Stance Adverbials in Spoken English Interactions: Insights from Corpora of L1 and L2 Elicited Conversations

This paper adopts corpus-based methods to investigate how adverbial stance markers are used in the Swedish and in the Italian EFL LINDSEI sub-corpora and in the L1 English LOCNEC comparable corpus (Gilquin et al. 2010). Inspired by research on corpus pragmatics, conversation analysis, learner corpus research and EFL pedagogy, the quantitative and qualitative investigations focus on stance adverbials in turn-initial and final position. The results reveal that the Italian learners use such markers significantly less frequently than the native speakers and the Swedish learners in both positions. However, differently from the Swedish and the Italian groups, the L1 group employs the adverbials mainly in final position and makes use of a wider repertoire of them to make more meanings. The study also explores how stance adverbials can contribute to discourse management in combination with other linguistic elements, and discusses implications of these and other findings for EFL pedagogy.


Introduction
The expression of stance through adverbials in spoken interaction has been studied extensively in various domains of linguistic research, including corpus data (see Section 4), native use should more realistically represent an aspect of supposedly successful interactive behaviour that illuminates EFL learner data and helps interpret them. This holds particularly true for stance adverbials, the meaning and function of which in conversation can be difficult to describe and interpret. The research questions guiding the study are the following: 1) Are there quantitative differences concerning the use of stance adverbials in turn-initial and turn-final position across the three sub-corpora? 2) Which specific stance adverbials are used and how does their use differ across the corpora? 3) What types of meanings do they make? 4) Do the three populations employ discourse markers, filled pauses and/or conjunctions before stance adverbials in turn-initial position or after those in turn-final position as cohesive and conversation-management tools? Are there differences among the three groups with regard to the use of these items? 5) Can any evidence be found in the data that stance adverbials in turninitial or turn-final position also contribute to discourse management and cohesion? After reviewing relevant studies of English stance adverbials (Section 2), turntaking and discourse management (Section 3), the paper will provide detailed information about the datasets explored (Section 4) and explain the methods adopted to carry out the study (Section 5). It will then present and discuss the main findings, with a view to applying them to EFL pedagogy (Section 6).

Stance Markers in Spoken English
Speaker's stance or attitude towards segments of discourse can be expressed in a variety of ways in English, including the use of core modal and semi-modal verbs (e.g. may, have to), lexical verbs (e.g. believe, suggest), adjectives (e.g. probable, likely), nouns (e.g. probability, likelihood), and stance adverbials (e.g. probably, definitely) (Carter and McCarthy 2006: 222-223;284-286;493;628ff). The primary function of stance adverbials (henceforth SA) is "commenting on the content or style of a clause or a particular part of it" (Biber et al. 1999: 853). They thus act as pragmatic markers and share some core similarities with other such markers (e.g. discourse markers and interjections), including the fact that they work outside of the structural limits of the clause (Carter and McCarthy 2006: 208).
Contrastive PragmaticS 4 (2023) 243-273 Biber et al. (1999: 854-857) groups SAs into three main groups according to semantic criteria: epistemic stance adverbials (e.g. evidently, generally), attitude adverbials (e.g. fortunately, hopefully), and style adverbials (e.g. honestly, seriously). The first group includes the sub-categories "doubt and certainty", "actuality and reality" and "source of knowledge", those of the second group can be restated with that-clauses and adjectives describing attitudes (e.g. it is fortunate that …), and those in the third one can be glossed as "I am being X [serious] when I say …". In conversation, epistemic adverbials of doubt/certainty (mainly probably, maybe, perhaps, of course) and epistemic adverbials of actuality (mainly really, actually, in fact) are the most common meaning sub-categories, followed by style and attitude adverbials (Biber et al. 1999: 859;869). Besides conveying these meanings, these adverbials perform other functions, including their contribution to cohesion and to the "interactive nature of conversation" (Biber et al. 1999: 874). Stance adverbials are ambiguous, as they can belong to other adverbial classes as well. Notably, really can also be used as a degree modifier, and in fact does not only signal actuality but also reinforces the point just made (Biber et al. 1999: 857-858). With regard to their position in the clause, stance adverbials tend to occur more frequently in medial position in all registers. However, conversation "has a higher percentage of stance adverbials in final position and a lower in initial position" (Biber et al. 1999: 872).
In a similar vein, Downing and Locke (2006: 73-74) break down SAs into four main categories: "epistemic", "evidential", "evaluative", and "style and domain". They gloss the first group with "Do you believe me? Of course I do", the second with "Apparently, the picture is a fake", the third with "Amazingly, he was a gold medal", and the fourth with "Quite frankly, it seems to me a lot of bullshit". Quirk et al. (1985: 612-631) refer to SAs as "disjuncts" and keep them distinct from "subjuncts" and "conjuncts". In the authors' words, "they are syntactically more detached and in some respect "superordinate", in that they seem to have a scope that extends over the sentence as a whole" (Quirk et al. 1985: 613). The authors categorise them according to the meaning they make as "style" disjuncts (e.g. truthfully, bluntly) and "content" disjuncts (e.g. certainly, wisely). For most disjuncts, the normal position in the clause is the initial one, yet they can appear at almost any place (Quirk et al. 1985: 627-628). By contrast, for content disjuncts expressing doubt (e.g. probably, possibly), the medial position is the favorite one, also in the case of direct and indirect questions (Quirk et al. 1985: 628). Halliday and Matthiessen (2004: 129-132) take a different approach to the description and classification of SAs, which they call "comment adjuncts". Such items are not closely tied to the grammar of mood, i.e. their scope is the Contrastive PragmaticS 4 (2023) 243-273 whole clause and they are realised as a boundary between tone groups. They are restricted to indicative clauses and express the speaker's attitude towards either the proposition as a whole (the propositional/ideational type) or a particular speech function (the interpersonal type). The former type only occurs in declarative clauses in various positions: in thematic position, medially, or in final position as afterthought (e.g. [Unfortunately,] the doctor [, unfortunately] hasn't [unfortunately] left an address [unfortunately]). The latter type, by contrast, can occur with either declarative or interrogative clauses and favours either the initial or the final position. In declarative clauses, comment adjuncts express the speaker's angle, while in interrogative clauses they elicit the listener's angle (e.g. [Frankly,] were you surprised at the outcome [, frankly]?).
Waters (2008) reports on a sociolinguistic investigation into the use that native speakers from the Toronto area in Canada make of actually, really, in fact, as a matter of fact, and indeed. The 52 male and 63 female participants are aged between 18 and 92 and have at least high school education. The data suggest that indeed and as a matter of fact are disappearing from Toronto English, as the few instances of these SAs are produced only by people aged 40-92. By contrast, actually and really appear to be on the increase, as they are frequently used by people aged 18-39. While actually occurs more frequently in clause-initial position, really tends to be used in clause-medial position. In clause-initial position, actually indicates disagreement, transition in discourse (elaboration or clarification), slight change of topic, and unexpectedness, whereas in medial position it signals realness, disagreement and emphasis.
Corpus-based investigations have shed further light on the meanings and functions of SAs. Aijmer (2002), for instance, finds that in the London-Lund Corpus of Spoken English in fact is more assertive and formal than actually and that it typically occurs in discussions and in public and prepared speech. In fact reinforces the opposition to the preceding proposition, while actually seems to suggest that some "slight reinterpretation or adjustment of perspective needs to take place which may be seen as a shift in common ground" (Aijmer 2002: 255-256). She also notices that in discussion actually occurs mainly in "initial or mid-position with the specialised function of emphasizing the individual position of the speaker or author in opposition to other possible positions" (Aijmer 2002: 261). Finally, she adds that in final position, the only function that the marker seems to have is "to soften what has been said by treating it as an opinion or evaluation" (Aijmer 2002: 275). Diani (2008) focuses on the use of really in corpora representing spoken and written academic genres. She points out a substantial ambiguity in its functions in relation to its syntactic role/scope. In initial position really mainly Contrastive PragmaticS 4 (2023) 243-273 acts as a sentence adverb, while in mid-position it may be interpreted both as a modifier and as an adverbial of stance. Furthermore, "its scope may even extend beyond the sentence, participating in a macro-textual pattern. This is the case with really having a strong anaphoric use, linking the writer's assertion to the preceding sentence" (Diani 2008: 316). Gómez-Moreno (2015) comes to similar conclusions after his analysis of adverbials taking up sentence final position (e.g. absolutely, exactly, obviously) in the Corpus of Contemporary American English. In his data, end-position is not only the locus in which afterthoughts can be expressed, but it often helps speakers to reinforce their arguments and to explicitly reassure their attitude concerning ongoing features of discourse. Furthermore, it serves to attain cohesion and enhance coherence. In other words, his findings suggest that in final position SAs can contribute to the unfolding discourse.
The textual and cohesive role of SAs shows their relevance to discourse management and turn-taking in L1 and L2 conversation.

Stance, Turn-Taking and Discourse Management in L1 and L2 Conversation
Research into EFL pedagogy has stressed the importance of stance adverbials for successful spoken interaction, and has explored ways to best teach L2 learners to use them effectively. As noticed by Pérez-Paredes and Díez-Bedmar (2019: 253), some CEFR can-do statements include "stance-related aspects that learners are supposed to display when speaking". Stance manifests itself differently across levels, yet it is more visible from B1 level onwards. According to the self-assessment grid for speaking at level B1, learners at this level should display the ability to "give reasons and explanations for opinions and plans" (Council of Europe 2001: 26). With regard to English, the operationalisation of this and other CEFR statements can involve the appropriate and increasingly frequent use of adverbs expressing various types of stance (Pérez-Paredes and Díez-Bedmar 2019: 273). Jones (2016) devotes a study to the acquisition of a selection of English SAs by EFL learners: admittedly, basically, fortunately, hopefully, in fact, no doubt, obviously, to be honest, and surprisingly. The author starts from the premise that "the function of marking stance would also seem to be an important one for most learners, as it has been shown to be highly frequent in conversational discourse" (Jones 2016: 83). Furthermore, SAs are optional in discourse and their meaning is not always transparent, which makes teaching intervention advisable (Jones 2016: 88). He explores whether explicit receptive practice (i.e. listening to a dialogue, answering comprehension questions and filling in gaps with the missing SAs) and explicit productive practice (i.e. listening to a dialogue, noticing SAs, discussing their meanings and comparing them to the participants' first languages) can help EFL students improve on this aspect of language proficiency. By administering receptive and productive tests immediately after the teaching interventions, the author finds that both types of practice can be beneficial in the short term, with learners exposed to receptive practice even outperforming the others in free discussions. However, when it comes to the long-term effects measured by "delayed" tests, the students' performance is generally characterised by a regressive pattern. The author thus concludes that "learners may benefit from hearing the stance markers […], in order to develop a 'feel' for when and how to use them prior to production", yet the pragmatic and optional nature of SAs "may require a greater length of treatment to have a clearer effect" (Jones 2016: 92-94).
A number of recent studies based on learner corpora have explored how EFL learners use SAs in spoken interaction and the factors that might contribute to their (un-)successful performance with regard to them. Pérez-Paredes and Díez-Bedmar (2019), for example, explore the effect of proficiency level, task type and L1 background (European and Mexican Spanish, Italian, Hindi and Chinese) on learner use of the certainty adverbs actually, really and obviously in the Trinity Lancaster Corpus. Their findings reveal that as proficiency increases so does the average frequency of use of really and actually, while obviously shows much lower frequencies. Really is used as an emphatic device in clausal medial and phrasal position, while actually displays meanings such as factualness, elaboration and clarification mainly in initial and phrasal positions and to a lesser extent in final position. Actually and really are favoured by dialogic tasks, with actually used more frequently by the Hindi speakers and significantly more present in the Mexican learner data than in the European Spanish one.
Pérez-Paredes and Bueno-Alastuey (2019) study the patterns of use of these same three adverbs in the German, Spanish and Chinese LINDSEI components, in LOCNEC (see Section 3) and in another corpus of interviews with British undergraduate students. The authors control for the task variable, as they only consider the language produced during a picture description task, and find that obviously and actually are more frequently used by the native speakers. With regard to the clausal position of actually, the Chinese favour the clause-initial position, the native speakers and the German learners the medial position, and the Spanish ones the final one. As for obviously, the native speakers opt for the medial position more frequently, while the Germans are divided among various positions.
Conversation can be defined as a joint production co-constructed by the interlocutors. Rühlemann (2007: 49-51) stresses that co-construction is central to any conversation, in combination with discourse management, shared context, real-time processing and relational management. From a linguistic perspective, interactive discourse can be managed in various ways, including the following: the use of backchannels such as yeah, ok, really (e.g. Clancy et al. 1996) to indicate that the other speaker(s) can continue with their turn; the use of filled pauses (mm and uhm) to indicate that one's turn is not complete (Tottie 2015); the overlapping of turns; the use of conjunctions (e.g. and, so) as turn-initial or turn-final items; and the completion of syntactic units by different speakers across turns (Clancy and McCarthy 2015).
Researchers in the field of L1 conversation analysis (e.g. Sacks et al. 1974 andShegloff 1996) have suggested that analytical attention should be paid to the succession of turns the interlocutors are involved in rather than to the sentences, clauses and phrases they produce. Therefore, the analytical approach they put forward involves breaking down conversation into "Turn Constructional Units" (TCU), that is units delimited by transition relevant places in which "the transfer of speakership from one party to another becomes relevant" (Clancy and McCarthy 2015: 433). Clift (2001) makes the additional point that also discourse markers, particles and stance markers should be studied with regard to their position in TCUs. Specifically, the author shows how actually takes on various meanings and performs different functions in relation to its position in TCUs in spontaneous conversation. She then goes on to say that "reference to TCUs treats actually as an interactional object in its own right, and the shift in focus also allows for a widening of scope to encompass […] the construction of the turn itself" (Clift 2001: 289).
Studies of L2 spoken English have explored the contribution of discourse management to successful communication, along with that of linguistic competence, pragmatic competence and strategic competence. Jones et al. (2018: 14-15), for instance, define discourse competence as "the ability to recognise and link language across extended conversational turns […] the ability to link ideas together into a coherent narrative, allowing the listener 'space' to react and respond". The authors operationalise the somewhat vague CEFR can-do statements devoted to turn-taking, flexibility, thematic development and coherence/cohesion (Council of Europe 2001: 124-125) by exploring aspects of discourse management in the "UCLan Speaking Test Corpus" and in LINDSEI (see Section 3). Specifically, they investigate the use of a selection of frequently occurring anaphoric referencing items (e.g. we, this), discourse markers (e.g. well, you know), and special chunks such as "and er" (Jones et al. 2018: 109-132).
Various other learner corpus-based studies are devoted to these and other aspects of discourse management. For instance, Castello and Gesuato (2019) Contrastive PragmaticS 4 (2023) 243-273 report on an investigation into the use of lexical backchannels in short turns (e.g. I know, of course) in the Trinity Lancaster Corpus, while Götz (2019) studies the occurrence of filled pauses (e.g. erm, ehm) in the same corpus.
The present paper investigates the Swedish and the Italian components of LINDSEI, which are introduced in the following Section.

4
The LINDSEI and the LOCNEC Datasets The Database of Spoken English Interlanguage (LINDSEI) (Gilquin et al. 2010) is a collection of transcribed spoken data from learners of English with different L1 backgrounds, including Chinese, Dutch, French, German, Italian, Norwegian, Spanish and Swedish. Each L1 subset comprises data from fifty interviews between an EFL university undergraduate and an expert English (native-) speaker examiner. Each interview consists of three parts: set topic, free discussion and picture description. In the set-topic phase, the examiner asks the candidates to describe an experience, a film/play or a country that taught them something or impressed them for some reason. During the free discussion, the candidates answer a series of questions not only about the chosen topic but also about other subjects. Finally, in the picture description part the candidates are asked to look at four pictures, describe what they see and make up a story. The present paper compares the Swedish and the Italian components of LINDSEI to each other and to a corpus of interviews with British English native speaker students elicited using the same prompt: the Louvain Corpus of Native English Conversation (LOCNEC) (De Cock 2004). The LINDSEI and the LOCNEC interactions were transcribed following the same transcription conventions (Gilquin et al. 2010: 13-18;65), which ensures the comparability of the data. Gilquin et al. (2010: 47) provide specific information about the context in which the learners contributing data to the LINDSEI project had learnt English, including an overview of the opportunities they had had to learn English extramurally. They review the typology and availability of English-language media in each country at the time the data was collected (late nineties, early two-thousands), as well as the opportunities they had to spend a reasonably long time in an English speaking country before the interviews.
With regard to Sweden, they write that: Television shows and films in English are subtitled, not dubbed, unless they are aimed at young children. Students are exposed to English via music, the Internet and computer games. Newspapers and books (e.g. Contrastive PragmaticS 4 (2023) 243-273 paperback novels) in English are generally available, although this type of reading is less common among young people. A majority of the students also come in contact with English as lingua franca during short holiday trips.
[…] Students at the University of Gothenburg have had the opportunity to study in Brighton for a term or more, which has been quite popular. Stays in other parts of the UK, Australia or the USA are not uncommon, whereas students at lower levels in general do not stay in English-speaking countries for long periods of time (more than one month). Gilquin et al. 2010: 56 The description of the Italian situation is rather different and reads as follows: Newspapers, television and radio programmes are in Italian. All foreign TV programmes and films are dubbed, which means that students have no exposure to English through the national media. English newspapers, books or audio-visual materials are available in shops and through the Internet.
[…] University students are encouraged to take part in exchange programmes (such as the Erasmus programme), especially if they wish to become language specialists, but only a minority of them actually spend a long period of time in English-speaking countries. Gilquin et al. 2010: 52 The LINDSEI Swedish learners are thus likely to be more proficient than the Italian ones with regard to spoken interaction, due the broader range of opportunities the former had to get in contact with spoken English. As suggested by the findings of studies on extramural English learning (e.g. Sundqvist 2009, Pavesi andGhia 2020), informal contact with English can indeed impact positively on both learner vocabulary expansion and oral proficiency. Furthermore, the professional rating of a sample of interactions suggests that, while most of the Swedish learners qualify as advanced, the Italians turn out to be at intermediate levels of proficiency (Gilquin et al. 2010: 10).

Data and Methods
The method adopted in this study is Contrastive Interlanguage Analysis, i.e. an approach to the analysis of learner corpus data put forward by Sylviane Contrastive PragmaticS 4 (2023) 243-273 Granger (e.g. 2012Granger (e.g. , 2015 which consists in comparing "either two learner varieties (L2 vs. L2) or one learner variety and one native (or expert) variety (L2 vs. L1)" (Granger 2012: 18). The utterances making up the turns produced by the 50 LOCNEC native speakers' interviewees and those of the 50 Swedish learners and the 50 Italian learners were separated from their interviewers' contributions, and uploaded onto the online query system Sketch Engine (Kilgariff et al. 2014).1 The tags marking the beginning of the learners/interviewees' turns -the tags <B> and </B> respectively -as well as other tags added to the transcripts were excluded from the count. These comprise the tags <overlap /> and <laughs>, used to indicate the beginning of overlapping discourse and laughing, respectively.
Subsequently, the lists of word types and tokens related to the three subcorpora were generated. Table 1 shows the numbers of tokens and types for the LOCNEC native-speaker (NS) sub-corpus as well as those for the two LINDSEI components: the Swedish (SW) and the Italian (IT) ones. As can be seen, the number of word tokens and types in the NS component is remarkably higher, while the Italian data scores the lowest values. The combined total of running words is 249,810.
Sketch Engine tags corpora for parts of speech automatically once texts are uploaded onto the system. This feature made it possible to create lists of adverbs for the four components, which were subsequently inspected to identify the most frequently used adverbs in the sub-corpora. The lists of stance markers provided in the works reviewed in Section 1 were also consulted to find other potential adverbial candidates. The adverbials thus identified were mainly adverbs ending in *ly (e.g. actually, hopefully, definitely), as well as perhaps, maybe, of course, in fact and indeed.
Using the Corpus Query Language (CQL) search option available in Sketch Engine, concordance lines were collected for all the instances of adverbials used as SAs in turn-initial and turn-final position in the sub-corpora. Specifically, the CQL searches conducted aimed to retrieve the instances of adverbials used 1 https://auth.sketchengine.eu/#login (last visited on 27/08/2022). at the beginning of turns, optionally preceded by up to five words, and those of adverbials employed at the end of turns, optionally followed by up to two words. The concordance lines thus obtained were downloaded as Excel spreadsheets, manually inspected and sorted out. Concordance lines for adverbials that did not represent SAs in turn-initial or turn-final position were deleted. Furthermore, instances of SAs making up short turns on their own were not taken into account, as the consulted literature considers them lexical backchannels rather than stance markers (e.g. Castello and Gesuato 2019). Finally, only the concordance lines starting with an SA and those in which an SA is preceded or followed by discourse markers and/or conjunctions within a five-word span were included in the database. Extracts 1) to 3) exemplify SAs produced by NS interviewees: example (1) shows that the SA obviously is used at the very beginning of the turn, example (2) that hopefully is preceded by a conjunction, while example (3) that actually is followed by a discourse marker at the end of a turn: (1) <B> obviously she's not very pleased with it . […] (2) <B> and hopefully I'll come home for Christmas as well […] (3) […] they were [really nice] actually yeah </B> It must be noticed that the identification of turns in real conversations is not always straightforward. To say it with Tottie (2015: 389), sometimes "we are obviously up against a problem of speaker intention and actual outcome". In some cases, decisions had to be made with regard to what counts as a turn, especially when backchannels, filled pauses and/or overlapping speech occur (Tottie 2015: 389-393). Specifically, in the three sub-corpora under investigation one challenging issue is the presence of some interviewees' turns starting with an SA and preceded by interviewers' turns2 consisting of filled pauses (e.g. mhm) and/or backchannels (e.g. yes) and/or overlapping speech. Extract (4), taken from the transcript of a Swedish conversation, illustrates this phenomenon: It can be claimed that the interviewee's second turn might in fact be considered the continuation of the preceding one and not a completely new turn. However, for the present study also SAs used after interviewer's "phatic" turns were included in the count (e.g. of course in extract 4). The main reason for this choice is that also such turns can have the potential to influence the (re-)wording of the interviewee's upcoming discourse, including the use of SAs in strategic positions. Furthermore, in some cases it would have been difficult to distinguish between full and phatic turns reliably.
Most of the analyses carried out for this study compared the language produced by the NS interviewees to that of the EFL learners as whole groups (i.e. the NS, the SW and the IT sub-corpora). Since the sub-corpora are of different sizes, the data were normalised to the basis of 1,000, according to the total number of tokens in each sub-corpus.
In order to take inter-speaker variation into account and at the same time investigate whether the effect of SA use is not due to chance (e.g. Brezina and Meyerhoff 2014), the non-parametric inferential Kruskal-Wallis test (H values) was used. Specifically, this test made it possible to explore across-group variation focusing on the consistency in the use of SAs in the speech produced by each individual speaker. The test does not assume that the underlying distribution of the variable under investigation is normal (i.e. symmetrical with regard to the mean) and takes into account ranks in multiple groups (Brezina, 2018: 195-199). The test was performed with IBM SPSS (version 27),3 and involved the normalization of the data to the basis of 1,000. Bonferroni error correction was used to make adjustments for multiple testing.

6
Data Analysis

The Combined Use of SAs in Turn-Initial and Turn-Final Position
In total, there are 589 instances of SAs used in turn-initial and turn-final position in the three datasets: 287 SAs were produced by the NS interviewees, while 223 and 79 were uttered by the Swedish and Italian EFL learners, respectively. In order to explore the first research question, the combined total of the relative frequencies of SAs employed in initial and final position was calculated for each individual speaker and normalised per 1,000 word tokens. The SW learners scored the highest average relative frequency for SAs (M = 3.04, SD = 2.23), followed by the NS interviewees (M = 2.66, SD = 1.97), and the IT learners (M = 1.37, SD = 1.38). The highest relative frequencies of SAs in the four components are 7.72 (NS), 9.36 (SW) and 6.59 (IT), respectively. Fourteen Contrastive PragmaticS 4 (2023) 243-273 Italian, two Swedish and three NS participants did not use any SAs in turn-initial and turn-final position. Figure 1 is a boxplot showing the distribution of the relative frequencies of SAs per speaker in the three sub-corpora. As can be seen, there are only a few cases of outliers, i.e. speakers presenting an unusually high frequency of SAs, who were attested in the IT component only.
As described in Section 5, the Kruskal-Wallis test was performed with a view to exploring whether the differences between the frequencies of SAs in the three sub-corpora are statistically significant or not, that is whether the variable L1 background and context has an effect on the use of SAs in turninitial and turn-final position. The test showed a significant effect of L1 background and context on the performance of this aspect of spoken interaction (H(3) = 19.94; p = .000). Specifically, it provided evidence of a difference between the IT group and the NS group (p = .001), as well as between the IT group and the SW group (p = .000). In other words, the Italian group proved to employ turn-final and turn-initial SAs significantly less frequently than the native and the Swedish participants. Table 2 shows the raw frequencies of SAs used at turn-initial and turn-final position, as well as the values normalised to the total number of tokens in each sub-corpus per 1,000 tokens. It is apparent that the Swedish group scored the highest normalised frequencies for both positions. Furthermore, while the two Figure 1 Relative frequency per 1,000 tokens of SAs in combined turn-initial and turn-final position across the corpora groups of EFL learners used more SAs in turn-initial position, the NS interviewees favoured the turn-final position to express their stance through SAs. That is, the NS group tended to rely more frequently on expressions of stance couched in afterthoughts, while the EFL learners were rather keen to guide their interlocutors in the interpretation of their unfolding discourse using adverbials in initial position. It should also be noted that the NS preference is in line with the findings reported in Biber et al. (1999: 869-872). Extract (5), taken from the Swedish sub-corpus, illustrates an EFL interviewee's use of SAs in turn-initial position (i.e. the underlined instance of actually, definitely, really) during a discussion about her plans for the near future: . what about are you worried about the discipline a= . at all </A> <B> not really because the= they are . between sixteen and nineteen years old so I don't think it will be . a problem really . it seems to be a very nice school </B> Table 2 Raw and normalised frequencies of SAs in turn-initial and turn-final position across the corpora The extract also features the interjection yeah and the negative adverb not occurring before the first occurrence of definitely and of really, respectively. The NS interviewee in extract (6)  Multifunctional really takes on the stance meaning of "in truth" (e.g. Biber et al. 1999: 857-858;Diani 2008: 301-302) and reinforces the statements made before, while definitely expresses certainty about the speaker's preference. These extracts show that the interviewees use turn-initial and/or turn-final SAs as a way to express their stance or comments on the propositional content of their own discourse, often in response to the interviewers' questions or remarks.

SA position NS (‰) SW (‰) IT (‰)
The analyses reported on so far have explored Research Question 1. They have revealed differences and similarities between the groups, regarding both the combined use of SAs and their specific use in turn-initial and turn-final Contrastive PragmaticS 4 (2023) 243-273 position. The following sections address the other research questions and investigate the specific SAs employed, their frequencies, the meanings they make and their contribution to discourse management and cohesion.

Types of SAs Used in Turn-Initial and Turn-Final Position and Their Meanings in Context
This Section explores the 319 instances of SAs detected in turn-initial position and the 270 ones in turn-final position. Tables 3 in Section 6.3.1 and Table 4 in Section 6.3.2 below break them down into SA types and tokens employed in turn-initial and turn-final position, respectively.
Mainly following Downing and Locke (2006) and Biber et al. (1999) (see Section 2 above), each SA type was associated with one of the following semantic categories: Epistemic, Evaluative, Evidential, Style (attitudinal), or Multifunctional.
In order to gain a deeper insight into the epistemic attitude of the speakers, the Epistemic category was further divided into "tentative" and "certain". The specific associations are: -Epistemic (tentative): maybe, probably, perhaps, possibly and presumably; -Epistemic (certain): of course, definitely and certainly. The remaining SAs types were matched with the other categories as follows: -Evaluative: basically, unfortunately, necessarily, luckily, surely, unluckily and correctly, absolutely, fortunately and strangely; -Evidential: apparently, obviously, generally and normally; -Attitudinal: hopefully, funnily, seriously; -Multifunctional: really, actually, in fact and indeed. Although most, if not all, SAs are multifunctional and can take on (slightly) different meanings according to the context of use, the Multifunctional category comprises items that the literature review has shown to be specifically challenging to categorise and whose meaning can vary largely from situation to situation. Table 3 displays the types of turn-initial SAs in the order of their Total (1) frequencies of use, and presents raw and normalised values for each one of them. A look at the upper part of the table reveals that maybe is by far the most frequently used SAs in turn-initial position across the board. Proportionally, the two learner populations use it more extensively than the NS interviewees to express their tentative epistemic evaluations. To this end, the Italians often use probably, the Swedes tend to rely on perhaps, while the native speaker participants are the only ones who also employ possibly. The Swedish learners Contrastive PragmaticS 4 (2023) 243-273 produced the highest number of instances of of course and proportionally also of really. They almost equal the natives with regard to the use of definitely, two items signalling epistemic certainty. Furthermore, they outperform the other two groups with regard to the use of actually, with the native group scoring the lowest normalised value for it. Finally, the Italians use in fact more frequently than the other groups.

Turn-initial SAs NS (‰) SW (‰) IT (‰) Total (1) (‰)
Excerpt 7 illustrates how a native speaker interviewee skilfully deploys attitudinal funnily and multifunctional actually at the beginning of two subsequent turns when talking about his/her experience as a student of English literature: (7) <B> well fun . funnily enough it's actually <X> the[i:] English literature course that I did at A-level <\B> <A> mhm <\A> <B> actually it helped me because <overlap/> I studied . I studied Chaucer <\B> (NS004) S/he also uses actually in clause-and turn-medial position. Extract 8, taken from a discussion about the film "Shakespeare in Love", exemplifies the use of tentative epistemic maybe and probably in the Italian sub-corpus: (8) <A> and what do you think do you think that maybe he could have been a woman </A> <B> probably he was a man (er) maybe it could be a woman (eh) in the sense of (er) the feeling he: he had (eh) this (eh) as I said before (eh) it is full of passion full of love and I think th= that these are strong feelings . more linked to a woman probably than a man .. in this sense </B> (IT028) The interviewer's use of maybe in clause-medial position might have prompted the learner's turn-initial expression of stance with probably as well as the repetition of the two SAs in the middle of her long turn. As seen in Section 3, interlocutors continuously co-construct and negotiate discourse across extended conversational turns. The linguistic items that enable them to achieve a coherent "narrative" include discourse markers and conjunctions or combinations thereof. Filled pauses (e.g. erm, er), on the other hand, are variously used to reduce planning pressure, hold or yield the floor or emphasise the following word (e.g. Tottie 2015; Götz 2019).
The specific LINDSEI and LOCNEC data under investigation indicate that the learners and the native speakers often use discourse markers, filled pauses and conjunctions before SAs at the beginning of their turns. Figure 3 provides the percentages of such items used turn-initially before SAs for each semantic type and for the lack of them ("No item"). As can be seen, the NS group relies extensively on conjunctions (22.56%), while the IT group mainly opts for discourse markers (41.94%). Furthermore, the SW participants tend not to add any of them before SAs, and refrain from doing so more frequently than the other groups (50.81%).
As illustrated by extracts 9) and 10), the conjunctions the NS favour are and and so, while the discourse markers the Italians mainly rely on are yes/yeah and well. The first example shows that the native interviewee links his/her upcoming turn to his/her previous one by means of so, and then makes a comment through really before carrying on telling about his/her recent past. The second illustrates how an Italian EFL learner answers her interviewer's specific question affirmatively, and then uses maybe before starting to talk about her relationship with her father. As seen above, only rarely do the Swedish learners use cohesive items and filled pauses before their SAs, as illustrated by the following extract: (11) <A> you lived with your parents in a house <overlap/> or is it </A> <B><overlap/> no in a flat </B> <A> it's a flat </A> <B> yeah </B> <A> if you could choose would you choose to live in a house rather than a flat </A> <B> (mm:) no not really </B> <A> you don't think it's .. (erm) it's </A> <B> maybe when I get older </B> <A><overlap/> <starts laughing> yes <stops laughing> </A> <B> but at the moment (em) .. I mean no </B> <A> (uhu) . <overlap/> a flat in </A> <B><overlap/> a flat would do fine for me </B> (SW004) The conversation revolves around the learner's preference for currently living in a flat rather than in a house. In the fourth <B> turn, however, the interviewee opens up to the possibility of living in a house in the future. The pivotal role played by discourse markers and conjunctions in achieving cohesion in discourse is indisputable. However, the observations made in this section back up Diani's (2008) and Gómez-Moreno's (2015) suggestion that also SAs can contribute to cohesion (see Section 2). Specifically, it can be claimed that the repetition of the same SAs and/or the subsequent use of SAs belonging to the same semantic typologies across turns and speakers has the potential to contribute to discourse cohesion, as some of the examples given in this section illustrate. Table 4 shows that overall the most frequent SAs used at turn-final position are the multifunctional items really and actually. The former marker is mainly used by the NS interviewees and the latter by the Swedish group. What is interesting to notice it that the NS participants employ both actually and really more frequently in turn-final position than in initial position (.228‰ vs. .042‰ and .566‰ vs. .177‰, respectively). Furthermore, it is the use of these two items that largely contributes to the native speakers' high score on turn-final SAs, as pointed out in Section 6.2 above. Turn finally, the native speakers often also employ definitely, obviously, probably and basically, while the Swedish participants are more prone to use maybe, perhaps and of course. As already pointed out, the Italians take little advantage of the end-of-turn position, and the few times they do so they go for really, maybe, of course and probably. Figure 4 captures the types of meanings made at the end of turns through SAs. As can be seen, the values on the multifunctional category scored by both the NS and the SW groups (i.e. .82‰ vs. .81‰) are mainly due to the many instances of multifunctional actually and really they produce. The frequent use of these items clearly tells these two sub-corpora apart from the Italian sub-corpus. The Swedish participants frequently express their epistemic tentativeness through maybe, perhaps and probably, while the native ones tend to make all the identified meanings, including attitudinal stance with hopefully.

SAs in Turn-Final Position
Extract (12) exemplifies the recurrent use of actually in the NS data at the end of a turn. In the example, actually also appears internally at the end of a previous clause, which confirms that SAs are often repeated in various turn positions. Discourse markers and conjunctions are also used after SAs at the end of some turns. Although the instances retrieved from the data are rare, their presence nonetheless hints to their potential contribution to cohesion. Excerpt 13), for instance, shows how an NS interviewee's turn ends with obviously followed by the discourse marker you know. This marker is likely used by the speaker to gain time before completing his/her utterance, but it could also signal to the interviewer that s/he can take over if s/he wishes to do so. The study first explored quantitative differences across the three sub-corpora (Research Question 1). The inferential statistical analysis performed revealed that the Swedish learners' use of SAs in turn-initial and turn-final position does not differ significantly from that of the native speakers, while the Italian performance does. However, a more focused quantitative investigation showed that both learner groups opted more frequently for the turn-initial position, while the native speakers for the turn-final position. This confirms the NS behaviour detected by previous corpus-based studies of L1 English (e.g. Biber et al. 1999), and to a certain extent tells apart the NS from the SW and IT groups. Furthermore, the NS students also employed a larger repertoire of SAs than the Swedish and the Italian group did.
The study then tackled Research Question 2 and focused on the specific SAs employed and how their use varies across the corpora. The most frequent SAs in both positions are actually, really and maybe across the board. However, detailed quantitative investigations into their specific use pointed to the fact that the three populations exploit the initial and final positions in quite different ways. Firstly, while the Swedish learners frequently used actually in turn-initial position, the native participants employed it much more frequently in turn-final position. Secondly, both the NS and the SW groups showed a tendency to end their turns with really rather than to start them with this adverbial. The Italians hardly used these two SAs, while they often used maybe in both positions, thus confirming the results of previous studies pointing to the overuse of this item by Italian EFL learners (e.g. Castello and Gesuato 2019). They also employed in fact more frequently than the other two groups, which might be the effect of transfer from L1 Italian false friend infatti (e.g. Bruti 1999, Philip 2000.
Research Question 3 guided the investigation of the meanings made through SAs in turn-initial and turn-final position. In this respect, both learner populations mainly expressed epistemic tentativeness in both positions, while the native speakers gave voice to the range of identified meanings more equally. On the one hand, this might be due to the EFL learners' cautious interactive behaviour and, on the other hand, to the native interviewees' higher degree of self-confidence and related calmness during the interviews. Another reason for this difference might be the longer time that the EFL learners devoted to the picture task, which involved making tentative suppositions about what was represented in the pictures.
The exploration of the contribution that SAs and other cohesive items preceding them can make towards discourse management (Research Questions 4 and 5) revealed that the English native speakers mainly use conjunctions in turn-initial position before SAs, while the Italians mainly go for discourse markers. Furthermore, the qualitative analyses of some extracts showed that the Contrastive PragmaticS 4 (2023) 243-273 strategic use of SAs can guide the interlocutor's interpretation of subsequent discourse and that the repetition of some SAs in various positions, including across turns and speakers, can itself represent an element of cohesion.
These findings have to be taken with caution as some differences might be due to the specific features of the datasets. Although the data were collected and transcribed following the same guidelines, the different teams that worked on their compilation might have made slightly different choices regarding their transcription. Specifically, the transcribers might have been more or less accurate at indicating some turns, especially phatic ones. Furthermore, the LOCNEC interviews are generally longer than the LINDSEI ones, which might impact on the amount and range of SAs used. Finally, as said above, the parts of the interviews devoted to the picture description task tend to be shorter in LOCNEC than in the LINDSEI components.
At any rate, the results suggest that the appropriate use of SAs in turn-final as well as in turn-initial position can be regarded as a sign of high proficiency in spoken interaction. Specifically, what appears to be especially revealing is the strategic deployment of multifunctional SAs (notably really and actually), as well as of other adverbials making various types of meanings (e.g., definitely, possibly). Finally, the addition of cohesive items before turn-initial SAs can also represent a feature of advanced conversational behaviour.
As seen in Section 3, mastering the use of SAs is possibly less of a priority in communication, as these are optional discourse elements that carry pragmatic and textual meanings rather than ideational ones. However, their usage is arguably less easily controllable and might thus require a greater amount of practice. While both receptive and productive practice have proved to bring about short-term effects with regard to the use of SAs, achieving long-term pedagogical effects represents a much more challenging objective (Jones 2016). An alternative approach that could be explored to hopefully achieve long-term effects is "data-driven learning", which consists in engaging learners with authentic corpus-derived data directly or indirectly (e.g. Boulton and Vyatkina 2021). EFL learners could be guided, for example, in the exploration and comparison of lists of SAs and of relevant extracts from the transcripts. They could also be invited to study the datasets and spot quantitative and qualitative differences between them. Finally, they could be asked to choose the best SAs for given stretches of interactions, place them strategically in discourse and then try to explain their specific functions in relation to their position. EFL learners would also benefit from listening carefully and repeatedly to the recordings of some of the interactions, as these represent discourse which they themselves might need to co-produce.
The use of SAs in elicited conversations could be explored by further research. More fine-grained studies could additionally be based on corpora of transcribed Contrastive PragmaticS 4 (2023) 243-273 interactions in the learners' first languages, with the aim of exploring the impact that the learners' L1 can have on their actual choice of SAs and their placement in discourse. These investigations could focus on the correspondences between English SAs and their equivalents in the L1s (e.g. Simon-Vanderbergen andAjimer 2007, Fedriani andSansò 2017), and on the impact of cognates and false friends, such as Italian infatti and English in fact, on EFL learner discourse (e.g. Bruti 1999, Lauwes et al. 2012, Buysse 2020. Finally, it would be worth compiling a corpus of more recent interactions, with a view to investigating the current state of extramural English language learning in given countries (e.g. in Sweden and Italy), with special attention to the new media affordances.