Open Research Online Communicative styles, rapport and student engagement: An online peer mentoring scheme

This article investigates the communicative styles of three different peer mentors in the context of online language learning, and considers their effect on student engagement. A key objective is to show how an innovative corpus-based technique, keyword analysis, can be used as a first step towards identifying communicative styles. We view communicative style as a linguistic means by which rapport is managed amongst participants (Spencer-Oatey 2008). Our primary data include 685 forum posts, of which 273 (over 26,000 words) were by the mentors at the heart of our study. We show that the three mentors have different communicative styles: different rapport management orientations are achieved in different ways. Furthermore, we bring together multiple data sources, including participants’ posts and self-reported perception data. This allows us to find evidence on if and how communicative styles impact on student engagement and perception. We discovered that rapport enhancement aligns with increased active participation, especially if a self-effacement strategy is used, and positive student perception, but that the lack of such rapport does not automatically imply negative student perception.


INTRODUCTION
Online distance language learning has been increasingly accepted by the mainstream academic community, and this has opened educational doors to many learners who might not otherwise have had the opportunities to study another language due to time and location restrictions (Garrison et al. 2000). However, it also presents a range of challenges, including social isolation, and anxiety (Hurd 2005(Hurd , 2007. To an extent, this has been offset by increasingly advanced technology that has made it possible for language learners and tutors to communicate with each other asynchronously and synchronously, thereby mitigating feelings of isolation (Lamy 2013a;Kan and McCormick 2014;Delahunty et al. 2014). Nevertheless, the mere existence of tools is not enough: proactive human intervention is required. Good distance teaching institutions actively 'try to take account of the socialization needs of students, recognizing this as key to student achievement in non-campus-based learning' (Lamy 2013b: 226). A means of meeting such needs is peer support, also known as 'peer coaching' and 'peer mentoring', which has been in use for a long time in higher education institutions (Giles and Ody 2015;Chilvers 2016).
The Open University, where one of the authors is based, is the leading distance learning institution in the UK. It provides a staff-moderated asynchronous online student forum for each module. In 2014, in an attempt to improve retention, a series of dedicated peer mentoring threads were introduced into module-wide student fora to provide an additional element of support to all Level 1 language students on 10 modules at critical points in their learning journey. For each module, one student, who had recently studied the same module, was given a mentoring role to provide moral support and practical advice online in the dedicated peer mentoring threads. At the end of the academic year 2014-2015, the scheme was evaluated with overwhelmingly positive feedback (Fayram et al. 2018). One noticeable result from the evaluation was the large difference in the number of posts by students: in one module only 28 per cent of the total number of posts was from students, whereas there were 62 per cent and 71 per cent in the other modules. We hypothesized that the nature of the communicative styles of the mentors had an impact on student engagement in posting.
One of the aims of this article is to clarify the nature of a communicative style. An even more important aim is to show how keyword analysis, a technique used in corpus linguistics, can help operationalize the study of communicative style. Hitherto, as Spencer-Oatey (2008: 28) notes, there has been no agreement on how to identify the clusters that constitute a communicative style. We argue and demonstrate that keyword analysis can help to identify such a style, though it is only a first step. The communicative side of such a style relies on the identification of communicative functions in context. Statistically significant keywords do not automatically perform a communicative function, and, when they do, it is necessary to know what function they are performing. Thus, a second step was undertaken, namely a functional analysis, which was informed by Spencer-Oatey's 'rapport management framework ' (e.g. 2008), a framework that has been used to analyse online discussion fora (e.g. Gonzalez 2013;Hopkinson 2014). Finally, we investigate the way in which these rapport-oriented communicative styles align with student engagement, particularly in terms of the nature of participation (i.e. postings) and student perceptions of each mentor.
The following section reviews the literature on online learning and discussion fora, thereby providing a backdrop for our own forum data. In order to capture the linguistic characteristics of the mentor contributions, it also builds a definition of a 'communicative style', starting with the notion of style, then identifying communicative style, and finally describing, with reference to Spencer-Oatey's (2008) rapport management framework, the communicative strategies from which communicative styles are constituted. Turning to methodology, our forum data sources and methods of analysis are described, focusing in particular on the innovative use of corpus-based text analysis tool to identify features of a mentor's style, but also noting subsequent steps that need to be taken to identify communicative function (Section 3). The analyses and results are presented in two sections from both the mentor and student perspectives (Sections 4 and 5). The final section offers discussion and conclusions.

LITERATURE REVIEW
2.1 Online learning and discussion fora Moore and Kearsley (1996) highlighted the role played by technology in separating distance education from conventional education, and highlighted the importance of providing motivational support to distance learners to make them active participants in the learning process. Baumann et al. (2008) and Murphy et al. (2010) also argue that the success of learning languages at a distance depends on how well learners are supported because, in addition to challenges such as social isolation, most distance learners have work and family commitments that compete with their study time (Hurd 2005;Murphy et al. 2010). One form of such motivational support is to build a sense of belonging to an online learning community amongst learners via institutionalized staff-moderated online discussion fora.
A particular difficulty, however, is the 'challenges of innovation' (White 2014: 548) in distance language learning course environments, which is also noted by Hampel (2014: 17): 'communication using digital media can be cognitively challenging, especially when it takes place in still relatively unfamiliar environments and involves additional mediation compared with face-to-face interaction'. It is, therefore, not surprising that forum participation and interaction is often low (Fayram et al. 2018). Much research has noted different kinds of participation, ranging from initiating a conversation, reading and commenting on others' posts, to just reading others' posts. In most public online discussion fora, it is observed that 90 per cent of people simply 'lurk' rather than contributing content (Wilkerson 2016). Many researchers define forum 'participation' as the posting of an online message (Dennen 2008;Ng et al. 2012), whereas others regard just 'reading' as a form of 'passive participation' (Kan and McCormick 2014). In this study, we use the term forum 'participation' to refer to anybody who visits a forum thread, reading and/or contributing content through posting a message. In our view, the term 'lurking' may under-rate the forum participation; we consider 'lurking' as a form of participation. To distinguish forum 'interaction' or 'active participation' from 'passive participation', we follow Ingram and Hathorn (2004) and Weinberger and Fischer (2006) in stipulating that a comment or question is a prerequisite for the former terms to be used.
The most widely used method in previous studies to analyse forum discussion is content analysis, namely 'a variety of textual analyses that involve comparing, contrasting, and categorizing a corpus of data in order to test hypotheses' (Schwandt 2007: 41). For example, the study conducted by Ng et al. (2012) examines techniques used in two peer facilitated discussion fora in blended-learning graduate courses (i.e. a mixture of online and face-to-face delivery), looking for evidence of the social construction of knowledge. However, Ng et al. give no linguistic analysis of communicative styles to align with the 'seven good techniques' they examine (e.g. 'Questioning' and 'Fostering interaction'). This is not to say that usually there is a total absence of linguistic analysis of forum posts. Liaw and English (2013), for instance, attempt to identify specific linguistic characteristics of forum participants using Halliday's social semiotic perspective (1978). Their analysis reveals how participants address each other, manage turn-taking as well as lexical choices. However, unlike our study, they do not align their textual findings with data on participant engagement.

Communicative styles
It is clear from even a cursory glance at our data that the mentors had their own individualized ways or styles of mentoring the students. The notion of 'style' has been much discussed within a number of disciplines, including literary studies and sociolinguistics. Four aspects, in our view, are crucial to thinking about linguistic style.
1 Style works as a whole: It is not constituted by one feature or dimension.
Social meanings of style are said to reside in 'constellations of features which are interpreted together' (Auer 2007: 12; see also Ervin-Tripp 1972). An atomistic approach to style which picks out some features but not others would not present a complete picture. 2 Style is lent meaning by contrasts with other styles: As Irvine (2001: 22) puts it, styles 'are part of a system of distinction, in which a style contrasts with other possible styles, and the social meaning signified by the style contrast with other social meanings'. For example, a colleague of one of the authors sometimes adopts a specific north-west British accent in order to mark identity differences with academics speaking with southern British accents. 3 Style is contextually relative: Enkvist (1964: 29) suggests that 'style is concerned with frequencies of linguistic items in a given context, and thus with contextual probabilities'. In other words, styles are constellations of features that correlate with particular contexts. In fact, this is a necessary feature of a style, otherwise it would be impossible to delimit any particular one. 4 Style has a two-way relationship with context: In other words, 'contexts can influence choice of style, but also choice of style can influence context' (Semino and Culpeper 2011: 301). Traditionally, style has been conceived of as being shaped by contextual constraints (e.g. a formal meeting produces a formal style). But more recently, scholars have recognized that the use of a style can engineer a change in context (e.g. an informal style can change a formal meeting into an informal one). Gumperz's (1982) work on 'contextualization cues' has been pioneering in this regard.
In this article, we take a mentor's style to be his or her total set of distinctive linguistic choices relative to those of others in the same context.
We are interested in not only how the mentors' styles differ, but also how they differ in communicative terms, and thereby differ in promoting or otherwise student interaction, and how they are differentially perceived by students. What is communicative style? In a general sense, all style is communicative, but that is obviously not the intended sense when scholars use the expression. Instead, the function of the modifier communicative is to restrict the notion of style to a subset of styles that contribute to the constitution of particular communicative activities in which they perform particular communicative functions. Selting spells out the relevant aspects: In relation to an activity type or genre that can be kept constant as a tertium comparationis, meaningful different ways of constituting this activity type or genre can be described as different communicative styles. Styles suggest additional social or interactional meanings which often have to do with self-presentation, definition of the situation, definition of the relationship between speaker and recipient, framing of activities and situations, etc. (2009: 21) Online language teaching and learning is just one such 'activity type'. Our goal is to investigate the communicative styles of mentors, styles that contribute to the constitution of that activity type and perform particular communicative functions within it.
A starting point for describing the building blocks of communicative styles is the over 50 'politeness strategies' listed in Brown and Levinson (1987) and other works, and also works on impoliteness (e.g. Culpeper 2011), because these lie at the heart of social communication, especially relationships. The term 'strategy' here is used to denote routinized ways in which communicative functions are achieved (e.g. 'could you X' is a highly regular formula for achieving a moderately polite request in British cultures). Indeed, Spencer-Oatey (2008: 21-28) devotes space to lists of such strategies because they relate to rapport management, which we will introduce in the next subsection. There is no uncontroversial finite set of strategies or of dimensions along which one might plot them. Nevertheless, there are a few dimensions along which strategies vary that are regularly mentioned in (im)politeness studies and crosscultural/intercultural studies. Drawing on Spencer-Oatey (2008: 28-31), we summarize these briefly below, partly as a way of presenting the array of relevant communicative strategies, rather than tying our later analyses to a specific set of dimensions. We give the labels used in Spencer-Oatey (2008), with slight adjustment.

Expressiveness-restraint
This dimension is variously labelled by researchers; for example, it subsumes the 'positive politeness-negative politeness' of Brown and Levinson (1987), and the 'involvement-independence' dimension of Scollon andScollon (2001[1995]). 'Expressiveness' typically includes, for example, attending to the hearer, expressing approval or sympathy, claiming in-group membership or common-ground, and using given names and nicknames. 'Restraint' typically includes, for example, giving the hearer options, minimizing impositions, apologizing, distancing the speaker and/or the hearer, and using family name and titles.

Directness-indirectness
A dimension that underpins classic work in politeness (e.g. Brown and Levinson 1987;Leech 1983), and cross-cultural pragmatics (e.g. Blum-Kulka et al. 1989). 'Directness' typically includes, for example, the use of mood (e.g. imperatives carry out requests) and performative verbs (e.g. 'I order you to'); 'indirectness' typically includes hints. Conventional indirectness (e.g. ability questions to do requests, such as 'Can you pass the water?') lies somewhere in the middle of the dimension. There are also various devices for softening a message (e.g. hedges) or strengthening it (e.g. taboo words).
Self-enhancement-self-effacement Spencer-Oatey (2008: 31) cites Ting-Toomey (1999: 107-8), who states that: 'The self-enhancement verbal style emphasizes the importance of boasting about one's accomplishments and abilities. The self-effacement verbal style, on the other hand, emphasizes the importance of humbling oneself via verbal restrains, hesitations, modest talk, and the use of self-deprecation concerning one's effort or performance.' Needless to say, a description that a linguistic expression is expressive, direct, self-enhancing, and so on is not simply a description of linguistic form, but an interpretation that a certain linguistic form in a certain linguistic context (e.g. activity type) is performing a certain communicative function.

From facework to rapport management
The notion of 'face' has been discussed in studies of interpersonal communication for decades. It is incorporated into Spencer-Oatey's (2008) rapport management, which is why we introduce it and explain its relevance here (we will not comment on Spencer-Oatey's notion of 'sociality rights', which form part of rapport management, because they are less relevant). Rapport management, in common with other relational frameworks, does not follow Brown and Levinson's (1987) notion of face but shifts back to Goffman's original definition: the positive social value a person effectively claims for himself by the line others assume he [sic] has taken during a particular contact. Face is an image of self delineated in terms of approved social attributes. (1967: 5) Note that Goffman's notion is dependent on others: face is a person's positive value claims as reflected in the assumptions made about them in interaction. Thus, thinking of oneself as intelligent is not a matter of face, but meeting one's face claims to be intelligent by being treated as an intelligent person would be. This interdependence is what makes it different from Brown and Levinson's (1987) notion, where it is much more the psychological property of the individual. Given its dependence on what others do in interactions, face is dynamic and emergent. All this is pertinent to learning situations. Not knowing something or getting something wrong are not at all likely to be a 'positive social value'-they are potentially face-damaging. Moreover, learning contexts, such as a course online discussion forum, are often public and thus have the potential for heavy loss of 'face'. Indeed, a learner may conclude that saying nothing is better than risking face loss. Consequently, a tutor must constantly demonstrate in their interactions a 'line' that supports the kind of face the learners might wish to claim. In other words, they must engage in 'facework', 'the actions taken by a person to make whatever he [sic] is doing consistent with face' (Goffman 1967: 12).
There are four ways of orienting actions to face-or, put differently, four face-related communicative functions that linguistic material, including strategies, might perform-within Spencer-Oatey's (2008: 32) rapport management framework: 1 rapport enhancement orientation: a desire to strengthen or enhance harmonious relations; 2 rapport maintenance orientation: a desire to maintain or protect harmonious relations; 3 rapport neglect orientation: a lack of concern or interest in the quality of relations; and 4 rapport challenge orientation: a desire to challenge or impair harmonious relations.
Rapport enhancement is in tune with Leech's (1983) model of politeness that accommodates acts that simply enhance politeness or face (e.g. compliments), perhaps to strengthen social relations. One might reasonably expect that tutors, or student mentors in our case, deploy 'face-enhancing actions' (e.g. welcoming actions at the beginning of a session) as a means of creating a facesupportive atmosphere. Rapport maintenance could be simply a matter of performing routine politeness behaviour where it is expected, or, in tune with Brown and Levinson (1987), a matter of restoring relations in the light of face threatening behaviour (e.g. a request). Note that the difference between maintenance and enhancement is that without effort to maintain rapport, rapport will be threatened; enhancement, in contrast, is a relatively free gift. As tutors inevitably have some kind of directive role (e.g. requesting students to complete tasks), one might hypothesize that face maintenance will be important. Moreover, pointing out errors and suggesting corrections are likely to be highly face-threatening acts which require significant face counter work. Rapport neglect, as the term 'neglect' implies, has negative connotations: it is the neglect of face support or redress where face threat or loss has occurred. Spencer-Oatey's suggestion that a 'focus on self' may lead to the neglect of another's face is plausible in many situations, but perhaps less so in the tutor-student situation where the tutor's role is very much focused on the other, that is, the student. What is more likely is that the tutor is not aware of the face support or redress needed or miscalculates how much is needed. Rapport challenge accommodates impoliteness (see e.g. Culpeper 1996). This, we assume, is not generally relevant to learning situations. However, highly critical feedback or an admonishment for lack of attendance can easily slip into rapport challenge. To these orientations, we add one other, rapport neutrality. In analysing the data, it became apparent that we needed a category for items that had relatively little to do with Spencer-Oatey's four kinds of rapport; in fact, rapport was not salient. Typically, these involve statements of information concerning aspects of the topic and the specific learning environment (e.g. telling students where they can upload their assignments).
These five general rapport orientations or functions help capture the interpersonal nature of the communicative strategies that comprise communicative styles. However, it should be remembered that they are general. As will be seen in our analysis, different strategies can achieve the same rapport functions but in different ways. An analysis will need to attend to the specifics of these strategies, and not least their linguistic realization. Moreover, a crucial point for this article is that rapport management is not simply achieved by individual communicative strategies but by clusters of communicative strategies constituting communicative styles (see Spencer-Oatey 2008: 28, for a list of scholars who make this claim). The problem, however, as Spencer-Oatey (2008: 28) notes, is that 'there is no consensus as to how clusters of these features are best grouped and labelled'. Though this point was made in 2008, this lack of consensus continues. In the following subsection, we will argue that a method in corpus linguistics, 'keyness analysis', is a method that can provide the analyst with a robust foundation for describing those very clusters.

Combining corpus techniques, rapport, and learning
The central corpus linguistic notion we will be deploying in this article is the 'keyword'. The term keyword is not to be confused with lexical items that are key because they are of particular social, cultural, or political significance (e.g. Williams 1988). It is simply another term for statistically based style markers (Enkvist 1973). Keyword analysis, a corpus technique, caters for the first and second points we made about style in 2.2, namely, that style is constituted by all its features and that it is lent meaning by contrasts with other styles, contrasts that make it distinctive. The power of this approach in the analysis of discourse has been repeatedly demonstrated (e.g. Bondi and Scott 2010; Mike Scott's bibliography at http://www.lexically.net/publications/publications.htm). In Section 3.3, we give details on both how keywords are extracted and analysed.
The idea of combining corpus techniques with communicative or pragmatic phenomena is not new-witness the advent in 2017 of the new journal Corpus Pragmatics. The same can be said of corpus techniques and learning, as evidenced by the advent of many learner corpora. However, the idea of combining corpus techniques with rapport management in the exploration of online learning situations is very rare. To our knowledge, Ä del (2011) is the only study that can claim to have done this. We follow Ä del (2011), in examining online student discussion forum data (Ä del also examines some face-to-face study group discussions). However, she stresses that her study is primarily qualitative (2011: 2939). Her corpus technique is to generate word frequency lists, and then manually scrutinize the items above a certain cut-off frequency. This manual scrutiny was achieved by examining concordances (i.e. lists of the examples representing the high-frequency item in the corpus, along with the words that immediately surround them). Through this, she identifies those words that seem to be 'rapport building', and then supplies a label for the particular kind of rapport building, which then feeds into the development of a taxonomy. Ä del acknowledges some limitations: 'only the most frequent and most salient expressions are captured ' (2011: 2939). A keyword analysis, in contrast, encompasses all the items in the data. Furthermore, there are important differences between Ä del's study and ours in the understanding and operationalization of salience. Raw frequencies, even if restricted to the most frequent items in a dataset, do not necessarily display what is distinctive about a particular dataset. For example, unsurprisingly the word the is at the top of Ä del's online data frequency list, it is after all the most frequent word in English and dominant in many genres. This word is not discussed by Ä del because it did not survive the second step of her approach, namely, the manual identification of items that are potentially rapport-building through qualitative analyses. It is here, presumably, that Ä del made a judgement about which of the high-frequency items were salient. In contrast with Ä del's focus on high-frequency items, our corpus method, keyword analysis, computes all the items in a dataset and identifies what is statistically distinctive-that is, statistically salient-in one relative to another (see Section 3.3 for a fuller explanation). Then we proceed qualitatively, not unlike Ä del, scrutinizing the words both individually and as a whole.
The differences in approach between our study and Ä del's partly reflect differences in research goal. Ä del focusses on the most frequent expressions as a way to cover 'expressions that are central and not merely peripheral to the specific speech events under investigation ' (2011: 2939). Our focus, which flows from our definition of communicative style is specifically on what is distinctive in one individual mentor's style compared with the other two. For example, in Ä del's two datasets, one concerning online data and the other face-to-face, the first-person pronoun I is in second and first position in the frequency ranking of the two datasets respectively. This particular fact is not discussed by Ä del, though it is clear from her discussion of rapport management functions that the first-person pronoun participates in a number of them. Had we been conducting a comparison of the same datasets it is unlikely that the first-person pronoun would have come into consideration in relation to our research focus. This is because our statistical method identifies only what is significantly different amongst the datasets (the mentors' contributions), not what is similar. There are in fact ways in which a keyword analysis could be developed to capture similarities, and we will mention these at the end of this article in the discussion of future research.

Data context
Language students at the university in this study come from diverse backgrounds, a wide age range and different levels education, and the majority of them are in employment (Table 2). All Level 1 language modules in Chinese, French, German, Italian, Spanish, and English for Academic Purposes are distance learning courses supported by face-to-face and synchronous online tutorials, as well as asynchronous communication, over a period of 37-study weeks. Each module has a module-wide student forum where students communicate asynchronously in English with each other and the academic team via text messages organized into thematic 'threads'. At beginners' level, most posts are in English as the majority of the students are UK based and their level of the target language is too elementary to conduct meaningful conversations. This forum contains no time-tabled learning activities and opens three weeks prior to the start of the module until the end of the module. Forum participation is voluntary. In the academic year 2014-2015, as part of the module, students were required to submit, via an online system, four tutor-marked assignments (TMAs) and one end-of-module writing assignment (EMA) by a cut-off date. Of those five assignments, two were speaking and involved using an audio tool to record the student's submission. Two weeks leading up to the submission date of each assignment, an appointed student mentor initiated a 'Student Buddy support' thread in the forum. These mentors were students who had recently completed the same module and were selected based on their forum activities in the previous year. They were given training regarding the mentor role remit, namely providing moral support and practical advice (Fayram et al. 2018). Their work was unpaid.

Data sets
Data sets were collected during the 2014-2015 academic year from three beginners' language modules. The three modules are labelled Module 1, Module 2, and Module 3. To protect the anonymity of the three mentors, the language names of the modules are withheld. Each of the three mentors (coded Mentor 1, Mentor 2, and Mentor 3) is responsible for the mentor-led threads of one module. The following five datasets were obtained:

Mentor posts
About 273 forum posts by the three mentors were exported as three text files and analysed for keywords by a corpus-based text analysis tool (see Section 3.3). Summary information on the posts are shown in Table 1, which indicates that Mentor 2's posts had the most words because the majority of this person's posts had at least five steps or bullet points.
The three mentors were all new to online mentoring, but experienced students at this university. They included one native English speaker and two Europeans with 'expert level' of English; one male and two females; aged between 38 years and 58 years. To protect their anonymity, detailed personal information is excluded.

Student posts
A total of 412 forum posts from the three modules by students were exported as three text files (one for each module), and manually categorized into common themes such as 'asking for practical advice', 'asking for reassurances/moral support', 'sharing learning journey/resources', etc. They provide contextual information and the nature of the interaction. Our analysis focused on understanding the student perspective: their anxiety and concerns, and needs and feelings. For demographic information of registered students of the three modules in our study, see the three emboldened columns in Table 2.  The demographic information of both the registered students and the survey respondents for each module is given in Table 2 for easy of comparison with the corresponding module profile. Module 1 had most registered students. Module 2 had more male students in comparison with the other two modules. Module 3 had the highest number of respondents aged 60 years or above (19 per cent), and both Modules 2 and 3 had 50 per cent of students with undergraduate or/and post-graduate degrees. The survey respondents in the three modules had a similar gender spread, but in terms of age and level of education, they were, overall, older with a higher level of education than the registered students on the modules. This could be because of self-selection as the older students are mostly retired, and those with higher-level education struggled less with their study, possibly giving these two groups more time to complete surveys.
Of the survey respondents, Module 3 had the highest number of respondents aged 60 years or above, and Module 2 had the highest level of education (50 per cent undergraduate and 18 per cent postgraduate). Later, we consider how these demographics might influence forum participation and student perception of each mentor and the usefulness of the mentoring scheme. However, the respondents to the survey are a relatively small sample of the registered students, so any findings from these data should be interpreted with caution.

Interview data
The three mentors were interviewed, to explore how their understanding of the mentor role influences their communicative styles. In addition, all students who took part in the online survey were also invited for an interview, but only two female students (aged over 40 years) from Module 2 volunteered. Both of them indicated in the survey that they only read the posts without contributing. Nevertheless, they were interviewed to find out why they did not contribute. All the five interviews, each lasting for about 30 min, were conducted on Skype, recorded and later transcribed.

Moodle data
Descriptive statistics captured by Moodle regarding forum participation were obtained for all three modules (see Figure 1 for Module 1 as an example).
The bars in Figure 1 indicate the first four weeks when assignments were submitted. Week 34 was the submission week for the final assessment (EMA). The online Moodle system captures all student participants of a particular week of a module, no matter which thread is clicked by a student. On average 44 per cent of students in Module 1 visited the module-wide student forum, 47 per cent in Module 2, and 40 per cent in Module 3. Of those participants, about 85 per cent on average visited the mentor-lead threads. For each thread, the system records all the posts with times, dates and their authors. It also captures the readers (as well as those who read and post) of each thread, but not of particular postings. From this information, the raw figure of all participants (including 'read-only'), and the percentage of passive participants and active participants (i.e. who posted messages) in each mentor-led tread can be obtained (Tables 6 and 7). As the system cannot distinguish between those who read and post, and those who read only, in order to obtain the read-only figures, we manually counted the students who posted and subtracted this number from the overall total number of participants. Note that the system does not capture forum subscribers who read the messages only via e-mail.

Keywords
In the context of corpus linguistics, the notion of keywords and the practice of keyword analysis has most notably been developed and popularized by Mike Scott, through the Keywords facility of his program WordSmith Tools (Scott 2016a), which is designed for the computational analysis of corpora. It performs the kind of statistical analysis required to identify keywords, by conducting a statistical comparison between the words of a corpus (or wordlist) and a reference corpus (that is usually bigger), in order to identify words that are unusually frequent or unusually infrequent. The choice of the reference corpus will affect the nature of the keyword results. As Culpeper (2009: 35) puts it, 'the closer the relationship between the target corpus and the reference corpus, the more likely the resultant keywords will reflect something specific to the target corpus'. In our study, we compared, in turn, all the posts of each mentor against all the posts of the other two mentors combined.
According to the WordSmith Tools Manual (Scott 2016b), a word is 'key' if: (a) it occurs in the text at least as many times as the user has specified as a minimum frequency; Figure 1: Module 1 student participation in the Student forum week by week for 40 weeks Note: The forum opened three weeks before the module start date (represented by the minus sign) (b) its frequency in the text when compared with its frequency in a reference corpus is such that the statistical probability as computed by an appropriate procedure is smaller than or equal to a p-value specified by the user (see below); (c) in addition, the strength of keyness must be at least as great as the minimum log ratio set by the user (see below).
Keyness, then, is a matter of being statistically unusual relative to some norm. The statistical operations involved here-a cross-tabulation, a significance test (typically log-likelihood)-are amongst the most basic in statistics, and common in the world of corpus linguistics. The addition of log ratio is to accommodate effect size-the size of the keyness and not simply evidence of its unusualness (see http://ucrel.lancs.ac.uk/llwizard.html, for more on relevant statistics, and especially 'log ratio', a statistic devised and labelled by Andrew Hardie). These days, many corpus analysis tools will carry out keyness analysis. For reasons largely of convenience, we used the program WMatrix (Rayson 2009). The minimum frequency was set at 5, and the minimum log-likelihood value at 6.63 (both typical values). The resulting unusually frequent words were ordered according to log ratio (i.e. the words higher on the list would account for greater differences between the data sets). Unusually frequent keywords are sometimes referred to as 'positive' keywords, and contrast with 'negative' keywords, unusually infrequent keywords. In this article, not unlike many keyword analysis articles, we concentrate solely on positive keywords. We will, however, make a few comments on negative keywords in the final discussion and conclusions section. Keywords may be taken as symptoms of a style, but they are not in themselves a communicative style, which, as we remarked in Section 2.2, can be taken to be a subset of the features that constitute a style. Less still can it be assumed that they are a communicative style constructing and reflecting rapport management. This is where the interpretative capability of the human analyst is essential. Pragmatic meanings, of which rapport management meanings are a part, are mediated in context. Some kind of more qualitative analysis is called for. Thus, as is typical of studies that bring keyness analysis to bear in the pursuit of discourse analysis (see, e.g. Baker 2006), we scrutinized concordances of every single keyword. We noted any repeated micro-pragmatic contexts these occurrences of the keyword participated in, especially the rapport-sensitive contexts displaying the strategies and features discussed in the literature. In many cases, those repeated contexts are the only micro-contexts the keyword instances participated in. Keywords that shared closely related contexts were grouped into specific communicative styles, and then we labelled these styles according to their rapport orientation. The results of these analyses are displayed in Tables 3, 4 and 5 in the section below.

THREE MENTORS: KEYWORDS, COMMUNICATIVE STYLES, AND RAPPORT ORIENTATIONS
This section presents the keywords for each of the mentors, and shows how groups of keywords create particular communicative styles with particular orientations to rapport management. All personal names are anonymized. Naming the precise language being learnt on the module (e.g. French, Chinese) is avoided, again to maintain anonymity (instead, we write: [language name]). Furthermore, both pronouns 'she/he' and possessive pronouns 'her/his' or 'her/himself' are used to protect the three mentors' identities.

Mentor 1
The following are the keywords for Mentor 1 (in rank order): [language name], Betty, hello, 'm, hope, module, helps, good, am, studying, luck, all, that, are, and I. These are categorized in Table 3. Here, and in the other keyword tables below, strict rank order makes way for the fact that keywords which belong to the same communicative strategies, or even collocate with each other (as indicated by a plus sign '+'), are placed in the same cell. Bold indicates keywords in the examples.
Overall, the keywords show that Mentor 1 has a communicative style that is strongly oriented to rapport enhancement, with occasional orientation to rapport neutrality when she/he gives practical module-related information. The communicative style largely consists of strategies to do with involvement, good wishes, positive evaluations, building common ground, and self-disclosure. One communicative strategy, advice giving, is ambiguous between enhancement and maintenance.
Of the 81 posts by Mentor 1, 25 of them (30 per cent) contain self-disclosure or personal information. She/He often uses emoticons, altogether 47 smileys used, to convey encouragement. Her/His overall rapport enhancement style is evidenced in the interview: From personal experience I know that straight after Christmas there is a little bit of a dip, . . . it is really hard to get back to studying after having two weeks off. So I started a thread just describing my feelings about getting back to studying after Christmas and how I struggled and the students they responded.
One effect of this style was to encourage students to reveal their own concerns and worries (e.g. 'if I'm honest I feel a bit stupid when it comes to learning a language, it's definitely out of my comfort zone' [Student A, EMA thread]; 'I have always felt that my lack of [a] degree was held me back career wise. . .' [Student B, TMA3 thread]). The analysis of student posts shows that the most common topic was 'asking for reassurances/moral support' (see Section 5.1).

Mentor 2
The following are the keywords for Mentor 2 (in rank order) : please, new, tip, save, already, files, open, word, into, click, email, page, book, file, which, task, name, useful, read, point, online, then, sentences, audio, English, or, in, your, tutor, and the. These are categorized in Table 4. The keywords show that Mentor 2 has a communicative style that is strongly oriented to rapport maintenance and neutrality. In particular, it consists of strategies relating to directive discourse (suggestions, requests, commands, etc.) and information-giving discourse, the latter sometimes being mixed with the former. Apart from the first and last post in each thread to announce her/his presence and congratulate students on completing the assignment, almost all of Mentor 2's posts are tips on assignments and technical advice on how to use tools. She/He often ended her/his post by saying 'watch out for my tips on [. . .]'. She/He used in total only eight smileys, as opposed to 47 used by Mentor 1. As Mentor 2 did not disclose any personal information and did not ask students how they felt about things, there was not a single post from students seeking emotional support. Amongst the 30 posts by students, 25 were about practical issues to do with assignments or technology. This was supported by the survey data where 100 per cent respondents reported that they visited Mentor 2's threads for practical advice. Mentor 2's overall rapport style is further evidenced in the interview: I focused on quite mechanical tips, very practical things for people to use rather than . . . social chat or reassurance. [. . .] I focused on practical support. So maybe people just read what I posted and then applied it themselves. [. . .] I tend to do things with bullet points and factual, try to get to the point, business like.

Mentor 3
The following are the keywords for Mentor 3 (in rank order): a bit, 'm, 'll, 's, may, until, 've, who, last, better, say, code, well, like, me, did, hope, everyone, there, they, hi, was, been, know, n't, do, just, be, that, it, and I. Most of these keywords are categorized in Table 5, except for last, say, well, did, they, been, n't, be, that, and it, because they have particularly varied functions, and in some cases occur infrequently, and so are difficult to categorize. It should also be noted that some keywords are misleading when considered out of context, and this happens more than for the other two mentors' lists. For example (Table 5, final row), just seems to be a classic 'minimizer', a strategy by which rapport can be Used to pack in extra information.
Neutral maintained especially when threatened in requests (e.g. 'Could I just borrow your pen a moment?'). However, in context it also performs rapport enhancement.
The keywords show that Mentor 3 has a communicative style that is mixed in terms of rapport management. As with Mentor 1, there is a rapport enhancement pattern involving strategies expressing positive wishes and feelings, building common ground, performing self-disclosure and so on. But there is also a rapport neutral pattern, involving strategies giving information about duration, course-related specifics, as well as the 'packing in of information'. However, unlike Mentor 1 and more like Mentor 2, there is a degree of rapport maintenance in the reminders and warnings. Moreover, unlike either Mentor 1 or 2, the distinctive feature of Mentor 3's communicative style lies in how rapport enhancement is achieved-it often involves a strategy of self-effacement. That self-effacement generally has a creative and humorous touch (e.g. 'sometimes I just rebel and watch Netflix', 'I sound like an old car engine. . .'). In response, there were posts by students with humour (e.g. '. . .Amy does her tutorials in her pyjamas' [Student C, TMA2 thread]). In the 112 posts by Mentor 3, 61 emoticons were used, 47 smileys with a few grins and a few sad faces when she/he revealed her/his own difficulties. Mentor 3's overall rapport style is further evidenced in the interview: It was nice to have that mutual support with each other so you could say 'well I am listening I promise you.' [. . .] I did tips for submission of audio files [. . .] you get a bit repetitive if you are just saying 'come along guys you can do it, you're doing all right.' I wanted to vary it a bit [. . .].
What this suggests is that Mentor 3 was taking a more indirect and creative line in supporting and encouraging learning, which encouraged students to share learning stories and resources (see Section 5.1). The survey data supported our analysis in that the top reasons for visiting Mentor 3's threads were for practical advice (90 per cent) and moral support/reassurances (72 per cent).

Keyword(s) Communicative strategies
Additional comments Rapport orientation point, your Giving information: your understanding of that grammar point. . . The point I wanted to emphasize is to. . .
Point is used to focus information; your is used to refer to aspects connected to the students.

Participation and interaction in mentor-led threads
As explained in Section 3.2.5, Moodle statistics report that of all the students who were registered in Week 1, on average 43 per cent visited the modulewide student forum (M1: 44 per cent; M2: 47 per cent; and M3: 40 per cent). Although the participants gradually declined, it is clear from Figure 1 that numbers peaked in the subsequent five assignment submission weeks for Module 1. A similar pattern occurred for Modules 2 and 3. Of those students who visited the module-wide fora, on average across the three modules, 85 per cent participated in the mentor-led threads (including 'readers' and 'posters'). As the Moodle system could not capture forum subscribers who read the posts via an e-mail alert, the analysis misses this population. Table 6 reports both the raw figures and the percentage of active participants who posted and passive participants who read only in each mentor-led thread by module, which reveals the often found fact that the majority of students were passive participants. Between 80 per cent and 90 per cent of participants across the three modules only read without posting, which is in agreement with Wilkerson's (2016) claim that 90 per cent forum participants read without contributing. We made the point earlier (in Section 2.3) that online learning environments are risky from a 'face' point of view; it might be deemed better to say nothing, risk no face exposure, and just 'listen'. Another reason may be that learners 'at a distance' are time-poor, as on average 40 per cent of the students in this study were in full-time employment and a further 18 per cent in part-time employment (Table 2). . . . just one week to go Self-disclosing (often with humour): sometimes I just rebel and watch Netflix Directive discourse. Both reminders and warnings can be construed as something that is of benefit to addressees but not directly the addressors (mentors do not set the deadlines; they are not responsible for them). just also seems to be part of a self-effacing style.

Enhancement/ maintenance
The results suggest that Mentor 3, who adopted a communicative style that combines rapport enhancement, especially involving self-effacement, with neutrality and a small degree of maintenance rapport, had the greatest impact on active participation-over 22 per cent, on average, of participants contributing to content. Mentor 1 follows with around 15 per cent; her/his rapport orientation was a mixture of enhancement and neutrality. Mentor 2, whose style was oriented towards maintenance and neutral rapport, ranks some distance behind the other two, with around 9 per cent. Furthermore, the higher levels of student active participation in Modules 1 and 3 are strikingly reflected in the large number of posts contributed by students; more posts per poster, as well as the rich content of posts, as displayed in Table 7.
Data from both mentor and student posts also indicate that enhancement rapport tends to create more opportunities for interaction (mentor to student, as well as student to student), which we suggest may facilitate the building of an online community where the moral support and reassurance comes from both the mentor and fellow students. This is illustrated in the following interaction (Mentor 1's TMA1 thread, 16-17 October 2014) (we have trimmed the extract due to lack of space): Mentor 1: . . .What is your top tip? Student D: . . .try to do a little bit on most days rather than doing nothing for a few days and then doing a marathon session . . . Notes: Some posters post in different threads. Each time he/she posts in one thread (one or more than one post) it counts as one poster in that thread.
Mentor 1: . . .Great tip! Student E: . . .break the assignment question down into all the points you have to cover and tick them off as you cover them. . . In contrast, Mentor 2's contributions, mixing maintenance and neutral rapport, produced dialogue between the mentor and students, rather than between students; there were no interactions between students and students on this forum. Below is a typical interaction between Mentor 2 and a student (Mentor 2's TMA 2 thread, 29 November 2014) (we have trimmed the extract due to lack of space):

Mentor 2:
The Problem: Do you find the speech too quick to follow on some of the audio tracks? Do you find the gaps too short, leaving you too little time to repeat what has just been said? Do you find some of the sentences too long to remember and repeat? [. . . The mention of Mentor 2's lengthy posts might suggest a negative effect on student engagement. In addition, the posts in Mentor 2's threads covered less in topic areas than in the other two mentors' threads (see Table 7).
It is worth remembering that factors such as age and levels of education may impact on the forum participation. In Module 3, the higher level of active 1) sharing learning journey/ resources 2) asking for reassurances/ moral support 3) asking for practical advice (technology and assignment format related) 4) joking with each other 5) expressing gratitude participation may be attributable to a higher number of older students, who perhaps have more time (as indicated earlier); and students with higher levels of education who perhaps are more confident (see Table 2).

Students' perceptions of their mentors and the usefulness of the scheme
The survey asked each respondent to give two words/expressions to describe their mentor. They were also asked to evaluate the mentoring scheme in terms of its helpfulness. Below the findings for each mentor are discussed.

Mentor 1
Of the words given, the top two used to describe Mentor 1 were helpful and friendly. The rest were pleasant, professional, encouraging, clear, approachable, feel not alone, and ideal. There was one negative word irrelevant, to which the student added 'I just do not see the need for it'. These perceptions are in line with our keyword analyses of Mentor 1 in Section 4.1, showing that Mentor 1 engages in enhancement rapport and is friendly and approachable. In addition, 80 per cent of respondents in Module 1 thought the mentoring scheme useful because it is good to 'bounce ideas' off someone 'who has been there before', and it makes one 'feel less alone with the studying' Spencer-Oatey (2008: 31).

Mentor 2
The top two words were helpful and knowledgeable. The rest were useful, dedicated, methodical, thorough, efficient, talkative, supportive, hardworking, praiseworthy, and descriptive.

Mentor 3
The top two words for Mentor 3 were the same as for Mentor 1: helpful and friendly. The rest were informative, useful, reassuring, supportive, organised, fun, knowledgeable, open, caring, and encouraging. Mentor 3's mixed rapport management style (enhancement, especially involving self-effacement, with neutrality and a small degree of maintenance) together with her/his humorous touches seems to have paid off. This is illustrated by the fact that all the words used to describe Mentor 3 were positive, and all the respondents in Module 3 (100 per cent) thought that the mentoring scheme was useful, for example, because 'it boosts moral when you know you are not alone', and that Mentor 3 was 'very supportive for those struggling and lots of extra ideas for those who are doing well' [Survey open comments]. As 60 per cent of survey respondents were over the age of 50 in Module 3, one might speculate that they might have been more generous with personal evaluation, and perhaps needed more help with technology and hence found the scheme useful. The survey data seem to suggest that different communicative styles had an impact on students' perception of the mentor as a person, but not as much as the impact on forum interaction (as discussed in Section 5.1). Although there were two negative terms used to describe Mentor 2, 90 per cent respondents of Module 2 thought the mentoring scheme was useful because of the good tips offered. However, as our survey data sample is small, it can only be used in conjunction with other four datasets (as described in Sections 3.2.1, 3.2.2, 3.2.4 and 3.2.5) as additional information.

DISCUSSION AND CONCLUSIONS
This article began by discussing the notion of style and more particularly communicative style, proposing an operationalizable definition. It explained how empirical evidence of styles could be derived through a corpus-based technique, namely keyword analysis. Hitherto, identifying the clusters of features that constitute styles had been an area of doubt. Instead, researchers, even Spencer-Oatey (e.g. 2008), had resorted to lists of possible communicative strategies, rather than addressing what in the lists might be pertinent to creating a particular style. The results of a keyword analysis, however, are only a first step: this may give evidence of a set of linguistic features constituting a style, but not necessarily a communicative one. Also, it will obviously enrich our findings further if the kind of communicative style can be identified. Hence, our study examined the functions in context of the occurrences of every keyword, starting by scrutinizing concordances. Our functional analysis here was informed by the classic (im)politeness strategies, along with strategies that have been discussed in studies of facework, and indeed relational or rapport management. Having characterized the communicative styles of the mentors thus, the analysis enabled us to relate them to one of the four rapport orientations-enhancement, maintenance, neglect, and challenge-suggested by Spencer-Oatey (2008), with the addition of our proposed orientation of rapport neutrality.
The article revealed not only the differences in the linguistic substance of the mentors' communicative styles, but also how those styles: (i) were characterized by particular communicative functions; (ii) pointed towards certain general rapport orientations; (iii) impacted on student participation and interaction; and (iv) were perceived by students. Mentors 1 and 3 had fairly similar communicative styles, both oriented to rapport enhancement and, to a slightly lesser degree, neutrality. This rapport enhancement involved strategies expressing positive wishes, evaluations, intentions, and sympathy; self-disclosure; building common ground. Their neutrality revolved around giving information. In addition, both mentors performed a mixed rapport category of enhancement and maintenance in the performance of giving personalized advice, reminding, and warning. What particularly distinguished Mentor 3 from Mentor 1 was the use of a strategy of self-effacement as part of a communicative style performing rapport enhancement, often coupled with a touch of humour. Mentor 2 stood apart from both of the other two mentors in having a style that oriented to maintenance and neutrality rapport in almost equal measure. Neutrality rapport again primarily involved a strategy of giving information, whereas maintenance rapport involved requesting, instructing, and advising.
These communicative styles arise from words that occurred significantly more frequently in one mentor's contributions compared with that of the others. Such unusually frequent words are so-called positive keywords. To have also discussed negative keywords (significantly infrequent words), would have required much more space. Moreover, most of our conclusions are apparent from the positive keywords analysis alone, not least because, as might be expected, positive and negative keywords are related: a particular positive keyword in one mentor may mean a negative keyword in the others. Looking at the three sets of positive keywords alongside those for negative keywords provides further evidence of the contrasts we have observed. Nine (60 per cent) of the words that constitute Mentor 1's 15 positive keywords (hello, hope, module, helps, good, am, luck, all, and I) also appear in the list of Mentor 2's negative keywords. Only 12 (39 per cent) of the words that constitute Mentor 3's 31 positive keywords ('ll, 's, who, well, like, me, hope, there, been, know, it, and I) appear in the list of Mentor 2's negative keywords. This suggests that Mentors 1 and 2 have more sharply contrasting styles than those between Mentors 3 and 2. No such sharp contrast appears between Mentors 1 and 3: none of the words that constitute Mentor 1's positive keywords appear in the list of Mentor 3's negative keywords or vice versa.
The communicative style of each mentor aligned with a different level of student participation. Enhancement rapport stimulated higher levels of active participation. Mentor 3's mixed rapport management communicative styleenhancement, especially involving self-effacement, combined with neutrality and a small degree of maintenance-achieved the highest level of active participation. The analysis and results of mentor-led threads in Section 5.1 indicated how enhancement rapport generated opportunities for multidirectional interactions involving mentors and students. Furthermore, students had very positive perceptions of Mentors 1 and 3. This is consistent with earlier studies showing that affective elements play a major role in online language learning (Hurd 2007); good support is key to the success of learning languages at a distance (Baumann et al. 2008;Murphy et al. 2010); online fora facilitate socialization and the building of the online learning community (Lamy 2013b). However, a surprising finding, at least at first glance, is the fact that the absence of rapport enhancement, and its affective and socializing role, did not create an overall negative impression of Mentor 2 for the students. Although students were not facilitated in expressing their concerns and worries, they appreciated the practical advice and good tips, finding them useful and reassuring. The majority of the queries from students were about practical aspects of study, in particular technical issues, as technology is one of the main challenges in online language learning (Hampel 2014;White 2014). Mentor 2 excelled in meeting this need.
As with any study, there are a number of areas that would benefit from further research. We stressed in the literature review that our focus in this study was on the distinctive styles of individual mentors. A further study could compare all three of our individual mentor datasets against other datasets (e.g. the British National Corpus, face-to-face teaching discourse, mentors from other online modules), in order to highlight distinctive features of the mentor style in general. Alternatively, some scholars have begun to suggest ways of identifying what is statistically similar across datasets, thus, in a sense, doing a 'reverse' keywords analysis (see Taylor 2018, for an overview). This could have highlighted similarities across the three mentor datasets in this study. This might reveal whether there are particular styles for particular types of teaching; for example, those relating to in distance mode. Finally, perhaps the main area that would benefit from further study is the students' contributions. These could, for example, be contrasted with the mentors' contributions or different students' contributions could be contrasted with each other using keyword analysis. That may reveal different preferences regarding communicative styles, both in general and connected to teachers or mentors more specifically. An additional relevant factor is that of gender, as face or politeness has sometimes been discussed in relation to 'male' versus 'female' communicative styles (e.g. Holmes 1995).

SUPPLEMENTARY DATA
Supplementary material is available at Applied Linguistics online.