2.1 Theoretical perspectives

As followers of one of the two theoretical perspectives, interactionist researchers define and classify CSs only by observing the outputs of the speakers, while psycholinguist researchers take into account the cognitive processes the speakers have undertaken while using CSs and define CSs by focusing on the communicative problems underlying their use. The L1 SPM (Levelt, Reference Levelt1989) has been the most widely accepted theoretical framework among psycholinguists and has served as the foundation for numerous experimental studies (Dörnyei & Kormos, Reference Dörnyei and Kormos1998; Poulisse, Reference Poulisse, Schreuder and Weltens1993). It covers the time frame from conceptualizing the communicative intention of the speaker into a preverbal message and on to the articulation of the overt speech. In the model, there are three monitor loops (for the representation of Levelt’s model, see Dörnyei & Kormos, Reference Dörnyei and Kormos1998: 352) whose functions are to check if anything is amiss with the speaker’s outputs at various levels of the speech production process. When the speaker detects something wrong with their output after or before articulation, they could start over in coding their message, change some or all of their message or repair it. The speaker could also repair it when they cannot negotiate with their interlocutor. Since Levelt’s model was originally designed for L1 speakers, in this study it was necessary to adopt an L2 SPM. De Bot (1992) designed his L2 SPM by adapting it from Levelt’s L1 SPM; therefore, this study was based on both Levelt’s and de Bot’s SPMs. Poulisse and Bongaerts (Reference Poulisse and Bongaerts1994) claim that de Bot’s model is the only one that fully describes the L2 speech production process. In contrast to Levelt’s model, in de Bot’s model, the L2 speaker has two conceptualizers (by which the language of the message is specified) and two formulators (by which the syntactic, morphological, and phonological processes of the language to be used are determined). Since there is an interaction between the languages of an L2 speaker, the CSs such as tip-of-the-tongue phenomenon, code-switching, use of similar-sounding words and many other CSs could be explained via de Bot’s model.

As Dörnyei and Kormos (Reference Dörnyei and Kormos1998) based their study on Levelt’s L1 SPM from the psycholinguistic perspective, their CS taxonomy was adopted here. In compliance with their theoretical view, they designed a comprehensive CS taxonomy by associating the L2 communication problems and the means used to address those problems (i.e. communication strategies) with the speech production process. To properly identify and classify the CSs, Dörnyei and Scott (Reference Dörnyei and Scott1997) identified four main problem sources (resource deficit, processing time pressure, perceived deficiency in one’s own language output, perceived deficiency in the interlocutor’s performance), which Dörnyei and Kormos integrated into their taxonomy. Moreover, Dörnyei and Kormos integrated each problem-solving mechanism (PSM) into their taxonomy by matching them with the relevant speech production phase in Levelt’s model. Using Levelt’s model as the foundation, Dörnyei and Kormos aimed to investigate at what stage and why the speakers may have encountered a communicative problem during speech production, which in return led to the accurate identification and classification of communication strategies.

2.2 Previous communication strategy research

Considerable research has been conducted to investigate CS use, which can be categorized according to the context in which CSs were studied.

3.1 Setting and participants

The study was conducted at the ELT department of a Turkish state university. The data were gathered from the 30 seniors (fourth and final year of study) who took the course “Materials Evaluation and Design” in the 2019–2020 fall term. The participants’ mean score on the Foreign Language Test (YDS) administered in Turkey was 86 out of 100, which corresponds to 103 points on the TOEFL IBT and C1 CEFR proficiency level, according to the Turkish student selection and placement center’s equivalency table (Türkiye Cumhuriyeti ÖSYM Başkanlığı, 2017). The background questionnaire revealed that the majority of participants were familiar with SCMC environments: 90% of them used VC tools and 86.7% used VWs prior to the study.

Neither gender nor age was examined as factors since the age range of the participants was quite narrow (from 21 to 22 years old) and because a large majority were female, making a comparison between males and females unrealistic given the relatively small sample size.

3.2 Data collection instruments

The data were collected via audio and video recordings of three opinion-exchange tasks (see Appendix E, in the supplementary material, for example task), a background and post-task questionnaire, and interviews consisting of retrospective think-aloud (RTA) protocols. In the background questionnaire, there were 34 questions (18 multiple choice, nine open ended, six demographic, and one matrix). In the post-task questionnaire, there were 16 questions (14 open ended and two multiple choice). The data obtained from the VC, VW, and F2F environments include only the audio and video chat recordings, not the text chat recordings. In order to maintain the parallel nature of the communication across the three modes, the participants in VC and VW were instructed not to make use of text chat during their communication. While the role of text chat in VC and VW modes has been examined by a number of scholars such as Hung (Reference Hung2012) and Zhao (Reference Zhao2010) in VC mode and Peterson (Reference Peterson2006) in VW mode, no other study has examined audio- and video-based interaction comparatively in VC and VW.

3.3 Data collection procedures

The data collection tools were first piloted with seven participants in the 2018–2019 spring semester. After revisions, Zoom was selected as the VC tool (Figure 1) for the main study, and Second Life was selected as the VW tool (Figure 2). For the F2F environment, a quiet room at the ELT department was used (Figure 3).

Figure 1.

A screenshot of Task 1 in Zoom

Figure 2.

A screenshot of Task 2 in Second Life

Figure 3.

A screenshot of Task 3 in F2F environment

The 30 participants were divided into six groups of five individuals, with each group choosing its own members. Each group completed one task per week, with each task taking place in a different environment (Task 1 in VC, Task 2 in VW, and Task 3 in F2F). In total, each group completed three tasks in three different environments (Table 1). All tasks were designed as opinion-exchange tasks so as not to add the task type effect as a variable. The content of the tasks was consistent with the “Materials Evaluation and Design” course, with each task referring to a different topic in the course syllabus and including open-ended questions to enable free exchange of opinions. In total, 18 audiovisual recordings (six per environment) were obtained from the discussion tasks. Therefore, every group spent around 45 minutes on each task and, overall, of all the six groups in three environments, 14 almost hour-long recordings were included in total. One researcher from the researcher team attended all discussion tasks as a non-participant observer (Creswell, Reference Creswell2013). To minimize the researcher’s impact, the researcher turned off video and audio in the VC, used an invisible avatar in the VW and left the F2F room with video cameras on. While Task 1 (in Zoom) was recorded using Zoom’s own recording feature, Task 2 (in Second Life EduNation III ^{Footnote 1} ) was recorded using an external screen recording tool, Camtasia, and Task 3 (F2F) was recorded with two video cameras, one main camera and one backup camera.

Table 1.

Data collection process

Total duration: 8 weeks

The post-task questionnaire investigated the participants’ perceptions toward the three environments and whether their perceptions influenced how they dealt with communication problems. Therefore, the post-task questionnaire was used to better understand the reasons behind their CS use, which in return helped to more clearly identify the CSs in the data.

During the one-to-one interviews conducted with each participant (N = 30), one researcher from the researcher team showed preselected video and audio recordings of the tasks in which the participants used CSs. The researcher then asked the participants to describe the communication problems they encountered and how they addressed them. The RTA protocol is utilized to comprehend the reasons behind the participants’ CS use in order to accurately identify the CSs. RTA protocols have been employed in numerous CS studies and are recommended for use in CS research (Dörnyei & Scott, Reference Dörnyei, Scott, Turley and Lusby1995; Færch & Kasper, Reference Færch and Kasper1983).

3.4 Data analysis methods

The video recordings of the tasks were transcribed and analyzed using the qualitative data analysis software MAXQDA 2020. The transcriptions included not just the verbal language but also the prosodic features likely to affect or change the meaning. In order to represent the verbal and nonverbal behaviors with precision, Jefferson’s (Reference Jefferson and Psathas1979) transcription conventions were used when necessary to identify the CSs clearly. To code the CSs, Dörnyei and Kormos’s (1998) taxonomy was used and when CSs not included in their taxonomy were discovered in the data, they were coded utilizing other researchers’ taxonomies (Bejarano et al., Reference Bejarano, Levine, Olshtain and Steiner1997; Chen, Reference Chen1990; Dörnyei & Scott, Reference Dörnyei, Scott, Turley and Lusby1995, Reference Dörnyei and Scott1997; Færch & Kasper, Reference Færch and Kasper1983; Smith, Reference Smith2001; Tarone, Cohen & Dumas, Reference Tarone, Cohen and Dumas1976). In addition, the CSs that were revealed in the data but did not exist in any of the aforementioned literature were also included in the codings (Appendix B). A code was assigned to each CS identified and was introduced into the MAXQDA system. The coding procedure was repeated for the data obtained from each environment (VC, VW, and F2F). Then, the frequency of CS occurrence in each environment was calculated, and the values were compared to determine if the participants’ use of CS varied between environments. To ensure the coding reliability, the CS coding taxonomy and a random sample of 20% of the data were independently coded by two experts in the field. The intercoder reliability was calculated via Cohen’s kappa statistics to be .81, indicating almost perfect agreement (Cohen, Reference Cohen1960).

For each of the communication problems that occurred in the data, several key steps were undertaken:

1. 1. One of four sources for each problem was identified based on Dörnyei and Kormos (Reference Dörnyei and Kormos1998).

2. 2. The strategy (or sometimes multiple strategies) used to resolve each of these problems was determined based on the taxonomies from the aforementioned literature, resulting in the identification of 67 varieties (e.g. umming and erring, restructuring, error repair, etc.) of strategies utilized (Appendix A2).

3. 3. The frequency of strategies was calculated, with 8,870 instances found overall (Table 2).

   Table 2.

   Frequency of communication strategy (CS) use in three environments

   

The data obtained from the open-ended questions in the background and post-task questionnaires were analyzed using the constant comparison method (Glaser & Strauss, Reference Glaser and Strauss1967), while the closed-ended questions were calculated via frequency analysis. The interview data were audio recorded and analyzed via MAXQDA after transcription.

4.1 CSs in VC, VW, and F2F environments

As shown in Table 2, the highest number of strategies used was in VC (Zoom: 3,401 times, 38.34%) followed by F2F (3,102 times, 34.97%), while the least number used was in the VW (Second Life: 2,368 times, 26.69%).

As seen in the distribution pattern across environments in Appendix A2, a wide variety of strategies were used in each of the three environments. This finding is consistent with previous studies (Hung, Reference Hung2012; Kost, Reference Kost and Magnan2008; Smith, Reference Smith2003) conducted to determine the CSs used by language learners in CMC environments.

Among the 67 strategies used, 61 were used at least once in VC, 60 in F2F and 56 in VW. Of the 67 strategies on the strategy coding list, the emojis and gestures strategy is not applicable in F2F and VC, and nonverbal strategy markers are not applicable in VW.

One of the most striking findings was that umming and erring and self-repetition strategies were the two most frequently used strategies in all three environments (Appendix A1). According to Dörnyei and Kormos (Reference Dörnyei and Kormos1998), the most frequently used strategies including these two are mainly caused by the same problem source: “processing time pressure” (VC: 1,870; VW: 1,190; F2F: 1,555). This shows that the participants needed to gain time while planning their speech or when they encountered a communication problem. Dörnyei (Reference Dörnyei1995) indicated that the most basic reason behind this problem source is that during speech production, processing time is not adequate for L2 learners. This could be because the L2 language production process is not as automatized as L1 (de Bot, Reference de Bot1992). As suggested by Uztosun and Erten (Reference Uztosun and Erten2014), even though the participants in this study were proficient English language learners, they may have had difficulties with fluency during the discussion tasks due to the limited opportunities to speak English outside of the classroom in their daily lives.

In addition, Hung (Reference Hung2012) noted that participants may experience “processing time pressure” in SCMC and F2F environments due to reasons such as feeling the need to respond to their interlocutors or plan their speech instantly. Therefore, due to these possible pressures, the speakers may have used time-gaining strategies (unfilled pauses, fillers, or self-repetition) more than any other strategy types in these environments (Appendix A1).

The reason why the participants used the strategies related to “perceived deficiencies in the interlocutor’s performance,” also known as “meaning negotiation strategies,” most infrequently (VC: 52; VW: 47; F2F: 48) is that they mostly attempted to solve their communication problems without expecting any assistance from their interlocutors. This confirms the view of the researchers (Bialystok, Reference Bialystok1990; Kellerman, Reference Kellerman, Phillipson, Kellerman, Selinker, Sharwood Smith and Swain1991; Poulisse, Reference Poulisse, Schreuder and Weltens1993) that the speaker’s role in CS use is also important and that speakers can use CSs without collaborating with their interlocutors. Stating that their findings highlighted the non-interactional aspect of CSs, Yarmohammadi and Seif (Reference Yarmohammadi and Seif1992) also encountered the same issue in their own study.

In addition, there were some factors affecting the CS use of the participants: proficiency level, task type, and the study context. Many researchers such as Ghout-Khenoune (Reference Ghout-Khenoune2012), Rosas (Reference Rosas2018), and Smith (Reference Smith2003) argue that free discussion tasks encourage a wide variety of CSs, while more guided tasks like jigsaw demand less variety of CS use. In this study, the opinion-exchange task was used, a type of free discussion task, and it was seen that the participants relied on various CS types. Consequently, it seems that opinion-exchange tasks might be influential in the participants’ wide variety of strategy use. Further, the intense use of strategies related to “processing time pressure” confirms that the participants mostly had communicative problems in terms of fluency rather than accuracy, which shows that their proficiency level in English might have been one of the factors affecting their CS use. Also supporting the probable connection between proficiency level and CS use, various researchers (e.g. Bialystok, Reference Bialystok, Færch and Kasper1983; Bialystok & Fröhlich, Reference Bialystok and Fröhlich1980; Fernández Dobao, Reference Fernández Dobao2001; Hsieh, Reference Hsieh2014; Rosas Maldonado, Reference Rosas Maldonado2016) suggest that proficient learners use more cognitively challenging strategies such as self-repair (i.e. error repair, appropriacy repair, different repair, and rephrasing repair). In addition, it was observed that the use of three different environments might have influenced the frequency and types of CS, as also found by Zhao (Reference Zhao2010). Presented below is a participant response from the post-task questionnaire that highlights the various affordances of the environments:

This issue carries over to facial expressions. The heads-up display (HUD) did include facial expressions such as SURPRISE, but those choices were limited. In addition, in VC and F2F, the participants could see the full nuance of facial expressions, which are simply not available in VW at this time (although virtual environments are beginning to use webcams to express the facial expressions and gestures of their users for their avatar forms). Therefore, even though, as noted in Figure 4 and also in Appendix A1, emojis and gestures was one the 10 most common CSs, some participants still did not feel that this was the case:

Figure 4.

Ten most frequent communication strategies in three environments

Although there are similarities in strategy use across all environments, the findings indicate there are also differences (Figure 4). Among the 10 most frequent strategies overall, seven (umming and erring, self-repetition, error repair, use of all-purpose words, appropriacy repair, different repair, fillers) were common in all three environments. However, it is noteworthy that although the participants can see each other’s faces in VC, nonverbal strategy markers were not among the top 10 frequent strategies in this environment. In addition, although stuttering strategy can be used with or without seeing each other’s faces, it is another notable finding that the stuttering strategy was among the most frequent strategies in F2F discussions, whereas it was not as prevalent in the other environments.

The frequent use of some repair strategies, such as error repair, appropriacy repair, and different repair, across all environments is likely associated with the participants’ higher proficiency level in English. As Levelt (Reference Levelt1989) indicated, to utilize repair strategies, the speaker’s monitor mechanism should be competent enough to realize that there is a communication problem waiting to be fixed so that a signal can be sent to the conceptualizer. In this way, the conceptualizer could start over the speech production process to fix the communication problem. Therefore, as Rosas Maldonado (Reference Rosas Maldonado2016) notes, repair strategies are more cognitively challenging and proficient learners employ these strategies more, so it is no surprise that the participants in this study used repair strategies so frequently. As a result, the findings are consistent with others (Bialystok, Reference Bialystok, Færch and Kasper1983; Bialystok & Fröhlich Reference Bialystok and Fröhlich1980; Fernández Dobao, Reference Fernández Dobao2001; Hsieh, Reference Hsieh2014), revealing that language proficiency level has an effect on the types of strategies employed. In line with our findings, studies conducted in CMC environments (Hung, Reference Hung2012; Kost, Reference Kost and Magnan2008; Lee, Reference Lee2001, Reference Lee2002) also showed that repair strategies were frequently used. Consequently, it seems that not just F2F but also CMC environments encouraged the use of repair strategies. As Jepson (Reference Jepson2005) revealed, speaking encourages more negotiation and repair moves, which is one of the affordances of the two CMC environments in this study. This explains why the participants frequently used the repair strategies in CMC environments as well. It is important that these three environments encourage the use of repair strategies because according to Swain’s (Reference Swain, Gass and Madden1985) output hypothesis, for the learners to produce meaningful output, they need to be able to monitor their production, realize their errors, and revise their output by repairing it.

Also, it is essential to discuss the strategies whose use varied across environments to determine the impact of the environments on strategy use. Strategies that differed across environments include indirect appeal for help, unfilled pauses, responding, nonverbal strategy markers, emojis and gestures, framing, giving assistance, lexical retrieval/word search, unasked-for help, and stuttering strategies. Most of the differences originated from the various affordances of the environments. In VW, for instance, participants are represented by avatars; therefore, their use of gestures and facial expression is limited compared to VC and F2F. Unlike in F2F or VC, where gestures may be more “automatic” in nature, participants in VW were required to click a button on their HUD, and were limited to 12 emojis and six gestures, such as hand up, nod, clap, LOL, and woo! (Figure 2).

Unfilled pauses were used less frequently in VW because it was observed that in some cases, whenever a participant used an unfilled pause, other participants assumed their interlocutor was having a technical problem and took the turn. The fact that in VW the speakers favored umming and erring strategy (605 times) more intensely than unfilled pauses (63 times) may support this. Smith (Reference Smith2001) also found that in a text-based CMC environment where participants cannot see each other as well, they showed intolerance toward unfilled pauses and preferred to use fillers instead. Another important finding was related to indirect appeal for help strategy. Even though the participants could see each other in VC as well as in F2F discussions, they used this strategy much less frequently in VC. Although gestures and facial expressions are visible in VC, it is hard for the participants to figure out who the speaker is making eye contact with and from which member specifically they are asking for indirect help. As nonverbal cues are crucial for the interlocutor to understand the signals of the speaker’s indirect request for help, this might explain why the participants used this strategy less frequently.

4.2 New strategies found in this study

This study uncovered 10 new strategies (Appendix B) that did not previously exist in the literature: stuttering, lexical retrieval/word search, unasked-for help, insertion, lexical substitution, managing the turns, phonological code-switching, recap of previous turns, use of synonyms, and lexical wavering. Although only one of these (stuttering) was included in the top 10 strategies, each of them still had an impact during the group discussions. Below, each strategy is defined, along with the cause of the communication problem, and how the strategy may be inserted into Dörnyei and Kormos’s (1998) taxonomy. Transcription samples for each strategy are given in Appendix C, and the strategy table along with definitions, problem sources, PSMs and example sentences from the data are given in Appendix B.

Although the theoretical framework of this study was mainly driven by the psycholinguistic perspective, CSs that enhanced communication were also included among the strategies found in this study in addition to the CSs that were discussed based on the problem sources introduced by Dörnyei and Kormos (Reference Dörnyei and Kormos1998). Although it is certainly the case that many studies that have focused on communication problems are both important and informative, strategies to enhance communication have often been ignored. However, as often shown by interactionist researchers, speakers do not always use CSs because they have a communication problem, but just to enhance communication and “act as better communicators” (Bejarano et al., Reference Bejarano, Levine, Olshtain and Steiner1997: 205).