Epistemic Stance in Spoken L2 English: The Effect of Task and Speaker Style

Abstract

The article discusses epistemic stance in spoken L2 production. Using a subset of the Trinity Lancaster Corpus of spoken L2 production, we analysed the speech of 132 advanced L2 speakers from different L1 and cultural backgrounds taking part in four speaking tasks: one largely monologic presentation task and three interactive tasks. The study focused on three types of epistemic forms: adverbial, adjectival, and verbal expressions. The results showed a systematic variation in L2 speakers’ stance-taking choices across the four tasks. The largest difference was found between the monologic and the dialogic tasks, but differences were also found in the distribution of epistemic markers in the three interactive tasks. The variation was explained in terms of the interactional demands of individual tasks. The study also found evidence of considerable inter-speaker variation, indicating the existence of individual speaker style in the use of epistemic markers. By focusing on social use of language, this article seeks to contribute to our understanding of communicative competence of advanced L2 speakers. This research is of relevance to teachers, material developers, as well as language testers interested in second language pragmatic ability.

INTRODUCTION

Speaking is among the most challenging tasks faced by L2 learners and users. In particular, interactive communication such as conversation requires the ability to participate appropriately in a rapidly developing discourse which involves drawing on diverse cognitive and linguistic resources. So L2 users, in addition to lexical and morpho-grammatical knowledge, need to have a good range of pragmatic skills to allow them to take their parts in co-constructing the interaction in a manner appropriate to the social context, interlocutor roles, and purpose of the exchange. This requirement for understanding and productive use of these different aspects of discourse makes pragmatic skills difficult to acquire and apply, even for advanced L2 speakers (Romero-Trillo 2002; Fordyce 2014).

Despite the inherent challenges, speaking is also one of the skills typically viewed as essential to acquire, as documented by a multitude of language courses targeting speaking or communicative skills. However, compared with other areas of learner language such as grammar, the pragmatics of spoken communication is still an under-researched area in second language acquisition (Bardovi-Harlig 2013; Callies 2013). This article, therefore, focuses on one of the core pragmatic skills in speaking, the expression of epistemic stance. Despite the vital role of epistemic markers (EMs) in everyday communication (Biber et al. 1999; Kärkkäinen 2003; Bygate 2009), they have not been studied systematically in L2 spoken production, with studies on the use of pragmatic markers (PMs) by L2 speakers having been typically based on one type of speaking task and carried out with L2 speakers from one linguistic background (cf. e.g. Müller 2005; Fung and Carter 2007). It therefore remains unclear whether the resulting findings represent performance typical of a particular task and a particular group of learners or whether they are reflective of more general trends in L2 spoken production.

The study used the Trinity Lancaster Corpus (TLC) of spoken L2 production to examine systematic variation in the use of three types of EMs (adverbial, adjectival, and verbal expressions of epistemic stance) across different speaking tasks and across speakers representing different L1 and cultural backgrounds. TLC represents a large collection of semi-formal speech elicited in an institutional setting as part of an examination of spoken English. By focusing on social use of language, this article seeks to contribute to our understanding of the communicative competence of advanced L2 speakers. Findings of this research can be thus of relevance to teachers, material developers, as well as language testers interested in second language pragmatic ability.

Epistemic stance

Epistemic stance, as part of evaluative discourse, fulfils three major interconnected functions in the interaction: (i) expressing opinion, (ii) maintaining relations between the interlocutors, and (iii) discourse organization (Hunston and Thompson 2000). In this way, it acts as both a subjective (self-expressive) and intersubjective device, allowing speakers to position themselves towards their propositions (express their doubt or certainty about a statement) as well as towards the other interlocutor(s) (Kärkkäinen 2006). When we speak, we not only convey propositional content but also communicate a variety of subjective meanings that pertain to our value systems, identity, confidence, and so forth (Hunston and Thompson 2000; Bucholtz and Hall 2005). The following example, taken from the TLC, demonstrates the interaction between propositional content and epistemic stance [transcription symbols: <.> short pause (up to three seconds); tha= truncated words; <unclear = for> unclear speech]:

(1) I like graffiti it’s er it’s ar= I think it’s art it’s just a different kind of art which is not very understood by society nowadays.

In Example (1), the speaker started making a statement about graffiti which most likely would have been ‘it’s art’. However, before he finished, he modified the proposition by inserting ‘I think’. As a result of this modification, this utterance does not only contain the statement itself but also communicates the speaker’s position with respect to this statement (lower degree of certainty); ‘I think’ also acts as a hedging device, reducing the impact of a potentially face-threatening move (cf. Kärkkäinen 2003; Kaltenböck 2010). ‘I think’ is just one of many forms investigated in this article that express epistemicity in discourse. There is a large variety of forms that indicate epistemic stance (Holmes 1988; Biber et al. 1999; Biber 2006; Englebretson 2007; Brezina 2012). In spoken communication, epistemic stance is most commonly expressed by lexical verbs (I think, I know, I guess), adverbs (maybe, actually, certainly), and modal verbs with epistemic meaning (may, could, might) (Biber et al. 1999).

Epistemic stance in second language

Research on epistemic stance expressions by L2 speakers has so far focused mostly on written language (e.g. Kärkkäinen 1992; Hyland and Milton 1997; Aijmer 2002; McEnery and Kifle 2002). Although these studies provide important insights into L2 speakers’ expressions of stance, we also need to learn more about how language users negotiate their position in spoken discourse, with its limited possibility of editing and greater demands on online processing (Bygate 2009).

The research on stance-taking in L2 speech focused both on the forms and functions of EMs. With respect to epistemic forms, previous research identified preference of lower-level learners for using cognitive verbs and adverbials rather than modal verbs (Salsbury and Bardovi-Harlig 2000; Fordyce 2009). In particular, two expressions, ‘maybe’ and ‘I think’, appeared to dominate learners’ expression of stance. This pattern was attributed to linguistic constraints in L2, with Salsbury and Bardovi-Harlig (2000) showing that the range of EMs increased with the development of L2 proficiency. They argued that adverbials and cognitive verbs convey the epistemic information explicitly and unambiguously, which makes them easier to use than modal verbs (Salsbury and Bardovi-Harlig 2000; Fordyce 2014). Research on functional range of epistemic expressions showed that while L2 speakers used EMs for both subjective and intersubjective purposes (Baumgarten and House 2010; Mortensen 2012), in some cases they used EMs in a way that differed from native speakers (Zhang and Sabet 2016). For example, Baumgarten and House (2010) reported that while L2 speakers primarily used ‘I think’ and ‘I don’t know’ to indicate certainty or lack of knowledge, L1 speakers used these expressions also for discourse organization and as a pragmaticalized verbal routine.

These findings are in line with the evidence from research on pragmatic markers in general, which showed differences in the range as well as function of PMs used by L1 and L2 speakers (e.g. Hasselgren 2002; Müller 2005; Fung and Carter 2007). To date, however, no systematic analysis of EMs, such as that found in Biber et al. (1999) for L1 English, has been attempted for L2 production. This study therefore seeks to complement previous research by investigating patterns in the L2 expression of epistemicity according to three large lexico-grammatical categories (adverbial, adjectival, and verbal expressions) across four different speech types.

The effect of different speaking tasks on the use of pragmatic markers

The ability to distinguish between different interactional contexts and to adjust one’s language accordingly lies at the core of effective communication. PMs are highly context-sensitive and can therefore provide an indication of the developmental level of second language pragmatic ability. Evidence from several studies (e.g. Granger and Rayson 1998; Hinkel 2003, 2005; Ädel 2008;,Gilquin and Paquot 2008) suggests that even advanced L2 users struggle to adjust their linguistic choices according to the context or genre of the discourse. Examining learner performance across two broad modes of communication (speech and writing), these studies showed that L2 users tend to draw on the same set of linguistic means irrespective of the purpose and context of discourse. This is apparent, for example, from L2 speakers using informal, speech-like features in their academic written discourse. The lack of genre awareness found in these studies was attributed to various factors such as insufficient exposure to authentic target language (Dewaele 2002; Romero-Trillo 2002) or developing literacy skills (Gilquin and Paquot 2008).

More recently, researchers also started looking at variation in PMs used in different speaking tasks. The results of these studies present mixed evidence: some found that L2 speakers adjusted their speaking style in terms of the frequency and type of PMs according to different interaction types (e.g. Lam 2009; Liao 2009; Huang 2011; Wei 2011; Neary-Sundquist 2013; Gablasova and Brezina 2015), while others reported no such variation in L2 speech (e.g. Fuller 2003a). L2 proficiency could partly explain the mixed results, with more advanced L2 users found to be more sensitive to the interaction type than lower-level L2 speakers (Wei 2011; Neary-Sundquist 2013). When variation in PMs between different tasks was identified in the studies, it was attributed to a range of factors such as different degree of task interactiveness (Lam 2009; Huang 2011) or difference in speaker roles adopted by the L2 users in these tasks (Liao 2009; Gablasova and Brezina 2015).

The studies discussed above involved a wide range of pragmatic items (e.g. cohesive as well as interactional PMs), and in most cases focused on a set of functionally unrelated PMs. Also, many of them involved a relatively small number of L2 speakers which made it difficult to methodically analyse the complex interplay behind variation (or lack of it) found in L2 speech across different interaction types. A more systematic inquiry into the pragmatic aspects of L2 spoken production in different settings is needed to assist teachers and material developers in guiding L2 learners in appropriate use of epistemic stance markers according to the demands of different types of interaction.

Individual speaker style in epistemic stance-taking

Systematic variation in speech can be studied from a variety of perspectives, focusing on different factors that influence speakers’ linguistic choices (Eckert and Rickford 2001; Biber and Conrad 2009). In addition to social attributes (e.g. age or gender) that may affect one’s speech style, as speakers select from words and phrases in their linguistic repertoire according to a particular communicative need and contextual constraints, they also retain some typical, individual preferences in their linguistic choices. As Coupland (2007) argues, speakers differ not only in the vocabulary they know and can choose from, but also in their tendency to use some words more frequently or in particular situations. Individual preferences may be especially reflected in stance-taking, with its close ties to how speakers perceive themselves and others (Bucholtz and Hall 2005).

Investigating epistemic stance in L1, Kirkham (2011) and Brezina (2013) reported distinct idiolectal patterns in the speech of native English users. Brezina (2013), in a quantitative, corpus-based study of epistemic adverbial markers, such as certainly and maybe, found that speakers tended to repeatedly express the same sub-type of epistemicity (e.g. probability or certainty) and also showed a relatively consistent preference for specific forms. Kirkham (2011), in a qualitative study, observed that even if different speakers used the same set of epistemic forms, each of the speakers consistently used them to signal different meanings.

Regarding individual style in the L2 use of PMs, Liao (2009) studied six speakers who shared several important characteristics (e.g. had the same L1 and similar educational backgrounds). Liao (2009) argued that the variation in the choice of PMs by these L2 users was caused by individual differences in the perception of their identity and professional role. For example, one of the speakers consciously avoided the use of PMs more typical of informal speech, as she wished to emphasize her authority as a teacher. Further, Siegal (1996) showed how speakers consciously reflect on their pragmatic choices (e.g. the expression of certainty) in the face of linguistic and societal expectations which may be in conflict with their personal preferences.

Studying individual speaking style (idiolect) in L2 has been somewhat complicated by the fact that the investigation of style presupposes that speakers have a range of expressions at their disposal from which they choose according to their preference and communicative goal. In an L2, this cannot be always assumed due to the intervening variable of language proficiency (Dewaele 2008a). For example, Siegal (1996) describes a case when an L2 Japanese speaker used an epistemic modal indexing uncertainty as part of her politeness strategy inappropriately without being aware of the full range of meanings of the word.

The current study

The present study focuses on the expression of epistemic stance in the speech of advanced non-native speakers of English. It examines variation in the use of EMs related to two sources: difference in speaking tasks and difference caused by individual style. The study addresses the following research questions:

RQ1: Is there an effect of different speaking tasks on L2 speakers’ use of epistemic stance markers?

RQ2: Is there evidence of individual speaker style in the use of epistemic stance markers across different speaking tasks?

RQ1 investigated the effect of the speaking task as the independent variable by analysing (i) variation in the frequency of EMs across four speaking tasks, (ii) variation in the distribution of EMs indexing (un)certainty across the tasks, and (iii) the range of EMs in each task. RQ2 investigated intra-speaker variation and consistency of EMs use across the four tasks.

METHOD

Corpus description

The corpus used in this study is the advanced-level subset of the Trinity Lancaster Corpus of spoken L2 production which was collected as part of the Graded Examinations of Spoken English conducted by Trinity College London (2010). The advanced sub-corpus consists of approximately 0.5 million words from candidate–examiner interaction, with around 300,000 tokens produced by (exam) candidates and around 200,000 tokens by examiners. All examiners were native speakers of English; the L2 data came from 132 speakers and were collected in six different countries: China (23 speakers), Italy (31 speakers), Mexico (31 speakers), Sri Lanka (13 speakers), India (4), and Spain (30 speakers). The sub-corpus contains data from 81 female and 51 male speakers representing three broad age groups: 45 young speakers (aged 14–19 years), 44 young adults (20–30 years), and 43 mature speakers (31–55 years). The L2 speakers include active learners as well as those who use English in their professional life or for leisure.

Each L2 speaker took part in an examination which consisted of five tasks (four of which are included in this study)¹ and lasted for about 25 minutes. Each examination was transcribed and divided according to the different tasks, thus creating 132 cases of each task. Participants included in the advanced level sub-corpus were examined at Grades 10, 11, and 12, corresponding to levels C1 and C2 of the Common European Framework of Reference (Papageorgiou 2007). Only candidates who received at least a Pass in their exam and thus fulfilled the requirements of C1 or C2 level were included in the corpus.

Speaking tasks

L2 spoken production across four speaking tasks was examined in this study: one prevalently monologic task and three largely interactive tasks (Trinity College London 2010). In the first task, the presentation (PRES), the candidate (L2 speaker) talks about a topic of his or her own choice. The PRES is semi-formal or formal in nature. The three tasks following the PRES are dialogic and represent different interactive situations: Discussion (DISC) involves the examiner and the candidate discussing the topic of the PRES. In the interactive task (INT) the candidate is required to take a proactive role in the conversation with a series of questions or comments following a prompt from the examiner. The final task, conversation (CONV), involves talking about two topics of general interest. For Grades 10 and 11, these topics are selected from a list previously known to the candidate; for Grade 12, any topic can be chosen by the examiner. The four tasks thus represent a variety of settings in which spoken language is used with different degrees of formality, interactiveness, topic familiarity, and different interlocutor roles (for a more detailed account of the tasks, see Wall and Taylor 2014; Gablasova and Brezina 2015). Table 1 provides an overview of the characteristics of interaction types represented by each task, while Table 2 shows the average length of language samples elicited in each task and the size of each sub-corpora used in the study. Examples (short extracts) of each task taken from the corpus can be seen in the Supplementary material.

Table 1:

Speaking tasks: discourse characteristics

Task	Topic familiarity	Interlocutor roles	Interactiveness	Formality
PRES	Pre-selected topic	Candidate-led	Monologic	Semi-formal/formal
DISC	Pre-selected topic	Jointly led	Dialogic	Semi-formal
INT	General topic	Candidate-led	Dialogic	Semi-formal
CONV	General topic	Jointly led	Dialogic	Semi-formal

Table 2:

Speaking tasks: individual speaker contribution and sizes of sub-corpora

	Number of words per L2 speaker		L2 speaker turns (per 1,000 words)		Duration	Number of words in each sub-corpus
Task	Mean	SD	Mean	SD
PRES	655.67	122.59	21.85	16.12	up to 5 min	86,549
DISC	469.04	125.34	86.59	37.09	up to 5 min	61,913
INT	379.49	97.06	81.84	37.59	up to 5 min	50,093
CONV	684.71	162.32	82.22	31.84	up to 6 min	90,382

Identifying epistemic stance markers

Applying corpus methods to the study of pragmatic features is a challenging task. Corpus linguistics is primarily a quantitative approach to the study of language, which allows us to search through large data sets to describe patterns of language use. The difficulty with studying pragmatic variation using a corpus stems from the fact that while we can search the corpus for specific forms, we cannot, in the absence of suitable annotation, ensure that the forms function in the same way (Hunston 2007; Rühlemann 2011). PMs in particular are notoriously multifunctional (e.g. Müller 2005; Baumgarten and House 2010), yet the process of annotating such features in a large data set is often prohibitively time-consuming.

In this study, we adopted a multi-step approach to identifying epistemic stance. In a preliminary analysis, we identified a list of candidate forms that were then used as search terms in the study. To create the list, we first consulted existing lists of epistemic forms (Biber et al. 1999; Conrad and Biber 2000; Kärkkäinen 2003; Brezina 2012). This was complemented by a manual analysis of 10 transcripts in which forms considered to express epistemic stance were identified. Following this process, a comprehensive list of search terms (Table 3) was compiled.

Table 3:

Epistemic items searched for in L2 speech^a

Grammatical category	Search terms
Adverbial expressions	Actually; apparently; certainly; definitely; evidently; for sure; kind of; maybe; no doubt; obviously; perhaps; possibly; predictably; probably; roughly; sort of; surely; undoubtedly; without @ doubt
Adjectival expressions	Doubtful; impossible; improbable; likely; possible; probable; unlikely; I am @ certain; I am @ confident; I am @ convinced; I am @ sure; I’m @ certain; I’m @ confident; I’m @ convinced; I’m@ sure; I cannot @ sure; I can’t @ sure; I cannot @ certain; I can’t @ certain
Verbal expressions	Appear; appears; seem; seems; I @ assume; I @ believe; I @ bet; I @ doubt; I @ gather; I @ guess; I @ mean; I @ know; I @ presume; I @ reckon; I @ suppose; I @ suspect; I @ think

^aIn the search terms, @ symbol represents zero to two words to include modification or negation of epistemic phrases (e.g. ‘I am totally convinced’ or ‘I don’t suppose’).

The search terms in Table 3 represent items with a fairly stable epistemic meaning across different contexts (although they may also have other, secondary, non-epistemic functions), making them amenable to corpus analysis. Even so, it should be noted that in a limited number of contexts these search terms may produce false hits, that is, they may identify forms that are not used for epistemic positioning of the speaker. For this reason, a concordance for each of the forms in Table 3 was produced and a sample of concordance lines was manually coded for epistemic or non-epistemic use of the form, confirming that these forms were used predominately in epistemic way.² In addition, while we cannot fully control the fact that some of the search terms also return results that include non-epistemic uses of the markers, we provide all of our search terms so that the methodology is transparent and replicable.

It should be noted that, unlike the approach taken by previous researchers (e.g. Holmes 1988; Biber et al. 1999; Biber 2006), our list does not include modal verbs and nouns. Nouns were omitted from the analysis, as the preliminary search in the corpus found that only a few of the target forms had an epistemic function. This was attributed to the fact that some expressions (e.g. ‘conclusion’ or ‘hypothesis’) are more common in writing than in speech. Modal verbs were omitted because the preliminary analysis revealed that (interactive) speech made deontic and epistemic modality difficult to distinguish reliably. In speech, PMs fulfil a variety of functions whose meaning relies on contextual factors which help the interlocutors with interpretation; in writing, the intention of the writer is often made more explicit as the extra-textual support is limited. As a result of this difference, some search terms regularly used in corpus-based studies of writing cannot be reliably applied to spoken language.

Data analysis

All searches in this study were carried out using MonoConc Pro (Barlow 2003). Descriptive statistics were used to compare general tendencies in stance-taking across the tasks [i.e. the distribution of EMs indexing (un)certainty and the range of EMs in each task]. Inferential statistics were used to investigate across-task variation focusing on (i) the effect of the task [repeated-measures analysis of variance (ANOVA)] on the frequency of EMs and (ii) the consistency in the EM use in the speech of each L2 speakers (Pearson’s product-moment correlational analysis). Inferential statistics allow one to measure the strength of the effect of the independent variable while taking into account inter-speaker variation. It also allows one to establish that the effect is not due to chance (for a discussion of the advantages of including inter-speaker differences in corpus-based studies see Granger 2012; Callies 2013; Brezina and Meyerhoff 2014). To ensure fair comparison, data were normalized to the basis of 100,000 when dealing with larger sub-corpora and to 1,000 when dealing with individual speakers to best reflect the size of the samples.

RESULTS

The effect of speaking task

Table 4 shows the descriptive statistics for the use of three different categories of EMs, namely, adverbial expressions (ADV), adjectival expressions (ADJ), and verbal expressions (VERB) as well as the overall frequency of EMs in all four tasks taken together. The frequency of occurrence was normalized to 1,000 words. These results are visualized in Figure 1. A repeated-measures (within-subjects) ANOVA, with a Greenhouse-Geisser correction for sphericity, was used to compare the overall frequency of EMs across the four tasks.³ The test showed a significant main effect for task type [F(2.8, 367.9) = 114.122, p < .001, partial eta² = .466]. Post hoc Bonferroni comparisons revealed that there was a significant difference between PRES and every other task (all p < .001) and between the INT and the DISC (p < .05). No difference was found between the CONV and the DISC (p = 1.0) or CONV and the INT (p = .113).

Table 4:

EM use across four tasks

	ADV		ADJ		VERB		Total EMs
Task	Mean	SD	Mean	SD	Mean	SD	Mean	SD
PRES	3.25	3.67	0.57	1.18	4.59	5.05	8.40	6.73
DISC	7.45	7.31	0.47	1.19	14.27	8.69	22.19	11.74
INT	8.81	7.77	0.66	1.45	16.27	9.90	25.74	11.84
CONV	7.05	5.82	0.47	0.91	15.82	7.99	23.34	10.61

Figure 1:

Proportion of different types of EMs in each speaking task

As Figure 1 shows, most EMs were used by L2 speakers in the INT, followed by the CONV and DISC, with the smallest number appearing in the PRES. In terms of the type of the EMs, regardless of the task, most occurrences of epistemic stance involved verbs, followed by adverbial expressions. Adjectival expressions played only a small role in the expression of epistemic stance.

To better understand the nature of epistemic stance employed by L2 speakers in each of the tasks, we further examined how the tasks differ in the degree of certainty expressed by L2 speakers. The EMs best suited for this purpose were adverbial epistemic markers (AEMs) that could be considered to unambiguously express strong or weak certainty (their epistemic polarity cannot be changed by modification of negation, e.g. ‘certainly not’ still expressed strong certainty and ‘maybe not’ weak certainty). The following markers were selected as indicative of certainty: certainly, definitely, for sure, no doubt, obviously, surely, while the following markers were considered to express uncertainty: maybe, perhaps, possibly, and probably.

Table 5 shows the occurrence of each group of markers (normalized to 1,000 words) and the proportion of this group of AEMs out of all AEMs produced in each task.

Table 5:

AEMs indexing (un)certainty

	Certainty		Uncertainty		Other AEMs
Task	Frequency	Per cent	Frequency	Per cent	Frequency	Per cent
PRES	4.0	12.03	13.1	39.18	16.3	48.80
DISC	5.3	6.95	37.4	49.05	33.5	44.00
INT	2.6	2.86	55.4	61.98	31.5	35.17
CONV	5.4	7.42	42.4	58.03	25.2	34.55

As can be seen from Table 5, there were considerable differences in what degree of certainty was expressed by AEMs in each task. In terms of the proportion of the markers signalling (un)certainty, PRES, closely followed by CONV and DISC, gave rise to more markers of strong certainty, with the INT eliciting less than 3 per cent of these EMs. With respect to adverbials expressing tentativeness, the order was reversed with almost two-thirds of AEMs in the INT signalling uncertainty.

Next, the frequency of individual EMs in each task was examined. Tables 6–8 provide information on the frequency of all EMs found in the four tasks according to their lexico-grammatical category (the frequencies are normalized to 100,000 words; the absolute frequencies can be seen in the Supplementary material).

Table 6:

Adverbial expressions

Epistemic marker	PRES	DISC	INT	CONV	Totala
Maybe	94.2	319.4	442.2	363.9	287.2
Kind of	95.3	176.5	176.9	105.1	130.0
Actually	55.1	152.5	108.1	125.0	107.0
Probably	24.1	41.7	66.8	39.8	40.2
Sort of	10.3	6.4	29.5	21.0	16.2
Perhaps	8.0	11.2	41.2	16.6	17.2
Obviously	17.2	24.1	7.9	16.6	16.9
Definitely	5.7	22.5	9.8	18.8	14.1
Surely	9.2	3.2	2.0	7.7	6.2
Possibly	4.6	1.6	3.9	3.3	3.4
For sure	1.2	3.2		6.6	3.1
Certainly	3.4		5.9	4.4	2.4
Apparently	2.3			1.1	2.4
No doubt	3.4				1.0

^aThe number of adverbial expressions per whole examination session which includes all four speaking tasks and all examination cases.

Table 7:

Adjectival expressions

Epistemic marker	PRES	DISC	INT	CONV	Totala
I @ sure	10.3	22.6	26.0	15.5	17.2
Possible	24.1	12.9	26.0	12.2	18.2
Impossible	6.9	12.9	16.0	8.8	10.2
Likely	11.5			4.4	4.8
I @ convinced	1.1		2.0	3.3	1.7
Improbable	1.1				0.34

^aThe number of adjectival expressions per whole examination session which includes all four speaking tasks and all examination cases.

Table 8:

Verbal expressions

Epistemic marker	PRES	DISC	INT	CONV	Totala
I @ think	295.1	892.3	1061.2	1090.5	804.6
I @ know	75.8	232.7	306.6	232.2	198.5
I @ mean	29.9	138.0	155.2	127.2	105.3
I @ believe	29.9	38.5	23.6	36.5	32.7
I @ guess	10.3	17.7	45.2	21.0	21.3
Seem/s	8.0	19.3	23.6	8.8	13.4
I @ suppose	2.3	16.0	17.7	17.7	12.7
Appear/s	6.9	3.2	2.0	1.1	3.4
I @ bet			3.9	2.2	1.4
I @ doubt				2.2	0.7
I @ suspect				1.1	0.3

^aThe number of verbal expressions per whole examination session, which includes all four speaking tasks and all examination cases.

Overall, the L2 speakers in this study produced 31 different epistemic forms (of the 55 listed in Table 3). As can be seen, the PRES and CONV tasks elicited the widest range of EMs. In terms of the frequency of individual markers, of 31 EMs used by L2 speakers in this study, three specific markers account for more than 65 per cent of all EMs. These markers are ‘I @ think’ (42.5 per cent), ‘maybe’ (15.2 per cent), and ‘I @ know’ (10.5 per cent), while the remaining 28 EMs account for the remaining one-third (31.9 per cent).

The evidence of individual speaking style

RQ2 sought to establish whether the data provide evidence of the existence of an individual speaking style operationalized as a consistent individual preference in the use of EMs by L2 speakers. A Pearson’s correlational analysis was used to determine whether there was a relationship between the frequency of EMs used by L2 speakers in different tasks. The relationship in the frequency of EMs was assessed at three different levels: (i) overall frequency of EMs in each task, (ii) frequency of EMs from a particular lexico-grammatical category (adverbials and verbs), and (iii) frequency of specific EMs (‘I @ think’ and ‘maybe’). Adverbials and verbal rather than adjectival expressions were used in the correlation, as they provided sufficient data for the analysis. ‘I @ think’ and ‘maybe’ were chosen to represent a frequent verbal and adverbial expressions, respectively. The results are presented in Table 9.

Table 9:

Correlation: speaker style across task types

	DISC	PRES	INT
Total number of EMs
CONV	.337**	.197*	.526**
DISC		.417**	.379**
PRES			.283**
ADV
CONV	.533**	.395**	.493**
DISC		.359**	.564**
PRES			.204*
VERB
CONV	.389**	.116	.482**
DISC		.384**	.288**
PRES			.195*
Specific expression: ‘I @ think’
CONV	.495**	.209*	.470**
DISC		.352**	.298**
PRES			.242**
Specific expression: ‘maybe’
CONV	.602**	.490**	.525**
DISC		.434**	.589**
PRES			.379**

*p < .05; **p < .001.

Correlations in Table 9, ranging from modest (0.2) to moderate (0.6), indicate the presence of individual speaker style at all three observed levels, with respect to overall frequency of EMs as well as in relation to the more specific lexico-grammatical preferences. It is interesting to note that in most cases (with the exception of the relationship in the use of verbal EMs between the PRES and CONV), the connection in the EM use was also found between the three dialogic tasks and the PRES, despite the fact that this task is the most formal, least interactive, and is prepared beforehand.

These individual preferences are illustrated in Figure 2 which shows the use of ‘maybe’ (normalized to 1,000 words) by 10 Spanish L1 speakers taken from the corpus. Figure 2 shows that while adjusting their production according to the speaking task, the L2 speakers retain their more individual preferences for the frequency of the particular marker. For example, Speaker 7 uses ‘maybe’ quite frequently across the three dialogic tasks, while Speaker 4 uses it rather moderately and Speaker 6 does not use ‘maybe’ at all in any of the tasks. A similar pattern could be observed with respect to the overall frequency of EMs and with the preference for a particular lexico-grammatical class of EMs. Here, for instance, nine speakers used fewer than five adverbial EMs in the whole of their speech sample, while six speakers produced more than 100 adverbial or verbal expressions in theirs. These findings provide further evidence of the large variety of communicative strategies used by advanced L2 speakers.

Figure 2:

Individual variation: ‘maybe’ across four tasks

DISCUSSION

The effect of the speaking task on the use of EMs

Distribution of EMs across tasks

Overall, the results showed that there is a relationship between different speaking tasks and the number and type of EMs used by L2 speakers. Consistently, the most formal, monologic task (PRES) was associated with a specific pattern of stance-taking that differed from the three dialogic semi-formal tasks; however, differences were also found between the three dialogic tasks. These findings provide a strong indication that advanced L2 speakers adapt their speech style to reflect changing contextual demands of the spoken activity in which they are engaged.

The variation in the expression of epistemic stance in the four tasks (see Table 4) can be attributed to several interconnected factors. First, it appears to be strongly related to the varying degree of interactiveness in the tasks as demonstrated by the difference between a relatively low frequency of EMs in the monologic task compared with more frequent occurrence in the three dialogic tasks. EMs, and PMs in general, are more characteristic of interactive speech (e.g. Fuller 2003b; Huang 2011; Brezina 2012) where they are used to express subjectivity (self), co-manage the developing discourse, and manage the relationship between the speakers (e.g. Hunston and Thompson 2000; Kärkkäinen 2006; Fox Tree 2010; Mortensen 2012). By contrast, in prevalently monologic speech, even though the intended listener may be acknowledged, the intersubjective relationship does not need to be negotiated with a similar intensity. The following examples taken from the PRES (2) and the CONV (3) illustrate this distinction (S = L2 speaker; E = Examiner).

(2) S: er <.> another <.> another argument <.> against er this kind of food is tha= the there are different points of view depending on the countries for example United States is very for GM and <.> is no is <unclear=for long time> to <.> the consumers if the food have <.> er have made with GM er products or not but in Europe er there is a lot er because European people are more concerned about this kind of <.> of arguments and if they have more than one per cent of of content in GM products they have to <.> warn <.> they have to have <.> er another concern is that with this kind of crops <.> er we are <unclear=not seeing> biodiversity <.> er so that means er less <.> erm <.> less variety in the world <.> less capacity to to find new things so but in conclusion <.> er I I am for genetic modified foods because I think they are an answer to some problems that our society have

(3) E: but d-do people really think about the next generation or do they tend to think about just their own generation? cos I think certainly in the UK no one worries about what’s going to happen to their grandchildren their great grandchildren they want to know what’s happening now to themselves

S: no I think that people er who have children they care about them

E: mm

S: and they wo-worry about the situation because of their children

E: uhu

S: I think I think so

E: right yes yeah I think in the UK […]

Example (2) is characterized by longer turns, with no interruption from the other interlocutor, while Example (3) demonstrates a more typical conversation which consists of shorter turns and frequent back-channelling signals. In terms of stance-taking, the epistemic form in Example (2) signals the speaker’s opinion which concludes a series of viewpoints presented; in Example (3), the EMs in addition to conveying speakers’ opinions also function as intersubjective devices, helping the speakers to manage the relationship in the conversation (e.g. show agreement or disagreement). It is due to this increased demand on negotiation of knowledge and interpersonal relationship management that the dialogic tasks require more EMs for effective communication.

Uncertainty and certainty across tasks

The difference in speakers’ stance-taking strategies was not limited to the monologic–dialogic dichotomy; variation related to both the frequency and type of EMs was also found across all four tasks. Closer analysis of AEMs revealed that L2 speakers expressed different degrees of (un)certainty about their statements in response to different interaction types, with PRES, CONV, and DISC giving rise to EMs of stronger certainty, whereas the INT was associated with a higher degree of tentativeness (see Table 5). This variation appears to be related to the changing interlocutor roles and topic control in the discourse, with the control of the topic being positively related to stronger certainty expressed by participants (Gablasova and Brezina 2015). Following the exam guidelines, in the PRES, the L2 speaker can speak in a virtually unchallenged manner and controls both the topic (which is prepared beforehand) and the speech flow without interruption from the examiner. The situation is to some extent similar in the DISC and CONV; the interaction is developed jointly, with examiner often asking open-ended questions that allow the L2 speaker to put forward and develop ideas, as demonstrated in Example (4).

(4) E: because in England for example if you are a teacher in London you get a higher salary

S: oh

E: do you think that’s the same in China?

S: er <.> I I believe that if you have ability you can <.> you c= you can get a good job so er you can have a home in Beijing

By contrast, in the INT, in which the L2 speaker is expected to take responsibility for the communication flow (Trinity College London 2010), he or she cannot control the topic, but rather has to negotiate it with the examiner. With respect to speaker roles, the INT is, in some ways, the opposite of the CONV and DISC, in which the examiner and L2 speaker talk about a general topic. In the INT, the L2 speaker has to comment on a topic that is presented as an issue of personal relevance by the examiner. As a result, when L2 speakers express their opinions, this frequently involves making evaluative claims about what is presented as the examiner’s actions, beliefs, or opinions. This is potentially a face-threatening situation, especially since the examiner arguably holds more power in the dialogue (as a native speaker and the person who will assess the candidate’s performance) and it necessitates strategies such as hedging and downplaying to be used in the non-native speaker’s questions, suggestions, and comments. This can be illustrated by two examples taken from the INT in which the L2 speaker expressed disagreement with the opinion presented by the examiner.

(5) S: I agree with this point but don’t you think maybe the ti= fact that times are changing is a good thing?

(6) S: yes I think that in that point you are right however er probably this teacher knows about erm his students’ possibilities to develop certain skill

As can be seen from Examples (5) and (6), the expression of disagreement is in both cases modified by an EM that acts as a hedging device. The handling of disagreements in the three dialogic tasks can serve as an illustration of the different question–answer dynamic, with the INT presenting a more face-threatening setting than the other two dialogic tasks: in the INT, an EM followed the expression of disagreement (‘but’ or ‘however’) in 24 per cent of the cases (130 of 543), while in the DISC and CONV, this number was consistently lower (17 and 19 per cent, respectively). In addition, while EMs indexing certainty appeared relatively often in the CONV and DISC, as shown in Table 5, in the INT they were almost entirely limited to the situations when the candidate expressed agreement with the examiner (Example 7).

(7) E: you can do anything just try <unclear=if you pres=> as he says do this be a space a space astronaut <.> ah for me there are limits

S: yeah I know I know and I certainly think that limits are important especially with teenagers for example as you as you said

It has been argued previously that the use of EMs by L2 speakers is somewhat limited (e.g. Aijmer 2004; Baumgarten and House 2010; Fordyce 2014). For example, Aijmer (2004: 188) reported that ‘learners use vague and uncertain markers to express uncertainty or hesitation and not for face-saving or to signal politeness’. However, while this may be true of L2 speakers with lower proficiency, this study showed that at more advanced levels of proficiency, the L2 speakers use EMs for multiple purposes and are able to exploit their multifunctionality in several respects. As illustrated by Examples (5) and (6), the L2 speakers in these interactions made use of the dual function of ‘probably’ and ‘maybe’ to communicate their less-than-full certainty about a specific matter as well as to downplay a disagreement with the examiner. The relatively high degree of tentativeness expressed by L2 speakers in the INT appears to reflect both their awareness of the social requirements of the situation as well as their ability to handle them linguistically.

Individual EMs

In line with previous research on epistemic forms used by non-native speakers (Salsbury and Bardovi-Harlig 2000; Fung and Carter 2007; Fordyce 2009, 2014), the results showed that (i) L2 speakers in this study favoured lexical verbs and adverbial expressions over another form of EMs (in this case adjectival expressions) and (ii) a small set of EMs prevailed in L2 production, accounting for almost two-thirds of all EMs identified in this study. In earlier studies, this pattern of use was interpreted as problematic and attributed either to insufficient genre awareness or to limited L2 proficiency (e.g. Fung and Carter 2007; Gilquin and Paquot 2008; Fordyce 2014). The reasons for the prevalence of these particular expressions may be, however, more complex.

While the speakers in the present study tended to favour a restricted range of EMs, a similar strategy has been repeatedly found in research on stance expression by native-speakers as well, with ‘I think’ being by far the most frequent epistemic item (e.g. Biber et al. 1999; Thompson 2002; Aijmer 2004). Kärkkäinen (2007: 183) reported that American English speakers in her study relied on ‘a limited set of high-frequency markers’, among them ‘I think’, ‘I know’, and ‘I guess’. Precht (2003: 240) concluded, following her findings on stance expressed by British and American English speakers, that ‘the resources of language enable a virtually unlimited number of ways in which we could express ourselves; however, my results suggest that we are culturally ‘programmed’ to use a very limited, very specific subset of these options’. This suggests that using a limited range of EMs may not be typical only of non-native speakers.

The findings also showed that, at least with respect to epistemic stance-taking, the advanced L2-speakers are not ‘mono-stylistic’ and their production shows sensitivity to different interactional requirements. Thus, while the study confirmed the prevalence of markers such as ‘I think’ and ‘I don’t know’ in the L2 speech, it also found that the speakers used considerably fewer of these markers typical of speech in the most formal and monologic task, the PRES. For example, whereas the three dialogic tasks resulted in very high occurrence of ‘I @ think’ per 100,000 words (CONV: 1090.5; DISC: 892.3; INT: 1061.2), the number was dramatically lower in the PRES (295.1).

Individual speaker style

This study showed that there was a systematic variation in the epistemic stance expressed by individual speakers. The variation was found with respect to the frequency of EMs, a specific word class or a particular expression indexing stance. These linguistic preferences could be interpreted as a result of limited linguistic repertoire; however, previous studies found that speakers’ choices reflected a more complex and dynamic decision-making process that combined L2 speaker’s mastery of the language, awareness of the social expectations, and speakers’ personal goals (e.g. Siegal 1995, 1996; Dewaele 2008b). Limiting our interpretation of L2 speakers’ choices to that of insufficient L2 proficiency, we would run the risk of categorizing L2 users as ‘failed native speakers’ (Cook 1999) and neglecting the rich evidence about the development of L2 pragmatic competence.

Because this study focused on relatively advanced L2 speakers, the internal consistency in the speech of individual speakers seems likely to reflect their preferred stance-taking patterns rather than a deficiency due to limited proficiency. The observed patterns thus appear to be indicative of L2 users’ preferences to express themselves with higher or lower degree of certainty, or to use a particular type of epistemic expression (cf. Brezina 2013 for individual preferences in native speech). This explanation appears probable, especially since stance-taking choices are part of the domain of language with close ties to individual and social identity (Bucholtz and Hall 2005). These epistemic preferences can be typical of groups of speakers defined by shared characteristics such as gender, L1, or culture, or by shared psychological traits such as extroversion (Dewaele 2004), but can also index a more complex, individual combination of factors. This was illustrated, for example, in Liao (2009) and Siegal (1996) where speakers’ pragmatic choices were strongly affected by their wish to uphold their professional status.

With lower-level learners, a preference for particular epistemic forms (‘I think’) has been attributed to limited proficiency and explained by the fact that these are easy to acquire due to their mono-functionality and explicit epistemic meaning (Salsbury and Bardovi-Harlig 2000; Fordyce 2014). However, this does not explain why words signalling strong certainty such as ‘I believe’ or ‘definitely’, which also fulfil these criteria and, as this study showed, are evidently in some learners’ lexicons, do not enjoy the same prominence (both with lower-level and advanced speakers). More research therefore needs to be done to investigate why speakers systematically prefer one stance marker over another. A further closer analysis, similar to the research conducted by Siegal (1996) and Liao (2009), would be needed to explore the motivation behind speakers’ epistemic choices that would enable us to more firmly connect the micro-level linguistic structures (e.g. stance markers) with the higher-order categories such as social identity.

Implications for teaching and testing of L2 pragmatics

This study gives rise to several important implications for pedagogical practice and research in the area of L2 pragmatics. First, the differences between the stance-taking strategies of the L2 users in this study challenge the view that L2 speakers of a certain proficiency level are a largely monolithic group in terms of their linguistic abilities, preferences, and needs. For example, our study identified speakers who used EMs with very high frequency as well as speakers who used them rarely regardless of the interaction type. As a result of this considerable variation, recommended pedagogical methods may not be relevant, effective, or suitable for all language users in a particular group. It is thus crucial to differentiate between speakers with distinct epistemic profiles and offer appropriate pedagogical support.

When offering the support, one of the challenges lies in establishing what constitutes a successful pragmatic performance. It is important to realize that while L1-based criteria may be useful for measuring grammatical accuracy in L2, a similar prescriptive approach may be of limited usefulness and validity in assessing pragmatic ability (Zhang and Sabet 2016). Rather than searching for ‘norms’ based on L1 performance, we should therefore focus on whether speakers react in an appropriate manner and whether they succeed in fulfilling their communicative goals. This implies shifting attention from (in)correctness of stance-making choices to ‘appropriateness’ as a more meaningful measure of effective communication (e.g. Dewaele 2008b; Pallotti 2009).

Secondly, given the complexity of social use of language, a simple pedagogical focus on range and frequency of EMs may not provide sufficient instruction in the dimension of language responsible for managing social relationships. Thus, while not disputing the benefits of programmes and methods aimed at assisting L2 speakers’ pragmatic ability through increasing their range of epistemic forms (e.g. Fordyce 2014), it is equally important to help these speakers with developing linguistic skills that would allow them to display their personality in a way they are comfortable with. It is possible that a limited mastery of pragmatic skill of this kind is to some extent related to bilingual speakers’ claims of feeling ‘fake’, ‘less logical’, or ‘less serious’ when expressing themselves through their second language (e.g. Dewaele and Nakano 2013). So, in addition to being given space to practise different epistemic forms, learners should also be offered guidance on presenting themselves through their L2 in a way that respects their personal and cultural preferences within the framework of social expectations of the target language and culture.

Finally, the findings of this study can provide valuable information for the development of tests that seek to assess speakers’ command of social aspects of language. One of the core characteristics of communicative competence is adaptability to the different types of discourse (Roever 2011; Harding 2014). The results showed that L2 speakers were able to recognize different interactional settings and modify their choice of EMs accordingly. This finding can be used for development of descriptors for tests of pragmatic ability that go beyond assessing isolated speech acts.

Limitations and future studies

Although there are other ways of expressing epistemic stance, this study focused on three major lexico-grammatical groups of EMs (verbal, adverbial, and adjectival markers). Rather than providing a broad and exhaustive description of different EMs, this study aimed at identifying patterns of stance-taking in different interaction types. The selected lexico-grammatical categories were deemed sufficient for this purpose. Additional research could be conducted on other major epistemic forms (e.g. modal verbs) to complement and extend the findings of the present study.

This study approached the investigation of epistemic stance from a quantitative perspective, which, while providing a good overall picture of EM use, did not allow for more fine-grained explorations of stance-taking strategies of individual learners or the way individual markers are used for different functions (e.g. Baumgarten and House 2010; Mortensen 2012). Further research, combining qualitative and quantitative methods, is needed to explore the complex nature of stance-taking in L2. The study demonstrated the advantage of taking individual differences into consideration when studying epistemicity in L2 speech. Additional individual factors that could be examined as predictors of pragmatic strategies in L2 are L2 proficiency and L1 background of the speakers.

Finally, this study investigated stance-taking in semi-formal discourse elicited in an institutional setting. This contextual frame is characterized by conventionalized roles of the speakers (e.g. the examiner and the test-taker) that to some extent determine the type of interaction. To gain a fuller picture of stance-taking in the speech of advanced users of L2, it is desirable to complement these findings with research on speech in informal, spontaneous settings.

CONCLUSION

The study found that the epistemic stance expressed by advanced L2 speakers was associated with complex interactions between the task type and speaker’s personal style. The findings show that advanced L2 speakers demonstrated sensitivity to social dimension of the task as well as to individual task requirements (e.g. limiting the frequency of oral features in the PRES) and adjusted their speaking styles accordingly. These findings are based on a relatively large speech sample (almost 300,000 words) and include speakers from a range of different L1 and cultural backgrounds. Having focused on four different speaking tasks, this research advances the agenda beyond studying writing in L2 corpora, and beyond comparing speaking and writing into more fully exploring the variation in spoken interaction types. These findings thus help us to reach a deeper understanding of stance-taking and subjectivity in second language and complement findings from L2 writing research. Finally, this study also demonstrated that with new advanced resources such as large corpora of different types of L2 spoken language, it is possible to move beyond a more basic analysis of L2 speech based on group means to a more fine-grained and comprehensive understanding of spoken pragmatic ability and communicative competence.

SUPPLEMENTARY DATA

Supplementary material is available at Applied Linguistics online.

Acknowledgements

The authors thank Alison Mackey, Anna Mauranen, and three anonymous reviewers for their valuable comments on different drafts of the article. They also thank their transcribers Ruth Avon and Alana Jackson. The research presented in this chapter was supported by the ESRC Centre for Corpus Approaches to Social Science, ESRC grant reference ES/K002155/1, and Trinity College London. All data used in this article are available in the Results section and through online Supplementary data. The whole Trinity Lancaster Corpus will be made publicly available in 2017 when the data is anonymised.

Conflict of interest statement. None declared.

NOTES

Footnotes

1 The data from the listening task were not used in the study, as the task requires only short answers from the L2 speakers.

2 For example, ‘definitely’ occurred 41 times in the L2 speakers’ production across all four tasks, with 39 of these instances (more than 95 per cent) being classified as epistemic.

3 Two assumptions were checked before performing the ANOVA test: normality of the distribution and sphericity. Because of violations of sphericity, Greenhouse-Geisser correction was used. Although the normality assumption was violated, the ANOVA test was still deemed suitable because of its robustness towards the violation of this assumption (Schmider et al. 2010).