Supporting Speaking Practice by Social Network-Based Interaction in Artificial Intelligence (AI)-Assisted Language Learning

Abstract

In recent decades, the rapid development of artificial intelligence (AI) technology has led to the increasing use of AI speaking apps in foreign language learning. This research investigates the impact of social network-based interaction on students’ English speaking practice with the assistance of AI speaking apps in China. During the summer vacation, 70 students from different Chinese universities and majors were recruited for the experiment. They were required to practice speaking skills with AI apps for five weeks and were divided into two groups. Participants in the experimental group were encouraged to engage in various interactive activities when practicing speaking with AI apps, while those in the control group were asked to use AI speaking apps without interaction. Data were collected through questionnaires and semi-structured interviews as well as pre-and post-tests. The results indicated that students generally held positive attitudes towards interactive activities when using AI apps to practice their spoken English. The finding also showed that social network-based interaction can effectively improve learners’ speaking skills in the AI context. This study contributes to the research on the implementation and promotion of AI speaking apps with social networking and extends the previous studies on network-based interaction to the AI-assisted learning environment. An investigation of interactions based on Chinese social network-based platforms such as WeChat can be further applied to other social networking platforms such as Facebook or WhatsApp in different cultural contexts for AI-assisted speaking practice.

1. Introduction

Interaction is defined as a mutual action requiring at least two objects or events [1]. Through interaction with teachers and peers, language learning is a social process in which learners acquire knowledge of the target language [2]. In the language classroom, the interaction involves the roles of teachers and students, who build a shared body of knowledge together [3]. Regarding teacher–student interactions, there are different patterns such as question-asking [4], clarification [5], directing drills [6], and corrective feedback giving [7,8]. In terms of student–student interaction, discussion [9,10], role-playing [11], and audio recording [12] are common tasks in the classroom discourse. It is widely believed that interaction is essential in the learning process [1,13,14,15]. It is considered crucial to communicative language teaching [13] and has positive effects on students’ language learning skills [14] and learning motivation [15]. With the development of mobile-assisted language learning (MALL), social network-based interaction in language learning and speaking practice has been widely investigated [16,17,18]. Accordingly, new forms of interactive tasks have been employed in the MALL context, such as likes and comments on posts [19], question-asking and answering through instant messaging [16], recording practice through voice messaging [18], and interactive videos and online quizzes in group chats [17].

Artificial intelligence (AI) is defined as computational systems that are capable of using data to perform human-like processes such as learning, self-correction, and solving complex tasks [20]. It has been widely applied in the field of education, especially in language learning [21,22]. As mobile-assisted language learning becomes more popular, AI-based speaking apps with speech evaluation systems have been employed for EFL speaking practice [23,24]. Many studies have shown that AI speaking apps could help improve speaking skills [25,26,27,28]. However, in the AI-assisted language learning context, the role of network-based interaction in speaking practice is under-investigated. Therefore, the current research aims to extend the studies from the mobile-assisted learning context to the AI-assisted learning context. Specifically, it focuses on learners’ perception of various interactive activities when practicing speaking skills with AI apps. Moreover, the effectiveness of the interaction in improving EFL speaking skills was also investigated.

2. Literature Review

2.1. Social Network-Based Interaction for Speaking Practice

Speaking skills refer to the capacity to communicate clearly by using appropriate vocabulary, grammar, and pronunciation [29], which are a crucial part of EFL teaching and learning [9]. Evaluation of EFL speaking skills includes employing language in an accurate context, requiring competence in grammar, discourse, and learning strategy [30]. Teachers’ feedback has a significant effect on this process [31]. Furthermore, many studies have found that students may achieve better speaking learning outcomes when they actively engage in an interactive atmosphere [13,32,33].

With the popularity of mobile devices, the role of mobile learning in language instruction has been highlighted in much research [34,35,36]. In the MALL context, there are a few studies focused on network-based interaction in the EFL context, exploring various aspects of EFL learning such as writing skills [37,38], listening skills [19], speaking skills [16,17], and vocabulary learning [39]. With regard to EFL listening practice, Read et al. [19] explored the role of social networking on Facebook by comparing two versions of the Audio News Trainer (ANT) app. In the experiment, 90 students were evenly divided into two groups practicing for nine weeks. Students in the experiment group were required to use the app version connected to a Facebook page while those in the control group practiced with no connection to social networks. The participants in the FB group with connections to the social network could carry out interactive activities such as viewing others’ practice and liking or commenting on other FB posts. The results showed that learners in the experimental group (FB) were more motivated to practice listening skills with the ANT app and develop learning habits better. In addition, most students enjoyed the functions of liking, commenting, and other interactive activities on Facebook.

In terms of EFL speaking practice, Tragant et al. [16] examined how using instant messages in WhatsApp groups could help with language learning beyond the classroom. In a five-week summer course at a Spanish university, teachers initiated a few learning activities, including oral tasks, in the WhatsApp group. Students were asked to engage in a series of language learning tasks in the WhatsApp group. Eight teacher-initiated tasks could be categorized into four kinds: guessing activities, question and answer, information giving, and drills. Both the on-task and off-task messages from teachers and students were analyzed to explore students’ engagement in the learning activities. The study found that task-based interaction could motivate students to engage in language practice, while informal communication was also productive in promoting learners’ practice, though it did not explore the impact of interactions on improving learners’ language abilities.

In addition to the motivational effect of social network-based interaction on speaking practice, a few studies explored how interactive activities effectively improve learners’ speaking skills. For instance, Mykytiuk et al. [17] investigated the effectiveness of Facebook-assisted interaction in developing the EFL speaking skills of undergraduate students in Ukraine. In the experiment group, the Facebook platform was incorporated into the academic course with contextual activities while traditional teaching was adopted in the control group. The interactive activities in the Facebook group included a series of language input activities and communicative output activities. In the language input activities, content-oriented input activities (introducing learning topics with pictures, interactive videos, and quizzes), speech shadowing activities (listening to the speech in a video and repeating the content), and structured output activities (filling information gaps in the materials posted on the FB group) were included. The communicative output activities also included activities for expressing ideas, role-playing (acting out an interview following a FB link), story completion (posting relevant vocabulary related to a story), and digital story narration (orally presenting ideas in posted videos). To evaluate the effectiveness of Facebook interaction, a pre-test and a post-test were conducted to measure the improvement of oral skills in terms of vocabulary and grammar, pronunciation, interactive communication, and discourse management. Furthermore, participants’ perceptions of their improvement in speaking skills were also examined based on the four criteria mentioned above. The results showed that the EG achieved significantly higher scores than the CG, thus indicating the effectiveness of the interactive activities. Additionally, based on students’ perceptions, a large number of students believed that the interactive activities on Facebook were helpful for developing their speaking skills, particularly in increasing vocabulary and grammar knowledge [17].

Meanwhile, the emergence of WeChat (a Chinese social communication app similar to WhatsApp) has not only met the need for diverse interpersonal communication and enriched the way people interact with each other but has also helped learners to improve their language skills [40,41,42,43]. In terms of EFL speaking, for example, Xu et al. [18] explored EFL learners’ perception of WeChat interaction in speaking practice. Thirty-five Chinese university students were asked to produce voice messages on WeChat on a series of video-based tasks for a semester. In the WeChat group, EFL teachers would provide individualized feedback on improving their speaking skills based on students’ oral production, which can be considered a form of corrective feedback. Data from the questionnaire, interviews, and students’ reflections showed that students had positive attitudes toward mobile-assisted feedback on WeChat. Students’ perceptions revealed that the provision of feedback could motivate them to participate in the learning activities actively, thus helping them develop their speaking skills in the long run. Nevertheless, the study did not investigate the effectiveness of mobile-assisted feedback in language assessments (pre- and post-tests). Therefore, in the current research, both learners’ perceptions and the improvement of their speaking skills are further examined.

Moreover, other services can be linked to social network platforms to provide more convenient learning tools, such as Google Docs [44]. For example, Alharbi (2020) [45] found that the real-time peer editing platform Google Docs can improve the writing skills of EFL learners by increasing their interest and providing convenient services. In terms of vocabulary learning, Liu et al., (2014) [46] explored whether adding Google Docs to learning could influence students’ self-regulated vocabulary strategy. In this study, Tencent Docs, the Chinese version of Google Docs, was also implemented along with WeChat to improve the vocabulary of learners, which is one of the aspects of speaking skills.

2.2. AI for Speaking Practice

Artificial intelligence (AI) is a means of embedding human intelligence into computer programs that can think, work, and make judgments as humans do [47]. Recently, AI has been increasingly used in the field of education [48], and the application of AI in language learning has received considerable attention [49]. Currently, powered by AI technology and automatic speech recognition (ASR) technology, there are various language learning tools for EFL speaking practice, such as ASR-based websites [25,50], intelligent personal assistants (IPAs) [51], and AI chatbots [52,53]. As mobile-assisted language learning gains popularity [54], many AI mobile apps have been developed for EFL speaking practice, such as Duolingo [55], Liulishuo [24], and EAP Talk [56], which are powered by speech evaluation technology and natural language processing. Two types of speaking tasks are common in AI apps with speech evaluation systems, which are “reading aloud” and “presentation” [54,56]. AI speaking apps provide many benefits. For instance, they can also help teachers save time, thus improving the quality of teaching to a certain extent [57]. In terms of speaking feedback, Hwang et al. [58] noted that AI could facilitate teaching and learning by providing personalized feedback to students. Different aspects of learners’ speaking performance, including pronunciation, grammar, vocabulary, and fluency, can be evaluated [59,60]. Various kinds of feedback can be provided by the AI apps, including scores for speaking performance [61], highlighted colors for correctness [62], and textual evaluation for practice suggestions [54].

Moreover, many studies have investigated the effectiveness of AI speaking apps for improving speaking skills [63,64]. For example, in Loewen et al.’s [23] research on second language acquisition, the researchers recruited a group of students (n = 9) to learn Turkish during one semester to evaluate the effectiveness of Duolingo. Based on the results of the language test at the university, the study found that speaking skills were improved with the assistance of the AI learning app. Studies also show that students had positive attitudes toward AI apps for their speaking practice [25,56,58]. For instance, Li and Zou [56] investigated users’ perceptions of utilizing AI speaking apps, including the benefits and drawbacks. In their study, 101 Chinese undergraduate students completed the questionnaire. The results indicated that most learners held a positive attitude toward using AI speaking apps in general, and they thought that AI apps could improve their speaking skills in various ways, such as pronunciation, fluency, and oral rhythm.

Nevertheless, most studies focused on the interaction between AI technology and EFL learners but neglected the role of social network-based platforms in facilitating communication between students and teachers when practicing speaking skills with AI apps. For studies that explore social network-based interactions in the AI context, Liu et al.’s [64] study is an example. In their research, 25 students were required to practice their speaking skills with an AI app called Liulishuo with the teacher’s guidance on a WeChat group. The various types of interaction in the WeChat group included daily reminding, question-answering, process monitoring, experience sharing, and feedback giving. The teacher also provided supplementary materials on students’ dubbing practice on the AI app to facilitate the flexible use of language. The findings indicated that teacher guidance effectively improved the learners’ speaking skills when practicing with the AI app. However, there are a few flaws in the research design. For instance, the researchers did not include a control group as the number of participants was small. Based on the results of the previous studies, it is noted that in MALL environments, social network-based interaction has effective and motivational effects on learners’ speaking practice. Nevertheless, in the field of AI-assisted language learning, few studies have explored the role of social network-based interaction in EFL speaking practice. Therefore, this study intends to explore this area by investigating learners’ perceptions and the effectiveness of social network-based interactions. The research questions are as follows:

• What are EFL learners’ perceptions of social network-based interactive activities when practicing oral English with AI speaking apps?
• How effectively can social network-based interactions help EFL learners to practice oral English with AI speaking apps?

3. Methodology

3.1. Participants

With a purposive sampling strategy, 70 undergraduate students from various majors and universities in China who endeavor to practice spoken English with AI apps participated in this study. They were recruited through convenience and snowball sampling and were divided equally into the experiment group (EG, N = 35) and the control group (CG, N = 35). The participants were randomly assigned into one of two WeChat groups. Students who volunteered to participate in the experiment were informed of the confidentiality and signed the consent forms.

3.2. Procedures

The study lasted for five weeks during the summer vacation of 2022. Figure 1 shows the detailed experimental steps.

3.2.1. Instructors and Apps

Three researchers of this study worked as instructors of the experimental learning, with multiple responsibilities such as providing guidance and feedback for the participants when they practiced speaking skills with AI apps. The instructors encouraged students in both the EG and CG to practice speaking English daily with the AI speaking apps. They were introduced to a few AI speaking apps for speaking practice, including English Liulishuo, IELTS Liulishuo, EAP Talk, and Yidian English. Though different Chinese companies or institutions developed the AI apps, they are all AI speech evaluation systems designed for Chinese EFL learners. Their basic functions are similar, including evaluation of reading aloud and presentation skills by giving scores and feedback based on pronunciation, fluency, and grammatical accuracy as well as range and accuracy of vocabulary. In addition, all the apps are easily accessible to Chinese students. Therefore, the students were allowed to choose freely among those four AI speaking apps according to their preferences; thus, their willingness to participate may have been enhanced to some extent. The instructors also created guidance for each AI app and recommended useful functions for speaking practice.

3.2.2. Interactive Activities

Participants in both the EG and CG received basic information about the arrangement of the experiment, guidance on the four AI apps, and daily task reminders for practicing in the two WeChat groups. However, to answer the research questions, a series of interactive activities were held only in the EG’s WeChat group. The difference in opportunities for interactive practice between the EG and the CG was an essential variable in exploring the research questions. The list of interactive activities for the EG is shown in

Table 1. Interactive activities in the EG.

	Interaction	Platform	Type	Description
1	WeChat communication	WeChat Group	Teacher–student/student–student interaction	Students can ask questions about AI learning apps and skills for practicing spoken English and receive a timely reply.
2	Punch cards/record the learning process	WeChat Mini Program	Teacher–student/student–student interaction	Students are advised to punch cards/record their learning process on a WeChat Mini Program called “Xiaodaka”. They can either upload screenshots of the AI learning app or write a few words.
3	Liking or commenting on others’ punching cards	WeChat Mini Program	Teacher–student/student–student interaction	Students can like or comment on others’ punch cards on the WeChat Mini Program. Meanwhile, they can receive likes or comments from teachers and other students.
4	Uploading recordings and receiving professional feedback from teachers	WeChat Mini Program	Teacher–student interaction	Students are recommended to record their speaking and upload their recordings on the WeChat Mini Program. They can also receive timely individual feedback from teachers the day after uploading.
5	Viewing or editing a collaborative document	Tencent Docs (Similar to Google Docs)	Student–student interaction	Students can view and edit the collaborative document for sharing useful expressions and vocabulary in spoken English.

. Various interactive activities were set up to provide more chances for learners in the EG to participate in social network-based interactions and find their preferred ones. Additionally, as the experiment was conducted online, interaction efficiency may have been a little lower; thus, a variety of interactive activities were set up to encourage students’ willingness to participate and to explore which interactions are more engaging for students. When students in the EG asked questions about either technical issues of the apps or the content of the learning materials, they could receive timely replies from the instructors. Moreover, the instructors sent supplementary practicing materials to the WeChat members of the EG each day and reminded them to practice with the AI speaking apps during the experiment (Figure 2). In their post of punch cards, participants could either write down a few words or post images or screenshots, similar to Facebook’s posting function.

Suppose that students uploaded their recordings for speaking practice. In that case, they could receive timely feedback from instructors the following day, which can be considered a type of corrective feedback in teacher–student interactions (Figure 2). The feedback was given based on pronunciation, fluency, the accuracy of grammar, and the accuracy and range of vocabulary. The evaluation criteria were similar to those provided by the speech evaluation system. Supported by the functionality of the WeChat Mini Program, students could express ‘likes’ and comment on others’ posts. Meanwhile, they could also receive ‘likes’ and comments from instructors and other students (Figure 2). The interactive process is very similar to liking and commenting on posts on Facebook.

In addition, an online collaborative document was created to share vocabulary and useful expressions in spoken English, powered by Tencent Docs, a free online document platform for multi-user collaboration (similar to Google Docs). To help improve the learners’ vocabulary, which is one aspect of speaking skills, the document was sent to the WeChat group of the EG for students to view and edit the content collaboratively.

3.3. Data Collection

Data collection methods in this study consisted of questionnaires and semi-structured interviews. Additionally, quantitative data from the pre-test and the post-test were collected to examine improvements in the students’ speaking skills.

3.3.1. Questionnaires

Quantitative data from the questionnaire were collected to investigate participants’ perceived improvement and their perceptions of a series of interactive activities. The questionnaire aimed to collect feedback from participants on their AI speaking practice and their perception of interactive learning.

The questionnaire contained 31 items in four sections (see Appendix A). In the first section, there were four items to investigate users’ basic demographic information, including their university, year of study, gender, and major. Secondly, in terms of the construct of perceived improvement, seven items were adopted from Li and Zou [56] (Cronbach’s alpha = 0.974), including the aspects of oral fluency (PI1), grammatical range and accuracy (PI2), pronunciation (PI3), oral rhythm (PI4), idea organization skills (PI5), reading aloud skills (PI6), and presentation skills (PI7). In the current study, the Kaiser–Meyer–Olkin (KMO) value of the seven factors measuring perceived improvement was 0.85 (p = 0.00), while the Cronbach’s alpha value was 0.93, which indicated their reliability and validity. They were rated on a five-point Likert scale (1 = strongly disagree, 2 = disagree, 3 = neutral, 4 = agree, 5 = strongly agree). In Li and Zou’s [56] research, 7 items were employed to evaluate the improvement of specific factors of speaking skills with AI speech evaluation programs. The next section explores the learners’ use patterns, including the number of days for practicing and the length of use each day. Furthermore, detailed questions regarding various interactive activities were included especially for participants in the EG. In another section, a few items were concerned with attendance, reasons for absence, perceived usefulness, and preferences for the various interactive activities. In addition, the future intention in interactive learning was also examined.

3.3.2. Semi-Structured Interviews

Semi-structured interviews were employed as the second research method to explore students’ perception of interaction further when using AI speaking apps. Only participants from the EG were asked to take the interview because those in the CG were not designated to engage in interactive activities. Ten students from the EG volunteered to participate in the interview. The interview section asked which forms of interaction were more effective, how interaction could improve students’ speaking skills, and how satisfied students were with the combination of social network-based interaction and AI apps for developing their speaking skills. In addition, participants were asked about their future suggestions for interaction-assisted AI speaking practice (see Appendix B). The interview questions were employed to gain more details including three factors: corresponding reasons for their choice in the questionnaire, the impact of interactive activities, and their future suggestion.

3.3.3. The Pre-Test and Post-Test

In this research, a pre-test and a post-test for EFL speaking skills in the two groups were conducted and compared to examine how interaction could effectively improve speaking skills in the AI context. The tests were completed with the SpeechAce Speaking test (https://www.speechace.com/speaking-test) (accessed on 27 January 2023). Regarding the reliability and validity of the SpeechAce test, many randomly chosen IELTS test takers were used for evaluation. For re-test reliability, a Pearson correlation coefficient of 0.82 could be shown by comparing the converted IELTS scores generated by the evaluation algorithm with blind human scores from IELTS examiners. The grading criteria consist of pronunciation, fluency, vocabulary, and grammar. Every test has three questions concerning the main topic (e.g., sports. Question 1: Describe a sport you would like to learn, and why are you interested in this sport? Question 2: Do you think competitive sports are good for young children? Why or why not? Question 3: What are some of the pros and cons of the international Olympics?), and a score was given according to the IELTS. Both the EG and the CG participated in the pre- and post-tests. They needed to choose one topic and answer the three questions consecutively within the allocated time. Then, the results from the two groups were compared to examine the different learning performances within the two groups before and after the use of AI apps for speaking practice. The difference between the pre-and post-tests could illustrate the level of effectiveness of students’ learning.

3.4. Data Analysis

Both quantitative and qualitative data were analyzed to investigate the two research questions. Based on the first research question, descriptive data on students’ attendance, perceived usefulness of interaction, preferences for interactive activities, and future intentions were analyzed. With the help of IBM SPSS28, an independent t-test was also conducted to explore the group difference in the number of days for punching (recording of practice process). Similar to the questionnaire data, the interview data were also analyzed for these factors through coding. All interviews were audio-recorded and transcribed to computers to ensure accuracy. The interviewees were coded as S1, S2, S3, etc.

Regarding the second research question, there were several steps of analysis with IBM SPSS28. Firstly, for the two groups, an independent t-test was conducted to compare the perceived improvements in various speaking skills, which could complement the results of the previous steps on learners’ improvement of speaking skills. Secondly, paired-sampled t-tests were conducted to compare the pre- and post-tests’ results between the two groups to provide evidence to investigate whether the EG had better learning outcomes than the CG.

4. Results and Discussion

4.1. Students’ Perceptions of Interaction When Practicing Speaking Skills with AI

4.1.1. Interaction Based on WeChat and Its Mini Program

In terms of communication in the WeChat group (the first activity in Table 1), relevant data can be found in the questionnaire. Based on the results from Q14, 68% of the students asked questions about the applications or tips for practicing spoken English on WeChat. Specifically, most of the respondents (87.5%) held the view that communication in the WeChat group was helpful for their learning process (Q15). Regarding the daily reminders and additional learning materials in the WeChat group, an overwhelming majority of them agreed or strongly agreed that they were helpful for their speaking practice, with 92% and 84%, respectively (Q16 and Q17).

As for the reasons behind the positive attitudes, the interview data showed that WeChat interaction was perceived as an excellent mechanism to compensate for the lack of flexibility and practicality in AI learning. Firstly, the suggested tasks and additional videos shared by the instructors were seen as a ‘reminder mechanism’ (S2, S3, S4, and S10). As S3 explained: “As there are no reminding functions within the AI apps, the instructors’ reminding in the WeChat group could help me to keep practicing every day”. Moreover, some students believed that the materials shared in the WeChat group were quite informative and instructive (S1, S2, S3, S5, and S9). S9 highlighted this point: “In some cases, learning materials provided on WeChat are more recent than those in AI apps, particularly for IELTS speaking”.

Additionally, Q18 to Q23 in the questionnaire indicated results on the activity of punching cards (recording the learning process). According to the result from Q18, most of the students (88%) punched their cards at least once on the platform of the WeChat Mini Program (the second activity in Table 1). Regarding their attitudes, 80% of the participants strongly agreed or agreed that punching the cards could be helpful to motivate them to practice with AI apps (Q20). Therefore, it revealed that punching cards could benefit their speaking practice. Meanwhile, the results from Q21 and Q23 showed that 80% of them also claimed that they had viewed, liked, or commented on others’ punch cards, and 56% of the respondents strongly agreed or agreed that it helped with speaking practice (the third activity in Table 1).

The interview data showed further details of participants’ perceptions. The majority of them mentioned that punching the cards reminded and motivated them to practice continuously (S1, S2, S6, S7, and S8). As S6 noted: “Seeing others punching cards gives me a sense of community and potential stress which forces me to keep punching. Meanwhile, I may compare myself with others”. S1 also mentioned, “I don’t want to just stand still while others are progressing”. Therefore, it indicated that liking, commenting on, and browsing other people’s daily WeChat Mini Program punching would significantly increase their motivation to stay persistent.

Meanwhile, data regarding audio recording and feedback can be seen from Q24 to Q26 in the questionnaire. The results indicate that only 32% of the participants submitted their recordings and received instructor feedback (the fourth activity in Table 1). Regarding why they did not record their practice, 32% of them claimed they did not have time, while 24% said they overlooked the function (Q25). Some participants also believed that the interactive activities had few positive impacts on them. Therefore, they preferred to learn independently rather than interact with others. For example, S2 argued: “I am more internally driven and prefer to do my own thing. Other learners’ posts can be a distraction to me”. Despite this, most learners believed that it was helpful to record their practice and receive detailed individual feedback when practicing speaking skills with AI apps.

4.1.2. Interaction Based on the Collaborative Document

The results relating to the collaborative document (the fifth activity in Table 1) are shown as followed. Based on Q27 and Q29 in the questionnaire, 44% of the respondents claimed to have checked or edited the collaborative document, while the majority of them (92%) agreed or strongly agreed with its usefulness. The results above demonstrated that the collaborative document was helpful for most participants in developing their speaking skills. Interestingly, participants’ answers in Q28 indicated that the main reason why they failed to check or edit the collaborative document was that they did not notice the arrangement of the document in the WeChat group. This indicates that instructors may need to frequently remind students of the arrangement during the process.

In addition, interview data revealed the reasons behind the participants’ perceptions. Firstly, the activities provided a useful source of knowledge. For example, S1 mentioned: “It allowed me to build up a lot of expressions on topics within a short time systematically and the richness of the document content was impressive and acceptable”. Another reason was that the activities helped create a sense of community. One student gave an interesting analogy that people who study individually might go through ‘nine × nine Eighty-one sufferings’ (a Chinese proverb in Journey to the West; in the novel, if Tang Monk and his discipline wanted to succeed, they had to overcome 81 sufferings, meaning jump through hoops) on their own, whereas with this collaborative document or by sharing activities, the “81 sufferings” would be broken up into several or even dozens of small tasks that can then be shared to be easily solved (S6). Therefore, the results above illustrate that most students held positive attitudes towards interactions in online workshops and the collaborative document.

4.1.3. Students’ Preferences for Interactive Activities

The results from Q30 in the questionnaire also demonstrate students’ rankings for interactive activities, while the results from the interviews illustrate more details regarding reasons for preferences (see Figure 3). The results show that “uploading the recording and getting professional feedback” ranked first, which was followed by “punching cards (recording the learning process)” and “joining in an online workshop”. Moreover, students chose “WeChat communication” and “liking or commenting on others’ punching” as the two least helpful interactive activities.

In the interview, factors contributing to students’ preferences were revealed. Firstly, compulsory activities were more welcomed by students compared to optional activities. S1 highlighted the supervising role of the interactive activity, commenting that compared to joining an online workshop, the activity of punching cards had a more invisible force for her. Secondly, activities including experience sharing and individualized feedback were more welcomed. For instance, both S5 and S7 mentioned that they could receive tips on practicing speaking skills with AI apps from a personal perspective.

Interestingly, uploading recordings and receiving teachers’ feedback ranked first in the students’ preferences but had the lowest attendance rates. As mentioned before, despite the high level of perceived usefulness, there were a few obstacles in terms of learners’ participation in practice. This gap will be further explored in the discussion.

4.1.4. Students’ Future Intention in Interactive Learning

Students in the EG were also asked in the questionnaire about their future intention in interactive learning. The results from Q31 in the questionnaire revealed that most EG students (88%) felt they would continue participating in interactive learning in the future. Moreover, compared with students in the CG, the EG students performed better in punching cards (recording their learning process). The results of the independent sample t-tests revealed that students in the EG punched on more days than those in the CG, and there was a significant difference in scores for the EG (M = 2.60, SD = 1.16) and CG (M = 1.70, SD = 0.57) (t (43) = 3.18, p = 0.00, two-tailed) (

Table 2. The results of independent t-tests for the two groups (two-tailed).

		Mean	SD	T	p
Number of days for punching	EG	2.60	1.16	3.18	0.00
	CG	1.70	0.57

). The magnitude of the differences in the means (mean difference = 0.90; 95% CI: −0.33 to 1.47) was large (eta squared = 0.19). The total number of hours for which the two groups of students used the AI apps over a five-week period also differed. Compared with 57.25 h in total for the CG, or 0.63 h per student per week, the EG held 136 h overall with 1.05 h per student per week—nearly twice the values of their counterparts.

Based on the results above, interactive activities could help motivate learners’ practice and potentially affect their learning habits in the future.

Furthermore, in the interview, students’ future suggestions were explored, which could be summarized as task-based interaction, hybrid activity, and community building. Regarding task-based interaction, four students pointed out that specific tasks with clear goals could provide them with the practice direction to make their learning more efficient (S1, S6, S7, and S8). For instance, S1 noted: “if the direction of learning English does not fit with everyday use, it tends to increase my inertia”. The results indicate that learners prefer tasks with specific goals to practice speaking skills for IELTS or daily communication.

Secondly, some interviewees pointed out that they were discouraged from practicing due to the message delay on WeChat; thus live streaming or offline activities are preferable (S1, S3, and S8). This suggests that asynchronous communication might prevent learners from practicing their speaking skills with AI apps, which could be potentially compensated for with synchronous and offline interaction. Therefore, a hybrid approach is preferable to increase learners’ motivation, especially their willingness to produce oral outputs.

In terms of community building, the interviewees placed a high value on collaborative interaction, so that the best learning outcomes could be achieved with a minimum investment of time and effort (S2, S3, S4, S6, S7, S9, and S10). For this, S6 made the interesting point that “the app is not alive, but the people are. The interpersonal interaction could help create a positive atmosphere for practicing speaking skills”. As a result, various interactive activities should be effectively integrated to build a learning community and facilitate virtual engagement.

4.2. Effectiveness of Interactions in the Context of AI-Assisted Language Learning

There are two ways to examine the effectiveness of network-based interaction. Firstly, the researchers compared the results of the perceived improvements of the two groups. Secondly, the scores on the pre- and post-tests between the EG and CG were compared with independent t-tests to explore how interaction can help improve speaking ability when using AI apps.

4.2.1. Comparison of the Perceived Improvement (PI)

Data from the questionnaire were analyzed to explore how interaction influenced the perceived improvement of speaking skills in the two groups (see

Table 3. The results of independent t-tests for the two groups (two-tailed).

		Mean	SD	T	p
PI_Average	EG (N = 25)	3.92	0.47	8.63	0.00
	CG (N = 20)	2.35	0.74
PI1	EG (N = 25)	4.04	0.93	6.03	0.00
	CG (N = 20)	2.35	0.57
PI2	EG (N = 25)	3.72	0.93	3.92	0.00
	CG (N = 20)	2.05	0.87
PI3	EG (N = 25)	4.08	0.95	6.52	0.00
	CG (N = 20)	2.25	0.91
PI4	EG (N = 25)	4.20	0.40	7.99	0.02
	CG (N = 20)	2.45	1.00
PI5	EG (N = 25)	3.40	0.82	4.22	0.00
	CG (N = 20)	2.40	0.75
PI6	EG (N = 25)	4.52	0.87	9.82	0.00
	CG (N = 20)	1.90	0.91
PI7	EG (N = 25)	3.92	0.86	5.67	0.00
	CG (N = 20)	2.25	1.12

). The results of the independent sample t-test indicate that there was a significant difference in the average scores for the EG (M = 3.92, SD = 0.47) and the CG (M = 2.35, SD = 0.74) (t (43) = 8.63, p = 0.00, two-tailed). The magnitude of the differences in the means (mean difference = 1.57; 95% CI: 1.20 to 1.93) was large (eta squared = 0.63). In addition, the results indicate that students in the EG held significantly stronger views than those in the CG that the AI speaking apps helped them practice their oral fluency (PI1), grammatical range and accuracy (PI2), pronunciation (PI3), oral rhythm (PI4), idea organization skills (PI5), reading aloud skills (PI6), and presentation skills (PI7) (Table 3). The results could illustrate the role of interaction in helping EG learners improve their spoken English skills while using AI apps.

4.2.2. Comparison of the Results of the Pre-Test and Post-Test

In addition to comparing the perceived improvements of the two groups from the questionnaire, the pre-test and post-test scores for the two groups were calculated and analyzed. There were two steps of analysis with independent sample t-tests. Firstly, the researchers examined whether the two groups were comparable in terms of their oral skills in English speaking before the experiment. Then, the post-test scores were measured and compared to explore the effectiveness of the social network-based interactions in objectively improving speaking skills.

Regarding the learners’ performances in the pre-test, the relevant data are shown in

Table 4. The results of the pre-test and post-test for the two groups.

Group	Pre-Test				Post-Test
	Mean	SD	t	p	Mean	SD	t	p
EG (N = 25)	5.42	2.13	0.37	0.71	6.58	1.44	2.06	0.05
CG (N = 25)	5.24	1.09			5.84	1.07

. The mean scores for the EG (M = 5.42, SD = 2.13) were slightly higher than those for the CG (M = 5.24, SD = 1.09). However, the results of the independent-sample t-test indicated that there was no significant difference between the two groups in the pre-test (t = 0.37, p = 0.71, two-tailed; mean difference = 0.18; 95% CI: −0.79 to 1.14; eta squared = 0.01). Therefore, the EG and CG were comparable in their speaking efficiencies before the experiment.

With respect to the scores in the post-test, as shown in Table 4, the EG obtained higher scores in the post-test than the CG did (EG: M = 6.58, SD = 1.44; CG: M = 5.84, SD = 1.07). The results of the independent sample t-test showed a significant difference between the groups despite a small effect size (t = 2.06, p = 0.05, two-tailed; mean difference = 0.74; 95% CI: 0.02 to 1.47; eta squared = 0.04). Similar to the findings from learners’ perceived improvement, the results from speaking tests also indicated that the EG achieved better learning outcomes with interactive activities than the CG did. The finding on the outperformance of the EG further suggested that interactive activities play an important supporting role in developing speaking skills with AI apps.

5. Discussion

The first research question explored the learners’ perceptions of interaction in the context of AI-assisted language learning. Firstly, students held more positive attitudes toward the interactive activities on WeChat, including messaging and instructors’ reminding, and the learning materials provided in the WeChat group. It is regarded as an efficient mechanism to complement the speaking practice with AI apps. For the WeChat Mini Program, most participants believed that punching cards positively affected their motivation, while about half of them believed that viewing others’ posts was also motivating. Students also considered that “uploading the recording and getting professional feedback” and “punching cards” were the most helpful activities in terms of speaking practice. These findings on WeChat interaction echo previous studies by Chai [40], Xu et al. [18], and Zou et al. [43] regarding the positive attitudes toward WeChat interaction for speaking practice. In practice, however, participants did not prefer to upload their audio recordings for various reasons such as shortage of time and unwillingness. A similar issue was also found in Xu et al.’s [18] study, in which some students felt uncomfortable when recording their audio on the WeChat platform.

For interactions based on the collaborative document, though the attendance for both activities was relatively low, participants had a positive perception of the usefulness of the activity when practicing speaking skills with the AI apps. The research thus expands on existing studies exploring the effect of Google Docs or other collaborative documents in the field of writing [42] or vocabulary [46] into speaking practice.

Students’ future intentions in interactive learning were also examined. Firstly, a majority of the students in the EG stated their willingness to continue engaging in interactive learning with AI apps. Secondly, compared with students in the CG, those in the EG punched cards for much longer, which indicates the role of interaction in motivation and habit development. The findings above are consistent with Read et al.’s [19] study regarding the positive role of interaction in motivation and learning habits. Additionally, the findings outlined future suggestions for interactive learning in the AI context, which the participants provided. In the future, activities with AI apps are recommended to be task-based, assisted by hybrid approaches, and held in a learning community. The finding corresponds with the studies by Mykytiuk et al. [17] and Tragant et al. [16], which indicated that task-oriented interaction could motivate students to engage in mobile-assisted EFL learning. Moreover, the results revealed that effective interactions could help EFL learners practice speaking skills in the AI context. Firstly, regarding students’ perceived improvement in speaking ability, students in the EG group held a stronger belief that the AI speaking app helped them practice speaking fluency, grammatical range and accuracy, pronunciation, speaking rhythm, idea organization skills, reading aloud skills, and presentation skills compared to students in the CG group. The results could illustrate the role of interaction in helping learners improve their spoken English skills when using AI speaking apps. Compared with the CG, the EG witnessed significantly higher scores in the post-tests. This echoes Liu et al.’s [64] research findings and further complements their study in terms of the experimental setting.

6. Conclusions

This study aimed to explore learners’ perceptions of interactive activities when practicing speaking skills with AI apps and the effectiveness of interaction in improving their speaking abilities. Based on the questionnaire and interview data, the findings revealed students’ positive attitudes and preferences for various interactive activities. Additionally, the research indicates that the social network-based interaction effectively improved their speaking skills in the AI context. Quantitative data showed that the participants in the EG obtained significantly higher scores in the post-test. They also had a stronger belief that AI apps could help them develop various speaking skills, including oral fluency, grammatical range and accuracy, pronunciation, oral rhythm, idea organization skills, reading aloud skills, and presentation skills. Therefore, throughout the learning process with AI apps, interactive activities based on WeChat can provide learners with more chances to communicate with instructors and peers, thus motivating learners’ speaking practice and helping them achieve a better learning outcome.

7. Limitation

Admittedly, there are some limitations in the current study. Firstly, the research was limited by the implementation period (five weeks), and the number of participants was also limited. In this regard, future studies can be conducted with a larger group of students for the whole semester. Secondly, the participants were recruited through snowball sampling. Though it is a small sample, the participants came from different majors from more than ten Chinese universities, which is relatively representative. Thirdly, due to the epidemic, the research was conducted online, resulting in a sense of distance and unfamiliarity among the students. Therefore, there was relatively less student–student interaction than student–teacher interaction. Nevertheless, the findings of this study can still enrich the research on AI-assisted language learning in the Chinese context and provide some recommendations for integrating social networking into AI-assisted learning on a global scale.