The 3DVWLL achievements are attributed to three key characteristics. The first involves embodiment (Gorham et al., 2019; Jusslin et al., 2022; Privas-Bréauté, 2015; Vasalou et al., 2007; Winn, 2003). Learners amplify self-focused attention during the creation of avatars and exploration in 3D virtual environments (Privas-Bréauté, 2015). Their awareness of individuating properties is amplified through expressing the feeling of ‘outwards to the virtual community through the connection between body and mind (Johnson et al., 2000; Vasalou et al., 2007). Second, 3DVWLL research focuses on learners’ presence in social cognition (McKerlich et al., 2011; Steinkuehler, 2008; Traphagan et al., 2010; Wang et al., 2015). Due to intensive interaction in multiple-user environments, the skills for meaningful negotiation were naturally acquired (Steinkuehler, 2008). Virtual residents establish sociocultural coordination by practicing simulated scenarios. Linguistic knowledge related to social interaction, such as indirect speech acts, is entrenched when learners pay close attention to the sociocultural contexts of engagement (Skyes et al., 2010; Peterson, 2016). Third, 3DVWLL research shows strength in promoting learners’ psychology. Park and Baek (2011) reported that 3DVWs enhanced self-efficacy, as this mode provided students with engaging learning experiences through learning-by-doing activities that incorporated the goals of investigation, discovery, and creation—such a mode empowered learners with more effective learning strategies (Henderson et al., 2012). Godwin-Jones (2014), Hung et al. (2018), Reinhardt & Sykes (2012), and Xu et al. (2020) mentioned that 3DVW game-based learning has the potential to generate EFL learners’ intrinsic motivation based on communication and confidence in using languages due to the factors of the authentic simulation of real-life situations, probability of success, high interest, challenge, and low anxiety. As for autonomy, learners’ L2 production improved in linguistic complexity and accuracy while completing 3D CALL-based tasks, with L2 learning progress aligning with the linguistic characteristics of the input they received in 3DVWs (Collentine, 2011).

3DVWLL studies must consider the targeted pedagogical purposes in implementing virtual technology, as discussed in 3DVWLL, in conjunction with collaborative activities (Dickey, 2003) and task design (Song et al., 2023). The 3DVWLL achievement has provided less justification for these challenges when considering the relationship between its theoretical orientations and implementation, e.g., teaching methods and learning activities, particularly when researchers aim to understand the differences between 3DVWLL and other fields of TELL, such as MALL. The achievements attested in previous reviews encountered two problems. The first concerns the representativeness and exhaustiveness of core studies procured in the previous reviews. The confirmation of core studies might be biased due to researchers’ selection of core studies through keyword searches (Templier & Paré, 2015). Moreover, the selected studies mainly emerged from the quantitative category, while qualitative studies might be ignored and neglected potential issues (Song & Hopper, 2015). Inconsistent explanations for the achievements also impact. Rankin et al. (2006) and Pellas & Boumpa (2017) reported the disadvantages of 3DVWs, as learners felt inconvenienced by 3DVW activities, including disorienting virtual navigation and difficulties integrating extended materials. Beaumont et al. (2014) reported that 3DVWs showed low affordance for improper learning scenarios in social interaction with non-player characters. Moreover, divergent theoretical positions for 3DVWLL affordance occur between learner-centered approaches (Wehner et al., 2011; Pasfield-Neofitou et al., 2015) and the teacher-centered mode (María, 2013). The second problem concerns the academic ecological relation between 3DVWLL and SLA. Such concern becomes an inevitable pursuit because the development of CALL inherently seeks interdisciplinary support from related fields such as psychology and digital engineering (Marek & Wu, 2014; Zhang et al., 2016; Xu & Yang, 2023). Since 3DVWLL researchers are inclined to integrate theories from SLA, education, and psychology, a qualified review must clarify theories and learning activities, especially when dealing with unaddressed issues in previous systematic reviews.

Throne & Smith (2011) mentioned the pitfall of focusing on the technology without attending assiduously to core issues of second language acquisition (SLA). They advocated continuous dialogue between CALL and SLA researchers, focusing on the reunition of theory, course design, and practice, e.g., Sykes et al. (2010). Kreshen (2014) explicitly argued against improperly integrating technology in SLA learning design because such an imprudent learning design could be disastrous due to misunderstanding the theories underlying foreign language acquisition. SLA research emphasizes a consistent link between practice and theory (Ur, 1996), e.g., the implementation of sociocultural theory (Lantolf & Poehner, 2014) and input processing (Vanpatten, 2007) in learning activities. The field shows a trend of increasingly integrated technology with theoretical underpinnings that span learners’ cognition and psycholinguistic perspectives (Zhang, 2019), especially on language forms (Swain & Lapkin, 2001; Nunan, 2004). In 3DVWLL, Lan (2021) discussed the theory-practice connection that focuses on sociocultural theory and embodiment, which concords with the previous reviews. However, the relationship between technological affordance and the acquisition of grammatical representations, skills, and sociocultural literacy has yet to be fully elucidated. For example, it is unclear how immersion and contextualization facilitate learners’ English grammatical competence carrying strategically communicative ability, as shown in Peterson (2016).

To solve the problems, the current study examined the interconnectedness of core studies within the existing literature sample to establish a consensus and thorough evaluation of findings related to 3DVWLL. A comprehensive literature review uses data-driven analysis to map the knowledge structure and identify key publications within the 3DVWLL domain. To achieve the research objective, this study aimed to delineate the 3DVWLL intellectual structure by utilizing co-citation analysis (Small, 1973), a method that has been adopted in educational technology (Tang et al., 2016) and TELL (Bibauw et al., 2019). The goal is to identify (i) the central and peripheral themes within 3DVWLL as revealed by co-citation patterns and (ii) the temporal evolution of the core research network. In more detail, co-citation analysis reveals foundational works in 3DVWLL that align with the collective view of the discipline, fostering scholarly conversations that sustain a progressive research direction (Sawyer, 2006) while also highlighting emerging areas of inquiry (White & Griffith, 1981). The present study addresses three research questions centered on the research goals:

Research question 1: What are the core research issues within the 3DVWLL domain, and how do they interrelate within the intellectual framework of the field when juxtaposed with the diverse findings?

Research question 2: How have the intellectual structures evolved based on the content of research streams with a specific concern on the linking of theories and implementation?

Research Question 3: What are the interdisciplinary connections between 3DVWLL and SLA regarding their intellectual structures? Moreover, what specific insights can be gained from the comparative analysis regarding the shared and non-shared research core issues?

The study used index studies to procure core studies through SNA and DCA calculations. Index studies represent a collection of high-quality research articles with titles, abstracts, or keywords relating to 3DVWLL language learning, language education, and language teaching. They were selected from the literature of the four review works of 3DVWLL from 2000 to 2018 (Lin & Lan, 2015; Reisoğlu et al.,2017; Avgousti, 2018) and AUTHOR & COLLEAGUE, 2019). The study further procured potential index studies whose research topics mainly focus on theories and implementation in 3DVWLL after 2018 by conducting a multi-keyword quarry in WOS with search keywords of 3D virtual worlds*, 3D virtual environment*, avatar*, language learning*, foreign language learning*, language acquisition*, *language policy, and language teaching*—the wildcard symbol (*) helps the search obtain the different naming conventions in the database. During this phase, researchers excluded irrelevant keywords, such as those related to chemistry, communication studies, or other STEM-related fields. The selection stage identified 63 studies, comprising 56 studies based on previous review articles and seven journal articles obtained through WOS searching by November 2024. Next, researchers with experience in 3DVWLL and co-citation analysis verified the candidates of index studies, focusing on empirical studies directly dedicated to 3DVWLL, and excluded those that did not meet the criteria. For example, Chua & Chua (2017) were excluded because they did not target language learning but focused on the topic of education on e-leadership quality in 3DVWs. The data selection phase procured 44 index studies to identify core studies as a representative intellectual base.

The core studies constituted a large data pool of intellectual structures regarding the research streams and the evolution within the field, as analyzed through DCA and SNA (McCain, 1990). The data screening stage utilized 44 index studies to identify cited studies in WOS and then retrieved 1,751 non-repeated studies cited by these 44 index studies. To investigate the core studies of 3DVWLL's intellectual structure, the current study utilized VOSviewer (Van Eck & Waltman, 2020) to analyze the dataset for co-citation calculations. The researchers use the co-citation matrix to generate results, specifically the link, i.e., a connection or relation between two references, and the strength, i.e., the number of cited references that two publications have in common. Therefore, the final step confirmed the joint reference-hood of source articles by the same citing article (Small, 1973). The researchers selected data from April 16, 2020, to March 3, 2024. Figure 2 illustrates the hierarchical relation among index, co-cited, and cited studies, along with the illustration of stream congregation.

To analyze the highly cited studies inside the field, DCA was treated as the method to match all possible co-cited pairs of research studies (Small, 1973). The analysis step built a symmetric matrix for the 1751 articles cited by the index studies. The matrix was conducted to calculate co-citation frequency and confirm the joint reference-hood of source articles by the same citing studies. When a pair of core studies are referenced in the same citing studies, the frequency of co-citation for that research pair was then reckoned as one. For example, Jauregi et al. (2011) and Liou (2012) have been cited 193 and 203 times, respectively, with six follow-up studies co-citing both works. Endorsed by these six citing studies, the value of co-citation for the research pair of Jauregi et al. (2011) and Liou (2012) in the co-citation matrix was coded as six, corresponding to the number of co-citation counts. Considering the transition of the 3DVWLL research focuses in 2014, as reported in Table 1, the DCA was analyzed based on three phases: (i) the whole stage between 2000 and 2024, (ii) the initial stage from 2000 to 2014, and (iii) the posterior stage between 2014 and 2024. The method aimed to present co-citation patterns and delineate intellectual structure changes (Small, 1974).

SNA functions to present dominant documents in a literature network and depicts the ecological relations through close observation of communication within an academic field (Scott, 1991). Unlike the directed graph of bibliographic study, the SNA framework reveals the co-cited relations amongst paired documents, collectively recognized by citing authors (Small, 1973). The current study employed VOSviewer for the visualization of co-citation, with further information on the algorithm for drawing networking diagrams available (Van Eck & Waltman, 2010). After retrieving the information based on VOSviewer networking diagrams, the current study employed Latent Dirichlet Allocation (LDA) to identify shared research topics, including practical approaches and methods, as well as the theoretical foundation of the core literature in each stream—the LDA model’s formulas compared the following four elements, as illustrated below.

Document-Topic Distribution:

The probability of topic z given document d is modeled as:

where θd,z represents the parameter of the topic distribution for document and d, and α is the Dirichlet prior parameter for the topic distribution.

Topic-Word Distribution:

The probability of word ω given topic z is given by:

where βz,ω is the parameter of the word distribution for topic z, and η is the Dirichlet prior parameter for the word distribution.

Dirichlet Distribution:

The Dirichlet distribution, which serves as the prior distribution for both the document-topic and topic-word distributions, has the probability density function:

where θ are the parameters of the topic distribution, α are the Dirichlet prior parameters, and K is the total number of topics.

Gibbs Sampling:

During the Gib sampling process in LDA, the document-topic and Topic-word distribution parameter are updated iteratively. The process of update for the topic assignment z of word ω in document d is proportional to:

where P(z|d) is the posterior probability of topic z in document d, and P(ω|z)i the probability of word ω in topic z.

The four formulas are central to the LDA model, defining its structure and the distribution of its parameters. In practical applications, the parameters of the LDA model were typically estimated using Marko Chain Monte Carlo (MCMC) methods such as Gibbs sampling. Notably, the current study employed a hybrid methodological approach in analyzing the results of SNA, utilizing (i) the LDA model for the initial identification of shared research topics and theories within core co-cited literature and (ii) a subsequent validation through the researchers’ examination to ensure a comprehensive and unbiased interpretation of core co-cited literature.

The document co-citation analysis (DCA) was treated as an initial step to calculate the number of co-citations of each 3DVWLL pair in the co-citation matrix. According to Small (1973), the co-citation frequency of a research pair in the matrix is counted when two target studies are jointly cited in the same citing document. The current study conducted the co-citation analysis based on 44 index studies selected and their citing references. Specifically, 1751 studies were listed in both rows and columns of a symmetric matrix, and the number of co-citations for the pairs of core articles in each cell was determined and counted for all references. By matching the reference data from 1751 cited studies, 203 co-cited studies tied to the 44 index studies were identified in the co-citation matrix. The dataset comprises 203 core studies from 75 journals, along with a pool of cited studies consisting of 1,751 works. The co-citation counts for each research pair were calculated based on matching the 1751 cited articles’ reference records.The core studies were paired document by document. Eight highly co-cited pairs were identified with the co-cited links over 5, which was considered core and central literature. The results further demonstrated a thrillingly developed disciplinary development in this field. The literature of these co-citation pairs was published in Computer Assisted Language Learning (CALL), British Journal of Educational Technology (BJET), European Association for Computer Assisted Language Learning (ReCALL), and Computer Assisted Language Instruction Consortium (CALICO).

The highest-referenced research tie was Jauregi et al. (2011) and Liou (2012), with six co-citations and a total link strength of 193 and 203, respectively. The two papers commonly discussed issues related to task design principles, communication competence, and differences between 3DVWLL and other CALL platforms, such as web-video communication. Jauregui et al. (2011) focus on four language learning issues: affective enhancement, communicative efficiency, affordance in the teaching phase, and inter-subjectivity. Liou (2012) further pointed out that the platform Second Life functioned as a complex pedagogical system from an ecological perspective. Table 2 summarizes further information on the other highly co-cited links.

Table 1 displays the specific linkage patterns of the topics raised in the highly co-cited literature, as illustrated in Fig. 3. Such patterns among the commonly mentioned issues reflect a picture of central consensus among the research of 3DVWLL (from 2000 to 2024). A further investigation identifies three strata of linkages among the issues commonly mentioned by the highly co-cited works of literature, according to evidence of Table 2: (i) the frequency of co-citation pairing works of literature and (ii) the occurrence of a specific research issue across the eight pairs of co-cited literature.

In contrast to highly co-cited pairs, the DCA also identified low co-cited pairs. This pair type included published works with low link strength, e.g., Gillen (2009) with 22 links and Steinkuehler (2007) with 8 links. Moreover, the low co-cited pairs also contained the high-low pattern of link strength, e.g., Peterson (2006) of 183 links paired with Svensson (2003) of 33 links.

The first stratum of linkage involved the pedagogical affordance of 3DVWs, encompassing both the learning and teaching phases, which comprised the appropriateness of task design and the other five issues: social interaction, affective enhancement, spatial cognition, collaborative learning, and communicative efficiency. The second stratum of linkage deals with interaction in the 3DVWs, including the linkages covering social interaction and two other issues: Conversational strategies and methods of communication. The third stratum of linkage involves specific learning approaches adopted in 3DVWs and appears in the correlation between discursive structures and collaborative learning. It was noteworthy that three particular issues played significant roles regarding the association across strata, i.e., (i) Task design of class activities, (ii) social interaction among the 3DVWs, and (iii) learners’ conversational strategies. As for the clustering of co-occurring research issues shown in Table 1 and Fig. 3, a close examination of the literature content revealed three key facts about 3DVWLL.

First, task design co-occurs with learners’ communicative efficiency. Moreover, the main direction of the discussion among the highly co-cited studies centers on the correlation between communicative competence, social interaction among users and learners, and learners' solidarity in virtual communities, as well as the affective enhancement or familiarity among learners, and spatial configurations or modules designed in 3D virtual worlds (3DVWs). For example, Dalgarno & Lee (2010) and Jauregi et al. (2011) highlighted the positive correlation between task types, such as decision-making tasks, and learners’ meaning negations, as well as the maintenance of coherent discourse. At the same time, both works further indicated that task-based language learning used in 3DVWs yielded significant achievements in a combination of collaborative learning.

Second, learners’ social interaction in 3DVWs involved conversational strategies and methods of communication. That is, learners’ sense of community increased their willingness to communicate by using paralinguistic features such as avatars’ gestures or movement (Cooke-plagwitz, 2008) or conversational strategies such as meaning negotiation, types of turn-taking and politeness strategies (Peterson, 2006).

Third, the design of spatial cognition in 3DVWs influenced learners’ communication efficiency. This ability influences learners’ communicative efficiency during verbal interaction and their learning affordance when exploring virtual environments (Peterson, 2006; Jauregi et al., 2011). Dalgarno & Lee (2010) pointed out that 3DVWs represent a spatial cognitive model that enhances interactive functions in learning activities by adopting the situated learning approach (Brown et al., 1989; Lave & Wenger, 1991).

To summarize, the results of DCA identified eight highly co-cited research pairs of 3DVWLL, representing a view in the field of 3DVWLL. By examining the issues shared by the eight research pairs, the current study attested to three strata of issue linkage: pedagogical capacity, learners’ interaction, and learning modes. Explicitly speaking, three issues play a pivotal role in connecting the three strata. They are (i) principles of task design as well as the implementation approaches (Dalgarno & Lee, 2010; Peterson, 2010; Jauregi et al., 2011; Liou, 2012; Lan et al., 2013), (ii) social interaction in 3DVWs as a cyber ecosystem (Peterson, 2006; Cooke-plagwitz, 2008; Deutschmann et al., 2009; Lan et al., 2013; Lan, 2014), and learners’ conversational strategies (Peterson, 2006; Deutschmann et al., 2009; Peterson, 2010; Liou, 2012).

This study conducted a social network analysis of the 203 core documents. Using the computation function of network visualization provided by VosViwer (Van Eck & Waltman, 2020), the layout of the 203 core documents was represented as 203 nodes, with every co-citation tie denoting a link between each pair of nodes. Figure 4 illustrates the chronological distribution of all core documents between 2000 and 2024, where the core documents published between 2005 and 2012 represented 67% of the population, with an average of 5 co-citations.

Figure 5a visualizes the computation of SNA, which identified six main streams of 3DVWLL based on studies between 2000 and 2024. Studies of Streams from 1 to 6 focused on communication, learning affordance, teaching affordance, pedagogical approach, interaction, and technological affordance in 3DVWLL, respectively. Figure 5a illustrates the clustering and linking of the network, delineating the within- and cross-boundary relations among the research streams. Figure 5b illustrates the visualization of network density among the research streams. Graphically, the density visualization demonstrated different centrality between representative and peripheral studies in the entire network.

The two figures illustrate three sectors of connectivity among streams. Streams 2, 3, and 5 formed a sector of intense connectedness among themselves, while Streams 1 and 4 were positioned on the periphery of the network, separate from the highly connected area. As for Stream 6, it was relatively isolated. The 203 nodes in the network clusters include 0.09% (19 studies) with over 100 linking strengths of their own, 0.34% (69 studies) with linking strengths between 100 and 50, and 0.58% (115 studies) of the nodes with linking strengths below 50.

Table 3 provides a general overview of the main streams of SNA between 2000 and 2024, including illustrations of active years, total link strength, average link strength, and co-citations. The SNA algorithm mapped the whole co-citation network of the 203 core documents, which comprised an overall of 4527 citation ties. The computation identified six streams of 3DVWLL with specific terminological labels, which align with those of previous systematic reviews (Lin & Lan, 2015; Reisoğlu et al., 2017; Avgousti, 2018; AUTHOR & COLLEAGUE, 2020) and the results of DCA. The identification of each stream is based on the application of various approaches, methods, and theoretical foundations, which are computed using the LDA model and subsequently examined by researchers.

The transition of research focus between the 2000–2014 network and 2015–2024 reveals sustained growth and evolution over time. Although teaching and learning affordances are essential in non-change streams, new elements constantly join the later network with increasing linking nodes and changes in stream content. The phase between 2010 and 2014 emphasizes the pedagogic dimension of comprehensible input and affect (Wehner et al., 2011; Jauregi et al., 2011; Liou, 2012), while the latter pays further concern to comprehensible output and learners’ cognition (Downey et al., 2012). Researchers of the early stage endeavor to build a virtual learning environment for fluent interaction, while researchers of the latter focus on learners’ learning cognition, such as deep learning (Hitosugi et al., 2014) and embodiment (Pasfield-Neofitou et al., 2015), meaningful language usage (Guy, 2010; Peterson 2006, 2010), and linguistic creativity (Dalgarno, 2010).

Central and less-centralized studies demonstrate the connections between theories and implementation, revealing potential future applications. First, the theoretical orientation beyond interactionalist and usage-based is encouraged to be explored. Recent studies have explored other theoretical possibilities, such as cognitive-based orientations (Song et al., 2023). 3DVWLL could further explore the interaction modes of the person-environment and person-object-person modes (cf. Lan et al., 2013). This research direction further substantiates Godwin-Jones’s (2023) findings on XR, specifically regarding the pedagogical strengths of virtual technology. Following this line of thought, problem-solving, and targeted exploration activities can further explore their pedagogical strength under various modes of interaction. Although the affordance of social-based exploration has been demonstrated, applying a game-based approach in 3DVWLL also warrants further exploration, especially regarding role-play learning activities. Moreover, 3DVWLL demonstrates the potential in specific linguistic areas, such as business or medical languages, i.e., language learning for specific purposes (LSP).

In a word, the identified research streams provide the following guiding principles for researchers and practitioners in research and course design:

3DVWLL’s intellectual structure, different from SLA’s between 1997 and 2018 (Zhang, 2019), prioritizes technological implementation and integrates theories and frameworks from education and psychology disciplines. For example, stream 1 of the second phase addresses the practical issues of implementing virtual educational environments, such as technical requirements, student distraction, and adequate equipment. On the contrary, SLA research focused on linguistic representations (Biber et al., 2004), cognitive complexity, and task sequencing (Robinson, 2005).

Zhang’s (2019) bibliometric analysis of SLA research underwent the following changes. The early stage focuses on acquiring linguistic representations and cognitively ontogenetic factors, while the later research stage has increased technological Applications. First, early studies primarily focus on the four skills of listening, speaking, reading, and writing. The later stage discusses the influence of social communication and vocabulary learning, particularly in relation to fixed phrases and formulaic language. Individual differences have become a hot topic of the later stage of research, with a focus on motivation, learning strategies, aptitude, and language anxiety. Second, the early stage of theoretical orientation focuses on cognitive approaches, such as input processing theory, while the later stage emerges with the application of sociocultural theory. Later developments in SLA studies show the emergence of a growing connection between technology and language teaching.

The current study observes a converging development between the SLA and 3DVWLL intellectual structures over the past two decades, despite 3DVWLL and SLA exhibiting divergent educational and ecological profiles in terms of their research inclinations and theoretical pursuits. 3DVWLL has increasingly explored language learners’ cognitive development and grammatical knowledge, e.g., Chen (2016), while SLA seeks to complement sociocultural and technology-assisted trajectories (Larsen-Freeman, 2018). As shown in Streams 4 and 5 of Table 4b, the later stage of core studies developed a research focus on the research topics, including actual language output (Cheng, 2010; Lan et al., 2013; Zheng et al., 2009), which had been the research traditions of SLA core studies. Therefore, the answer to RQ3 is clear: Research integrating technology as a component in language learning does not significantly diverge from SLA intellectual structure, particularly in its philosophical and theoretical aspects (Thorne & Smith, 2011), even though it originated from different ontological roots. The recent development of these two fields addresses several key points, and both fields adopt different perspectives in understanding the essence of language learning. The notion of affordance, as demonstrated in the results of SNA (cf. Tables 4a and 4b), plays a key role during the convergence of such a trajectory. That is, the two fields endeavor to offer the learners the maximum affordance by utilizing 3DVWs as a pedagogical resource (Lippman & Matthews, 2020). Figure 7 summarizes the phenomenon.