Metadiscourse in spoken interaction in ESL A multilingual perspective Pia Resnik Metadiscourse can fulfil various functions, two of them being to clarify and interpret what was said. These are highly relevant in spoken discourse in a language other than one’s L1, where miscommunication can easily happen. Due to globalisation and, along with it, migration, English is more and more being used in these contexts to bridge gaps when communicating with speakers of other L1s, who then run the risk of being misunderstood. Therefore, the present paper investigates the use of metadiscourse as a strategy to solve or prevent instances of miscommunication among multilingual speakers who communicate in their L2 (English). An analysis of 24 in-depth interviews with multilinguals from different language backgrounds (L1 German or L1 Mandarin Chinese) on verbalising emotions in various languages show different metadiscursive strategies to maintain mutual understanding. Quantitative analyses reveal that the L1 of the interlocutors had a significant effect on the overall frequency of using metadiscourse in general and, in particular, otheroriented forms. With regard to the underlying functions, no effect of the L1 can be shown. Furthermore, the bilingualism index proved to be a good predictor of metadiscursive instances in general as well as for those directed at the other interlocutor to clarify or interpret what was said. Furthermore, as the interlocutors were all individuals with more than one language at their disposal, code-switching (CS) instances on a metadiscursive level were also taken into account, which shows that CS was only used strategically in cases where both interlocutors shared the same L1 to restore comprehensibility. 1. Introduction When we ask ourselves what makes us fundamentally human, one of the core components mentioned by many is language. Even though communication as such is not unique to us human beings, the way we communicate via language is incomparable to the way other species do. Not only AAA - Arbeiten aus Anglistik und Amerikanistik Band 42 (2017) · Heft 2 Gunter Narr Verlag Tübingen Pia Resnik 190 are we able to talk about people or items that are not present or do not even exist, but we can also refer to past events and future developments via language. With a limited repertoire of sounds we can create an infinite number of utterances in a given language and we can also create entirely new expressions. Hockett (1960) identified his 13 well-known design features, which, according to him, in their entirety are unique to human language. One important feature contributing to our fundamentally human way of communicating is being able to reflect on language. Reflexivity is also crucial to spoken interaction as “[w]hen we communicate, we not only talk about the world and about ourselves (many people’s favourite topic), we also use language to talk about talk” (Ädel 2006: 1). This phenomenon is also referred to as metadiscourse, which is, as the term denotes, “meta, i.e. discourse about discourse” (Aguilar 2008: 58). As it can be realised in various ways ranging from morphemes to several sentences and is highly context-sensitive (cf. Ädel 2006), metadiscourse is difficult to categorise. Thus, it is crucial to focus on both form and function when analysing metadiscursive instances (cf. Aguilar 2008). This is what the present paper sets out to do with a particular focus on spoken, dialogic interaction in an LX, which refers to any language acquired after one’s first (L1), i.e. after the age of 3 (cf. Dewaele 2017). As most investigations into metadiscourse have analysed the written mode so far (cf. Ädel 2006), it is crucial to also focus on the spoken mode, as was already noted by Schiffrin almost 40 years ago (cf. Schiffrin 1980), because it employs different realisations and functions of metadiscursive instances. This holds especially true for dialogic situations, in which meaning is frequently negotiated metadiscursively in order to restore or maintain mutual understanding (cf. Mauranen 2010). As doing so is even more challenging in an LX, this paper seeks to provide a better understanding of the strategies employed by L2 users of English in dialogic situations. Due to different L1s of the interlocutors in the 24 interviews analysed for this paper, the use of metadiscourse can, furthermore, be analysed with regard to ELF (English as a lingua franca) contexts (cf. Jenkins 2007, 2009, Seidlhofer 2011) and contexts where the interlocutors communicate in their L2, but share the same L1. This way, we can gain insights into the likelihood, frequencies and functions of metadiscourse (cf. Penz 2011). Furthermore, a link to linguistic multicompetence will be established (cf. Cook 1991, 2016) in this paper, for which ELF contexts are said to be the perfect arena (cf. MacKenzie 2012). The paper is organised as follows: the literature review starts with defining the term metadiscourse and then continues with a discussion of its relevance in multilingual contexts. The subsequent section focuses on the methodology applied in the study and includes a description of the participants, procedure, independent and dependent variables as well as the research questions. Afterwards, the results from statistical analyses are Metadiscourse in spoken interaction in ESL 191 presented, which are discussed in the light of previous studies. The findings are then summarised and elaborated on in a conclusion. 2. Defining metadiscourse The term metadiscourse was first used by Zellig Harris in 1959 (cf. Hyland 2005), but was only theorized more comprehensively about 30 years ago (cf. Vande Kopple 1985, Crismore 1989). Two different research traditions have emerged: integrative and non-integrative approaches (cf. Mauranen 1993). These differ in their conceptualisations of metadiscourse (cf. Ädel/ Mauranen 2010). The former, also referred to as interactive models (Ädel 2010: 70), define it in a broad way, placing emphasis on “textual interaction” (Ädel/ Mauranen 2010: 2). Proponents classify hedges, for instance, as metadiscourse. In non-integrative models, also known as reflexive models (Ädel 2010: 70), these would not be considered metadiscourse (cf. Ädel/ Mauranen 2010), but the determining feature of metadiscourse in this approach is reflexivity (cf. Hockett 1960). Thus, proponents of the reflexive model take a comparatively narrow approach to defining the term and to, consequently, categorising metadiscursive instances (cf. Ädel 2010). Just as two different ways of conceptualising the term have emerged, so have different methods applied to identifying metadiscursive instances. Ädel and Mauranen (2010: 2) refer to them as ‘thick’ and ‘thin’ approaches. These mostly correspond to the above-mentioned research traditions. Proponents of ‘thin’ approaches usually analyse data purely quantitatively with regard to the frequencies and the distributions based on pre-defined lists of forms. Thus, these approaches usually lack taking variation in functions of the same form into account and are highly intuitive. Still, one of their core strengths is enabling automatised analyses of large corpora. Applying a ‘thick’ approach, on the other hand, requires a rather qualitative orientation, in which emphasis is placed on the contextdependence and dynamic nature of metadiscursive instances. Even though researchers usually also start their analyses with potentially reflexive forms in this case, the context is, furthermore, crucial as the metadiscursive unit of analysis is considered larger, taking, for instance, its function into account (cf. Ädel/ Mauranen 2010). The approach taken in this paper is in line with the latter. I will nevertheless also use quantification in order to be able to investigate possible effects of independent variables on the frequency and function of the metadiscursive units that were categorised depending on context. This way, a new perspective on the phenomenon will be provided. To put it simple, just as metalanguage is language about language (cf. Mey 2001), also metadiscourse is “discourse about the ongoing discourse” (Mauranen 2010: 16). As the focus in this paper will be on metadiscur- Pia Resnik 192 sive instances in spoken interaction, the definition adopted here is “speaking about the discourse at hand” (Mauranen 2010: 16). Even though most research on metadiscourse has focused on the written mode so far (cf. Ädel 2006), the importance of also investigating its uses in spoken interaction was already highlighted in 1980 (cf. Schiffrin 1980). In spoken interaction, metadiscourse clearly helps to monitor the current discourse in that it offers us the possibility to explicitly comment on the discourse as such (cf. Carter/ McCarthy 2006, see also Mauranen 2012). It is indeed a very effective way of negotiating meaning interactively on the spot in spoken conversations, which are simultaneous and transient. Some of these instances are easy to identify as the metadiscursive function of these forms is salient and, therefore, easily codable (e.g., you + BE + saying, you + mean). Other instances are less salient though and in these cases the context is often the key to understanding their metadiscursive function (cf. Mauranen 2010). In the present paper, metadiscourse will be analysed in interactive communication. This is important as the functions of metadiscursive instances in dialogues clearly differ from monologic uses. In conversations, metadiscourse is a particularly useful tool of negotiation (e.g., to clarify what was said) and thereby contributes to maintaining mutual understanding (cf. Hyland 2005, Mauranen 2010, Penz 2011, Schiffrin 1980). To conclude, investigating dialogues in this respect provides us with a better understanding of how discourse is managed strategically to, for instance, avoid conversation breakdowns and, more generally, to make points more explicit and precise (cf. Mauranen 2010). As Mauranen (2010: 20) puts it, “discourse reflexivity plays crucial roles in negotiating the flow of discourse between participants.” According to her, it can be directed at the self, but also at other people’s utterances and can be used for various functions in dialogic speech events, such as clarification, interpretation or as springboard to move on to another topic in a conversation (for a detailed discussion of the different functions, see section 4.4). How these instances are applied in communication in an LX will be the focus of the study included in this paper. Before moving on to the description of the methodology, I will explain the relevance of investigating the use of metadiscourse in LX users of English first. 3. Metadiscourse in multilingual contexts - the LX user When investigating metadiscourse in multilingual contexts, it is important to be aware of the following: LX users differ from L1 users in that they have more than one language at their disposal (cf. Ortega 2009). The languages a multilingual knows furthermore mutually influence each other, which means multi-competent (cf. Cook 1991, 2016) LX users will always differ in their language use from (idealised monolingual) L1 users Metadiscourse in spoken interaction in ESL 193 or native speakers of a language. I will therefore take a holistic, multilingual view on LX users (cf. Cook 2016, Cook/ Singleton 2014), which acknowledges multilinguals as language users in their own right. Of the approximately 2 billion users of English worldwide, multicompetent LX users of English by far outnumber its L1 users (cf. Graddol 2006, Mauranen 2012). For them, English today often is “the communicative medium of choice, and often the only option” (Seidlhofer 2011: 7) when communicating with someone whose L1 they do not share. It is thus regularly used to bridge communicative gaps in these contexts (cf. Phillipson 2008). Linking the idea of linguistic multi-competence to these typical lingua franca contexts, it becomes apparent though that ELF users are also skating on thin ice when communicating with each other as they “cannot rely on much shared linguistic or cultural knowledge with their interlocutors” (Mauranen 2010: 13, see also Firth 1996). As metadiscourse is a good way of monitoring dialogic speech events in general to ensure mutual understanding, this paper tries to provide a better understanding of how and to what extent multi-competent LX speakers use it to, for instance, avoid conversation breakdowns. Furthermore, I will examine whether the functions and frequencies of using metadiscourse differ in conversations among ELF users, who do not share the same L1, and non-ELF speakers, who communicate in their L2, but share the same L1 and, consequently, cultural and linguistic knowledge with their interlocutors. Gaining a deeper understanding of metadiscourse in such scenarios is important, as research on metadiscourse among ELF and LX users is, in general, still in a nascent stage (cf. Mauranen 2010). Even though previous studies show that metadiscursive strategies are used by both L1 and lingua franca users (cf. Mauranen 2007), no study to date has investigated the above-mentioned scenarios systematically. This is the objective of my contribution. 4. Methodology The focus of the present study is an investigation into both selfand other-centred metadiscourse in 24 interviews between L2 users of English. Similar to Mauranen (2010), the focus here will thus be on dialogic events, which have not been taken into account sufficiently in previous studies. 4.1 Participants The participants in this study were 24 multilinguals living in London at the time of the interviews. They were all L2 users of English. Among the interviewees, females were predominant (75%, N = 18). A female predominance is not rare in SLA research (cf. Dewaele 2010, Hammer 2017), Pia Resnik 194 an explanation for which might be females’ general predominance in language-related jobs and, consequently, greater interest in and willingness to participate in these studies (if they involve self-selection). The participants were on average 32 years old (min. = 20, max. = 53; SD = 9.812) with 14 aged 20 to 29, 3 between 30 and 39, 6 between 40 and 49 and one above 50. They were all well educated: 2 had finished their PhD, 19 their MA, 2 a BA and 1 had taken his A-levels. The interviewees differed in their L1s and were, consequently, divided into two groups: the 12 participants of group 1 shared German as their L1, and the 12 interviewees of group 2 shared L1 Mandarin Chinese. The participants’ age of onset of acquiring English as L2 was, on average, 10 years (min. = 4; max. = 20; SD = 3.919). While 7 participants started acquiring it between the age of 4 and 8, the majority did so between the age of 10 and 13 (N = 14). In the case of one participant, the age of 16 marked the starting point of acquiring English and another one started acquiring it for work at age 20. Their lengths of residence in London varied and ranged from 7 months to 21 years. Only one of them was bilingual. The remaining 23 were either trilingual (N = 9; 37.50%), quadrilingual (N = 11; 45.83%) or pentalingual (N = 3; 12.50%). They rated their proficiency in the L2 as relatively high on a scale from 1 to 5 in the different skills (reading, writing, speaking, comprehension). On average, the overall selfreported proficiency amounted to 15.67 (min. = 6; max. = 20; SD = 4.469). In the L1, the self-rated proficiency was considerably higher (mean = 19.29; SD = 1.429). Still, it ranged from a total of 16 to 20 points, which clearly demonstrates that not everyone perceived themselves as maximally proficient in their L1, which makes sense when taking the length of residence in the UK into account. 4.2 Procedure The original purpose of the structured interviews was to investigate multilinguals’ verbalisation and perception of emotions in various languages and only later re-analysed with a focus on metadiscursive instances. They all took place in London, UK and were conducted by myself, an L1 speaker of German with L2 English living and studying in London at the time the study was conducted. The participants were approached via forums for multilinguals living in London and via colleagues I knew. The main strategy used in this study was snowball sampling. The sampling was, furthermore, not random as research into SLA requires participants who are users of at least two languages and meet further criteria depending on the research questions (cf. Wilson/ Dewaele 2010). In this case, two important selection criteria were the following: the participants’ L1 and their being L2 users of English, i.e. they started acquiring their L2 after the age of 3 (i.e. they were sequential bilinguals). Due to the original research topic’s sensitivity, the inter- Metadiscourse in spoken interaction in ESL 195 viewees’ needs and wants were considered important and the interviews were, consequently, conducted one-on-one either at the interviewer’s or the interviewee’s homes or in quiet seminar rooms at different universities in London (Birkbeck College, University of London, School of African and Oriental Studies, University of London, University of Roehampton). The interviewees were, initially, not aware of the research topic as the procedure also included further test tools and informing them about the focus might have led to a distortion of the findings. The interviewer and interviewees did not share any interpersonal background knowledge. The interviews’ duration ranged from 15 minutes to 25 minutes each. They were all conducted in English due to cross-linguistic variation in emotion and emotion-laden words and their occasional non-translatability. In 12 instances (group 1), the interviewer and interviewees shared the same L1. This is not as uncommon as it might seem at first sight as there are many situations in which multilinguals with the same L1 need to communicate in an LX, for instance, in the workplace if other interlocutors who do not share this language are also involved. In group 2, the L1 of the interlocutors differed. These scenarios thus represented typical ELF contexts. Even though the situation was to some extent artificial in group 1, it allowed for examining not only individual differences in various types of metadiscursive instances, but also inter-group differences without creating so much noise in the data that the reasons for using metadiscourse could no longer be disentangled. The interviews were transcribed in full length, which amounted to 65,850 words, and the metadiscursive instances were subsequently coded manually for orientation (self-directed versus other-directed) and function (clarification, interpretation, springboard) (cf. 4.4 below). Automatic annotation was not possible due to the context-dependence of metadiscursive elements (cf. Ädel 2006). The interviewees had additionally filled in a short questionnaire in which they were asked for information on their demographic background, language learning history as well as present language use. This information was also used for the current study. The quantitative data were analysed using SPSS 24. 4.3 Independent variables L1 The effect of the participants L1s will be investigated as it is one important aspect in which they differed. In this way, a better understanding of the use of metadiscursive elements in ELF-contexts versus contexts in which the LX users shared the same L1 should be achieved. As mentioned above, group 1 refers to interviews in which both speakers had L1 German and L2 English. In group 2, the participants’ L1s differed, with the L1 of the interviewer being German and the L1 of the interviewees being Pia Resnik 196 Mandarin Chinese. This analysis sheds light on possible effects of shared background knowledge on the use of metadiscourse to negotiate meaning. Bilingualism index I will also examine whether language proficiency shows any effect on the use of metadiscourse. In the questionnaire, in which the interviewees provided information on their language learner history and present habits, they were asked to rate their self-perceived proficiency in speaking, comprehension, reading and writing on a scale from 1 to 5. As previous studies have shown, self-rated proficiency measures are good indicators of actual language proficiency (cf. Dewaele 2010, MacIntyre et al. 1997). As the idea of linguistic multi-competence (cf. Cook 1991, 2016) is adopted in this paper, according to which the languages an LX user knows are not to be seen as isolated from each other, the participants’ self-reported bilinguality is reflected in a second-order variable, referred to as bilingualism index (cf. Resnik forthcoming). Based on Dewaele and Stavans’ global multilingualism measure (2014: 10), the bilingualism index acknowledges the mutual influence of the knowledge of the L1 and L2 in an L2 user’s mind in the present study. Thus, the bilingualism index here refers to a speaker’s overall self-rated proficiency in the L1 and L2. The participants rated their bilinguality on average as 34.96 on a scale from 0 to 40, ranging from a minimum of 22 to a maximum of 40 (SD = 5.328). Overall length As the lengths of the interviews differed, I will also examine whether the overall duration of an interview had any effect on the overall number of metadiscursive instances. 4.4 Dependent variables The two main dependent variables are frequency and type of metadiscourse use. So in addition to counting metadiscursive instances, the latter were also categorised for two dimensions: orientation and function. These categorisations will be illustrated with examples from the corpus below. Self-oriented versus other-oriented metadiscourse As a first step, the metadiscursive instances in the corpus were classified according to their orientation, i.e. whether they refer to an utterance by the current speaker or by another participant (cf. Mauranen 2012). Selforiented metadiscourse can help speakers to clarify what they said, e.g.: Metadiscourse in spoken interaction in ESL 197 As I said… Well, that’s not exactly what I meant Other-directed metadiscourse, on the other hand, can be used to offer or seek clarification of what someone else said, e.g.: Do you mean linguistically? It’s about German words you are talking now? Functions of metadiscourse In a second step, metadiscourse were classified according to their functions. Mauranen (2010), in her study of metadiscourse in the ELFA corpus (Spoken English as a Lingua Franca in Academic Settings, www.eng.helsinki.fi/ elfa), identified three functions: clarification, interpretation and springboard. I also used these for my own data. Clarification refers to instances where a speaker wants the addressee to repeat, confirm or elucidate something the latter has said (cf. Mauranen 2010), e.g.: Do you mean linguistically? It’s about German words you are talking now? What do you mean? Do you mean generally or only in the L2? Yeah, but were you saying society? In interpretation, the speaker interprets the meaning of another utterance. Typical examples of this type are: As you said before... So you’re saying... Clearly, a speaker could theoretically also interpret their own utterance or, in self-talk, try to clarify what they meant. None of these functions were included in the data though. The springboard function is the last one identified by Mauranen (2010). It refers to instances where an interlocutor paraphrases another speaker’s previous utterance as a start for a new direction in the conversation. These instances can be evaluative and can be used to criticise what someone else said, but also to support it. No clarification from the other interlocutor is expected in these cases. Typical examples of this type are: If you swear… I think you do because you said so before... So, when swearing... And just because you mentioned “I love you“... Pia Resnik 198 I think it’s true what you’re saying that women and men… Also, speaker-oriented metadiscourse can function as a springboard though. In the present corpus, some instances of self-directed metadiscourse clearly worked in a similar way as illustrated in the following example by Emma (L1 German, L2 English, L3 Italian, L4 Kurdish), in which she referred to something she had previously mentioned, which served as a starting point for a different topic: I was just saying that with my friend in Boston we normally speak English, but… eh… I just realized that… Thus, not only you + BE + saying instances (cf. Mauranen 2010) were counted in this respect, but also self-directed forms. All these categories were, of course, quantified, too. 4.5 Research questions The following research questions were formulated in order to investigate metadiscursive elements in spoken interaction between L2 users of English: 1. Are metadiscursive strategies more common in ELF conversations than in conversations between L2 users of English who share the same L1 because they are more eager to ensure mutual understanding? 2. Do the strategies employed differ qualitatively? 3. Do speaker-related differences, such as self-rated proficiency, of L2 users affect the frequency of metadiscursive strategies? 4. Does the length of the conversation correlate positively with the frequency of metadiscourse use? 5. Results In the following, the results from various statistical analyses will be presented. In case the data showed normal distribution, independent samples t-tests and regression analyses were run. The former allow for testing inter-group differences by testing differences in mean values (e.g., L1based differences in the number of metadiscursive instances in the interviews). Linear regression analyses are also used to calculate possible relations between an independent and a dependent variable. In the following, one aim is to also test if the independent variable taken into account, which is in this case the bilingualism index, is a good predictor of the use of (various types of) metadiscursive instances. In case the data were not Metadiscourse in spoken interaction in ESL 199 normally distributed, Mann-Whitney tests were used as non-parametric equivalents of t-tests. 5.1 Inter-group differences Frequency of metadiscursive instances: inter-group differences As a first step, possible inter-group differences in the overall frequency of metadiscourse use are analysed. This analysis is based on the overall use of metadiscursive instances in each interview and includes the interviewer herself, as the focus here lies on both interlocutors’ use of metadiscourse in each dialogue to co-construct meaning. Thus, including those of the interviewer does not lead to a distortion of the findings as the instances were coded for each interview separately and not for each interlocutor. As a series of Shapiro-Wilk tests showed normal distribution (group 1 total : p = .096, group 2 total : p = .199), an independent samples ttest was run. It showed a significant difference between the groups (p = .030; t = -2.411; df = 14.518): the interviews in group 2 (different L1s) showed on average a greater number of instances of metadiscourse (group 2: N = 12, mean = 18.75, SD = 9.430) than those of group 1 (group 1: N = 12, mean = 11.67, SD = 3.882). A comparison of the mean values shows that the interviews in group 2 contained on average 60 percent more metadiscursive elements than those of group 1. The overall distribution of frequencies for each group is found in Figure 1. All in all, the interviews in group 1 included 140 elements of metadiscourse, while the interviews of group 2 included 225 such instances. Speakerversus other-oriented metadiscourse The data with regard to speakerversus other-oriented types were also normally distributed (group 1 speaker-oriented MD : p = .080, group 2 speaker-oriented MD : p = .271; group 1 other-oriented MD : p = .091, group 2 other-oriented MD : p = .104). An independent samples t-tests shows that, as illustrated in Table 1, both speakerand other-oriented instances of metadiscourse occurred on average more frequently in group 2 than in group 1: on average, the interviews in group 2 contained almost double the amount of otherdirected instances than the interviews in group 1 (mean group1 = 7.5; mean group2 = 13.08). Also the use of self-oriented metadiscourse was used on average more frequently in the interviews in group 2. Here, the difference was not as great as in the previous case (mean group 1 = 4.17; meangroup 2 = 5.67) and it was not significant (cf. Table 1). Figure 1 illustrates the overall frequency of use of metadiscourse for each group, but also includes illustrations of inter-group differences in the sum of selfand other-oriented forms (group 1: total = 140, other-oriented = 90, selforiented = 50; group 2: total = 225, other-oriented = 157, self-oriented Pia Resnik 200 = 68). In general, other-directed instances were more frequently applied in both groups than self-directed elements. When investigating possible links between the overall number of metadiscursive elements and the two different forms investigated, a regression analysis showed that the former was only a strong predictor for self-directed forms in ELF contexts (p = .004; F = 14.158; adjusted R 2 = .586; regression coefficient b = .400; β = .766). Variation in the overall number of metadiscursive instances explains 58.6% of the uses of self-directed forms. This does not apply to group 1 (p = .140; F = 2.573; adjusted R 2 = .125; regression coefficient b = .353; β = .452). For other-oriented forms, the overall numbers of metalinguistic elements used in each interview were a good predictor in both groups and explained 41.1% in group 1 and 73.8% in group 2 (group 1: p = .015; F = 8.666; adjusted R 2 = .411; β = .681; regression coefficient b = .647; group 2: p < .0001; F = 31.936; adjusted R 2 = .738; β = .873; regression coefficient b = .600). Figure 1: Instances of metadiscursive elements per group. Other-oriented instances N Total p t df Group 1 Group 2 24 .016 -.903 22 N mean SD N mean SD 12 7.50 3.631 12 13.08 6.487 Speaker-oriented instances N Total p t df Group 1 Group 2 24 .376 -2.602 22 N mean SD N mean SD 12 4.17 2.980 12 5.67 4.924 Table 1: Inter-group differences in speakerversus other-oriented metadiscourse. Metadiscourse in spoken interaction in ESL 201 Functions of metadiscursive elements As stated above, the data were also analysed with regard to the functions underlying the metadiscursive instances. As illustrated in Figure 2, the metadiscursive elements were most often used to clarify or interpret what was said (group 1: clarification = 72, interpretation = 62; group 2: clarification = 107, interpretation = 114). The number of instances serving a springboard function was comparatively low (group 1: springboard = 6; group 2: springboard = 4). As Shapiro-Wilk tests only showed a normal distribution of the metadiscursive instances serving the purpose of interpreting what was said in the interviews (group 1: p = .255; group 2: p = .081), in this case only an independent samples t-test was run. Even though the mean values (group 1: 5.12; group 2: 9.50) show that group 2 used this strategy more frequently, this difference remains on the level of tendency only and is not significant (cf. Table 2 for details). With regard to clarifications (group 1: p = .044; group 2: p = .026) and springboards (group 1: p < .001; group 2: p < .001), nonparametric equivalents were used. Instead of t-tests, Mann-Whitney-tests were run here. In both cases, no significant differences were observable (cf. Table 2 for details). Figure 2: Functions of metadiscursive elements per group. Pia Resnik 202 Interpretation N Total p t df Group 1 Group 2 24 .054 -2.052 18.744 N mean SD N mean SD 12 5.12 3.950 12 9.50 6.157 Clarification N Total p Z U 24 .076 -1.775 41.500 Clarification: mean ranks Mean rank: Group 1 Mean rank: Group 2 9.96 15.04 Springboard N Total p Z U 24 .613 -.506 65.000 Springboard: mean ranks Mean rank: Group 1 Mean rank: Group 2 13.08 11.92 Table 2: Inter-group differences in functions of metadiscursive elements. 5.2 Bilingualism index A regression analysis showed that the bilingualism index is a good predictor of the overall amount of metadiscursive instances in the conversations. The results show that the higher the bilingualism index, the lower the number of metadiscursive instances (see regression coefficient, Table 3). The same effect can be seen in other-oriented metadiscourse, which is also used significantly less often, the higher the self-ratings concerning one’s competencies in L1 and L2. As the adjusted R 2 values indicate, 14.7% of the overall number of metadiscursive instances can be explained by variances in the bilingualism index. In the case of otheroriented forms, the percentage is much higher and 32.1% of these can be explained this way. This does not apply to self-directed metadiscourse, as a regression analysis did not reveal a clear relation here (for a summary of statistics, see Table 3). The effects of self-rated L1 and L2 proficiency were also calculated as post-hoc tests to identify which of these shows an effect regarding the three dependent variables when taken into account separately. As can be seen in Table 3, participants’ self-rated proficiency in the L1 is a good predictor of the overall number of metadiscursive instances and of otherdirected forms. While it explains 14.3% of the former, variances in L1 proficiency explain 26.9% of the latter. In both cases, the following relation is shown: the higher the self-rated proficiency in the L1, the less frequently metadiscursive instances were used in general and so were other-directed forms of it. Self-rated proficiency in the L2 is only a good predictor of the frequency of other-directed forms: the higher the selfratings in L2 proficiency, the lower the number of metadiscursive in- Metadiscourse in spoken interaction in ESL 203 stances. Variance in self-perceived L2 proficiency explains 24.9% of those. Thus, both show an effect on the frequency of use of other-directed metadiscursive elements in L2 dialogic speech events. Still, the bilingualism index is seen here as the more realistic approach and, as demonstrated, is also a better predictor for other-directed metadiscourse. Bilingualism index Variable N p Regression coefficient b (Adjusted) R2 β F Total instances 24 .037 -.637 .147 -.429 4.957 Other-oriented 24 .002 -.654 .321 -.592 11.887 Speaker-oriented 24 .919 .017 .000 .022 .011 L1 proficiency Variable N p Regression coefficient b (Adjusted) R2 β F Total instances 24 .039 -2.352 .143 -.425 4.844 Other-oriented 24 .005 -2.258 .269 -.549 9.484 Speaker-oriented 24 .878 -.094 .001 -.033 .024 L2 proficiency Variable N p Regression coefficient b (Adjusted) R2 β F Total instances 24 .071 -.665 .102 -.375 3.610 Other-oriented 24 .008 -.698 .249 -.531 8.624 Speaker-oriented 24 .864 .033 .001 .037 .030 Table 3: The effect of the bilingualism index, L1 proficiency and L2 proficiency on the frequency of metadiscursive elements. 5.3 Overall number of tokens produced I was also interested in whether the overall length of the interviews has an effect on the use of metadiscourse. A Pearson correlation reveals that the overall number of tokens produced in an interview does not correlate with the number of times metadiscourse is used (p = .159; N = 24). Thus, the overall length of the interviews is not linked to the frequency of metadiscursive elements. 5.4 Code-switching in metadiscourse A very interesting aspect, only observable in group 1 and thus only elaborated on descriptively, is code-switching (CS) in metadiscourse for the purpose of clarifying what was said. This strategy was employed 19 times by the discourse participants in the 12 interviews between L1 German users of English. The following extract, in which Hanna (L1 German, L2 English, L3 French, L4 Italian) was interviewed, illustrates the aforementioned: Pia Resnik 204 I: Okay. And now a few questions on language, emotion and gender - so if you now think of women and men when talking: Have you ever had the impression that they differ in the way they express their feelings? Such as, that women, for example, show love more openly and men… I don’t know… express anger more openly? H: Ehm, it’s just the… ehm… you mean now… Meinst, du, das hat jetzt nichts damit zu tun mit den verschiedenen languages, sondern nur zwischen Mann und Frau? [‘Do you mean differences between men and women in general and not across languages? ’] These instances, in which CS was used in metadiscourse to demand clarification, were all related to LX proficiency. In cases where it was either difficult for the interviewed LX user to express their ideas in the LX, they resorted to the L1 as they were aware of mutual understanding. In group 2, interlocutors were forced to put more effort into circumscribing and paraphrasing original ideas in these cases, as there was no common L1 and competence in the speaker’s own L1 could not be expected. The interlocutors in group 1, furthermore, took advantage of the possibility of switching to the L1 when they did not understand a question or particular word in the L2 as is the case in the following instances, in which both Mia (L1 German, L2 English, L3 Italian) and Alex (L1 German, L2 English) reacted in similar ways in the two interviews: I: And, if you swear… A: If I swear? I: Yeah. A: Ah, okay. I: What language do you swear in most of the time? A: What do you mean by swearing? Fluchen? […] M: What does it mean, to swear? Fluchen? It is interesting to note that both interviewees’ proficiency in English was not only the lowest on the level of self-perceived proficiency, but stories they were asked to produce in the L2 (for a different study) also demonstrated their comparatively low proficiency. Even though in 18 instances, CS happened in a controlled way to ensure mutual understanding, Alex’ low proficiency in the L2 clearly also led to communication breakdowns. Not only did he misunderstand a previously mentioned question, but he clearly misunderstood a follow-up question included in the following extract, too: I: Your personal opinion. What do you think? A: (5) Hm, I don’t understand this question. Too... eh... difficult for me. I: Shall I say it in German? Metadiscourse in spoken interaction in ESL 205 A: Ich versteh die Frage nicht ganz. [‘I do not understand this question.’] All in all, switching between codes was, in general, not very frequent in the interviews and switches to the L1 only occurred in group 1, where participants shared the L1. This shows the important role the awareness of the languages an interlocutor knows plays with regard to the frequency and likelihood of CS (cf. Dewaele 2010). Furthermore, CS happened comparatively often in interviews where the interviewee’s knowledge of English was relatively low. Thus, proficiency showed to be influential, too. The relatively low number of CS instances in this study, unfortunately, did not allow for any statistical analyses though. 6. Discussion The results from this study demonstrate that L2 users of English make use of metadiscourse as a means to negotiate meanings and to ensure mutual understanding in spoken interaction. This is in line with findings from previous studies, the results of which also indicate that L1 as well as LX users both apply these strategies in spoken (cf. Mauranen 2007, 2010, 2012), but also in written communication (cf. Ädel 2006). By means of analysing interviews from two groups which differed in their L1s (either sharing the same L1 or not), I also show that LX users of English made use of these strategies significantly more frequently in ELF contexts than in contexts where the interlocutors communicated in their L2 (English), but still shared an L1 (German) and, linked to it, cultural background knowledge. This suggests that metadiscourse seems to be needed and, consequently, used more frequently to monitor dialogic speech events effectively in ELF contexts. Thus, in cases where the interlocutors share less linguistic and cultural knowledge, clarification and interpretation of what is being said and, consequently, explicit coconstruction of meaning seems to occur more frequently. Also, the results seem to point to ELF users’ awareness of the impossibility of relying on common ground (cf. Mauranen 2010), which is why they actively engage in achieving mutual comprehensibility. To put it in Mauranen’s (2012: 7) words, “they seem to be prepared for the possibility of misunderstanding, and take steps to pre-empt that, which in effect results in misunderstandings being rare (cf. Mauranen 2006, Kaur 2009).” Cook’s idea of multicompetence (1991, 2016) furthermore offers an explanation for the increased use of metadiscourse in dialogic speech events among ELF users: as the knowledge of the L1 influences the use of the LX (and vice versa), two interlocutors with the same L1 are likely to be more similar in their LX usage than interlocutors who differ in their L1s. In case of the latter, this difference is likely to lead to greater differences in LX usage, which leads to a greater potential for miscommunication, too. ELF users seem to Pia Resnik 206 employ metadiscourse more frequently to overcome these obstacles and seem to do so effectively. The data furthermore revealed that metadiscourse is used in both selfand other-directed ways in dialogic situations. Other-directed instances occurred much more frequently though, which highlights the relevance of studying metadiscourse in interaction, as this function is irrelevant in monologic texts. The findings support Mauranen’s (2010: 24) suggestion that “it seems discourse reflexivity is not so entirely speaker-oriented as we are used to thinking.” The analyses furthermore revealed significant inter-group differences in the frequency of other-directed instances: ELF users generally draw upon these to a much greater extent, mostly if there are problems of comprehension. This possibly points to an awareness of the presence or absence of common ground or, at least, to using metadiscourse to maintain mutual understanding. Self-directed instances were much less common in general and inter-group differences were not significant, which means sharing or not sharing knowledge of the L1 did not affect speaker-oriented use of metadiscourse in this study. Linked to the previously mentioned awareness of common ground is another finding, which was only analysed descriptively though, namely code-switching. CS intended to discuss the current discourse was only used by interviewees sharing an L1 with the interviewer. This agrees with previous studies of CS in different contexts, which showed that familiarity with the interlocutor’s language background played a decisive role in the frequency of doing so (cf. Dewaele 2010, Resnik forthcoming). In group 1, participants thus switched to their L1 when they ran the risk of miscommunication because they knew the interviewer also spoke the language. This might be linked to a conscious awareness and supports Grosjean’s (2010) assumption of CS occurring mostly in a controlled way - at least in situations in which we are not emotionally aroused. Furthermore, the CS instances might be explainable by means of Grosjean’s Complementarity Principle (2008, also 2001), saying that bilinguals’ languages are always activated to different degrees and people always need to make a choice which language to use at a particular point. If we are aware that we are speaking to someone in an L2 who shares the same L1, the L1 might still be activated to a greater extent than when we are speaking to someone in an L2 who does not know our L1. As far as the functions of metadiscourse is concerned, no significant differences between the groups were found. This suggests that LX users employ metadiscourse for similar purposes in ELF and non-ELF contexts in conversations, namely mostly to clarify or interpret an utterance or as a springboard for a change of topic (cf. Mauranen 2010). The main difference here clearly seems to be the extent to which they do. Not only was the L1 background of the interlocutors influential, but so was their proficiency. To show this, I also examined the effect of the bilingualism index, a second-order variable which reflects a speaker’s over- Metadiscourse in spoken interaction in ESL 207 all self-rated proficiency in the L1 and L2 (cf. Resnik forthcoming, see also Dewaele/ Stavans 2014). It was a good predictor of the frequency of metadiscursive instances in general and of other-directed forms. The higher the bilingualism index, the lower the number of metadiscursive instances. As these were mostly used for interpretation and clarification purposes in these interviews, this again demonstrates the crucial role of metadiscourse in the co-construction of meaning (cf. Mauranen 2012). It was furthermore demonstrated that variance in the bilingualism index is a stronger predictor than the self-rated proficiency in the L1 or L2 as such. As the interviews differed in length, the effect of their overall length on metadiscursive instances was also analysed. These factors did not correlate though. 7. Conclusion To sum up, the present paper investigated distributions and functions of metadiscourse in 24 interviews of L2 users of English, which were also conducted in the L2. In half of these interviews, both interlocutors shared an L1 and the concomitant linguistic and cultural background knowledge. In the other interviews, the participants spoke different L1s (German and Mandarin Chinese) and could therefore not resort to common linguistic (or cultural) ground (cf. Firth 1996, Mauranen 2010). This difference led to significant differences in the frequency of using metadiscourse: in ELF contexts, metadiscourse was used significantly more frequently, mostly to ensure mutual understanding, which can possibly be explained by LX users’ awareness of being on dangerous ground when using English as a contact language (cf. Jenkins 2009) with speakers who do not share their L1. Apparently, using “language to talk about talk” (Ädel 2006: 1) is a frequently employed strategy to negotiate and co-construct meaning (cf. Mauranen 2012) and to avoid conversation breakdowns in dialogic ELF contexts, which interlocutors of group 2 mostly applied effectively. The underlying purposes for using metadiscourse were similar in both groups: the interlocutors used it mostly for clarification and interpretation purposes. Not only common knowledge proved to be influential though, but also self-rated proficiency. The less proficient the LX users were, the more they made use of metadiscursive strategies. This again points to an awareness of possibly running the danger of miscommunication and demonstrates the importance of metadiscourse in LX users’ conversations for monitoring problematic contexts. In case the interlocutors shared the same L1, CS to the L1 was a commonly used strategy, too. Even though CS is a reality in most multilinguals’ lives and is commonly used (cf. Cook 2008) - consciously as well as without being aware of it - in meta- Pia Resnik 208 discourse it seems to be applied strategically (for a detailed discussion of CS, see Grosjean 2008). All in all, it has been shown that metadiscourse is an important strategy applied by LX users to communicate successfully in an LX. Further research in multilingual contexts is needed to gain a deeper understanding - for instance, also with a focus on individual differences in LX users. More studies of the use of metadiscourse in interaction are also needed as “dialogue brings out new facets and different emphases on the functions of self-referential language in communication as compared to written text analysis” (Mauranen 2010: 37; see also Ädel/ Mauranen 2010: 6). A final point needs to be made: even though reflexivity is seemingly easy for us human beings to apply, the term metadiscourse apparently is not easy to define. Just as researchers’ understanding of it differs, so do the methodologies applied and the ways of classifying uses of metadiscourse. This makes results from various studies difficult to compare and also poses challenges with regard to frameworks for analysis. Due to its context-dependence, metadiscursive instances have been coded and analysed manually in most studies to date, also in this one. Computerassisted methods clearly allow for analysing more material though and would thus help to obtain more generalisable results (cf. Ädel 2006). This might be one way of overcoming obstacles and increasing the validity and reliability of the results in future studies. References Ädel, Annelie (2006). Metadiscourse in L1 and L2 English. Amsterdam: John Benjamins. Ädel, Annelie (2010). “Just to give you kind of a map of where we are going: A Taxonomy of Metadiscourse in Spoken and Written Academic English.” Nordic Journal of English Studies 9(2). 69-97. Ädel, Annelie/ Anna Mauranen (2010). “Metadiscourse: Diverse and Divided Perspectives.” Nordic Journal of English Studies 9(2). 1-11. Aguilar, Marta (2008). Metadiscourse in Academic Speech. A Relevance-Theoretic Approach. New York: Peter Lang. Archibald, Alasdair/ Alessia Cogo/ Jennifer Jenkins (eds.) (2011). Latest Trends in ELF Research. Newcastle upon Tyne. Cambridge Scholars Publishing. Carter, Ron/ Michael McCarthy (2006). Cambridge Grammar of English. Cambridge: CUP. Cook, Vivian James (1991). “The Poverty-of-the-stimulus Argument and Multicompetence.” Second Language Research 7(2). 103-117. Cook, Vivian James (2008). Second Language Acquisition and Teaching. 4 th ed. London: Hodder Education. Cook, Vivian James (2016). “Premises of Multi-competence.” In: Cook/ Li (2016). 1-25. Cook, Vivian James/ Wei Li (eds.) (2016). The Cambridge Handbook of Linguistic Multi-Competence. Cambridge: CUP. Metadiscourse in spoken interaction in ESL 209 Cook, Vivian James/ David Singleton (2014). Key Topics in SLA. Bristol: Multilingual Matters. Crismore, Avon (1989). Talking with Readers. Metadiscourse as Rhetorical Act. New York: Peter Lang. Dewaele, Jean-Marc (2010). Emotions in Multiple Languages. London: Palgrave Macmillan. Dewaele, Jean-Marc (2017). “Why the Dichotomy ‘L1 versus LX user’ is Better than ‘Native versus Non-native Speaker’.” Applied Linguistics, DOI: 10.1093/ applin/ amw055. Dewaele, Jean-Marc/ Anat Stavans (2014). “The Effect of Immigration, Acculturation and Multicompetence on Personality Profiles of Israeli Multilinguals.” International Journal of Bilingualism 18(3). 203-221. Firth, Alan (1996). “The Discursive Accomplishment of Normality. On ‘Lingua Franca’ English and Conversation Analysis.” Journal of Pragmatics 26. 237-259. Fløttum, Kjersti (ed.) (2007). Language and Discipline Perspectives on Academic Discourse. Cambridge: Cambridge Scholars Publishing. Graddol, David (2006). English Next? London: British Council. Grosjean, François (2001). “The Bilingual’s Language Modes.” In Nicol (2001). 1- 22. Grosjean, François (ed.) (2008). Bilingualism: Language and Cognition. Danvers: CUP. Grosjean, François (2010). Bilingual: Life and Reality. Cambridge: Harvard University Press. Hammer, Kate (2017). “Bilingual Cogito: Inner Speech in Acculturated Bilinguals.” International Journal of Bilingual Education and Bilingualism. DOI: 10.1080/ 13670050.2017.1285862. Hockett, Charles F. (1960). “The Origin of Speech.” Scientific American 203(3). 88- 96. Hyland, Ken (2005). Metadiscourse: Exploring Interaction in Writing. London: Continuum. Jenkins, Jennifer (2007). English as a Lingua Franca: Attitude and Identity. Oxford: OUP. Jenkins, Jennifer (2009). World Englishes: A Resource Book for Students. 2 nd ed. Milton Park: Routledge. Kaur, Jagdish (2009). “Pre-empting Problems of Understanding in English as a Lingua Franca.” In: Mauranen/ Ranta (2009). 107-123. MacIntyre, Peter/ Kimberly Noels/ Richard Clément (1997). “Biases in Self-ratings of Second Language Proficiency: The Role of Language Anxiety.” Language Learning 47(2). 265-287. MacKenzie, Ian (2012). “English as a Lingua Franca in Europe: Bilingualism and Multicompetence.” International Journal of Multilingualism 9(1). 83-100. Mauranen, Anna (1993). Cultural Differences in Academic Rhetoric: A Textlinguistic Study. Frankfurt a.M.: Peter Lang. Mauranen, Anna (2006). “Signalling and Preventing Understanding in English as a Lingua Franca Communication.” International Journal of the Sociology of Language 177. 123-150. Mauranen, Anna (2007). “Hybrid Voices: English as the Lingua Franca of Academics.” In: Fløttum (2007). 244-259. Mauranen, Anna (2010). “Discourse Reflexivity - a Discourse Universal? The Case of ELF.” Nordic Journal of English Studies 9(2). 13-40. Pia Resnik 210 Mauranen, Anna (2012). Exploring ELF: Academic English Shaped by Non-native Speakers. Cambridge: CUP. Mauranen, Anna/ Elina Ranta (eds.) (2009). English as a Lingua Franca. Studies and Findings. Newcastle upon Tyne: Cambridge Scholar Publishing. Mey, Jacob L. (2001). Pragmatics. An Introduction. Oxford: Blackwell. Nicol, Janet (ed.) (2001). One Mind, Two Languages: Bilingual Language Processing. Oxford: Blackwell. Ortega, Lourdes (2009). Understanding Second Language Acquisition. Oxon: Routledge. Penz, Hermine (2011). “‘What do we mean by that? ’ Metadiscourse in ELF Project Discussions.” In: Archibald/ Cogo/ Jenkins (2011). 185-201. Phillipson, Robert (2008). “Lingua franca or lingua frankensteinia? English in European Integration and Globalisation.” World Englishes 27(2). 250-267. Resnik, Pia (forthcoming). Differences in Feeling - Feeling the Difference: Multilinguals’ Verbalisation and Perception of Emotions. Bristol: Multilingual Matters. Schiffrin, Deborah (1980). “Metatalk: Organizational and Evaluative Brackets in Discourse.” Sociolinguistic Inquiry: Language and Social Interaction 50. 199-236. Seidlhofer, Barbara (2011). Understanding English as a Lingua Franca. Oxford: OUP. Vande Kopple, William J. (1985). “Some Exploratory Discourse on Metadiscourse.” College Composition and Communication 36. 63-94. Wilson, Rosemary/ Jean-Marc Dewaele (2010). “The Use of Web Questionnaires in Second Language Acquisition and Bilingualism Research.” Second Language Research 26(1). 103-123. Pia Resnik Department of English University of Vienna Austria