‘Well it’s not very ideal…’ The pragmatic marker 'well' in learner English

‘Well it’s not very ideal…’ The pragmatic marker well in learner English Lieven Buysse KU Leuven Published in: Intercultural Pragmatics 12(1), pp. 59-89, 2015. Doi: http://dx.doi.org/10.1515/ip-2015-0003 Abstract The pragmatic marker well has received a lot of attention in studies on native speaker discourse and has served as an interesting testing ground for theories accounting for the multifunctionality of pragmatic markers. In the rapidly expanding body of research on pragmatic markers in learner English well has also claimed a prominent position, but so far no comparison has been made of how learners of varying mother tongue backgrounds use well. This article offers a Contrastive Interlanguage Analysis (cf. Granger 1996) in scrutinising well as a pragmatic marker in the Dutch, French, German, Spanish and Chinese components of the Louvain International Database of Spoken English Interlanguage (LINDSEI) and comparing these results with Aijmer’s (2011) findings for the Swedish component of LINDSEI and a comparable native speaker corpus. Well is shown to be more prevalent overall in each of the learner corpora than in the native corpus, except for the Chinese (in which well displays a marginal incidence). This overall discrepancy between the learners and native speakers only holds for the speech management functions of well; its attitudinal functions are significantly less common in the learners’ discourse than in the native speakers’. The observed differences are attributed to a complex interplay of factors, involving a.o. the learners’ limited inventory of pragmatic markers, their extensive exposure to well, L1 interference, and the speech context. Keywords pragmatic markers, well, learner English, learner corpora, Contrastive Interlanguage Analysis 1. Introduction Pragmatic markers, such as you know, I mean, so, well, like and now, have been studied at great length since the 1980s. They can be defined as linguistic elements “which do not contribute to the propositional content of the utterance which they modify [and that] are frequent in conversation, where they express the speaker’s attitudes to the addressee, negotiate background assumptions, express emotions and contribute to coherence” (Aijmer and Simon-Vandenbergen 2003: 1123). Of these well is probably the one that has received most scholarly attention (see e.g. Schourup 1985, 2001; Schiffrin 1987; Watts 1989; Jucker 1993; Smith and Jucker 2000, 2002; de Klerk 2005; García Postprint version - [1] Vizcaíno and Martínez-Cabeza 2005; Müller 2005; Cuenca 2008; Lam 2010b). Apart from its high frequency in spoken English (O’Keeffe et al. 2007: 33-35), this is undoubtedly due to its versatility and elusiveness, confounding its circumscription and leaving a lack of agreement among researchers on the meaning and use of well (as deplored by e.g. Aijmer and Simon-Vandenbergen 2003: 1124). All studies of well have pictured it as a pragmatic marker with a highly complex functional scope, which begs the question how learners of English cope with such a “recalcitrant marker (…) which varies its function depending on the situation” (Aijmer 2013: 71). Over the past decade the interest in how foreign language learners make use of pragmatic markers has surged (e.g. Romero Trillo 2002; Fuller 2003; Müller 2005; Fung and Carter 2007; Mukherjee 2009; Polat 2011; Buysse 2012). Unfortunately, most of these scrutinise corpora drawn from different speech contexts and learners from varying mother tongue backgrounds and language learning tracks, making it difficult to compare the findings of these investigations. The present study aims to address this problem by taking as its object of study five components of the Louvain International Database of Spoken English Interlanguage (LINDSEI), a collection of comparable corpora of learner speech (Gilquin et al. 2010). Each of the selected components minimally groups 50 English interviews with native speakers of a single language, viz. Dutch, French, German, Spanish and Chinese. All instances in which well occurs as a pragmatic marker will be categorised according to Aijmer’s (2011) functional classification of the marker, drawn up for an analysis of the Swedish component of LINDSEI and the latter’s native speaker reference corpus. The result is what Granger (1996) has dubbed a “Contrastive Interlanguage Analysis” (1996: 43-44), viz. a comparison of corpora of different non-native varieties of a language and a comparison of these varieties with a native variety of the same language. 2. Functional scope of well as evidenced in previous research Well has been approached from many different angles. Jucker’s (1993) and Schourup’s (2001) analyses are rather theoretical, as they are based on reflections on previous analyses and categorisations of well. Others have conducted corpus analyses to tease out the functions of well inductively, such as Schiffrin (1987) and Aijmer (2002, 2013). In some cases specific contexts or functions are targeted, as with Norrick (2001), who focuses on oral narratives, or with Jackson and Jones (2013), who only discuss well-prefaced self-initiated repairs. Also specific varieties of English have received attention (e.g. Xhosa English in de Klerk [2005]; Hong Kong English in Lam [2010a,b]), as well as contrastive analyses that have looked into equivalents of well in other languages (e.g. Aijmer and Simon-Vandenbergen 2003 on Swedish and Dutch; García Vizcaíno and Martínez-Cabeza 2005, and Cuenca 2008 on Spanish). The result is a multifarious list of functions for well, many of which appear to overlap, if not coincide. What most analyses share, though, is their observation that Postprint version - [2] well’s functions can be divided into those that are situated in the textual or speech management domain, and those in the interpersonal or attitudinal domain. The former relates to the production and structuring of discourse, whereas the latter is involved with expressing the speaker’s attitudes to the discourse and to interlocutors. Aijmer’s (2011) functional classification of well, based on a corpus analysis of native and learner English, will serve as the basis for our overview of the functions most commonly attributed to well in prior research, because it is highly pertinent for the present study. She distinguishes between five speech management and two attitudinal functions. With a term borrowed from Allwood et al. (1990), Aijmer (2011: 237-241) describes well in its first speech management function as ”choice-related”, in that it reflects the choices the speaker is faced with in continuing the turn. The speaker is planning ahead what to say next, and this is reflected by well, which simultaneously indicates their desire to hold the turn as well. The same function has been attested by Jucker (1993: 447), who considers well a ”delay device” in such contexts, by de Klerk (2005: 1191), who explains it as the speaker’s need ”to contemplate”, by Müller (2005: 109), who describes it as “searching for the right phrase”, and by Lam (2010a: 272; 2010b: 668-669), who includes it in a broad “processing” category. Aijmer’s (2011) second speech management category for well (“change”) is modelled on Allwood et al.’s (1990: 11) change-related function ”enabling the speaker, on the basis of various feedback processes (internal or external) to change already-produced content, structure or expressions”. Müller (2005: 111-113) too distinguishes a ”rephrasing/correcting” function for well, as do Schiffrin (1987: 123) and García Vizcaíno and Martínez-Cabeza (2005). Cuenca (2008: 1384) notes its capacity for ”reformulation” and to ”change topic”, the latter of which has also been observed by Jucker (1993) and Lam (2010b). Third, Aijmer (2011: 241) refers to tokens of so that mark ”deliberation or pausing” at the start of a turn, pointing forward to a forthcoming answer, as ”prospective well”1. Schiffrin (1987) dubs well a ”marker of response” (1987: 102) in such instances, but notes that it may be used to ”defer the ideational core of their answers” (1987: 110) in that it may preface introductory brackets or contextualizing segments (e.g. in a narrative). As such speakers can comply flexibly with the conversational demand of responding to a question or request by delaying an immediate response. The fourth speech management function in Aijmer’s (2011) classification is that of marking stages in a narrative, which is in line with Norrick’s (2001) observation that in narratives well signals the start of a new episode or of the narrative altogether, or prefaces a summary coda. Likewise, 1 Note that this category of well does not comprise all of its functions that have ”prospective orientation”, as defined by Lenk (1998: 54) and Aijmer (2002: 37), viz. those utterances pointing forwards to the upcoming message. This definition would apply, for example, also to the choice function and the quotative function in Aijmer’s (2011) classification. Postprint version - [3] Müller (2005) distinguishes categories for well as moving to the main story, introducing the next scene in a story and providing a conclusion to a story description. The fifth, and final, function of well in the speech management domain according to Aijmer (2011) occurs when well appears at the start of a stretch of direct speech (cf. Schiffrin 1987: 124; Jucker 1993: 446; Smith and Jucker 2000: 216; García Vizcaíno and Martínez-Cabeza 2005: 80; Müller 2005: 113-115; Lam 2010b: 663), and therefore fulfils a quotative function. Attitudinal functions of well ”express attitudes or feelings to the hearer or to the preceding discourse” (Aijmer 2011: 245). The first of these in Aijmer’s (2011: 246) classification is meant to ”mitigate the speaker’s opinion”, which corresponds with Smith and Jucker’s (2000: 216) attestation of well to ”downgrade a claim” and Pulcini and Furiassi’s (2004: 114) use of well to ”mitigate the strength of a statement”. The second attitudinal function, marking disagreement, ”signals a correction, explanation, or clarification if the preceding discourse is challenging or makes the wrong assumption” (Aijmer 2011: 246). As such it typically prefaces a dispreferred response such as (partial) disagreement, scepticism, rejection of an offer, and criticism (cf. Schiffrin 1987; Jucker 1993; Smith and Jucker 2002; Pulcini and Furiassi 2004; de Klerk 2005; Müller 2005; Cuenca 2008; Lam 2010b). It, therefore, clearly has a ”downtoning” (Cuenca 2008: 1385) role as a ”face-threat mitigator” (Jucker 1993: 444; García Vizcaíno and Martínez-Cabeza 2005: 79). Given this broad functional scope of well many have attempted to account for its multifunctionality, with two dominant strands. The first is that which advocates a monosemy perspective, i.e. a core, invariant meaning can be posited for well from which individual instantiations are derived through pragmatic processes (cf. Fischer 2006: 12-14). Schourup (2001), for example, describes the marker as ”conveying that the speaker is actively considering whatever it is relevant to consider in determining what should now follow” (2001: 1058). In observing that well often marks insufficiency, Jucker (1993) and Smith and Jucker (2002) similarly qualify it as a signal ”that the context created by an utterance may not be the most relevant one for the interpretation of the next utterance” (Jucker 1993: 450). More recently such approaches have been criticised, in particular for failing to account for variation in the use of pragmatic markers across speech contexts (see e.g. Aijmer 2013: 11), as the function and interpretation of well has been shown to be highly context-dependent (see e.g. Aijmer and Simon-Vandenbergen 2003; Aijmer 2013). As an alternative, polysemy approaches have been gaining ground. These advocate that one form can occur with different interpretations “that are assumed to be related” (Fischer 2006: 13) without being traceable to a single core meaning. Cuenca (2008), for example, distinguishes eight functions for well that can be grouped in two broad categories, viz. that of a modal focus (e.g. disagreement) and that of a structural focus (e.g. Postprint version - [4] reformulation). In the resulting structure meanings may be closely related to one another or rather peripheral or only related to a single meaning in the network. In the same vein Aijmer (2013) posits that well cannot be attributed a fixed meaning but rather “a meaning potential available as a communicative resource to language users” (2013: 30) in the form of several core aspects from which new functions can be derived in specific situations. 3. Well in contrastive studies The use of well in English has on occasion been compared to pragmatic markers in other languages, although truly contrastive investigations have been rare. In a study with a monolingual French focus Beeching (2002: 128-129) suggests that the French markers enfin, bon, ben and c’est-à-dire (que) can all be expressed by well in English, each fulfilling a different function. Enfin, for example, marks a reformulation, c’est-à-dire que acts as a face-threat mitigator, bon ben indicates word-searching, etc. In Spanish bueno, bien and pues have been found to be the most common equivalents of well (among a whole host of options), all of which have been reported to be highly frequent, both in a small-scale contrastive study of a dubbed film (Cuenca 2008: 1376) and in monolingual analyses (Bellés Fortuño 2006: 224; Cuenca and Marín 2009: 907). García Vizcaíno and Martínez-Cabeza’s (2005) contrastive analysis of Spanish and English corpora with a focus on English well and Spanish bueno has demonstrated considerable overlap in the functional scope between these. They both serve to make a transition in a conversation less abrupt, mitigate a face-threatening act (attenuation), mark self-correction, and introduce direct speech. The attenuation function is, however, taken on three times as often by well as by bueno. Contrary to bueno, well is said to have additional functions such as act as a preface to an answer, but Cuenca’s (2008) results suggest that pues assumes this role instead. Bueno, on the other hand, can signal that the speaker accepts the contents contained in the previous speaker’s turn. Aijmer and Simon-Vandenbergen’s (2003) fairly small-scale investigation of an English-Dutch and English-Swedish translation corpus of fictional texts has shown that well has a plethora of translation equivalents in both languages, most notably nou/nu and (t)ja in Dutch, and ja, tja and men in Swedish. One notable difference between the two languages is that none of the Swedish forms stand out, whereas the Dutch corpus displays a slightly stronger preference for nou. As Aijmer and Simon-Vandenbergen (2003: 1156) themselves suggest, though, there is a Dutch particle, viz. wel, that shows a striking formal resemblance with well and shares many of its functions with its Postprint version - [5] English cognate but only does so in the Belgian variety of Dutch. This does not surface in their data, because they are exclusively based on texts from the Netherlands.2 The most common correspondences for well in Johansson’s (2006) investigation of the English-Norwegian parallel corpus are vel, nå, (t)ja, and nei, so here too “there is no clear counterpart in Norwegian corresponding to well” (2006: 118) and each Norwegian equivalent covers one or some of well’s functions. Johansson’s comparison with German in the Oslo Multilingual Corpus yields much the same picture, with a whole host of forms that take on functions of English well in German, most notably also, na (ja) and nun (ja). The odd one out in this overview is Chinese: Liu (2013: 157) points out that no functional or translation equivalents can be found for well in Chinese. It should be clear from this overview that most languages offer linguistic means that serve similar roles as well does in English, but that none have a device with the same broad scope as well. In consequence, language users of languages other than English need to resort to many more linguistic items to meet the same pragmatic demands than is necessary in English. 4. Well in learner language Well has been involved in many studies on the use of pragmatic markers in learner English, given its prominence in the speech of native speakers of English, and the wide attention it has received in the study of native speaker discourse. An overview of investigations of well in learner language yields a fairly mixed picture, depending largely on the learners’ language background and proficiency in English. Fuller (2003), for example, studied interviews, elicited narratives and casual conversations with native speakers of German, French and Spanish who had been residing in the United States for at least two years and had hence been exposed to English quite extensively in everyday life. These were found to be using well ”in a reasonably similar manner to the native speakers” (2003: 200). Without analysing their data in detail, Gilquin (2008) and Buysse (2010), on the other hand, observed French- and Dutch-speaking learners of English, respectively, using well overall considerably more frequently than native speakers in a comparable setting. In a similar vein, Müller’s (2004, 2005) participants in dyad conversations and story-retellings used well more than twice as often as native speakers. Statistically significant differences were attested for the functional categories ”searching for the right phrase”, ”conclusive well”, ”indirect answer” and ”continuing an 2 Hogeweg (2009) and Niemegeers (2009) do investigate Dutch wel but both restrict its scope to that of a modal particle, and neglect its uses as (what is sometimes described as) “an interjection” (Aijmer and SimonVandenbergen 2003: 1158), which would correspond with the functions that English well assumes. Postprint version - [6] answer or opinion” (2005: 138). Such cases of higher learner frequencies were attributed to the learners’ familiarity with well (due to its intensive use in English discourse and in learner textbooks), which makes it conveniently available for use in conversation. Additionally, Müller (2005: 251) suggests that learners had well take on some of the functions of the pragmatic marker so because they believed that the former sounds more English than the latter and they wanted to avoid another pragmatic marker, viz. so, due to its similarity to German also. Aijmer’s (2011) examination of the Swedish component of LINDSEI has shown that the Swedish learners use well significantly more frequently than native speakers in a similar setting. Strikingly, this difference is entirely down to speech management functions of well. In the attitudinal domain the learners even use well significantly less often than their native speaker peers, which might suggest that many learners remain unfamiliar with interpersonal communication strategies in the target language. Other studies have found learners using well considerably less often. For example, without comparing it to native speaker data, Hellermann and Vergun (2007) report almost negligible use of well (and most other pragmatic markers) in classroom interactions of beginning adult learners of English with varying L1 backgrounds, and they suggest that pragmatic marker incidence will rise with increasing proficiency. Yet, Romero Trillo (2002) demonstrates that Spanish university students turned to well only half as often as native speakers, and only 3 of the over 30 Japanese university students of different proficiency levels reported on by Hays (1992) used well (as opposed to and, but and so, which were used much more widely). A comparably low incidence has been observed for native speakers of Chinese: in Liao’s (2009) investigation only 1 teaching assistant studying in the United States used well about as frequently as native speakers, albeit still with a considerably restricted functional scope. Wei’s (2011) investigation of the speech of 141 Chinese students learning English likewise found that even the most proficient group of participants used well merely 0.028 times per 1,000 words, and in the speech of 49 final-year secondary school pupils in Hong Kong, where English plays a prominent role in society, well was virtually absent, according to Fung and Carter (2007). In sum, well has been shown to be present in learner discourse but to varying extents, with advanced learners and learners with Dutch, French, German or Swedish L1 using it all in all more frequently than native speakers of English (albeit not in all of its functions), and less experienced learners and learners with Spanish, Japanese and Chinese L1 using it considerably less frequently. Due to the widely diverging approaches and speech contexts adopted by these studies, comparison of their findings remains a thorny endeavour, though. Postprint version - [7] 5. Data and methodology 5.1 Corpus The corpus for the current study has been drawn from the Louvain International Database of Spoken English Interlanguage (LINDSEI), an international project that has pooled English interviews with learners from 11 different language backgrounds (Gilquin et al. 2010). From these I have selected five that have surfaced in previous research on well in learner English: (Belgian) Dutch (DU), (Belgian) French (FR), German (GE), Spanish (SP) and Chinese (CH). All learners were at least in their second year of studying English as a major subject at university, and they were interviewed by members of the teaching staff at the learners’ institution, most of whom were native speakers of English. All interviews were conducted along the same lines: first the interviewee talked for 1-2 minutes about a topic such as a book, a film, a travel experience, which sparked a conversation with the interviewer; at the end of each interview the interviewee was asked to tell a story based on four pictures. Apart from the Swedish component of LINDSEI, Aijmer (2011) also analysed the Louvain Corpus of Native English Conversation (LOCNEC), LINDSEI’s native speaker reference corpus. These interviews had the same setting as those conducted for LINDSEI, and the interviewees were all students at a British university. Although these data have been elicited for research purposes, they still comply with the prerequisite of authenticity traditionally imposed on language corpora. The interviews constitute speech produced by language users who are in full control of their linguistic behaviour within the constraints of the context. Similarly, job interviews, sales encounters, police interviews, retellings of silent films and courtroom questionings qualify as valid data, as long as their findings are not extended to the language community at large. Moreover, all participants can be expected to have been familiar with a context in which they talk face-to-face with an interlocutor who is more proficient in the language than they are (e.g. an oral exam in an English course). It can even be argued that most contexts in which foreign language learners speak English are artificial, such as roleplays or classroom conversations in which all participants share the same mother tongue and are forced to use a foreign language. Table 1 gives an overview of the size of the learner sub-corpora selected from LINDSEI and of LOCNEC. Postprint version - [8] Table 1 Size of the selected sub-corpora of LINDSEI (Gilquin et al. 2010) and of LOCNEC (Aijmer 2011: 233) DU FR GE SP CH Learner LOCNEC total interviews 50 50 50 50 53 253 50 total words 92,865 134,844 105,955 80,467 75,837 489,968 170,533 total words 74,589 83,294 81,181 62,101 58,523 359,688 125,666 1,492 1,666 1,624 1,242 1,104 1,422 2,513 interviewees av. words interviewees To enable comparison between the sub-corpora all absolute numbers have also been normalised to a relative frequency expressed in number of tokens per 1,000 words of interviewee speech. Differences in frequency have been tested for significance using the two-sample corpus frequency test wizard (http://sigil.collocations.de/wizard.html), which ”automatically chooses between chi-square (X²) and log-likelihood (G²), depending on which test is deemed to be more accurate for [the] data” (Hoffmann et al. 2008: 85). In the vast majority of cases the wizard selected the chi-square option. 5.2 Functional classification As most pragmatic markers, well is a form that can occur with meanings and functions that fall beyond the realm of pragmatics. All such tokens of well that do not perform a pragmatic marker function were identified and discarded from the analysis. Most of these were cases in which well was used as the adverb form of good and in combinations such as as well and as well as. Table 2 provides the overall incidence of well in each learner sub-corpus, that of all tokens in which well fulfils a pragmatic function, and the ”Index of Pragmatic Use” (Romero Trillo 2002: 776), i.e. the ratio of tokens with a pragmatic function vis-à-vis the total number of tokens of the item. By way of comparison, Aijmer (2011: 233) mentions 529 pragmatic tokens of well in LOCNEC, corresponding to 4.21 tokens per 1,000 words. As she does not provide any overall frequencies for well, the IPU cannot be calculated for LOCNEC. Postprint version - [9] Table 2 Total number of tokens of well in absolute numbers (n) and per 1,000 words of interviewee speech (‰), tokens of well with pragmatic marker function, and Index of Pragmatic Use (IPU) DU FR GE SP CH Learner total n ‰ n ‰ n ‰ n ‰ n ‰ n ‰ total well 869 10.91 1,086 11.88 574 6.68 541 8.71 73 1.15 3,143 8.74 PM well 754 9.47 1,031 11.28 479 5.57 408 6.30 48 0.76 2,720 7.56 IPU (%) 86.77 94.94 83.45 75.42 65.75 86.54 The pragmatic marker tokens of well were categorised following Aijmer’s (2011) classification of well into speech management functions and attitudinal functions, as introduced in Section 2 and presented schematically in Table 3. In Section 6 all functions will be expounded in greater detail. Table 3 Functions of well as distinguished by Aijmer (2011) and used in the present study Speech management functions Choice Planning of an upcoming utterance Change Reformulation, correction, topic change Prospective Deliberation in a response turn, pointing forward to upcoming utterance Stages in a narrative Start (a new episode in) a narrative Quotative Transition to a direct speech quotation Attitudinal functions Opinion Mitigation of speaker’s opinion disagreement Mitigation of (partial) disagreement, scepticism, criticism This classification was deemed most suitable for the analysis, because (i) it was developed while analysing a corpus identical in make-up to that studied here, and (ii) it facilitates comparison of the present findings to Aijmer’s for LOCNEC, as well as for the Swedish component of LINDSEI. A marginal proportion of tokens of well (n=9; 0.03 tokens per 1,000 words) was not classified because they could not be interpreted due to the unintelligibility of a part of the utterance, as in example (1), where two words were not transcribed in the interviewer turn that immediately precedes the well-prefaced utterance, and two immediately following well in the interviewee turn, which hampers the interpretation of well in a research study. Postprint version - [10] (1) <B> and a . a= and my wife may just as well be expecting . so you know <overlap/> <XX> <laughs> </B> <A> <overlap/> what do you mean she may just as well be exp= <XX> </A> <B> <laughs> well <XX> we don't know yet I mean it's it's a matter of a few days normally on on Wednesday but (er) </B> (FR34) 6. Results 6.1 Overall distribution of well The distribution of well as a pragmatic marker in the learner sub-corpora and in LOCNEC3 (Table 4) indicates that all learner groups use well significantly more often than their native peers, except for the Chinese learners, who use it to an almost negligible extent. The degree to which well is more common in the learner sub-corpora varies from only one third of tokens for the German sub-corpus, over half for the Spanish, to well over double for the Dutch and almost triple for the French. If additional chi-square tests are performed on the results for the learner sub-corpora, significant differences are revealed between all of these (p<0.001), except for the difference between the German and the Spanish sub-corpora. This yields the following hierarchy of frequency between the different sub-corpora: French > Dutch > German/Spanish > LOCNEC > Chinese. Table 4 The overall distribution of pragmatic well in the sub-corpora in absolute numbers (n) and in tokens per 1,000 words (rel.) subcorpus n rel. significance X²/G² +/- Int. Dutch 754 9.47 p<0.001 G² + 48 French 1,031 11.28 p<0.001 G² + 44 German 479 5.57 p<0.001 X² + 46 Spanish 408 6.30 p<0.001 G² + 42 Chinese 48 0.76 p<0.001 X² - 15 LOCNEC 529 4.21 50 3 Here and in all remaining tables in which numbers for LOCNEC are cited, these have been taken over from Aijmer (2011). Postprint version - [11] Interestingly, there are more learners in the Dutch and German sub-corpora who use well at least once than there are French learners, which signifies that the spread between learners who use it often and those who do not is wider in the latter than in the former. All interviewees in LOCNEC use well at least once (Aijmer 2011: 249). Aijmer (2011: 234-235) observed a difference in the position taken by well in the utterance, in that it most often occupies initial position in the native corpus and most often medial position in the Swedish learner corpus. A similar trend can be noted for the learner corpora currently under investigation (Figure 1), with ratios between 60 and 77 per cent for utterance-medial tokens, compared to a mere 42 per cent in LOCNEC. In a few rare instances well occurs in final position or on its own, typically “reflecting the fact that [the speakers] did not know how to continue” (Aijmer 2011: 235). 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Final/stand-alone DU 1.2 FR 2.4 GE 0.6 SP 2.5 CH 0.0 LOCNEC 0.7 Medial 77.6 73.5 70.4 69.1 60.4 42.0 Initial 21.2 24.1 29.0 28.4 39.6 57.3 Figure 1 Position of well in each sub-corpus (in percent) The analysis of well in the following sections will investigate whether the observed overall differences also hold for each of the specific pragmatic marker functions. 6.2 Speech management - Function 1: Choice In the choice-related speech management function well reflects the speaker’s planning efforts as well as a desire to hold the floor. Example (2) illustrates this word-searching function. Postprint version - [12] (2) <B> and I can speak English a lot when we go eating or something .. and stuff like that . and she . well she kinds of .. now=. nowadays she used to do .. she used to be able to speak Dutch very well . but nowadays it has really .. gone less and less . because . </B> (DU40) Often the interviewee’s processing effort surfaces even more explicitly when metalinguistic utterances are inserted, such as what do you call it in (3), indicating that the interviewee is aware of his inability to find the correct word, or I wouldn’t say in (4), which reflects the interviewee’s consideration of various alternative formulations. (3) <B> […] (em) I mean we we= we were swimming in this cave (em) . all of a sudden I saw this .. well (em) .. what do you call it (em) . somebody had (em) .. one tourist or I don't know who . (erm) had to go . very . badly to the[i:] (er) ..</B> (DU03) (4) <B> . (erm) .. so: . he starts a new picture . painting her a bit more .. well .. I wouldn't say <starts laughing> beautiful <stops laughing> </B> (GE14) In line with Müller’s (2005) and Aijmer’s (2011) observations that in its choice function well tends to be accompanied with other hesitation markers such as filled and empty pauses, two thirds of the ”choice” tokens of well indeed have a preceding and/or following pause, as illustrated in all of the examples given so far. This need not be the case, though, as (5) demonstrates: (5) <B> […] so that we can speak about (er) literature or culture and well different aspects that we want to: (eh) . to focus upon </B> (FR21) The choice function is the most frequent or second most frequent in the different learner sub-corpora (except for the Chinese), contrary to LOCNEC, where its use is marginal. Consequently, statistically significant differences can be noted between LOCNEC and all of these sub-corpora (Table 5), as well as between the Dutch and French sub-corpora on the one hand, and the German and Spanish on the other. This largely mirrors the observations for well as a whole. Postprint version - [13] Table 5 Distribution of well in the choice function in the learner subcorpora and LOCNEC sub-corpus n rel. % p X²/G² +/- Int. Dutch 309 3.88 41.0 p<0.001 X² + 36 French 340 3.72 33.0 p<0.001 X² + 32 German 153 1.78 31.9 p<0.001 X² + 34 Spanish 112 1.73 27.5 p<0.001 G² + 28 Chinese 10 0.16 20.8 n.s. X² (-) 3 LOCNEC 34 0.27 6.40 6.3 Speech management - Function 2: Change When it performs its change-related function, well may signal a need to reformulate an utterance that has just been produced, which may – but need not – also entail a topic shift. As Jackson and Jones (2013: 29) point out, rather than set straight a mistake, in such cases ”the speaker attends to ’on the record’ accuracy, with an original utterance being treated as essentially correct”. In example (6) the interviewee interrupts her utterance and reformulates it instantaneously: the painter does not continue a painting but starts a new one. In (7) the correction prefaced by well does not apply to the prior adjacent word, but entails a qualification of the first word of his claim (everywhere). The interviewee in excerpt (8) abandons the clause she had started to provide more apt a description than she probably had in mind first. (6) <B> then she sits down again and he continues well he makes a new portrait I think […] </B> (DU23) (7) <B> everywhere you go it's the language that most people speak </B> <A> (mhm) </A> <B> well not everywhere but </B> (SP32) (8) <B> […] it's kind of not . well it's not very ideal . to be on the beach cos it really it it . bothers the view you know </B> (DU49) Postprint version - [14] The self-interruption may serve not to correct a prior part of the utterance, but to insert background information the speaker deems vital for a good understanding of the upcoming part of the utterance, which Schiffrin (1987: 123) calls a ”background repair”. In (9) the interviewee stops to add the information that the film Lord of the Rings is a trilogy in order to make his reference to “the first part of the film” clear. (9) <B> . and (er) . the first thing is that the film tackles (er) an amazing scope of literary base . <overlap/> and . I'm </B> <A> <overlap/> (mhm) </A> <B> convinced that it's that it that it has done so quite well I mean that it has some shortcomings . (er) the first . well it's . it's a . trilogue in the first <overlap /> (er) . part of the film </B> <A> <overlap/> (mhm) </A> <B> . is . in my opinion really really good […] </B> (GE40) The change prefaced by well need not be a correction or clarification but may also be a topic change or may lead the discourse out of a digression (cf. Norrick 2001: 855), as in (10): the interviewee is interrupted by the interviewer and, after a brief question-answer sequence, continues the main point with well and a reiteration of the proposition he had initiated before the interruption. (10) <B> well I must adm= I preferred Namur . as a town </B> <A> yeah </A> <B> yeah it was <overlap/> warme= </B> <A> <overlap/> do you live here now </A> <B> yeah I do </B> <A> oh I see </A> <B> yeah . well Namur was warmer (er) it was (er) a really little town and .. I like the people there </B> (FR22) The change function is the first or second most frequent category for all sub-corpora, including the native speaker sub-corpus. Nonetheless, statistically significant differences hold between all learner sub-corpora and LOCNEC (Table 6), and between all learner sub-corpora (all of which at p<0.001, except for the difference between the Dutch and French corpus, at p<0.01, and Postprint version - [15] between the Dutch and Spanish, at p<0.05). Again this mirrors the overall frequency hierarchy for the sub-corpora, albeit that for the change function a significant difference can be seen between the German and Spanish sub-corpora as well. Table 6 Distribution of well in the change function in the learner subcorpora and LOCNEC sub-corpus n rel. % p X²/G² +/- Int. Dutch 248 3.11 32.9 p<0.001 G² + 40 French 363 3.97 35.2 p<0.001 X² + 39 German 140 1.63 29.2 p<0.01 X² + 34 Spanish 161 2.48 39.5 p<0.001 G² + 36 Chinese 13 0.20 27.1 p<0.001 X² - 8 LOCNEC 138 1.10 26.1 6.4 Speech management - Function 3: Prospective well Aijmer’s (2011: 241) “prospective well” allows the speaker to pause briefly before responding to the interlocutor’s question or request, in which case well points forward to an upcoming answer. The most common context in which this function surfaces in the corpus under investigation is at the start of the conversation, when the interviewee is asked to talk about one of the set topics. Often, as in (11), interviewees provide a preamble before they get to the actual topic. Similarly, in (12) the interviewee chooses to picture the context before responding to the question of what impressed him about Rwanda. (11) <A> okay <name of interviewee> if I can ask you to (er) talk on the topic that you've selected </A> <B> (mhm) well (er) I've selected topic (er) three and (er) I wanted to talk about a movie that I . saw (er) last week </B> (DU48) (12) <A> <overlap/> so what particularly impressed you </A> <B> well the thing is that I'm . I'm quite concerned about Africa as (er) because (erm) .. my: well I was born there first of all and (erm) </B> <A> whereabouts </A> <B> in (erm) Ivory Coast . in Abidjan </B> Postprint version - [16] <A> (mhm) </A> <B> and my: my father (erm) . goes there . at least four or five times a year </B> (FR25) The frequencies for the prospective function of well are slightly lower in the learner subcorpora than in LOCNEC, but with the exception of the Chinese sub-corpus none of these differences achieve statistical significance (Table 7). Table 7 Distribution of well in the prospective function in the learner sub-corpora and LOCNEC sub-corpus n rel. % p X²/G² +/- Int. Dutch 61 0.77 8.1 n.s. X² (-) 31 French 57 0.62 5.5 n.s. X² (-) 26 German 53 0.62 11.1 n.s. X² (-) 27 Spanish 40 0.62 9.8 n.s. X² (-) 20 Chinese 8 0.13 16.7 p<0.001 X² - 5 LOCNEC 107 0.85 20.2 6.5 Speech management - Function 4: Marking stages in a narrative Aijmer (2011: 243) observes that interviewees in the Swedish component of LINDSEI and in LOCNEC particularly have well mark stages in a narrative when they are telling the picture-based story towards the end of the interview, ”usually to introduce the main character” (2011: 244). This can be corroborated in the other components of LINDSEI, as in (13): (13) <A> […] so if you just take a look at that <XX> in that sequence and describe what is happening there </A> <B> (em:) . well there's a portrait painter . and there's a lady in a chair . who is not . that attractive I would say but that's also maybe cos I glanced . you know <laughs> at the bottom of the page </B> (DU27) In all learner sub-corpora (except the Chinese) the relative frequency for the stage-marking function is higher than in the native corpus, which yields statistical significance for all but the Spanish sub-corpus (Table 8). Other significant differences can be found between the Dutch sub-corpus on Postprint version - [17] the one hand and the German (p<0.01), Spanish (p<0.01) and Chinese (p<0.001) on the other, as well as between all other learner sub-corpora and the Chinese. Table 8 Distribution of well in the stage-marking function in the learner sub-corpora and LOCNEC sub-corpus n rel. % p X²/G² +/- Int. Dutch 46 0.58 6.1 p<0.001 X² + 25 French 37 0.40 3.6 p<0.001 X² + 22 German 23 0.27 4.8 p<0.05 X² + 18 Spanish 14 0.22 3.4 n.s. X² (+) 11 Chinese 2 0.03 4.2 n.s. X² (-) 2 LOCNEC 16 0.13 3.0 6.6 Speech management - Function 5: Marking a transition to a direct speech quotation When well introduces a stretch of direct speech – as is illustrated in excerpt (14) – the marker is presented as if it was mentioned in the quoted utterance, although it most likely was not, and it collocates with such verbs as say, think, believe, and feel (Schiffrin 1987: 124; Jucker 1993: 446; Smith and Jucker 2000: 216; García Vizcaíno and Martínez-Cabeza 2005: 80; Müller 2005: 113-115; Lam 2010b: 663; Aijmer 2011: 244-245). In employing quotative well the speaker appears to attribute a mitigating context to the quoted utterance, which may or may not be veritable. (14) <B> <overlap/> cos he: . he's not very fond <XX> he's going with me in (eh) July so I'm trying t= cos he's never been there he says well the: the food is quite disgusting I said well you've never been there so he doesn't know . and I will try and convince him <laughs> </B> (FR04) The differences in frequency between LOCNEC and the Dutch and German learner subcorpora are negligible (Table 9). With a relative frequency that is triple that of their native peers the French learners demonstrate a significantly higher incidence of quotative well compared to LOCNEC, but also compared to all other learner sub-corpora (FR-DU p<0.05, FR-GE p<0.01, FR-SP and FR-CH p<0.001). The Spanish learners clearly do not use this quotative structure often, with a mere 9 participants mentioning it only 12 times in total. Postprint version - [18] Table 9 Distribution of well in the quotative function in the learner subcorpora and LOCNEC sub-corpus n rel. % p X²/G² +/- Int. Dutch 29 0.36 3.8 n.s. X² (-) 17 French 55 0.60 5.3 p<0.05 X² + 21 German 25 0.29 5.2 n.s. X² (-) 18 Spanish 12 0.19 2.9 p<0.05 X² - 9 Chinese 5 0.08 10.4 p<0.001 X² - 3 LOCNEC 49 0.39 9.3 6.7 Attitudinal - Function 6: Mitigation of speaker’s opinion The first attitudinal function of well in Aijmer’s (2011) classification is when the marker serves to mitigate the speaker’s opinion or the force of the speaker’s upcoming statement. For example, in (15) the interviewee confesses to not being all that enthusiastic about her studies anymore. She seems to realise that this opinion may not go down well with her interlocutor, who is a member of her university’s teaching staff, and therefore mitigates the proposition with well and with not really, that and brief laughter. In this category well tends to be accompanied by other such mitigating devices like I guess, probably or modal must, as in excerpt (16). (15) <A> (mhm) . and do you like your studies so far . <X> </A> <B> (mhm) . well <overlap/> <laughs> </B> <A> <overlap/> <laughs> </A> <B> (mhm) I'm not really that enthusiastic anymore . I wasn't </B> (DU09) (16) <A> what do you think the painter is thinking here </A> <B> <laughs> well he must be surprised . because of . if someone asks for a portrait . you have to have what you have . you can't change something to please […] </B> (SP06) The only statistically significant differences with LOCNEC hold for the French learners, who use it to a greater extent, and the Chinese, who use it to a lesser extent (Table 10). All learner groups use well significantly more often than the Chinese, and it is also significantly more frequent among Postprint version - [19] the French, German and Spanish learners than the Dutch, the latter of whom find themselves at the same level as the Swedish learners in Aijmer’s (2011: 247) study. Table 10 Distribution of well in the opinion function in the learner subcorpora and LOCNEC sub-corpus n rel. % p X²/G² +/- Int. Dutch 29 0.36 3.8 n.s. X² - 21 French 74 0.81 7.2 p<0.05 X² + 28 German 52 0.60 10.9 n.s. X² (+) 28 Spanish 40 0.62 9.8 n.s. X² (-) 20 Chinese 7 0.11 14.6 p<0.001 X² - 5 LOCNEC 65 0.52 12.3 6.8 Attitudinal - Function 7: Disagreement In its capacity of marking disagreement well downtones dispreferred responses such as disagreement, scepticism, and criticism. In (17) the interviewee has just described her activities as leader in a youth movement. When the interviewer makes the assumption that the children in her group are not too young, the interviewee adjusts this view. (17) <A> no too sma= not too young then </A> <B> well </B> <A> and not too </A> <B> some of them are quite young the thirteen year olds can be very young but the fifteen year olds can be very . old as well so </B> (DU17) The interviewer in excerpt (18) does not make an assumption, but the interviewee believes the binary option to answer the question is not adequate since the situation described is not black or white. (18) <A> was the mother at home or was she also <overlap/> working </A> <B> <overlap/> (erm) well she was at home . but: when I started a few: month after she: (erm) . started working part-time . teaching dyslexic children </B> <A> (mhm) </A> Postprint version - [20] <B> but because they had five children she was actually busy driving them around to all their . <overlap/> <starts laughing> activities </B> (GE09) In this category well is used to avoid offending the hearer by disagreeing politely. In (19), however, what seems to be more at play is the interviewee’s desire to remain modest, which requires only a partial agreement with the interviewer’s assumption. Partial (dis)agreement may additionally be signalled by a collocation such as yes well, indicating the need to qualify an answer to a polar question more elaborately, as example (20) demonstrates. (19) <A> oh you're a very good swimmer then </A> <B> well . a little <overlap/> <laughs> </B> (SP23) (20) <A> <overlap/> <laughs> and you da= you're dancing in the streets </A> <B> yes well we had some sort of stage </B> (FR29) Finally, well can preface an utterance that displays the interviewee’s reluctance or inability to answer a question: (21) <A> and so the purpose of the work is to find out if it's successful or to make it more successful or: </A> <B> <starts laughing> well I I don't know what the purpose is </B> (FR31) The Dutch, German, Spanish and Chinese learner groups use well to express an attitude of disagreement significantly less often than the native speakers and the French learners in the study (Table 11). The latter turn to it even slightly more frequently than the native speakers but this difference does not yield statistical significance. Among the learner groups, the French are set off from the others with a significantly higher frequency (p<0.001), as are the Chinese with a significantly lower frequency (p<0.001). Postprint version - [21] Table 11 Distribution of well in the disagreement function in the learner sub-corpora and LOCNEC sub-corpus n rel. % p X²/G² +/- Int. Dutch 32 0.40 4.2 p<0.001 X² - 22 French 97 1.06 9.4 n.s. X² (+) 24 German 32 0.37 6.7 p<0.001 X² - 18 Spanish 26 0.40 6.4 p<0.001 X² - 16 Chinese 3 0.05 6.3 p<0.001 X² - 3 LOCNEC 120 0.95 22.7 7. Discussion The results outlined in the previous section are summarised in Table 12 in the form of the relative frequencies for each of the functions of well in each sub-corpus. To enable comparison with Aijmer’s (2011) findings, the frequencies that she has reported for the Swedish component of LINDSEI have been added (SW). Since she does not provide relative frequencies, I have calculated the numbers for SW and LOCNEC on the basis of the absolute numbers (Aijmer 2011: 248). Table 12 Relative frequencies (per 1,000 words) for all functions in each sub-corpus DU FR GE SP CH SW LOCNEC Speech management Choice 3.88 3.72 1.78 1.73 0.16 1.67 0.27 Change 3.11 3.97 1.63 2.48 0.20 1.29 1.10 Prospect 0.77 0.62 0.62 0.62 0.13 0.77 0.85 Stages 0.58 0.40 0.27 0.22 0.03 0.46 0.13 Quotative 0.36 0.60 0.29 0.19 0.08 0.14 0.39 Sub-total 8.70 9.32 4.58 5.25 0.60 4.33 2.74 Opinion 0.36 0.81 0.60 0.62 0.11 0.31 0.52 Disagree 0.40 1.06 0.37 0.40 0.05 0.81 0.95 Sub-total 0.77 1.87 0.98 1.02 0.16 1.11 1.47 Total 9.47 11.28 5.57 6.30 0.76 5.44 4.21 Attitudinal Postprint version - [22] Well is overall more prevalent in each of the learner sub-corpora selected for this study than in the native speaker corpus. The Chinese learners form the sole exception to this tendency, but they use well to such a marginal extent that this sub-corpus clearly occupies an exceptional position in the study. This finding is in keeping with Fung and Carter’s (2007), Liao’s (2009) and Wei’s (2011) findings, reported in Section 4. At the other end of the spectrum, significantly more tokens of well are found among the French learners than among the native speakers in all but two functional categories . The higher incidence in the Dutch and German sub-corpora is due to higher frequencies in the choice, change and stage-marking categories, whereas that in the Spanish sub-corpus is limited to the former two categories. Smaller numbers than in the native corpus are noted for the disagreement function in the Dutch, German and Spanish sub-corpora, the latter of which also displays a lower frequency for quotative well. On the whole, the learners make significantly greater use of well in the speech management domain than the native speakers, which can especially be attributed to higher frequencies in the choice and change functions. These are more common in all learner groups (except the Chinese) than in the native group, and the stage marking-function in all but the Spanish and Chinese groups. The overall significantly lower frequency of the attitudinal functions for the learners is particularly due to the disagreement category: well expresses disagreement significantly less often in all learner subcorpora than in LOCNEC, except in the French sub-corpus where it more closely approximates native speaker frequency. Also the shares taken up by the speech management functions as a group in the learner sub-corpora differ markedly from that in the native speaker corpus: the former range between 79 and 92 percent, whereas the latter amounts to 65 percent (Figure 2). Postprint version - [23] 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Attitudinal DU 8.1 FR 17.4 GE 17.7 SP 16.7 CH 20.8 SW 20.4 LOCNEC 35 Speech mgt. 91.9 82.6 82.3 83.3 79.2 79.6 65 Figure 2 Shares of speech management and attitudinal functions per sub-corpus (in percent) The average incidence of well in Aijmer’s (2011) study of the Swedish component of LINDSEI is lower than in most other learner sub-corpora. It closely approximates that of the German subcorpus, with which it does not differ for the relatively large choice and change categories. For the prospective, stage-marking and opinion functions the learner group closest to the Swedish in terms of frequency is the Dutch, whereas for the quotative and attitudinal functions the Swedish frequencies more closely approximate the Spanish. Aijmer (2011: 248) too notes more intensive use of all speech management functions, except for the quotative function, which occurred less frequently in her learner corpus, as did both attitudinal functions. How can these findings be explained? First, the remarkably higher incidence of well as a choice-related marker is likely to be a result of the learners’ less fluent discourse in English, creating more opportunities for such a marker to be inserted in the same way as learners have been reported to include more filled pauses (see e.g. Gilquin 2008). In the same vein, learners can be assumed to feel a greater need to set straight an utterance (change-related function), since they may be less able to provide accurate claims right away than native speakers. Moreover, as Müller (2004) has suggested, learners are more inclined to display a ”lack of confidence in their linguistic abilities” (2004: 1175). Second, if learners indeed feel a greater need to index choice and change slots, well appears to be a particularly popular means to cater for it. In general, learners have a less complex and more Postprint version - [24] restricted inventory of pragmatic devices than native speakers (Kasper and Blum-Kulka 1993; Gilquin 2008; Buysse 2010), and can be presumed to stick to those pragmatic markers they are most familiar with. Well is a likely candidate to take on the role of a “dummy” pragmatic marker, thanks to its high frequency in native discourse, and its functional versatility and elusiveness. The former intensifies the learners’ exposure to the marker, and the latter enhances its employability in any type of discourse. This is not to say that learners outstep the bounds of well’s functional scope, since all of its tokens in the learner data are fully compatible with native practice. It does signify, though, that instead of varying their choice of pragmatic markers that can fulfil similar or identical functions, learners appear to stick to one marker that they use with a higher incidence than native speakers. Moreover, Müller (2005: 250) has found that well occurs in English textbooks aimed at native speakers of German to a greater extent than so and to a much greater extent than you know and like. Likewise Mukherjee and Rohrbach (2006: 216) posit that “apart from well – all other discourse markers are notoriously underrepresented even in modern [ELT textbooks and] materials”. The dominance of well in teaching materials may be related to its high frequency in spoken English, the wide range of functions it can express (which enhances its employability in conversation and in textbooks), and its occurrence in formal and informal spoken registers alike. Further research is needed, however, to gauge the full extent of the influence of this factor in how language learners use a pragmatic marker like well. At present only a severely restricted number of the wide range of materials on offer has been covered, presenting information for only a select number of language learner backgrounds. It would also be interesting to test the hypothesis that speech management functions of well are more prevalent in the ELT materials than attitudinal functions, which could explain why the former occur more often than the latter in the learner data. This could further corroborate previous findings that learners tend to be exposed to well quite frequently in pedagogic settings, such as in the classroom, because pragmatic markers involved in speech management strategies have been claimed to be part and parcel of teacher talk, which is not necessarily the case for attitudinal (functions of) markers (Romero Trillo 2002; Fung and Carter 2007). A fourth factor might be the (un)availability of functionally equivalent items in the learners’ mother tongue. To fully gauge the extent of positive or negative L1 transfer in this study would of course necessitate a detailed comparative analysis of all possible equivalents of well in the languages involved, which obviously extends beyond our scope. As we have seen in Section 3, to date very few such systematic studies have been conducted. What is clear from prior research, though, is that – unlike for many other pragmatic markers – none of the languages concerned have a dominant equivalent form of well that has a similar functional range, while most of them do have linguistic devices that take on one or more of well’s roles. It can safely be assumed that if learners can appeal to a single pragmatic marker in English to fulfil the same function as an array of markers in their L1, Postprint version - [25] as appears to be the case for Dutch, French, German and Spanish (yet not for Chinese), it stands a good chance of being opted for time and again by learners. Hard and fast conclusions on the potential L1 influence on specific uses of well in learner English can, however, not be drawn. For example, the formal and (assumed) functional likeness between the (Belgian) Dutch marker wel and the English marker well may have had an influence on learner usage of the latter in English, but too little is known on Dutch wel to compare the two in any meaningful way. Likewise, there are no relevant contrastive analyses available between English and French or German. Well and bueno have been juxtaposed by García Vizcaíno and Martínez-Cabeza (2005, see Section 3), but their findings cannot assist in our search for explanations either. For example, quotative well is significantly less common in the Spanish learners’ discourse in spite of the similar function assumed by bueno. The significantly lower incidence of the disagreement function in the Spanish sub-corpus could be considered as in line with García Vizcaíno and Martínez-Cabeza’s observation that bueno’s attenuation function is three times less frequent than well’s. As we have seen in Section 3, though, the equally frequent Spanish marker pues has been reported to assume this role as well as an equivalent of well. Finally, the less prominent role given to well in its attitudinal functions may be related to their greater familiarity with the marker’s speech management function, but also to the conversational needs of language learners: not being fully proficient in the language, learners may be less concerned with the interpersonal, attitudinal aspects of the discourse than with getting an accurate, coherent message across to their interlocutor. This explanation does not cater for the frequencies found in the French sub-corpus. The French learners, however, use well to a greater extent in all categories, which buttresses previous claims that they use well whenever they can, as a ”pragmatic teddy bear” (Gilquin 2008: 129). Obviously these five factors are tentative suggestions that can help build an explanation for the extensive prevalence of well in the English discourse of learners with varying mother tongue backgrounds. As the body of research on pragmatic markers in learner language expands, these can further be falsified, e.g. in studies that contrast the present findings with similar analyses of other pragmatic markers, other analyses of well but with different types of learners, or in contrastive analyses of the languages involved. 8. Conclusion This investigation has applied Aijmer’s (2011) functional classification of the pragmatic marker well in a Swedish corpus of learner English to the Dutch, French, German, Spanish, and Chinese components of the LINDSEI corpus. The general tendency that has emerged from the analysis of 253 interviews, Postprint version - [26] across these five learner groups, is one of a higher incidence of well in learner than in native English discourse. There is, however, a discrepancy between functions related to speech management and those with an attitudinal role, the former of which tend to be significantly more frequent in the learner data and the latter significantly less frequent. Although no uses have been attested that are particular to the learner corpora, this seems to indicate that many learners of English are familiar with pragmatic functions of well, they may not have acquired the marker to the extent that they can use it to its full potential. Hence, this investigation provides further evidence to support calls to devote more attention to pragmatic features of the language in EFL courses (cf. Mukherjee and Rohrbach 2006; Lam 2010a; Mukherjee 2009; Aijmer 2011). The learner groups do not form a homogeneous population, though. Some of the functional differences that have been identified may simply be attributed to the highly specific nature of the function and its overall small frequency (e.g. the quotative function), and the differences between learners within a sub-corpus can be large (e.g. relative frequencies in the French sub-corpus range between 0 and 53). Yet, many of the observed differences between the sub-corpora are likely to be due to (an interplay of) other factors. First, the primary variable in the learner corpus is the learners’ mother tongue, so positive or negative L1 interference may influence learners’ use of well. Second, the degree to which learners in these countries are exposed to English in everyday life, in the media, and in educational settings tends to differ greatly (Gilquin et al. 2010: 48-56). Third, all participants were students of English at university when the corpus was compiled, but this does not necessarily imply that they had all reached the same level of general proficiency in English. Further research is, therefore, called for (i) to focus contrastively on the systems available to language users in these different languages to perform the functions identified for well, (ii) to expand the study of well in learner English to other text types (e.g. one in which learners may be more inclined to index interpersonal relations), (iii) to explore the five remaining components of LINDSEI, and (iv) to conduct similar studies on other pragmatic markers than well. Appendix: Transcription conventions <A> Interviewer turn starts <B> Interviewee turn starts </A> Interviewer turn ends </B> Interviewee turn ends … Empty pause: . (short), .. (medium), … (long) [] Phonetic annotations (e.g. the[i:], a[ei]) : Vowel lengthening (e.g. so:) Postprint version - [27] = Truncation <X> Unintelligible word <overlap> Overlapping speech <foreign> Foreign words <coughs> Non-verbal sounds and contextual comments are specified in angle brackets Notes * I am truly grateful to the three anonymous referees and the editor-in-chief for their most insightful comments on an earlier version of this article. References Aijmer, Karin. 2002. English discourse Particles. Evidence from a corpus. Amsterdam/Philadelphia: John Benjamins. Aijmer, Karin. 2011. Well I’m not sure I think… The use of well by non-native speakers. International Journal of Corpus Linguistics 16(2). 231-254. Aijmer, Karin. 2013. Understanding pragmatic markers. A variational pragmatic approach. Edinburgh: Edinburgh University Press. Aijmer, Karin & Anne-Marie Simon-Vandenbergen. 2003. The discourse particle well and its equivalents in Swedish and Dutch. Linguistics 41(6). 1123-1161. Allwood, Jens, Joakim Nivre & Elisabeth Ahlsén. 1990. Speech management. On the non-written life of speech. Nordic Journal of Linguistics 13. 3-48. Bellés Fortuño, Begona. 2006. Discourse markers within the university lecture genre: A contrastive study between Spanish and North-American lectures. Castellón: Universitat Jaume I Castelló PhD Dissertation. Beeching, Kate. 2002. Gender, politeness and pragmatic particles in French. Amsterdam/Philadelphia: John Benjamins. Buysse, Lieven. 2010. Discourse markers in the English of Flemish university students. In Iwona Witzcak-Plisiecka (ed.), Pragmatic perspectives on language and linguistics. Vol. 1: Speech actions in theory and applied studies, 461-484. Newcastle-upon-Tyne: Cambridge Scholars Publishing. Buysse, Lieven. 2012. So as a multifunctional discourse marker in native and learner speech. Journal of Pragmatics 44(13). 1764-1782. Cuenca, Maria-Josep. 2008. Pragmatic markers in contrast: The case of well. Journal of Pragmatics 40(8). 1373-1391. Postprint version - [28] Cuenca, Maria-Josep & Maria-Josep Marín. 2009. Co-occurrence of discourse markers in Catalan and Spanish oral narrative. Journal of Pragmatics 41(5). 899-914. de Klerk, Vivian. 2005. Procedural meanings of well in a corpus of Xhosa English. Journal of Pragmatics 37(8). 1183-1205. Fischer, Kerstin. 2006. Towards an understanding of the spectrum of approaches to discourse particles: Introduction to the volume. In Kerstin Fischer (ed.), Approaches to discourse particles, 1-20. Amsterdam: Elsevier. Fuller, Janet M. 2003. Discourse marker use across speech contexts: A comparison of native and nonnative speaker performance. Multilingua 22. 185-208. Fung, Loretta & Ronald Carter. 2007. Discourse markers and spoken English: Native and learner use in pedagogic settings. Applied Linguistics 28(3). 410-439. García Vizcaíno, María José & Miguel A. Martínez-Cabeza. 2005. The pragmatics of well and bueno in English and Spanish. Intercultural Pragmatics 2(1). 69-92. Gilquin, Gaëtanelle. 2008. Hesitation markers among EFL learners: Pragmatic deficiency or difference? In Jesús Romero-Trillo (ed.), Pragmatics and corpus linguistics: A mutualistic entente. Berlin: Walter de Gruyter. Gilquin, Gaëtanelle, Sylvie De Cock & Sylviane Granger. 2010. Louvain International Database of Spoken English Interlanguage. Louvain-la-Neuve: Presses Universitaires de Louvain. Granger, Sylviane. 1996. From CA to CIA and back: An integrated approach to computerized bilingual and learner corpora. In Karin Aijmer, Bengt Altenberg & Mats Johansson (eds.), Languages in Contrast, 37-51. Lund: Lund University Press. Hays, Paul R. 1992. Discourse markers and L2 acquisition. In Don Staub and Cheryl Delk (eds.), The proceedings of the twelfth Second Language Research Forum. April 2-5, 1992. Michigan State University, 24-34. East Lansing, Michigan: Papers in Applied Linguistics. Hellermann, John and Andrea Vergun. 2007. Language which is not taught: The discourse marker use of beginning adult learners of English. Journal of Pragmatics 39(1). 157-179. Hoffmann, Sebastian, Stefan Evert, Nicholas Smith, David Lee & Ylva Berglund Prytz. 2008. Corpus linguistics with BNCweb. A practical guide. Bern: Peter Lang. Hogeweg, Lotte. 2009. The meaning and interpretation of the Dutch particle wel. Journal of Pragmatics 41(3). 519-539. Jackson, Clare & Danielle Jones. 2013. Well they had a couple of bats to be truthful: Well-prefaced, self-initiated repairs in managing relevant accuracy in interaction. Journal of Pragmatics 47(1). 28-40. Postprint version - [29] Johansson, Stig. 2006. How well can well be translated? On the English discourse particle well and its correspondences in Norwegian and German. In Karin Aijmer & Anne-Marie Simon-Vandenbergen (eds.), Pragmatic markers in contrast. Amsterdam: Elsevier. Jucker, Andreas H. 1993. The discourse marker well: A relevance-theoretical account. Journal of Pragmatics 19(5). 435-452. Kasper, Gabriele & Shoshana Blum-Kulka. 1993. Interlanguage pragmatics: Introduction. In Gabriele Kasper & Shoshana Blum-Kulka (eds.), Interlanguage pragmatics, 3-17. Oxford: Oxford University Press. Lam, Phoenix W. Y. 2010a. Discourse particles in corpus data and textbooks: The case of well. Applied Linguistics 31(2). 260-281. Lam, Phoenix W. Y. 2010b. Toward a functional framework for discourse particles: A comparison of well and so. Text & Talk 30(6). 657-677. Lenk, Uta. 1998. Marking discourse coherence. Functions of discourse markers in spoken English. Tübingen: Gunter Narr Verlag. Liao, Silvie. 2009. Variation in the use of discourse markers by Chinese teaching assistants in the US. Journal of Pragmatics 41(7). 1313-1328. Liu, Binmei. 2013. Effect of first language on the use of English discourse markers by L1 Chinese speakers of English. Journal of Pragmatics 45(1). 149-172. Mukherjee, Joybrato. 2009. The grammar of conversation in advanced spoken learner English: Learner corpus data and language-pedagogical implications. In Karin Aijmer (ed.), Corpora and language teaching, 203-230. Amsterdam/Philadelphia: John Benjamins. Mukherjee, Joybrato & Jan-Marc Rohrbach. 2006. Rethinking applied corpus linguistics from a language-pedagogical perspective: New departures in learner corpus research. In Bernhard Ketteman & Georg Marko (eds.), Planning, gluing and painting corpora: Inside the applied corpus linguist’s workshop, 205-232. Frankfurt am Mein: Peter Lang. Müller, Simone. 2004. Well you know that type of person: Functions of well in the speech of American and German students. Journal of Pragmatics 36(6). 1157-1182. Müller, Simone. 2005. Discourse markers in native and non-native English discourse. Amsterdam/Philadelphia: John Benjamins. Niemegeers, Sofie. 2009. Dutch modal particles maar and wel and their English equivalents in different genres. Translation and Interpreting Studies 4(1). 47-66. Norrick, Neil R. 2001. Discourse markers in oral narrative. Journal of Pragmatics 33(6). 849-878. O’Keeffe, Anne, Michael McCarthy & Ronald Carter. 2007. From corpus to classroom. Language use and language teaching. Cambridge: Cambridge University Press. Postprint version - [30] Polat, Brittany. 2011. Investigating acquisition of discourse markers through a developmental learner corpus. Journal of Pragmatics 43(15). 3745-3756. Pulcini, Virginia & Cristiano Furiassi. 2004. Spoken interaction and discourse markers in a corpus of learner English. In Alan Partington, John Morley & Louann Haarman (eds.), Corpora and discourse, 107-123. Bern: Peter Lang. Romero Trillo, Jesús. 2002. The pragmatic fossilization of discourse markers in non-native speakers of English. Journal of Pragmatics 34(6). 769-784. Schiffrin, Deborah. 1987. Discourse markers. Cambridge: Cambridge University Press. Schourup, Lawrence. 1985. Common discourse particles in English conversation. New York/London: Garland Publishing. Schourup, Lawrence. 2001. Rethinking well. Journal of Pragmatics 33(7). 1025-1060. Smith, Sara W. & Andreas H. Jucker. 2000. Actually and other markers of an apparent discrepancy between propositional attitudes of conversational partners. In Gisle Andersen & Thorstein Fretheim (eds.), Pragmatic markers and propositional attitude, 207-235. Amsterdam/Philadelphia: John Benjamins. Smith, Sara W. & Andreas H. Jucker. 2002. Discourse markers as turns. Evidence for the role of intersubjectivity in interactional sequences. In Anita Fetzer & Christiane Meierkord (eds.), Rethinking sequentiality, 151-178. Amsterdam/Philadelphia: John Benjamins. Watts, Richard J. 1989. Taking the pitcher to the “well”. Journal of Pragmatics 13(2). 203-237. Wei, Ming. 2011. Investigating the oral proficiency of English learners in China: a comparative study of the use of pragmatic markers. Journal of Pragmatics 43(14). 3455-3472. Postprint version - [31]