Lin (2017) explained that VR, also called artificial environment, is generating a 3D virtual world with computer simulation and providing users with a sensory stimulation of sight, hearing, and touch which allows users to timely observe affairs and objects in the 3D space without restrictions. Richardson et al. (2017) pointed out that VR was the feeling of presence in a virtual environment, namely, the artificial environment. VR simulated a 3D scene for real-time response through human–computer interaction (HCI) so that people could feel the presence. Archer and Finger (2018) simply explained VR as creating a virtual world with computer graphics, allowing users to have authentic feelings. Herrera et al. (2018) mentioned that VR constructed the environment through computer equipment and integrated authentic or virtual pictures into simulated situations to present the authentic effect. The environment presented highly authentic interaction, allowing users to view various computer images or videos and operate or interact with the objects through HCI. Users could freely move in the space to appear senses of integration and participation to further feel the presence.

According to Wu and Chen (2018), VR covers 3 dimensions in this study.

There have been various works on the research of VR. For example, Lv et al. (2020) argued that VR technology could be used to design industrial security solutions. The constructed algorithm performed well in the attack response to the industrial control system in the VR simulation environment, which provided practical significance for the application of VR technology. Chang et al. (2020) proposed various indexes to accurately measure the disease level by reviewing the measurement of VR disease symptoms and causes. Emmelkamp and Meyerbröker (2021) used VR technology to diagnose and treat mental health diseases, and the research showed that the application of VR in phobia could be extended to the diagnosis and treatment of other mental health disorders.

Kao et al. (2016) indicated that immersion-based teaching promoted the idea of a language swimming pool, allowing students to cultivate their language sense in the environment with the pure TL. Students, either in class/after class or on/off-campus, would completely immerse themselves in the TL ocean, constantly receive the stimulation of distinct TL information, absorb TL knowledge, repeatedly practice TL skills, and continuously use TL for learning and living to eventually develop TL thinking pattern and expression ability naturally. Slater (2018) stated that interaction and communication were the major teaching forms, the classroom climate was relaxing and could fully induce students’ motivation for positive participation, and students could cultivate the habit of thinking with TL in the real immersive environment, break through their psychological barriers, and pick up some language skills. Bachen et al. (2016) explained the so-called immersion-based teaching that students, for about all or half of their time in schools, immersed in the English language environment, teachers taught language knowledge using English, and the English language was used for teaching non-language courses as well. In other words, the English language was the learning content as well as the learning tool. La Corte et al. (2019) regarded immersion-based teaching as a teaching method for learning English, which mainly used TL for general teaching activities. TL was the medium of teaching, rather than the major objective and object in classroom teaching. Hortensius et al. (2018) pointed out the core concept of immersion-based language teaching as using the function and objective of language for naturally learning language, as the initial function, and objective of language were for message communication. Immersion-based language teaching is expected to replace the NL with TL in thinking or language application environments; when learners were immersed in the language application environment for a long period, their sense of TL could be naturally cultivated through interaction with people using TL.

According to Lin et al. (2018), FLL should be content-based. Full exposure to the meaningful language input situation is the key to acquiring language. Language learning and teaching contain three dimensions.

To sum up, there are many studies on immersion-based language teaching. For example, Altun and Lee (2020) used meta-analysis to carry out immersion-based English language teaching. The research results showed that immersion-based learning technology had a positive impact on English language learning. Wang et al. (2021) employed immersive experience technology to improve the effect of undergraduate Chinese multilevel learning. The results suggested that students liked to explore experiential learning opportunities of English culture and language outside the classroom. Hayes et al. (2021) studied the immersion degree of interactive VL language learning and mapped the immersion level to the learning objectives related to language learning through replicable simulation and social interaction. Combined with the above research, the use of immersion-based teaching can effectively improve the effect of English teaching. Kim et al. (2019)’s research shows that robot programs allow direct interaction between human and computer.

Cerezo et al. (2019) described a mobile phone application, which helped Spanish children practice the pronunciation of basic English vocabulary, and the robot served as a virtual teacher and interacted with children by dividing them into the experimental group and the control group. According to the performance evaluation before and after the test, satisfaction survey, and children’s emotional analysis, the application significantly affected children’s English learning. Bin and Mandal (2019) pointed out that the application of computer technology (CT) and AI technology in the field of education would become a trend, even in the automatic scoring of English composition. Specifically, based on the integration of information technology and curriculum, the application of AI in middle school English teaching was explored by combining curriculum theory, literature analysis, and field investigation. Moreover, the implementation scheme of English assisted teaching system based on AI technology was proposed. The results also implied that AI technology could make the teaching system more humanized and further improve the quality and effect of English teaching. Park et al. (2020) studied the improved automatic SR system, and the recognition accuracy could be improved by 4% on noisy experimental test data sets. Ye et al. (2021) studied the dual-channel hybrid-level framework using SR; the system reduced the relative word error rate by 8–10%, showing that the proposed system might maintain the advantages of the hybrid system. In conclusion, the application of AI in SR has practical reference value for immersion-based English learning and teaching.

Hong et al. (2016) mentioned that teachers could understand the practice effect of teaching activity from students’ evaluation results and further inspect the teaching excellence for evaluating teaching. Teachers could understand students’ responses to different teaching methods by observing their learning responses and understanding their performance in the class or learning process. Isabelli-García and Lacorte (2016) regarded learning outcome (LO) as the practice of various types of evaluation or tests on learners for a period of time after learning activity to understand their LO on the learned content. Lull and Bushman (2016) pointed out LO as an indicator to evaluate students’ absorption of course content; and, teachers’ teaching effectiveness could be judged based on students’ test performance. Chrysikou and Thompson (2016) regarded LO as students’ change before and after receiving education. The real change was by deducting “entry behavior” before education from “terminal behavior” after education, which was the students’ “direct” LO. On the contrary, students’ “indirect” LO was the effect appearing a period after students had been educated. van Loon et al. (2018) stated that students’ LO was inspected through examinations to check whether students mastered the knowledge and skills preset in the teaching objectives and learned teaching activities or materials. The higher test performance revealed students’ better LO.

According to Tsai et al. (2018), the following factors in students’ LO are proposed in this study.

Meanwhile, concerning other factors affecting students’ academic performance, Liu et al. (2020) conducted a meta-analysis on the factors affecting students’ academic performance. The results corroborate that there was a medium relationship between socio-economic status and academic performance. Sadeqi et al. (2021) studied the impact of curriculum teaching on students’ learning effects. The research results revealed that teaching practical knowledge by comprehensive methods was of great help to students’ clinical experimental application.

Aiming at abstract concept learning and specific operation learning, Wenga et al. (2018) conducted the clinical teaching for experimental comparison. The experiment proved that learning anxiety and frustration could be easily alleviated when pictures were applied to operational abstract concept learning. Besides, the logical thinking in programmed learning was not simple, and actual active operation learning could acquire sensory experience of seeing and touch and appear better message processing than learners with abstract ideas. DeSmet et al. (2016) applied the characteristics of virtual space and further used innovative immersive interaction media to prove that virtual operation through users’ intuitive allowed users to not only acquire information from picture-based abstract concept learning but also integrate the interactivity of immersion-based VR. Wu and Chen (2018) discovered that the application of VR could make up for the inadequacy of other teaching methods through the interactive virtual learning environment; it could enhance learners’ enthusiasm and provide interactive teaching, as well as apply the space of virtual environment to promote immersion-based teaching on the spatial structure. Chen et al. (2021) used interactive spherical video to study the writing of VR Chinese composition teaching. The results proved that it could improve students’ writing ability. Accordingly, the following hypothesis is inferred.

Shin (2018) indicated that immersion-based teaching allowed learners to learn the course content through a TL environment, so it could eventually help learners become bilingual learners; for this reason, the use of TL between teachers and students was particularly emphasized. Madsen (2016) analyzed the effect of immersion-based listening and reading learning period on the learning effect of elementary pupils’ ESL learning. It was discovered that longer immersive courses for either English listening or reading appeared to produce a better performance on foreign language acquisition. Ferchaud and Sanders (2018) proposed that immersion-based English teaching could enhance elementary pupils’ learning attitudes and English test performance. It was, therefore, considered that immersion-based learning could assist in enhancing English learning effectiveness. Lin et al. (2018) discovered that pupils receiving entirely immersion-based English learning showed higher learning confidence, keener learning interests, and also stronger learning belief than those who did not. Chan et al. (2021) applied immersive VR to distance architectural history teaching. The research results signified that immersive visualization ability could be used as an essential distance teaching tool in educational institutions to improve students’ learning effects. In this case, the following hypothesis is inferred.

Zych et al. (2019) mentioned that VR stressed being on the scene, and the sensory stimulation of the taught was far more than picture and text message delivery on print media to deepen the topic memory and comprehension; it, therefore, could effectively promote LO. Pavone et al. (2016) indicated that a lifelike simulation situation could induce students’ learning motivation, implement the learning initiative, and control the interaction with the environment to achieve the skill learning and effectively enhance LO. Gorisse et al. (2017) stated that computer simulation technology offered learners with proper control; learners could control the learning sequence, content, and schedule and actively operate, explore, and recombine knowledge. It conformed to the spirit of constructive learning to promote LO. Tsai et al. (2018) indicated that a well-designed skill simulation system could simulate authentic operation situations, allowing learners to clearly learn important concepts through self-control, interaction, simulation, and observation, thereby effectively enhancing the LO. Han et al. (2021) conducted the inspection and research of nervous system teaching tools based on VR. The research results implied that VR technology could effectively improve students’ learning satisfaction. The following hypothesis is therefore inferred in this study.

Knox (2020) pointed out that AI played a great role in educational institutions. New Oriental Group, tomorrow advancing life (TAL), and squirrel AI institutions have made good achievements in the use of AI. They also stated that educational institutions were influential participants in national and regional AI development strategies. Cope et al. (2020) pointed out that AI would never take over the role of teachers, but it could be used to improve the potential of education in a certain range and make education more humanized. Lin and Wang (2021) used VR technology in English language learning. The research found that VR technology could be effectively integrated into English classrooms to promote students’ creative self-efficacy and intrinsic motivation. Therefore, the following assumption is deduced:

With the experimental design model, a total of 106 English-major students in a Chinese school were selected as the research objects for the 16-week (3 h per week, with a total of 48 h) quasi-experimental teaching research. The retrieved data are analyzed with computer statistics software to test various hypotheses. Combined with the research results of Schieber et al. (2018), the sample form of the designed questionnaire survey (QS) is manifested in Table 1. After the students’ study, the QS is distributed for statistics, and the learning effect is analyzed after unified recovery and processing.

Validity refers to a measurement tool to truly measure what a researcher intends to measure. Generally speaking, validity is divided into content validity, criterion-related validity, and construct validity. The QS design in this work has referred to domestic and international researchers’ research items that the QS shows certain content validity. Dimensions of VR, immersion-based language teaching, and LO in this work are tested through the overall structural causality with linear structural relation (LISREL). The data input is based on the correlation coefficient matrix of the above observatory variables. The analysis results with the LISREL model reveal that the overall model fit reaches the reasonable range that it presents favorable convergent validity and predictive validity. Item-to-total correlation coefficients are used for testing the construct validity of the QS content, i.e., reliability analysis. The calculated item-to-total correlation coefficients are used for judging the questionnaire content. The item-to-total correlation coefficients of dimensions in the designed QS are higher than 0.7, revealing a certain degree of the construct validity of the dimensions.

Reliability and validity analyses are conducted to further understand the reliability and validity of the QS. The higher Cronbach’s α shows the better reliability. The formal QS is developed according to the standard, and the measured Cronbach’s α appears in 0.75–0.90, apparently conforming to the reliability range.

After the content data of the students’ oral test are collected, the data are transformed to extract the characteristic value of the speech content conveniently. Through identifying the features of vocabulary, sentence, and text structure, and then correcting the final learning results, the effectiveness of immersion-based language teaching is tested.

The high-frequency words expressed in oral English are input into the intelligent system. With a full score of 10 points, the contents are scored by the system according to their correlation degree. The scores are recorded in the system for the following score collection.

With VR equipment, the students can have immersive learning independently in a classroom, in which there are several small spaces for the experiment. Therefore, each student can log in to the account before using it and perform voice interaction in the independent space without being disturbed by the external environment. After the students log in, the teacher can check the students’ learning and training at any time by logging into the background system to assist the students in completing the follow-up study.

The students can also learn through the mobile APP. The contents of the mobile APP can be synchronized with those in the experimental classroom system so that the students can learn relevant content at any time. During classes, VR equipment is used for immersion-based learning to consolidate what the students have learned. The specific environmental conditions are manifested in Figure 1.

In the LISREL model, the factor analysis is combined with the path analysis in traditional statistics, and simultaneous equations are added in econometrics for simultaneously calculating multiple factors and multiple causal paths. The evaluation of model fit can be evaluated from the preliminary fit criteria, overall model fit, and fit of the model internal structure.

The data are organized as shown in Table 2, and the preliminary fit criteria, internal fit, and overall fit of the model are explained as follows.

Table 2 shows that three dimensions (imagination, immersion, and interactivity) can significantly explain VR (t > 1.96, p < 0.05). The three dimensions (general characteristics, teaching strategy, actual application) can remarkably explain immersion-based language teaching (t > 1.96, p < 0.05); and two dimensions (teaching factors and environment factors) of can notably explain the LO (t > 1.96, p < 0.05). Apparently, the overall model reveals favorable preliminary fit criteria.

Regarding internal fit, VR presents positive and significant correlations with immersion-based language teaching (0.851, p < 0.01); immersion-based language teaching shows positive and remarkable correlations with LO (0.824, p < 0.01); VR appears positive and notable correlations with LO (0.836, p < 0.01); thus, H1, H2, and H3 are supported.

Regarding the overall model fit, the overall model fit standards χ²/Df 1.626 indicates it is smaller than the standard value 3; RMR 0.004 reveals the proper χ²/DF and RMR results. Furthermore, the chi-square value is sensitive to sample size, and so it is not suitable for directly judging the fit. However, the overall model fit standards GFI 0.974 and AGFI 0.932 are above 0.9 (the closer GFI and AGFI are to 1, the better model fit is); thus, this model presents better fit indices. The results of hygiensis test are shown in Table 3.

In the process of correction, the SR technology is used to extract the relevant content in the audio content of the oral test, including the characteristics of vocabulary, sentence, and text structure. Through the analysis and comparison with the system database, the students’ oral test is corrected, and the corresponding results and suggestions are given.

Among the 106 oral test audios submitted, the corresponding oral test scores are shown in Figure 2.

Figure 2 shows that most people scored 80–90, followed by 70–80. Moreover, there is a small score gap between the teacher’s correction and AI grading, which can prove that it is feasible and effective to use AI for correction.

Meanwhile, AI is used to make a correction, and the system also gives suggestions for revision, such as word pronunciation, sentence structure, and logic of the text. This requires students to focus on the pronunciation of words and the grammatical relationship of sentences when practicing oral English, thus further conveying the correct pronunciation to the other side while improving the expression ability and avoiding misunderstanding in the communication.

Simultaneously, with an extensive voice database, the AI system provides students with reading examples and rereading choices. Various functions, such as reading and accent correction, can also be realized. The research results indicate that a deep understanding of their deficiencies can be provided for the students of different English proficiency by the AI system within a short time. Meanwhile, after the students complete the corresponding learning tasks within a specified time, their learning results are recorded automatically and analyzed by the system. Then more targeted exams are arranged for them. Students can apply their knowledge through frequent tests of what they have learned. In this way, teaching and learning are completed. Besides, the system also provides students with analyses of focuses and difficulty. Through suggested answers to and remarks on important knowledge, the students can finish the training. Besides, homework search and problem analysis can be completed in the AI system’s homework management module so that the students can see their disadvantages clearly.