS) was developed by H. W. Huang (2005). The scale is used to measure motivation in learning a second foreign language with a five-point Likert scale (1 = strongly disagree; 5 = strongly agree). The scale is divided into three sub-scales measuring learning needs (LN), self-efficacy (SE), and achievement motivation (AM). Subsequent to the pilot test, even though the reliability and validity data of the questionnaire were within an acceptable scope, we readjusted the questionnaire structure in accordance with the Self-Determination Theory (SDT) and the hierarchical model of motivation proposed by Vallerand (1997), based on SDT to achieve more accurate conclusions. Specifically, we incorporated three additional factors (Identified Regulation (IR), Negative Motivation (NM), Situational Motivation (SM)) and streamlined the number of questionnaire items from 34 to 26.

With reference to the Common European Framework of Reference for Languages (CEFR), this research formulated a self-assessment scale for foreign language proficiency (FLP). The scale encompasses 20 items, which are categorized into four dimensions (listening, speaking, reading, and writing), and a five-point Likert scale is employed for scoring (1 = Strongly Disagree; 5 = Strongly Agree). The Foreign Language Proficiency Scale (FLPS) was used to assess an individual’s foreign language proficiency (FLP) and exhibited excellent reliability and validity.

Based on the Technology Acceptance Model (TAM), this study developed the AI Technology Acceptance for Foreign Language Learning Scale (ATA-FLLS). A self-developed questionnaire measures the usage and attitude of GenAI technology acceptance by measuring four areas of TAM perspective (PU, PEOU, ATU, and BI). It contains 20 items with a 5-point Likert scale ranging from 1 to 5, and demonstrates strong reliability and validity, as evidenced by the following factor analysis results.

Instrumentation

The research utilized the WJX online platform to collect questionnaires, and SPSS 31, Amos 31, and SmartPLS 4 were employed as data analysis tools. SPSS 31 was used to analyze the demographic data of the samples, reliability, validity, factor loading, normality, common method variance (CMV), multicollinearity diagnostics, and correlations; Amos 31 conducted confirmatory factor analysis (CFA) for all scales; SPSS PROCESS and SmartPLS 4 were applied to test the paths and fit of the structural equation.

Results

This section presents the results of the study, including demographic information of the formal sample, reliability and validity tests, factor loadings, normality, common method variance (CMV), multicollinearity diagnostics, correlations, confirmatory factor analysis (CFA), and the structural equation modeling (SEM) test for mediating effects.

Demographic Analysis

The demographic characteristics of the sample, including age and gender of the participants, were recorded in Table 1. The sample includes a total of 548 participants, with a higher number of females (72.8%) than males (26.6%), and no intention to disclose (0.5%). The participants’ ages range from 17 to 20 years (M = 18.36, SD = 0.606), with the majority (535 students, 97.6%) being in their first year of undergraduate studies.

Table 1.

Descriptive Statistics of Demographics of the Study (N = 548).

Demographics	Frequency	Percentage	Demographics	Frequency	Percentage
Gender	M = 1.74	SD = 0.452	Age	M = 18.36	SD = 6.06
Male	146	26.6	17 years old	10	1.8
Female	399	72.8	18 years old	355	64.8
No intention to disclose	3	0.5	19 years old	156	28.5
			20 years old	27	4.9

Note. M = Mean value; SD = standard deviation.

Reliability Analysis

Reliability analysis was performed on all study scales based on the formal sample (N = 548). Cronbach’s alpha statistics for all the constructs, as shown in Table 2, reveal good to excellent internal consistency. The Foreign Language Motivation Scale (FLMS) (26 items) has good reliability (α = .948), and the AI Technology Acceptance for Foreign Language Learning Scale (ATA-FLLS) (20 items) has excellent reliability (α = .972). The four subscales of the Foreign Language Proficiency Scale (FLPS) (listening, speaking, reading, and writing) also show high internal consistency (α = .892–.926).

Table 2.

Reliability Analysis of All Scales (N = 548).

Scale	α (.972)	CR	AVE	No. of Items
FLM	.948	.943	.408	26
LN	.919	.916	.610	7
AM	.907	.905	.615	6
SE	.918	.918	.691	5
NM	.940	.940	.839	3
IR	.829	.932	.627	3
SM	.965	.965	.933	2
ATA	.972	.972	.637	20
PU	.940	.941	.726	6
ATU	.952	.952	.741	7
PEOU	.894	.876	.638	4
BI	.822	.879	.708	3
FLP	.967	.967	.595	20
Listening	.892	.893	.627	5
Speaking	.926	.924	.710	5
Reading	.903	.904	.654	5
Writing	.909	.909	.668	5

Note.α = Cronbach’s alpha; CR = Composite Reliability; AVE=Average Variance Extracted; FLM = Foreign Language Motivation; ATA = AI Technology Acceptance for Foreign Language Learning; FLP = Foreign Language Proficiency; LN = Learning Needs; AM = Achievement Motivation; SE = Self-efficacy; IR = Identified Regulation; NM = Negative Motivation; SM = Situational Motivation; PU = Perceived Usefulness; PEOU = Perceived Ease of Use; ATU = Attitude Toward Using; BI = Behavioral Intention.

Table 2 also shows the composite reliability (CR) and average variance extracted (AVE) for all the constructs examined (N = 548). Consistent with the CR values, the CR values range from .876 to .972, all CR > .7, providing evidence of good internal consistency reliability among all variables. Simultaneously, the value range of AVE spans from .408 to .933. Only when FLM (AVE = .408) serves as a single factor, its AVE is lower than .5. For all other factors, their AVE values are higher than .5. According to Maruf et al. (2021), the AVE should be at least .50 or higher. However, if the composite reliability (CR) value is sufficient, an AVE value exceeding .40 is also acceptable. The CR and AVE values suggest that the constructs were measured reliably.

Convergent Validity

The convergent validity of all constructs in this study was evaluated using factor loadings, composite reliability (CR), and average variance extracted (AVE). As can be discerned from Table 2, both the CR and AVE values are at an outstanding level. Table 3 showcases the factor loadings of all items. With the exception of a small number of items in the Foreign Language Motivation Scale (FLMS), all factor loadings exceed the recommended value of .70 (Fornell & Larcker, 1981; Hair et al., 2014). Collectively, the entire questionnaire exhibits satisfactory convergent validity.

Table 3.

Factor Loadings of All Scales (N = 548).

Constructs	Items and code	Factor loading coefficient	Deleted items
FLM	26 items	KMO=0.941	8 items
Leaning needs	FLM1	0.804
	FLM2	0.780
	FLM3	0.707
	FLM4	0.810
	FLM5	0.841
	FLM6	0.732
	FLM7	0.818
Achievement motivation	FLM8	0.817
	FLM9	0.764
	FLM10	0.832
	FLM11	0.759
	FLM12	0.812
	FLM13	0.732
Self-efficacy	FLM14	0.851
	FLM15	0.859
	FLM16	0.863
	FLM17	0.802
	FLM18	0.784
Negative motivation	FLM19	0.927
	FLM20	0.909
	FLM21	0.913
Identified regulation	FLM22	0.815
	FLM23	0.876
	FLM24	0.667
Situational motivation	FLM25	0.971
Situational motivation	FLM26	0.960
ATA	20 items	KMO = 0.973	Nil
Perceived usefulness	ATA1	0.835
	ATA2	0.809
	ATA3	0.880
	ATA4	0.878
	ATA5	0.862
	ATA6	0.841
Attitude towards use	ATA7	0.836
	ATA8	0.873
	ATA9	0.893
	ATA10	0.843
	ATA11	0.881
	ATA12	0.870
	ATA13	0.832
Perceived ease of use	ATA14	0.742
	ATA15	0.826
	ATA16	0.817
	ATA17	0.824
Behavioral intention	ATA18	0.778
	ATA19	0.878
	ATA20	0.873
FLP	25 items	KMO = 0.971	Nil
Listening	LP1	0.717
	LP2	0.822
	LP3	0.832
	LP4	0.778
	LP5	0.803
Speaking	LP6	0.758
	LP7	0.865
	LP8	0.877
	LP9	0.898
	LP10	0.836
Reading	LP11	0.836
	LP12	0.789
	LP13	0.849
	LP14	0.800
	LP15	0.759
Writing	LP16	0.868
	LP17	0.869
	LP18	0.787
	LP19	0.783
	LP20	0.780

Note. FLM = Foreign Language Motivation; ATA = AI Technology Acceptance for Foreign Language Learning; FLP = Foreign Language Proficiency.

Discriminant Validity

The Fornell-Larcker criterion (Table 4) assessed discriminant validity. The Fornell-Larcker criterion indicated adequate discriminant validity because the square roots of the average variance extracted (AVEs) for most constructs (diagonal values) were greater than the correlations with all other constructs. For instance, Foreign Language Motivation (FLM) (√AVE = 0.771), Learning Needs (LN) (√AVE = 0.781), and Perceived Usefulness (PU) (√AVE = 0.852), all met this criterion.

Table 4.

Discriminant Validity of All Scales (Fornell-Larcker Criterion, N = 548).

Construct	FLM		ATA		FLP
FLM	0.771
ATA	0.343^***		0.798
FLP	0.439^***		0.554^***		0.639
FLM	LN	AM	SE	NM	IR		SM
LN	0.781
AM	0.752^***	0.784
SE	0.571^***	0.655^***	0.831
NM	0.100^**	0.084^**	−0.019	0.916
IR	0.760^***	0.669^***	0.456^***	0.153^***	0.792
SM	0.707^***	0.591^***	0.393^***	0.039	0.615^***		0.966
ATA	PU		ATU		PEOU	BI
PU	0.852
ATU	0.829^***		0.861
PEOU	0.764^***		0.789^***		0.799
BI	0.738^***		0.854^***		0.718^***	0.841
FLP	Listening		Speaking		Reading	Writing
Listening	0.792
Speaking	0.796^***		0.843
Reading	0.758^***		0.800^***		0.809
Writing	0.747^***		0.779^***		0.864^***	0.817

Note. FLM = Foreign Language Motivation; ATA = AI Technology Acceptance for Foreign Language Learning; FLP = Foreign Language Proficiency; LN = Learning Needs; AM = Achievement Motivation; SE = Self-efficacy; IR = Identified Regulation; NM = Negative Motivation; SM = Situational Motivation; PU = Perceived Usefulness; PEOU = Perceived Ease of Use; ATU = Attitude Toward Using; BI = Behavioral Intention. Gray shading values indicate the square root of the average variance extracted for each construct.

***

, ** represent p levels of 1% and 5%, respectively.

Additionally, the square root of the average variance extracted (AVE) for Listening (√AVE = .792) is slightly lower than the correlation coefficient between Speaking and Listening (r = .796), with a difference of 0.004; the √AVE for Reading (√AVE = .809) is slightly lower than the correlation coefficient between Reading and Writing (r = .864), with a difference of .055. According to Henseler et al. (2015), when using the Fornell-Larcker criterion to assess discriminant validity, if most constructs meet the standard but only a few √AVE values are slightly lower than the correlation coefficients, this situation may be considered acceptable in practice, especially when the degree of violation is minor.

Common Method Variance

To evaluate the potential effects of common method bias, we conducted Harman’s single-factor test. The first unrotated factor explained 36.657% of the total variance. This is well below the 50% benchmark, which is suggested as a rule of thumb (Baumgartner et al., 2021). Overall, this suggests that common method variance is not likely a concerning issue in the data as well as in the measurement model results, due to single-source bias.

Assessment of Multicollinearity and Normality

Tolerance Index (TI) and Variance Inflation Factor (VIF) are indicators used to detect multicollinearity in regression analysis. Regarding TI, a threshold of greater than 0.1 (TI > .1) is generally recommended, indicating that multicollinearity is not severe; if it is less than .1, it suggests a significant multicollinearity problem; some literature suggests a threshold of greater than .2 for a more conservative assessment (Marcoulides & Raykov, 2019). All the TI values in Table 5 are above .1, and most of them are above .2, with only two items (FLM 25, FLM 26) ranging between .1 and .2.

Table 5.

Multicollinearity Statistics TI and VIF (N = 548).

FLM	TI	VIF	ATA	TI	VIF	FLP	TI	VIF
FLM1	0.307	3.253	ATA1	0.318	3.145	LP1	0.454	2.202
FLM2	0.345	2.897	ATA2	0.341	2.936	LP2	0.293	3.414
FLM3	0.425	2.352	ATA3	0.249	4.008	LP3	0.324	3.088
FLM4	0.315	3.179	ATA4	0.255	3.916	LP4	0.403	2.481
FLM5	0.294	3.402	ATA5	0.268	3.728	LP5	0.342	2.923
FLM6	0.406	2.461	ATA6	0.292	3.427	LP6	0.375	2.665
FLM7	0.295	3.392	ATA7	0.283	3.535	LP7	0.269	3.720
FLM8	0.335	2.981	ATA8	0.226	4.422	LP8	0.253	3.960
FLM9	0.381	2.622	ATA9	0.212	4.725	LP9	0.238	4.204
FLM10	0.316	3.165	ATA10	0.292	3.427	LP10	0.300	3.332
FLM11	0.420	2.383	ATA11	0.226	4.418	LP11	0.291	3.431
FLM12	0.345	2.901	ATA12	0.241	4.145	LP12	0.367	2.723
FLM13	0.383	2.610	ATA13	0.306	3.273	LP13	0.288	3.469
FLM14	0.284	3.522	ATA14	0.435	2.300	LP14	0.355	2.814
FLM15	0.275	3.632	ATA15	0.361	2.766	LP15	0.398	2.513
FLM16	0.285	3.514	ATA16	0.377	2.656	LP16	0.278	3.595
FLM17	0.322	3.105	ATA17	0.353	2.836	LP17	0.274	3.647
FLM18	0.367	2.722	ATA18	0.380	2.634	LP18	0.337	2.967
FLM19	0.201	4.968	ATA19	0.273	3.661	LP19	0.382	2.616
FLM20	0.223	4.478	ATA20	0.263	3.798	LP20	0.316	3.165
FLM21	0.221	4.516
FLM22	0.325	3.074
FLM23	0.292	3.424
FLM24	0.516	1.938
FLM25	0.117	8.573
FLM26	0.118	8.467

Note. TI = Tolerance Index; VIF = Variance Inflation Factor; FLM = Foreign Language Motivation; ATA = AI Technology Acceptance for Foreign Language Learning; FLP = Foreign Language Proficiency.

For VIF, its value range extends from 1 to positive infinity. A value closer to 1 implies lower multicollinearity. Conventionally, a threshold of less than 5 or less than 10 (i.e., VIF < 5 or VIF < 10) is recommended, suggesting that the multicollinearity is acceptable. If the VIF value exceeds 10, it indicates severe multicollinearity (O’brien, 2007). As presented in Table 5, all VIF values are lower than 10, and the majority of them are below 5. Only two items (FLM 25 and FLM 26) exhibit VIF values ranging from 5 to 10. Overall, Table 5 demonstrates that there should be no multicollinearity issues among all the constructs.

Skewness and kurtosis are employed to assess the normality of a sample. A skewness value between −2 and +2 (or more strictly, −1 and +1) is considered acceptable as approximately normal, while values outside this range indicate significant skewness; similarly, for kurtosis (excess kurtosis), −2 to +2 or −1 to +1 is regarded as the threshold for approximate normality (Mishra et al., 2019). As presented in Table 6, all values of skewness and kurtosis fall within the range of −1 to +1. Consequently, all constructs can be deemed to follow an approximately normal distribution.

Table 6.

Normality Statistics Skewness and Kurtosis (N = 548).

FLM	S	K	ATA	S	K	FLP	S	K
FLM1	−0.707	0.963	ATA1	−0.62	0.336	LP1	−0.199	−0.368
FLM2	−0.515	0.271	ATA2	−0.525	−0.199	LP2	0.118	−0.532
FLM3	−0.327	−0.176	ATA3	−0.541	−0.005	LP3	−0.164	−0.406
FLM4	−0.514	0.175	ATA4	−0.566	0.02	LP4	−0.228	−0.234
FLM5	−0.49	0.192	ATA5	−0.429	−0.428	LP5	0.182	−0.286
FLM6	−0.409	0.032	ATA6	−0.499	−0.256	LP6	0.308	−0.507
FLM7	−0.583	0.19	ATA7	−0.471	−0.126	LP7	−0.102	−0.254
FLM8	−0.273	−0.146	ATA8	−0.437	−0.312	LP8	−0.139	−0.297
FLM9	−0.239	−0.571	ATA9	−0.43	−0.361	LP9	−0.065	−0.036
FLM10	−0.331	−0.401	ATA10	−0.422	−0.488	LP10	−0.116	−0.221
FLM11	−0.254	−0.216	ATA11	−0.494	−0.13	LP11	−0.135	−0.05
FLM12	−0.292	−0.043	ATA12	−0.504	−0.051	LP12	−0.154	0.015
FLM13	−0.222	−0.216	ATA13	−0.462	−0.192	LP13	−0.147	−0.225
FLM14	−0.131	−0.299	ATA14	−0.303	−0.493	LP14	−0.24	0.092
FLM15	−0.085	−0.306	ATA15	−0.335	−0.514	LP15	−0.171	−0.293
FLM16	0.022	−0.341	ATA16	−0.302	−0.614	LP16	−0.04	−0.085
FLM17	−0.067	−0.476	ATA17	−0.325	−0.618	LP17	−0.02	−0.238
FLM18	−0.041	−0.32	ATA18	−0.157	−0.483	LP18	−0.183	0.033
FLM19	−0.253	−0.5	ATA19	−0.407	−0.144	LP19	−0.186	−0.142
FLM20	−0.238	−0.679	ATA20	−0.495	−0.094	LP20	0.03	−0.429
FLM21	−0.205	−0.531
FLM22	−0.634	0.201
FLM23	−0.752	0.434
FLM24	−0.661	0.278
FLM25	−0.757	0.598
FLM26	−0.685	0.288

Note. S = Skewness; K = Kurtosis; FLM = Foreign Language Motivation; ATA = AI Technology Acceptance for Foreign Language Learning; FLP = Foreign Language Proficiency.

Confirmatory Factor Analysis (CFA)

Confirmatory factor analysis (CFA) was employed to assess the fit of the three measurement models: Foreign Language Motivation Scale (FLMS), AI Technology Acceptance for Foreign Language Learning Scale (ATA-FLLS), and Foreign Language Proficiency Scale (FLPS). Fit indices for each model are displayed in Tables 7 to 9 and depicted in Figures 2 to 4. The recommended fit indices are as follows: χ²/df < 5, CFI > 0.90, TLI > 0.90, RMR < 0.08, NFI > 0.90, GFI > 0.90, RMSEA < 0.08 (Byrne, 1994; Hu & Bentler, 1999). When the primary indices (RMSEA, CFI, TLI) are satisfied, even if some other indices (e.g., GFI or NFI) do not meet the criteria, the model can still be regarded as having an acceptable fit (Goretzko et al., 2024; Xia & Yang, 2019).

Table 7.

Model Fit Indices for the Foreign Language Motivation Scale (Six Factors, N = 548).

Model	χ²	df	χ²/df	CFI	TLI	RMR	NFI	GFI	RMSEA
Model fit	1,257.396	284	4.427	0.919	0.908	0.045	0.898	0.835	0.079

Note. Bootstrap = 5,000; CI = 95%.

Table 8.

Model Fit Indices for the AI Technology Acceptance in Foreign Language Learning Scale (Four Factors, N = 548).

Model	χ²	df	χ²/df	CFI	TLI	RMR	NFI	GFI	RMSEA
Model fit	577.162	164	3.519	0.961	0.955	0.019	0.947	0.903	0.068

Note. Bootstrap = 5,000; CI = 95%.

Table 9.

Model Fit Indices for the Foreign Language Proficiency Scale (Four Factors, N = 548).

Model	χ²	df	χ²/df	CFI	TLI	RMR	NFI	GFI	RMSEA
Model fit	821.595	164	5.010	0.931	0.921	0.040	0.916	0.858	0.086

Note. Bootstrap = 5,000; CI = 95%.

Figure 2.

CFA path diagram for the foreign language motivation model (six factors).

Figure 3.

CFA path diagram for the AI technology acceptance in foreign language learning model (four factors).

Figure 4.

CFA path diagram for the foreign language proficiency model (four factors).

Per Tables 5 and 6, all items of the Foreign Language Motivation Scale (FLMS) were found to follow a normal distribution, and collinearity diagnostics yielded favorable results. Accordingly, confirmatory factor analysis (CFA) was performed on FLMS using the Maximum Likelihood Estimation (MLE) in Amos, with the path diagram presented in Figure 2.

For the Foreign Language Motivation Scale (FLMS) (Table 7), the CFA results indicated χ²/df = 4.427, CFI = 0.919, TLI = 0.908, RMR = 0.045, RMSEA = 0.079. Overall, these values suggest an acceptable model fit, as most of the indices meet recommended thresholds. However, some improvement could be made to the NFI = 0.898 and GFI = 0.835 model fit values. Overall, the fitting indicators of this model are acceptable.

Consistent with FLMS, all items of the AI Technology Acceptance for Foreign Language Learning Scale (ATA-FLLS) were found to follow a normal distribution, and collinearity testing yielded favorable results. Accordingly, confirmatory factor analysis was performed on ATAS using Amos, with the path diagram results presented in Figure 3.

For the AI Technology Acceptance for Foreign Language Learning Scale (ATA-FLLS) (Table 8), the model was considered an excellent fit to the data, χ²/df = 3.519, CFI = 0.961, TLI = 0.955, RMR = 0.019, NFI = 0.947, GFI = 0.903, and RMSEA = 0.068. All indices met the suggested thresholds (Byrne, 1994; Hu & Bentler, 1999), suggesting a good fit for the ATA construct.

Consistent with FLMS and ATA-FLLS, confirmatory factor analysis was also conducted on the Foreign Language Proficiency Scale (FLPS) using Amos, with the path diagram presented in Figure 4.

The Foreign Language Proficiency Scale (FLPS) (Table 9) reported the following fit indices: CFI = 0.931, TLI = 0.921, RMR = 0.040, NFI = 0.916, and RMSEA = 0.086. While most fit indices (CFI, TLI, RMR, and NFI) meet acceptable requirements, the differences between the χ²/df, GFI, and RMSEA values and the recommended values are very small. Especially, the RMSEA = 0.086, which is the primary fitting index, has a gap of only 0.006 from the recommended value. It falls within the range of 0.08 to 0.1. MacCallum et al. (1996) pointed out that RMSEA between 0.05 and 0.10 indicates fair fit, while > 0.10 indicates poor fit. Hu and Bentler (1999) and other studies have expanded the threshold, suggesting that RMSEA between 0.08 and 0.10 is moderately fit (neither good nor bad), and implying that > 0.10 is unacceptable. Overall, the model can be considered acceptable.

Correlation Analysis

Correlation analysis serves as the cornerstone for the construction of Structural Equation Modeling (SEM) models. In the realm of SEM, parameter estimation and model fitting are generally carried out based on covariance matrices or correlation matrices. Initial correlation analysis is instrumental in discerning the linear relationships among observed variables. This provides guidance for the definition of latent variables and the selection of indicators (Hair et al., 2021).

Furthermore, in the domain of psychology, the classification of correlation strength often draws on Cohen’s (1988) guidelines. Within the context of Pearson’s and Spearman’s correlation coefficients, a correlation coefficient (r/ρ) with an absolute value ranging from 0.1 to 0.3 is considered to represent a weak correlation. When the absolute value of the coefficient falls between 0.3 and 0.5, it is regarded as indicative of a moderate correlation. Meanwhile, a coefficient with an absolute value of 0.5 or higher is recognized as denoting a strong correlation.

Figure 5 presents the correlations between FLM, ATA, and FLP. The correlations among the three factors were obtained by calculating the correlation coefficients using Spearman’s method on the means of the items within each factor.

Figure 5.

Heatmap of correlation coefficients of main study variables (N = 548).

Table 10 shows significant positive correlations between all study variables (p < .01). It is worth noting that the correlation coefficient between ATA and FLM is 0.570, which is greater than 0.50, indicating a strong correlation. Additionally, the correlation coefficient between FLM and FLP is 0.443, and that between FLP and ATA is 0.342, both reaching a moderately high level of correlation. This lays a solid foundation for the subsequent SEM modeling and analysis, which can preliminarily respond to H1 and H2.

Table 10.

The Correlation Among FLM, ATA and FLP (N = 548).

Construct	FLM	ATA	FLP
FLM	1^***	0.570^***	0.443^***
ATA	0.570^***	1^***	0.342^***
FLP	0.443^***	0.342^***	1^***

Note. FLM = Foreign Language Motivation; ATA = AI Technology Acceptance for Foreign Language Learning; FLP = Foreign Language Proficiency.

***

represent p levels of 1%.

Figure 6 depicts the correlations between LN, AM, SE, NM, IR, SM, and PU, PEOU, ATU, BI. The correlations among these factors were derived by computing the correlation coefficients of the means of the items within each factor using the Spearman’s method.

Figure 6.

Heatmap of correlation coefficients of all factors (N = 548).

Table 11 shows that, except for NM, all other factors of FLM and ATA manifested significant positive correlations (p < .01). It is noteworthy that the correlation coefficients between LN and PU, PEOU, ATU, BI fell within the range of 0.526 - 0.562. Since these values were greater than 0.50, they indicated a strong correlation.

Table 11.

The Correlation Between Factors of ATA and FLM (N = 548).

Factor	LN	AM	SE	NM	IR	SM
PU	0.526^***	0.447^***	0.293^***	0.077	0.522^***	0.462^***
ATU	0.562^***	0.485^***	0.321^***	0.054	0.522^***	0.502^***
PEOU	0.538^***	0.470^***	0.379^***	0.001	0.476^***	0.437^***
BI	0.536^***	0.480^***	0.329^***	0.022	0.484^***	0.473^***

Note. LN = Learning Needs; AM = Achievement Motivation; SE = Self-efficacy; IR = Identified Regulation; NM = Negative Motivation; SM = Situational Motivation; PU = Perceived Usefulness; PEOU = Perceived Ease of Use; ATU = Attitude Toward Using; BI = Behavioral Intention.

***

represent p levels of 1%.

Figure 6 and Table 12 present the correlations between LN, AM, SE, NM, IR, SM, and Listening, Speaking, Reading, Writing. The correlations between those factors were obtained by calculating the correlation coefficients using Spearman’s method on the means of the items within each factor. Except for NM, the correlation coefficients between the other factors of FLM and FLP exhibit significant positive correlations (p < .01). Simultaneously, the correlation coefficients of LN, AM, and SE with Listening, Speaking, Reading, and Writing are in the range of 0.3 to 0.5, suggesting a moderate level of correlation.

Table 12.

The Correlation Between Factors of FLM and FLP (N = 548).

Factor	LN	AM	SE	NM	IR	SM
Listening	0.322^***	0.398^***	0.464^***	0.102^**	0.230^***	0.251^***
Speaking	0.287^***	0.353^***	0.429^***	0.051	0.209^***	0.224^***
Reading	0.317^***	0.385^***	0.431^***	0.044	0.235^***	0.261^***
Writing	0.336^***	0.411^***	0.498^***	0.057	0.268^***	0.243^***

Note. LN = Learning Needs; AM = Achievement Motivation; SE = Self-efficacy; IR = Identified Regulation; NM = Negative Motivation; SM = Situational Motivation.

***

, ** represent p levels of 1% and 5%, respectively.

Figure 6 and Table 13 present the correlations between PU, PEOU, ATU, BI, and Listening, Speaking, Reading, Writing. The correlations between those factors were obtained by calculating the correlation coefficients using Spearman’s method on the means of the items within each factor. The correlation coefficients between each factor of ATA and FLP range from weak (0.249) to moderate (0.346), indicating a statistically significant but modest association (p < .01).

Table 13.

The Correlation Between Factors of FLP and ATA (N = 548).

Factors	Listening	Speaking	Reading	Writing
PU	0.319^***	0.276^***	0.346^***	0.271^***
PEOU	0.286^***	0.249^***	0.301^***	0.264^***
ATU	0.298^***	0.271^***	0.317^***	0.269^***
BI	0.278^***	0.251^***	0.281^***	0.272^***

Note. PU = Perceived Usefulness; PEOU = Perceived Ease of Use; ATU = Attitude Toward Using; BI = Behavioral Intention.

***

represent p levels of 1%.

SEM Model Fit and Path Indices

The fit of the structural models was assessed using several fit indices presented in Table 14. The main variables (FLM, ATA, and FLP) structural equation model (SEM) for all constructs exhibited a relatively poor fit (χ²/df = 5.059; GFI = 0.708; CFI = 0.751; TLI = 0.743; RMSEA = 0.086). Conversely, the multi-factor SEM demonstrated an excellent fit to the data (χ²/df = 2.680; GFI = 0.850; CFI = 0.900; TLI = 0.894; RMSEA = 0.055). This model not only categorizes the sub-dimensions but also maintains the conceptual relationships among ATA, FLM, and FLP. Additionally, the model fit coefficients span from acceptable to excellent.

Table 14.

Model Fit Indices for SEM of All constructs (N = 548).

Model	χ²	df	χ²/df	GFI	CFI	SRMR	TLI	RMSEA
Main variables	10,503.029	2,076	5.059	0.708	0.751	0.075	0.743	0.086
Multi-factors	5,384.057	2,009	2.680	0.850	0.900	0.049	0.894	0.055

Note. Bootstrap = 5,000; CI = 95%.

Despite the fact that the fit indices of the main variables’ structural equation model (Figure 7, Table 15) are less than satisfactory and we are unable to draw valid conclusions from its outcomes, as a simplified form of the multi-factor model, the research results obtained from this model still possess a certain degree of supplementary reference significance for the multi-factor model. (Hair et al., 2019; Kline, 2023).

Figure 7.

Mediation analysis path diagram (main variables, N = 548).

Table 15.

Mediation Path Analysis Results (Main Variables, N = 548).

Path	β	Mean	STDEV	T	p	Decision
ATA → FLM	.586	0.588	0.043	13.702	.000^***	Nil
FLM → FLP	.391	0.394	0.049	7.947	.000^***	Nil
ATA → FLP (Direct)	.122	0.124	0.048	2.526	.012^*	Nil
ATA → FLM → FLP (Indirect)	.229	0.232	0.031	7.376	.000^***	Nil
ATA → FLP (Total)	.352	0.355	0.043	8.118	.000^***	Nil

Note. Bootstrap = 5,000; CI = 95%.

***

, * represent p levels of 0.1% and 5%, respectively.

Table 15 (main variables) demonstrates that ATA has a positive effect on FLP (β = .122, t = 2.526, p < .05); ATA has a significant positive effect on FLM (β = .586, t = 13.702, p < .001); The indirect effect via FLM was positive (β = .229, t = 7.367, p < .001). The total effect of ATA on FLP is 0.352 (t = 8.118, p < .001). Although certain indicators have revealed the direct, indirect, and mediating effects among ATA, FLM, and FLP, owing to the unsatisfactory fit of the main variables model, this study is unable to draw corresponding conclusions based on Table 15. Nevertheless, by integrating the results in Table 10, we can obtain preliminary support for hypotheses H1, H2, and H3. But further verification through a multi-factor model is still required.

Table 16 describes the direct impact of ATA on FLM, as well as the mediating effect of ATA and FLM on FLP. Notably, the direct effect of ATA on LN is the largest, with R² = 0.337 (p < .001, F (4, 543)), and the mediating effect of ATA and FLM on Writing is the largest, with R² = 0.304 (p < .001, F (10, 537)), and effects of ATA on other factors are also significant. Furthermore, it must be elucidated that, constrained by the limitations of SPSS PROCESS in calculating mediating effects with respect to the number of factors, the factors presented in Tables 16 and 17 are computed as the means of the items within each factor. This approach overlooks the fact that the weights and loadings of each item vary. Consequently, subsequent to obtaining the regression coefficients of the model, additional in-depth analysis is indispensable.

Table 16.

Coefficients for the Mediation Effect Regression Model (Multi-factors, N = 548).

	LN	AM	SE	NM	IR	SM	Listening	Speaking	Reading	Writing
Sample	548	548	548	548	548	548	548	548	548	548
R ²	0.337	0.27	0.16	0.019	0.296	0.221	0.274	0.227	0.257	0.304
F	F (4, 543) = 69.011, 0.000^***	F (4, 543) = 50.106, 0.000^***	F (4, 543) = 25.783, 0.000^***	F (4, 543) = 2.644, 0.033^**	F (4, 543) = 57.041, 0.000^***	F (4, 543) = 38.549, 0.000^***	F (10, 537) = 20.225, 0.000^***	F (10, 537) = 15.77, 0.000^***	F (10, 537) = 18.564, 0.000^***	F (10, 537) = 23.51, 0.000^***

Note. Bootstrap = 5,000; CI = 95%. LN = Learning Needs; AM = Achievement Motivation; SE = Self-efficacy; IR = Identified Regulation; NM = Negative Motivation; SM = Situational Motivation.

***

, ** represent p levels of 1% and 5%, respectively.

Table 17.

The Mediation Effect Test.

Model	B ^c	a	p ^a	b	p ^b	B ^a×b	SE ^a×b	Z ^a×b	p ^a×b	CI 95%	B ^c'	p ^c'
PU=>IR=>Listening	.247	0.23	0.001^***	−0.238	0.001^***	−0.055	0.025	−2.233	.026^**	−0.11, −0.013	0.283	0.001^***
PEOU=>SE=>Listening	.138	0.434	0.000^***	0.374	0.000^***	0.163	0.044	3.723	.000^***	0.088, 0.26	−0.027	0.727
PU=>IR=>Speaking	.161	0.23	0.001^***	−0.213	0.007^***	−0.049	0.026	−1.877	.061^*	−0.114, −0.01	0.206	0.027^**
PEOU=>SE=>Speaking	.189	0.434	0.000^***	0.366	0.000^***	0.159	0.044	3.645	.000^***	0.083, 0.256	0.024	0.770
PU=>IR=>Reading	.203	0.23	0.001^***	−0.149	0.032^**	−0.034	0.021	−1.663	.097^*	−0.089, −0.004	0.243	0.003^***
PEOU=>SE=>Reading	.205	0.434	0.000^***	0.312	0.000^***	0.135	0.037	3.7	.000^***	0.073, 0.216	0.047	0.514
PEOU=>SE=>Writing	.196	0.434	0.000^***	0.394	0.000^***	0.171	0.043	3.98	.000^***	0.094, 0.263	−0.01	0.892

Note. Bootstrap = 5,000; CI = 95%; PU = Perceived Usefulness; PEOU = Perceived Ease of Use; SE = Self-efficacy; IR = Identified Regulation; a, b, c and c' and their superscripts represent the corresponding paths in the mediation model.

***

, **, * represent p levels of 1%, 5%, and 10%, respectively.

Table 17 shows several moderate pathways that merit attention. These results are derived from SPSS PROCESS analysis. In particular, the moderating effect of SE on the relationship between PEOU and FLM is highly pronounced. Moreover, IR appears to exert a competitive effect on the association between PU and FLM. Although the significance levels are not particularly high, with p-values ranging from .026 to .097, this aspect should be considered during further in-depth analysis.

Figure 8 illustrates the detailed mediation path diagram of ATAS, FLMS, and FLPS, which was analyzed using partial least squares structural equation modeling (PLS-SEM) and generated via SmartPLS 4. The correlation indices corresponding to the detailed paths are provided in Tables 18 and 19.

Figure 8.

Mediation analysis path diagram (multi-factors, N = 548).

Table 18.

Direct Path Analysis Results (Multi-Factors, N = 548).

Path	β	Mean	SD	T	p	Decision
PU → NM	.196	0.199	0.070	2.817	.005^**	Supported
PU → IR	.220	0.224	0.084	2.613	.009^**	Supported
PU → Listening	.233	0.239	0.092	2.537	.011^*	Supported
PU → Reading	.214	0.220	0.099	2.158	.031^*	Supported
PEOU → AM	.254	0.252	0.074	3.420	.001^**	Supported
PEOU → SE	.375	0.374	0.080	4.715	.000^***	Supported
PEOU → NM	−.169	−0.166	0.070	2.420	.016^*	Supported
PEOU → LN	.238	0.239	0.072	3.310	.001^**	Supported
BI → AM	.216	0.215	0.078	2.770	.006^**	Supported
BI → LN	.200	0.197	0.076	2.618	.009^**	Supported
SE → Listening	.393	0.392	0.058	6.790	.000^***	Supported
SE → Reading	.346	0.348	0.055	6.294	.000^***	Supported
SE → Speaking	.361	0.361	0.059	6.142	.000^***	Supported
SE → Writing	.425	0.428	0.054	7.812	.000^***	Supported
IR → Listening	−.190	−0.189	0.059	3.209	.001^**	Supported
IR → Reading	−.128	−0.127	0.061	2.109	.035^*	Supported
IR → Speaking	−.168	−0.168	0.063	2.676	.007^**	Supported

Note. Bootstrap = 5,000; CI = 95%. PU = Perceived Usefulness; PEOU = Perceived Ease of Use; BI = Behavioral Intention; LN = Learning Needs; AM = Achievement Motivation; SE = Self-efficacy; IR = Identified Regulation; NM = Negative Motivation.

***

, **, * represent p levels of 0.1%, 1%, and 5%, respectively.

Table 19.

Mediation Path Analysis Results (Multi-Factors, N = 548).

Path	β	Mean	SD	T	p	Decision
PEOU → Listening (Direct)	−.027	−0.027	0.071	0.381	.703	Nil
PEOU → Reading (Direct)	.050	0.050	0.074	0.674	.501	Nil
PEOU → Speaking (Direct)	.021	0.023	0.076	0.281	.416	Nil
PEOU → Writing (Direct)	−.008	−0.009	0.074	0.110	.912	Nil
PU → Listening (Direct)	.233	0.239	0.092	2.537	.011^*	Supported
PEOU → SE → Listening (Indirect)	.147	0.147	0.039	3.750	.000^***	Supported
PEOU → SE → Reading (Indirect)	.130	0.130	0.035	3.721	.000^***	Supported
PEOU → SE → Speaking (Indirect)	.135	0.135	0.037	3.680	.000^***	Supported
PEOU → SE → Writing (Indirect)	.159	0.160	0.040	3.959	.000^***	Supported
PU → IR → Listening (Indirect)	−.042	−0.042	0.021	2.012	.044^*	Supported
PEOU → Listening (Total)	.122	0.124	0.071	1.716	.086	Nil
PEOU → Reading (Total)	.201	0.201	0.074	2.713	.007^**	Supported
PEOU → Speaking (Total)	.163	0.165	0.074	2.205	.028^*	Supported
PEOU → Writing (Total)	.185	0.184	0.078	2.377	.018^*	Supported
PU → Listening (Total)	.206	0.211	0.095	2.172	.030*	Supported

Note. Bootstrap = 5,000; CI = 95%. PU = Perceived Usefulness; PEOU = Perceived Ease of Use; SE = Self-Esteem; IR = Identified Regulation.

***

, **, * represent P levels of 0.1%, 1%, and 5%, respectively.

Table 18 presents the more significant path analysis results. On the one hand, the factors PU, PEOU, and BI of ATA can positively predict the factors LN, AM, SE, and IR of FLM, but the relationship with NM remains unclear. Combining Tables 10, 11, 15, and 17, we can conclude that H1 is supported. On the other hand, the factor PU of ATA can positively predict Listening and Reading. Combining Tables 10, 13, 15, and 17, although some factors of ATA show a competitive effect on FLP, the overall results support H2. Those results indicate that when people feel that the acceptance of technology is more useful and easier to use in learning a foreign language, their motivation for learning the foreign language increases, and it also promotes the development of foreign language proficiency.

Table 19 illustrates the indirect effects of the interactions among ATA, FLM, and FLP. The results presented in the table suggest that between PEOU and FLP, SE serves as the sole mediator with a positive effect and robust statistical significance. More precisely, PEOU exerts a substantial positive indirect influence on FLP via SE, which indicates that when the system is more user-friendly, self-efficacy (SE) enhances the four dimensions of foreign language learning (listening, speaking, reading, and writing), thus boosting FLP. Furthermore, PU has a negative indirect impact on Listening through IR. However, the magnitude of the IR effect is considerably smaller. This implies that although PU inhibits Listening through IR, this effect is not highly significant (p = .044). In contrast to the highly significant positive effect of SE (p < .001), the inhibitory effect of IR is rather negligible. Significantly, LN, AM, IR, NM, and SM do not appear to significantly mediate the relationship between ATA and FLP, as their indirect effects lack statistical significance. Based on the results presented in Tables 10, 15, 16, 17, and 19, it can be concluded that H3 is supported. In particular, this indicates that enhancing self-efficacy (SE) is a more practical method for improving FLP.

Variance Accounted For

VAF (Variance Accounted For) is mainly used in mediation effect analysis to assess the contribution ratio of the mediating variable in explaining the total effect of the independent variable on the dependent variable. Furthermore, VAF < 20% means no mediation effect, 20% < VAF <80% means partial mediation, and VAF > 80% means complete mediation (Vinzi, 2010).

Table 20 shows the mediating role of SE in the relationship between PEOU and FLP. According to Vinzi (2010), SE completely mediates the relationship between PEOU and Speaking (VAF = 82.822%), also completely mediates the relationship between PEOU and Speaking (VAF = 85.946%). Meanwhile, it partially mediates the relationship between PEOU and Reading. Notably, the variance accounted for (VAF) of self-efficacy (SE) exceeds 100%. We examined the effect of perceived ease of use on listening in Table 19, and the value is −0.008, which indicates that the model is a suppression mediation model, meaning that self-efficacy greatly promotes the improvement of students’ foreign language ability through perceived ease of use, while suppressing the negative impact of perceived ease of use on listening. Furthermore, self-efficacy (SE) plays a competitive mediating role between perceived usefulness (PU) and listening (VAF = −20.388%). Specifically, the total effect of perceived usefulness (PU) on listening is positive and significant (β = .206, p < .05), indicating that learners who perceive GenAI as useful are more inclined to engage in listening activities. However, after introducing self-efficacy as a mediating variable, the direct effect increases (β = .233, p < .05), while there is a significant negative indirect effect (β = .042, p < .05).

Table 20.

The Analysis Results of VAF.

Path (Mediator SE)	β (Indirect)	β (Total)	VAF
PEOU → Listening	.147	.122	120.492%
PEOU → Speaking	.135	.163	82.822%
PEOU → Reading	.130	.201	64.677%
PEOU → Writing	.159	.185	85.946%
PU → Listening	−.042	.206	−20.388%

Note. PU = Perceived Usefulness; PEOU = Perceived Ease of Use.

Discussion

This research explored the influence of generative artificial intelligence (GenAI) on foreign language learning motivation (FLM) and foreign language proficiency (FLP) from the perspective of the Technology Acceptance Model (TAM). The research results demonstrate that the usage and attitudes of Chinese university students towards generative artificial intelligence (GenAI) can positively predict foreign language learning motivation (FLM), particularly self-efficacy (SE). The research findings indicate that there is a moderate to strong positive correlation between AI Technology Acceptance for Foreign Language Learning (ATA-FLLS), Foreign Language Motivation (FLM), and Foreign Language Proficiency (FLP). These findings echo the conclusion of Honarzad and Rassaei (2019), who pointed out that learners’ motivation, self-efficacy, and self-reliance are largely influenced by technology-enhanced language learning tasks. However, this research places greater emphasis on exploring the influence of generative artificial intelligence (GenAI) on users, with a particular focus on investigating the mediating effects among multiple factors.

The findings of this study suggest that foreign language motivation (FLM) can serve as a significant mediator in the relationship between AI Technology Acceptance for Foreign Language Learning (ATA-FLLS) and foreign language proficiency (FLP). The mediating effect in this study was explored similarly in J. Huang and Mizumoto (2024), who investigated the interaction between the L2 Motivational Self-System (L2MSS) and the Technology Acceptance Model (TAM) among 35 Japanese students after the introduction of ChatGPT, revealing a positive cycle where OL2 promotes motivation, technology acceptance, and language practice in a ChatGPT-assisted EFL environment. However, our research focuses more on the impact of proprietary AI models on second language acquisition and has a larger sample size (N = 548).

In light of the path analysis results, it can be postulated that self-efficacy (SE) serves as a potent mediator in the relationship between perceived ease of use (PEOU) and foreign language proficiency (FLP, encompassing listening, speaking, reading, and writing skills). In some instances, the mediation is complete (e.g., the link from PEOU to Speaking), while in others, it exhibits an inhibitory mediation effect (e.g., the connection from PEOU to Listening). Simultaneously, perceived usefulness (PU) functions as a competitive mediator in the association between identified regulation (IR) and listening. This indicates that the impact of foreign language proficiency (FLP) on the GenAI technology acceptance model (TAM) is accounted for by self-efficacy (SE), which is consonant with the research conducted by Kittredge et al. (2025) and Zhang et al. (2025). However, differing from the research results of Zheng et al. (2024), their findings indicate that SDT motivation might more effectively predict the relationship between behavioral intention (BI) and facilitating conditions. This disparity could potentially stem from differences in the theoretical models employed and the nature of generative artificial intelligence (GenAI) utilized. Nonetheless, both studies can establish the positive influence of motivation. Furthermore, this competitive mediation implies that although perceived usefulness (PU) directly boosts listening engagement, it might indirectly undermine it by diminishing self-efficacy (SE). This could be attributed to the fact that when Chinese university students utilize GenAI for foreign language listening learning, the requirements of complex and long-term listening tasks prove to be overly formidable.

These findings further suggest that the utilization and attitudes of Chinese university students towards generative artificial intelligence (GenAI) exert a substantial influence on their learning motivation, which subsequently impacts their academic attainments. This phenomenon may not be confined solely to the domain of foreign language learning but could potentially be extended to other fields.

Theory Implications

This research promotes the Technology Acceptance Model (TAM) by integrating the influence of generative artificial intelligence (GenAI) on foreign language motivation (FLM) and foreign language proficiency (FLP). It underscores the role of self-efficacy (SE) as a crucial mediator between perceived ease of use (PEOU) and components of FLP, showing different effects. For instance, it acts as a full mediator in speaking skills and a suppressive mediator in listening skills. This finding extends previous works (Honarzad & Rassaei, 2019; J. Huang & Mizumoto, 2024), emphasizing the role of GenAI in motivation, which differs from the predictive power of SDT motivation as discussed by Zheng et al. (2024). This indicates the subtle interaction between TAM factors and psychological mediators, which may extend beyond language learning to other educational domains, enriching the theory of technology-enhanced learning.

Practice Implications

Educators and policymakers are advised to advocate for the application of GenAI tools within foreign language curricula. This initiative aims to boost students’ motivation and proficiency levels. Emphasis should be placed on enhancing self-efficacy by means of user-friendly interfaces. Training programs can be designed to address the issue of perceived usefulness. This approach is intended to alleviate the suppressive effects in areas like listening and to encourage a well-balanced integration of AI in language learning. In the context of Chinese universities, the implementation of large-scale proprietary AI models may contribute to improved foreign language proficiency (FLP) outcomes. GenAI developers should give precedence to features that facilitate long-term language learning tasks and minimize the cognitive burden on users. These insights advocate for the application of GenAI in language education and can also be extended to other fields, such as language cognition, aphasia-assisted intervention, and intervention in depression, which have even more profound significance.

Limitations

This study has limitations, including that the research subjects were limited to Chinese university students, thus restricting its generalizability. Self-reported language proficiency and language learning motivation may be subjective. Additionally, this study did not fully consider the differences between GenAI specifically designed for foreign language learning and general-purpose GenAI, as well as the varying degrees of exposure to AI technology among university students, which could affect their attitudes towards GenAI. Future research should strive to overcome these limitations to reach more consistent conclusions regarding the application of generative artificial intelligence (GenAI) in foreign language learning.

Conclusion

This research is grounded in the Technology Acceptance Model (TAM) to explore the impacts of generative artificial intelligence (GenAI) on foreign language motivation (FLM) and foreign language proficiency (FLP) among Chinese university students. The research results demonstrate that students’ utilization and attitudes towards GenAI positively predict FLM, especially self-efficacy (SE). There exists a moderate to strong positive correlation among AI Technology Acceptance for Foreign Language Learning (ATA-FLLS), FLM, and FLP. FLM serves as a notable mediator between ATA and FLP, highlighting the bridge of motivation between technology acceptance and language skills. Path analysis shows that self-efficacy (SE) acts as a potent mediator between perceived ease of use (PEOU) and various components of FLP. For example, it completely mediates the aspect of speaking and has a suppressive mediation effect on listening. Moreover, perceived usefulness (PU) functions as a competitive mediator between identified regulation (IR) and listening, indicating that PU directly enhances listening but may indirectly reduce it by decreasing SE. This might be attributed to the challenges posed by complex and long-term listening tasks when using GenAI. Overall, these research findings suggest that attitudes towards generative artificial intelligence (GenAI) considerably influence learning motivation, which in turn impacts academic achievements. The potential applications of this study are not confined to foreign language learning but may also be extended to other educational domains.

Footnotes

Acknowledgements

Fei Qin was responsible for writing and analyzing the data, Bin Liu & Weibin Li were responsible for providing guidance on the framework and supervising the entire process, and Huiyuan Li & Jiayin Li were responsible for collecting the data, all authors approved the final manuscript.

ORCID iDs

Fei Qin

Bin Liu

Weibin Li

Ethical Considerations

This study was conducted after ethical approval from Liaocheng University (HE2024111901). Human participants were engaged through questionnaire surveys. The investigation strictly adhered to the ethical principles outlined in the Declaration of Helsinki (1964) and its subsequent amendments. Written informed consent was obtained from all participants prior to their involvement.

Consent to Participate

Informed consent was obtained from all participants before data collection. Participants indicated their consent by completing an online questionnaire that included an informed consent form, which explained the purpose of the study, the voluntary nature of participation, and the guarantee of data confidentiality. Completion of the questionnaire was regarded as obtaining informed consent.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Humanities and Social Science Youth Fund of Ministry of Education of China (25YJC740035); Shandong Province Humanities and Social Sciences Specialized Think Tank Key Project (2023ZKZD064); National Social Science Fund Funding Project for Thematic Academic Activities of Academic Societies of Social Sciences (23STA004); Liaocheng University Doctoral Research Initiation Fund for Social Sciences (321052324 & 321051733).

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The data underlying this study are available from the corresponding author upon reasonable request, subject to approval by all co-authors and compliance with ethical review procedures.

References

Adeshola

Adepoju

A. P.

(2024). The opportunities and challenges of ChatGPT in education. Interactive Learning Environments, 32(10), 6159–6172. https://doi.org/10.1080/10494820.2023.2253858

Almelhes

S. A.

(2023). A review of artificial intelligence adoption in second-language learning. Theory and Practice in Language Studies, 13(5), 1259–1269. https://doi.org/10.17507/tpls.1305.21

Baidoo-Anu

Ansah

L. O.

(2023). Education in the era of generative artificial intelligence (AI): Understanding the potential benefits of ChatGPT in promoting teaching and learning. Journal of AI, 7(1), 52–62. https://doi.org/10.61969/jai.1337500

Bandura

(1982). Self-efficacy mechanism in human agency. American Psychologist, 37(2), 122–147. https://doi.org/10.1037/0003-066X.37.2.122

Baumgartner

Weijters

Pieters

(2021). The biasing effect of common method variance: Some clarifications. Journal of the Academy of Marketing Science, 49(2), 221–235. https://doi.org/10.1007/s11747-020-00766-8

Byrne

B. M.

(1994). Structural equation modeling with EQS and EQS/Windows: Basic concepts, applications, and programming. SAGE.

Cohen

(1988). Set correlation and contingency tables. Applied Psychological Measurement, 12(4), 425–434. https://doi.org/10.1177/014662168801200410

Creely

(2024). Exploring the role of generative AI in enhancing language learning: Opportunities and challenges. International Journal of Changes in Education, 1(3), 158–167. https://doi.org/10.47852/bonviewIJCE42022495

Davis

F. D.

(1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319–340. https://doi.org/10.2307/249008

10.

Fornell

Larcker

D. F.

(1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50. https://doi.org/10.1177/002224378101800104

11.

Gao

Wang

(2024). Exploring EFL university teachers’ beliefs in integrating ChatGPT and other large language models in language education: A study in China. Asia Pacific Journal of Education, 44(1), 29–44. https://doi.org/10.1080/02188791.2024.2305173

12.

Goretzko

Siemund

Sterner

(2024). Evaluating model fit of measurement models in confirmatory factor analysis. Educational and Psychological Measurement, 84(1), 123–144. https://doi.org/10.1177/00131644231163813

13.

Hair

J. F.

Jr. Hult

G. T. M.

Ringle

C. M.

Sarstedt

(2014). A primer on partial least squares structural equation modeling (PLS-SEM). Sage Publications.

14.

Hair

J. F.

Hult

G. T. M.

Ringle

C. M.

Sarstedt

Danks

N. P.

Ray

(2021). An introduction to structural equation modeling. In Partial least squares structural equation modeling (PLS-SEM) using R: A workbook (pp. 1–29). Springer International Publishing. https://doi.org/10.1007/978-3-030-80519-7_1

15.

Hair

J. F.

Risher

J. J.

Sarstedt

Ringle

C. M.

(2019). When to use and how to report the results of PLS-SEM. European Business Review, 31(1), 2–24. https://doi.org/10.1108/EBR-11-2018-0203

16.

Henseler

Ringle

C. M.

Sarstedt

(2015). A new criterion for assessing discriminant validity in variance-based structural equation modeling. Journal of the Academy of Marketing Science, 43, 115–135. https://doi.org/10.1007/s11747-014-0403-8

17.

Honarzad

Rassaei

(2019). The role of EFL learners’ autonomy, motivation and self-efficacy in using technology-based out-of-class language learning activities. The JALT CALL Journal, 15(3), 23–42. https://doi.org/10.29140/jaltcall.v15n3.170

18.

Hsiao

J. C.

Chang

J. S.

(2023). Enhancing EFL reading and writing through AI-powered tools: Design, implementation, and evaluation of an online course. Interactive Learning Environments, 32(9), 4934–4949. https://doi.org/10.1080/10494820.2023.2207187

19.

Bentler

P. M.

(1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55. https://doi.org/10.1080/10705519909540118

20.

Huang

Wang

Zhang

(2024). Modelling generative AI acceptance, perceived teachers’ enthusiasm and self-efficacy to English as a foreign language learners’ well-being in the digital era. European Journal of Education, 59(4), e12770. https://doi.org/10.1111/ejed.12770

21.

Huang

H. W.

(2005). The relationship between learning motivation and speaking anxiety among EFL non-English major freshmen in Taiwan (Master’s thesis). Chaoyang University of Technology, Taiwan.

22.

Huang

Mizumoto

(2024). Examining the relationship between the L2 motivational self-system and technology acceptance model post ChatGPT introduction and utilization. Computers and Education: Artificial Intelligence, 7, 100302. https://doi.org/10.1016/j.caeai.2024.100302

23.

Jansen

Höft

Bahr

Fleckenstein

Möller

Köller

Meyer

(2024). Empirische Arbeit: Comparing generative AI and expert feedback to students’ writing: Insights from student teachers. Psychologie in Erziehung und Unterricht, 71(2), 80–92.

24.

Jeong

(2020). The social brain of language: Grounding second language learning in social interaction. npj Science of Learning, 5(1), 8. https://doi.org/10.1038/s41539-020-0068-7

25.

Kasneci

Seßler

Küchemann

Bannert

Dementieva

Fischer

Kasneci

(2023). ChatGPT for good? On opportunities and challenges of large language models for education. Learning and Individual Differences, 103, 102274. https://doi.org/10.1016/j.lindif.2023.102274

26.

Kim

H. I.

(2020). The effects of experience abroad, English self-efficacy, and proficiency on the L2 motivational selves: A study of Korean EFL university students. Innovation in Language Learning and Teaching, 14(3), 259–272. https://doi.org/10.1080/17501229.2019.1579218

27.

Kittredge

A. K.

Hopman

E. W.

Reuveni

Dionne

Freeman

Jiang

(2025). Mobile language app learners’ self-efficacy increases after using generative AI. Frontiers in Education, 10, 1499497. https://doi.org/10.3389/feduc.2025.1499497

28.

Kline

R. B.

(2023). Principles and practice of structural equation modeling (5th ed.). Guilford Press.

29.

Law

(2024). Application of generative artificial intelligence (GenAI)in language teaching and learning: A scoping literature review. Computers and Education Open, 6, 100174. https://doi.org/10.1016/j.caeo.2024.100174

30.

Lievertz

(2019). Artificial intelligence in education. In Artificial intelligence and machine learning for business for non-engineers (pp. 125–140). CRC Press. https://doi.org/10.1201/9780367821654

31.

Linh

P. M.

Starčič

A. I.

T. T.

(2022). Challenges and opportunities of education in the COVID-19 pandemic: Teacher perception on applying AI chatbot for online language learning. In International conference on innovative technologies and learning (pp. 501–513). Springer International Publishing. https://doi.org/10.1007/978-3-031-15273-3_55

32.

MacCallum

R. C.

Browne

M. W.

Sugawara

H. M.

(1996). Power analysis and determination of sample size for covariance structure modeling. Psychological Methods, 1(2), 130–149. https://doi.org/10.1037/1082-989X.1.2.130

33.

Marikyan

Papagiannidis

(2024). Technology acceptance model: A review. TheoryHub.

34.

Marcoulides

K. M.

Raykov

(2019). Evaluation of variance inflation factors in regression models using latent variable modeling methods. Educational and Psychological Measurement, 79(5), 874–882. https://doi.org/10.1177/0013164418817803

35.

Maruf

T. I.

Hazilah

Haque

A. K. M.

Suharni , . (2021). Factors affecting attitudes towards using ride-sharing apps. International Journal of Business, Economics and Law, 25(2), 60–70.

36.

Mishra

Pandey

C. M.

Singh

Gupta

Sahu

Keshri

(2019). Descriptive statistics and normality tests for statistical data. Annals of Cardiac Anaesthesia, 22(1), 67–72. https://doi.org/10.4103/aca.ACA_157_18

37.

Nilashi

Abumalloh

R. A.

(2025). I-TAM: A model for immersive technology acceptance. Education and Information Technologies, 30(6), 7689–7717. https://doi.org/10.1007/s10639-024-13080-5

38.

O’Brien

R. M.

(2007). A caution regarding rules of thumb for variance inflation factors. Quality & Quantity, 41, 673–690. https://doi.org/10.1007/s11135-006-9018-6

39.

OpenAI. (2023). ChatGPT. https://openai.com/blog/chatgpt

40.

Pack

Maloney

(2023). Using generative artificial intelligence for language education research: Insights from using OpenAI’s ChatGPT. TESOL Quarterly, 57(4), 1571–1582. https://doi.org/10.1002/tesq.3253

41.

Vallerand

R. J.

(1997). Toward a hierarchical model of intrinsic and extrinsic motivation. In Advances in experimental social psychology (Vol. 29, pp. 271–360). Academic Press. https://doi.org/10.1016/S0065-2601(08)60019-2

42.

Vinzi

V. E.

(2010). Handbook of partial least squares. Springer.

43.

Wang

Pan

Ortega-Martín

J. L.

(2024). The predicting role of EFL students’ achievement emotions and technological self-efficacy in their technology acceptance. The Asia-Pacific Education Researcher, 33(4), 771–782. https://doi.org/10.1007/s40299-023-00750-0

44.

Wang

Xue

(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

45.

Wei

(2023). Artificial intelligence in language instruction: Impact on English learning achievement, L2 motivation, and self-regulated learning. Frontiers in Psychology, 14, 1261955. https://doi.org/10.3389/fpsyg.2023.1261955

46.

Wang

(2024). “To use or not to use?” A mixed-methods study on the determinants of EFL college learners’ behavioral intention to use AI in the distributed learning context. International Review of Research in Open and Distributed Learning, 25(3), 158–178. https://doi.org/10.19173/irrodl.v25i3.7708

47.

Zhang

(2021). L2 learner cognitive psychological factors about artificial intelligence writing corrective feedback. English Language Teaching, 14(10), 70–83.

48.

Xia

Yang

(2019). RMSEA, CFI, and TLI in structural equation modeling with ordered categorical data: The story they tell depends on the estimation methods. Behavior Research Methods, 51, 409–428. https://doi.org/10.3758/s13428-018-1055-2

49.

Zhang

Nie

Fan

Liu

(2025). Exploring the dynamics of artificial intelligence literacy on English as a foreign language learners’ willingness to communicate: The critical mediating roles of artificial intelligence learning self-efficacy and classroom anxiety. Behavioral Sciences, 15(4), 523. https://doi.org/10.3390/bs15040523

50.

Zheng

Wang

Liu

K. S. X.

Jiang

M. Y. C.

(2024). Examining the moderating effect of motivation on technology acceptance of generative AI for English as a foreign language learning. Education and Information Technologies, 29(17), 23547–23575. https://doi.org/10.1007/s10639-024-12763-3

The Mediating Role of Motivation Between Generative Artificial Intelligence and Foreign Language Proficiency: A Technology Acceptance Model Perspective

Abstract

Plain Language Summary

Keywords

Introduction

Literature Review

Technology Acceptance Model

Generative AI Acceptance in Foreign Language Learning

Role of Students’ Motivation in Foreign Language Learning

UNIPUS AI: Foreign Language Learning Tool

Rationale

Hypotheses

Methods

Research Design

Sampling and Data Collection

Measurement

Instrumentation

Results

Demographic Analysis

Reliability Analysis

Convergent Validity

Discriminant Validity

Common Method Variance

Assessment of Multicollinearity and Normality

Confirmatory Factor Analysis (CFA)

Correlation Analysis

SEM Model Fit and Path Indices

Variance Accounted For

Discussion

Theory Implications

Practice Implications

Limitations

Conclusion

Footnotes

Acknowledgements

ORCID iDs

Ethical Considerations

Consent to Participate

Funding

Declaration of Conflicting Interests

Data Availability Statement

References