2.1Discourse markers and speech fluency

DMs have sometimes been negatively characterized as “a sign of disfluency and carelessness” (Brinton 1996Brinton, Laurel J. 1996 Pragmatic Markers in English: Grammaticalization and Discourse Functions. New York: Mouton de Gruyter. , 33). Nevertheless, Tottie (2011Tottie, Gunnel 2011 “ Uh and Um as Sociolinguistic Markers in British English.” International Journal of Corpus Linguistics 16 (2): 173–197. , 193) argues that the term disfluency is based on an idealized conception of fluent speech production and is “a rather negative and uninformative default term that says nothing about the discourse functions”. She then proposes the more positive term “planners”. When employed as a speech planning and monitoring strategy, DMs in spoken discourse have been shown to be helpful in spontaneous speech production; they contribute to speech fluency and smoother communication (Götz 2013Götz, Sandra 2013 Fluency in Native and Non-Native English Speech. Amsterdam: John Benjamins. ; Crible et al. 2017Crible, Ludivine, Liesbeth Degand, and Gaëtanelle Gilquin 2017 “The Clustering of Discourse Markers and Filled Pauses: A Corpus-Based French-English Study of (Dis)fluency.” Languages in Contrast 17 (1): 69–95. ; Rühlemann 2019Rühlemann, Christoph 2019 Corpus Linguistics for Pragmatics. Oxon: Routledge.; Wolk et al. 2021Wolk, Christoph, Sandra Götz, and Katja Jäschke 2021 “Possibilities and Drawbacks of Using an Online Application for Semi-Automatic Corpus Analysis to Investigate Discourse Markers and Alternative Fluency Variables.” Corpus Pragmatics 5: 1–30. ), especially when a lexical gap or speech difficulty emerges. In this regard, DMs may fulfill “potentially disfluent functions” by monitoring (checking for understanding, calling for help), punctuation (stalling, planning) and reformulation (paraphrase and actual corrective relations; Cribble 2017Crible, Ludivine 2017 “Discourse Markers and (Dis)fluency across Registers: A Contrastive Usage-based Study in English and French.” PhD diss., Université de Berne., 80).

As House (2009House, Juliane 2009 “Subjectivity in English as Lingua Franca Discourse: The Case of You Know .” Intercultural Pragmatics 6 (2): 171–193. , 187) notes, DMs can be a “gap-filler”, in the form of “a stock phrase mainly used to help speakers process and plan their output, and link spans of discourse”, which may fill the pauses and replace the disfluency. Tsai and Chu (2017)Tsai, Pei-shu., and Wo-hsin Chu 2017 “The Use of Discourse Markers among Mandarin Chinese Teachers, and Chinese as a Second Language and Chinese as a Foreign Language Learners.” Applied Linguistics 38 (5): 638–665. examine the DMs used by Chinese-speaking teachers and learners and report that individual speakers who very often use DMs display their fluency in the target language and decreasing numbers of incomplete utterances (false starts) per turn. Like filled pauses in speaking, DMs serve as the “elegant fillers” or “fluencemes” that occur frequently and can easily be used to fill a silence during speech processing and planning in a natural-sounding way, increasing “the length of a speech run (and thus the overall productive fluency) as well as the degree of naturalness of the output (and thus perceptive fluency)” (Götz 2013Götz, Sandra 2013 Fluency in Native and Non-Native English Speech. Amsterdam: John Benjamins. , 40). Such fluency-enhancing functions of DMs can be equivalent strategies for speakers to choose for coping with planning phases in spontaneous speech (Wolk et al. 2021Wolk, Christoph, Sandra Götz, and Katja Jäschke 2021 “Possibilities and Drawbacks of Using an Online Application for Semi-Automatic Corpus Analysis to Investigate Discourse Markers and Alternative Fluency Variables.” Corpus Pragmatics 5: 1–30. ). Learning how to use DMs can facilitate listeners’ interpretation of English. If speakers are able to produce DMs appropriately, the English that they produce will be more natural and fluent (Hoey 2002Hoey, Michael 2002 “Spoken Discourse.” In Macmillan English Dictionary for Advanced Learners of American English, ed. by Michael Rundell, LA16–LA17. Oxford: Macmillan Education.).

2.2Discourse markers in native and learner discourse

DMs have been investigated extensively and intensively in the speech of English NSs and learners across numerous L1s or cultural backgrounds, such as Chinese (Liao 2009Liao, Silvie 2009 “Variation in the Use of Discourse Markers by Chinese Teaching Assistants in the US.” Journal of Pragmatics 41 (7): 1313–1328. ), Dutch (Buysse 2012Buysse, Lieven 2012 “ So as a Multifunctional Discourse Marker in Native and Learner Speech.” Journal of Pragmatics 44 (13): 1764–1782. ), German (Müller 2005Müller, Simone 2005 Discourse Markers in Native and Non-Native English Discourse (Vol. 138). Amsterdam: John Benjamins. ; Götz 2013Götz, Sandra 2013 Fluency in Native and Non-Native English Speech. Amsterdam: John Benjamins. ), Spanish (Romero-Trillo 2002Romero-Trillo, Jesús 2002 “The Pragmatic Fossilization of Discourse Markers in Non-Native Speakers of English.” Journal of Pragmatics 34 (6): 769–784. ), Swedish (Aijmer 2011Aijmer, 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. ), French and Polish (Gilquin and Granger 2015Gilquin, Gaëtanelle, and Sylviane Granger 2015 “Learner Language.” In The Cambridge Handbook of English Corpus Linguistics, ed. by Douglas Biber and Rand Reppen, 418–435. Cambridge: Cambridge University Press. ) and Taiwanese (Lin 2016Lin, Yen-liang 2016 “Discourse Marking in Spoken Intercultural Communication between British and Taiwanese Adolescent Learners.” Pragmatics 26 (2): 221–245. ; Huang 2019Huang, Lan-fen 2019 “A Corpus-Based Exploration of the Discourse Marker Well in Spoken Interlanguage.” Language and Speech 62 (3): 570–593. ). Müller (2005)Müller, Simone 2005 Discourse Markers in Native and Non-Native English Discourse (Vol. 138). Amsterdam: John Benjamins. , for example, investigates the use of DMs so, like, well, and you know by NSs and German-speaking learners. It has been reported that overall in learner discourse DMs are under-represented,44.For language learning, it is reasonable to assume that the language produced by native speakers is taken as the target norm. The terms underuse and overuse are generally adopted by most learner corpus studies, implying that learners use a given target item too much or too little to sound like a native speaker. In studies of DMs, we would suggest as alternatives the neutral terms under- and over-representation for discussing differences in frequency across corpora. The underlying assumption of under- and over-representation is to keep frequency information as linguistic evidence in focus and avoid over-generalizing differences to learners’ performance; in particular, the use of DMs is contextually dependent and syntactically and semantically optional. with the exception of well. Aijmer (2011)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. , for example, reports an over-representation of well by Swedish learners in several different functional roles, with learners tending to use it as a fluency device mostly to cope with speech management problems and rarely to indicate attitude (e.g., by mitigating disagreement). Fung and Carter’s (2007Fung, Loretta, and Ronald Carter 2007 “Discourse Markers and Spoken English: Native and Learner Use in Pedagogic Settings.” Applied Linguistics 28 (3): 410–439. , 410) examination of learners in Hong Kong, based on a pedagogic sub-corpus from CANCODE, shows evidence that DMs serve as “useful interactional manoeuvres” to organize and structure speech on interpersonal, referential, structural and cognitive levels. Employing the same analytical scheme, Lin (2016)Lin, Yen-liang 2016 “Discourse Marking in Spoken Intercultural Communication between British and Taiwanese Adolescent Learners.” Pragmatics 26 (2): 221–245. examines the speech of British NSs and Taiwanese learners of English, based on a specialized corpus derived from an adolescent intercultural exchange program. Both studies show that NSs used a wider range of DMs for discourse pragmatic functions, whereas L2 learners’ use of DMs was more restricted. Although the above studies found significant differences between NSs and non-NSs, it remains unclear how the proficiency levels of learners influence the use of DMs. Learners’ proficiency levels remain a “fuzzy variable” in learner corpus research (Carlsen 2012Carlsen, Cecilie 2012 “Proficiency Level – A Fuzzy Variable in Computer Learner Corpora.” Applied Linguistics 33 (2): 161–183. ).

Several studies document the different uses of DMs across proficiency levels in L2 speech. Dumont (2018)Dumont, Amandine 2018 Fluency and Disfluency: A Corpus Study of Non-Native and Native Speaker (Dis)fluency Profiles. PhD thesis, Université Catholique de Louvain, Louvain-la-Neuve. finds that C1 learners use significantly more DMs than do B2 learners. Jones et al. (2018)Jones, Christian, Shelley Byrne, and Nicola Halenko 2018 Successful Spoken English: Findings from Learner Corpora. Oxon: Routledge. show that more proficient learners (equivalent to C1 in CEFR in the UCLan Speaking Test Corpus) used well more frequently than lower-level learners, whereas B1 learners employed the DM you know significantly more than B2 and C1 learners. Neary-Sundquist (2014)Neary-Sundquist, Colleen 2014 “The Use of Pragmatic Markers across Proficiency Levels in Second Language Speech.” Studies in Second Language Learning and Teaching 4 (4): 637–663. reports that DM use is positively correlated with proficiency; but even highly proficient learners could not reach native-like patterns of variation, and certain DMs, such as I think, were over-represented. Although some studies do not clarify whether a wider variety or higher frequency of DMs was predictive of higher proficiency levels (Wei 2011Wei, Ming 2011 “A Comparative Study of the Oral Proficiency of Chinese Learners of English across Task Functions: A Discourse Marker Perspective.” Foreign Language Annals 44 (4): 674–691. ), many studies have found that DMs are used more frequently by more advanced learners of a language (Müller 2005Müller, Simone 2005 Discourse Markers in Native and Non-Native English Discourse (Vol. 138). Amsterdam: John Benjamins. ; Hellermann and Vergun 2007Hellermann, John, and Andrea Vergun 2007 “Language Which Is Not Taught: The Discourse Marker Use of Beginning Adult Learners of English.” Journal of Pragmatics 39: 157–179. ) and the rate of DM use by advanced learners is similar to that by NSs (Neary-Sundquist 2014Neary-Sundquist, Colleen 2014 “The Use of Pragmatic Markers across Proficiency Levels in Second Language Speech.” Studies in Second Language Learning and Teaching 4 (4): 637–663. ). This implies that the use of DMs can be a mark of both NSs and successful users of English (Prodromou 2008Prodromou, Luke 2008 English as a Lingua Franca: A Corpus-Based Analysis. London: Continuum.). DMs indicate speakers’ membership “within cultural communities and project a ‘deep commonality’ amongst interlocutors” (O’Keeffe et al. 2007O’Keeffe, Anne, Michael McCarthy, and Ronald Carter 2007 From Corpus to Classroom: Language Use and Language Teaching. Cambridge: Cambridge University Press. , 76). It is further suggested that learners who wish to advance closer to near-native fluency should be exposed to and practice these distinctive features of spoken grammar (ibid.).

However, neither the levels of learners’ speaking proficiency nor the fluency levels of these studies are clearly or uniformly defined, making it difficult to compare them in understanding the development of DM use. Only Dumont (2018)Dumont, Amandine 2018 Fluency and Disfluency: A Corpus Study of Non-Native and Native Speaker (Dis)fluency Profiles. PhD thesis, Université Catholique de Louvain, Louvain-la-Neuve. and Jones et al. (2018)Jones, Christian, Shelley Byrne, and Nicola Halenko 2018 Successful Spoken English: Findings from Learner Corpora. Oxon: Routledge. used the CEFR to quantify learner levels, while other studies used a graded language course (Hellermann and Vergun 2007Hellermann, John, and Andrea Vergun 2007 “Language Which Is Not Taught: The Discourse Marker Use of Beginning Adult Learners of English.” Journal of Pragmatics 39: 157–179. ), investigator assessment (Fung and Carter 2007Fung, Loretta, and Ronald Carter 2007 “Discourse Markers and Spoken English: Native and Learner Use in Pedagogic Settings.” Applied Linguistics 28 (3): 410–439. ), a pre-study speaking proficiency test (Neary-Sundquist 2014Neary-Sundquist, Colleen 2014 “The Use of Pragmatic Markers across Proficiency Levels in Second Language Speech.” Studies in Second Language Learning and Teaching 4 (4): 637–663. ), or lacked a defined proficiency variable (Müller 2005Müller, Simone 2005 Discourse Markers in Native and Non-Native English Discourse (Vol. 138). Amsterdam: John Benjamins. ). In light of these ongoing discrepancies in the literature, the present study uses a multi-level learner corpus, evaluated post hoc, deploying CEFR fluency descriptors. It allows the results of developing DM use to be easily adopted in such practical contexts as language classrooms and assessments.

2.3Discourse markers and immersive experience

Socio-cultural integration and exposure to the target language environment have been shown to influence thinking, speaking and the acquisition of a new language. Previous L2 research has provided strong evidence that learners’ immersive experience with rich L2 input and opportunities for interaction in natural communicative contexts have a positive impact on the development of oral fluency. Mora and Valls‐Ferrer (2012)Mora, Joan C., and Margalida Valls-Ferrer 2012 “Oral Fluency, Accuracy, and Complexity in Formal Instruction and Study Abroad Learning Contexts.” TESOL Quarterly 46 (4): 610–641. , for example, investigated 30 learners at an upper-intermediate level (B2) who had failed to improve in fluency after six months of formal instruction in the participants’ home country, but after a three-month study-abroad program increased all measures of fluency other than accuracy and complexity. These observed fluency gains may have been due to the frequent employment of various fluency devices in speech, such as DMs.

Immersive experience may also promote the acquisition of DMs. Hellermann and Vergun (2007)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: 157–179. find that L2 learners who had more contact with NSs acquired three DMs (you know, well and like), whereas students with little or no use of DMs in their classroom talk all reported speaking their first language at least 50 per cent of the time spent outside the classroom. Liu (2016)Liu, Binmei 2016 “Effect of L2 Exposure: From a Perspective of Discourse Markers.” Applied Linguistics Review 7 (1): 73–98. notes that, for Chinese learners of English as a foreign language (EFL) who lived in the United States, both the increased exposure and increased socialization had significant positive effects on the frequency and variety of the DMs produced. Gilquin (2016)Gilquin, Gaëtanelle 2016 “Discourse Markers in L2 English: From Classroom to Naturalistic Input.” In New Approaches to English Linguistics: Building Bridges, ed. by Olga Timofeeva, Anne-Christine Gardner, Alpo Honkapohja, and Sarah Chevalier, 213–249. Amsterdam: John Benjamins. examines 554 EFL learners from the Louvain International Database of Spoken English Interlanguage (LINDSEI) and finds a general, significant increase in the use of DMs with the increased length of time spent studying abroad in an English-speaking country. Götz and Mukherjee (2018)Götz, Sandra, and Joybrato Mukherjee 2018 “Investigating the Effect of the Study Abroad Variable on Learner Output: A Pseudo-Longitudinal Study on Spoken German Learner English.” In Learner Corpus Research, ed. by Vaclav Brezina, and Lynne Flowerdew, 47–65. London: Bloomsbury. report a positive significant effect on the use of DMs of immersive experiences lasting more than one year, showing how the duration of English instruction and a period of stay abroad lead to an increased use of DMs and a reduced use of (un)filled pauses, resulting in speech that is more fluent. These studies demonstrate how increased exposure and socio-cultural integration through a study-abroad experience have measurable effects on the L2 development of DMs.

Although the above studies show a relationship between immersive experience, fluency and the use of DMs in learner groups of a particular level of L1, the present study proposes to further investigate the effect of immersive experience on learners across four CEFR fluency levels.

2.4Focal discourse markers

A wide variety of words or multiword units considered as DMs have been investigated, but some may be taken as uncontroversial and are classified as central DMs, such as well, you know and like. These three DMs are the most frequently used and have been extensively selected for analysis in both native and non-NS corpora (e.g., Müller 2005Müller, Simone 2005 Discourse Markers in Native and Non-Native English Discourse (Vol. 138). Amsterdam: John Benjamins. ; Hellermann and Vergun 2007Hellermann, John, and Andrea Vergun 2007 “Language Which Is Not Taught: The Discourse Marker Use of Beginning Adult Learners of English.” Journal of Pragmatics 39: 157–179. ; Polat 2011Polat, Brittany 2011 “Investigating Acquisition of Discourse Markers through a Developmental Learner Corpus.” Journal of Pragmatics 43: 3745–3756. ; Götz 2013Götz, Sandra 2013 Fluency in Native and Non-Native English Speech. Amsterdam: John Benjamins. ; Dumont 2018Dumont, Amandine 2018 Fluency and Disfluency: A Corpus Study of Non-Native and Native Speaker (Dis)fluency Profiles. PhD thesis, Université Catholique de Louvain, Louvain-la-Neuve.). Making the distinction between non-DM use and DM use of well and like is straightforward, by referring to the parts of speech that they represent. When analyzing you know, syntactical necessity was the criterion.

DM well serves multiple functions in spontaneous speech; one of its major functions is to indicate that the speaker is thinking about things (Carter and McCarthy 2006Carter, Ronald, and Michael McCarthy 2006 Cambridge Grammar of English. Cambridge: Cambridge University Press.). Aijmer (2011Aijmer, 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. , 235) describes it as “primarily a ‘mental state’ interjection”, which can be associated with the speaker’s deliberation. Biber et al. (1999Biber, Douglas, Edward Finegan, Stig Johansson, Susan Conrad, and Geoffrey Leech 1999 Longman Grammar of Spoken and Written English. Essex: Pearson Education Limited., 1086) also state that well “appears to have the general function of a ‘deliberation signal’, indicating the speaker’s need to give (brief) thought or consideration to the point at issue”. In such cases, the use of well can provide cognitive benefits that allow speakers to buy time for planning, processing and searching for alternative expressions. As Fung and Carter (2007)Fung, Loretta, and Ronald Carter 2007 “Discourse Markers and Spoken English: Native and Learner Use in Pedagogic Settings.” Applied Linguistics 28 (3): 410–439. note, DM well also serves the interpersonal function of indicating the speaker’s attitude and the structural function of indicating the shift to a new topic. It can be taken as a mitigator to soften disagreements, dispreferrred points, unexpected answers, etc. Some studies have reported an over-representation of well by French-speaking (Gilquin 2008Gilquin, Gaëtanelle 2008 “Hesitation Markers among EFL Learners: Pragmatic Deficiency or Difference?” In Pragmatics and Corpus Linguistics: A Mutualistic Entente, ed. by Jesús Romero-Trillo, 119–149. Berlin, Heidelberg and New York: Mouton de Gruyter.) and Swedish-speaking (Aijmer 2011Aijmer, 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. ) learners, compared to their native-speaking counterparts, while others report an under-representation of DM well, for instance by Chinese-speaking learners from Hong Kong (Fung and Carter 2007Fung, Loretta, and Ronald Carter 2007 “Discourse Markers and Spoken English: Native and Learner Use in Pedagogic Settings.” Applied Linguistics 28 (3): 410–439. ) and Taiwan (Lin 2016Lin, Yen-liang 2016 “Discourse Marking in Spoken Intercultural Communication between British and Taiwanese Adolescent Learners.” Pragmatics 26 (2): 221–245. ; Huang 2019Huang, Lan-fen 2019 “A Corpus-Based Exploration of the Discourse Marker Well in Spoken Interlanguage.” Language and Speech 62 (3): 570–593. ).

You know is commonly considered an interpersonal DM, signaling that speakers are sensitive to the needs of their listeners and are monitoring the state of shared knowledge in the conversation (O’Keeffe et al. 2007O’Keeffe, Anne, Michael McCarthy, and Ronald Carter 2007 From Corpus to Classroom: Language Use and Language Teaching. Cambridge: Cambridge University Press. ). But, while you know functions as an interpersonal DM, it may not always be the case that speakers and hearers have shared knowledge. Speakers occasionally use it for reformulating, repairing and exemplifying in a way that may replace pauses and disfluency and provide a coherence function in discourse (Polat 2011Polat, Brittany 2011 “Investigating Acquisition of Discourse Markers through a Developmental Learner Corpus.” Journal of Pragmatics 43: 3745–3756. ; Lin 2016Lin, Yen-liang 2016 “Discourse Marking in Spoken Intercultural Communication between British and Taiwanese Adolescent Learners.” Pragmatics 26 (2): 221–245. ). It can also be used to launch a new topic (O’Keeffe et al. 2007O’Keeffe, Anne, Michael McCarthy, and Ronald Carter 2007 From Corpus to Classroom: Language Use and Language Teaching. Cambridge: Cambridge University Press. ) and to highlight a particular point in an utterance (Fox Tree and Schrock, 2002Fox Tree, Jean E., and Josef C. Schrock 2002 “Basic Meanings of You Know and I Mean .” Journal of Pragmatics 34 (6): 727–747. ). However, House (2009)House, Juliane 2009 “Subjectivity in English as Lingua Franca Discourse: The Case of You Know .” Intercultural Pragmatics 6 (2): 171–193. argues that the functional use of you know by EFL learners and NSs is markedly different in that EFL speakers use you know predominantly as a self-serving strategy to improve coherence rather than inviting addressee inferences or cooperating with their interlocutors.

Like has been reported as the most prevalent DM in casual spoken interaction (Lin 2016Lin, Yen-liang 2016 “Discourse Marking in Spoken Intercultural Communication between British and Taiwanese Adolescent Learners.” Pragmatics 26 (2): 221–245. ). The DM like in spoken discourse is interpreted as serving several functions. One of the most frequent is to preface new information (Fuller 2003Fuller, Janet M. 2003 “Use of the Discourse Marker Like in Interviews.” Journal of Sociolinguistics 7 (3): 365–377. ; Hellermann and Vergun 2007Hellermann, John, and Andrea Vergun 2007 “Language Which Is Not Taught: The Discourse Marker Use of Beginning Adult Learners of English.” Journal of Pragmatics 39: 157–179. ). Studies have identified significant differences in the use of like between native and non-NSs, and have identified like as having the greatest disparity in usage between these two groups (Müller 2005Müller, Simone 2005 Discourse Markers in Native and Non-Native English Discourse (Vol. 138). Amsterdam: John Benjamins. ; Lin 2016Lin, Yen-liang 2016 “Discourse Marking in Spoken Intercultural Communication between British and Taiwanese Adolescent Learners.” Pragmatics 26 (2): 221–245. ). When like occurs with numeral expressions, it often serves as a vagueness marker, denoting the approximateness of the quantity and purposely suggesting uncertainty. O’Keeffe et al. (2007O’Keeffe, Anne, Michael McCarthy, and Ronald Carter 2007 From Corpus to Classroom: Language Use and Language Teaching. Cambridge: Cambridge University Press. , 177) state that “speakers frequently introduce approximators to downtone what might otherwise sound overly precise”. Like can also function as a filler, hesitation marker or discourse linking device, indicating the need for speech planning without giving up the floor (Polat 2011Polat, Brittany 2011 “Investigating Acquisition of Discourse Markers through a Developmental Learner Corpus.” Journal of Pragmatics 43: 3745–3756. ). Speakers search for the content or appropriate lexical information while thinking and speaking. This use of like occurs commonly with false starts, pauses and self-repairs, especially in language learners’ discourse.

4.1Corpus data under investigation

The learner corpus data were derived from 183 interviews, collected by two of the three authors. One hundred interviews were held with university English majors from the Czech (Gráf 2017Gráf, Tomáš 2017 “The Story of the Learner Corpus LINDSEI_CZ.” Studie z Aplikované Lingvistiky [Studies in Applied Linguistics] 8 (2): 22–35.) and Taiwanese (Huang 2014Huang, Lan-fen 2014 “Constructing the Taiwanese Component of the Louvain International Database of Spoken English Interlanguage (LINDSEI).” Taiwan Journal of TESOL 11 (1): 31–74.) sub-corpora of LINDSEI55.The first 11 sub-corpora were published in LINDSEI version 1 (Gilquin et al. 2010Gilquin, Gaëtanelle, Sylvie De Cock, and Sylviane Granger (eds.) 2010 LINDSEI Louvain International Database of Spoken English Interlanguage. Handbook and CD-ROM. Louvain-la-Neuve: Presses Universitaires de Louvain.). At the time of writing, there are 24 sub-corpora; see LINDSEI Partners on https://​uclouvain​.be​/en​/research​-institutes​/ilc​/cecl​/lindsei​-partners​.html. One of the criteria for selecting eligible participants was majoring in English. (Gilquin et al. 2010Gilquin, Gaëtanelle, Sylvie De Cock, and Sylviane Granger (eds.) 2010 LINDSEI Louvain International Database of Spoken English Interlanguage. Handbook and CD-ROM. Louvain-la-Neuve: Presses Universitaires de Louvain.). To expand our learner corpus data to lower proficiency levels, 83 interviews came from the supplemented version of LINDSEI (see Huang and Gráf 2021Huang, Lan-fen, and Tomáš Gráf (2021) “Expanding LINDSEI to Spoken Learner English from Several L1s across CEFR Levels.” Corpora 16 (2): 271–285. for more detail), which were collected in various university departments (English, Chinese, Business Management, Financial Management, International Trade, Tourism Management and Information Technology) in Taiwan and Finland. The interviewees’ ages ranged between 19 and 26 years, with an average age of 22.5 years in the Czech sub-corpus, 21.7 in the Taiwanese one and 20.7 in the supplementary data.

The LINDSEI interviews called for three major tasks. The first was a monologue on a set topic. There were three set topics in LINDSEI: (a) An experience you have had which has taught you an important lesson; (b) A country you have visited which has impressed you; (c) A film/play you’ve seen which you thought was particularly good/bad (Gilquin et al. 2010Gilquin, Gaëtanelle, Sylvie De Cock, and Sylviane Granger (eds.) 2010 LINDSEI Louvain International Database of Spoken English Interlanguage. Handbook and CD-ROM. Louvain-la-Neuve: Presses Universitaires de Louvain., 8). The participants chose one of the set topics. The lower-level students in the supplemented corpus were given ten simple topics (hobbies, school/major, daily routine, plans, family, a person you admire, good friends, favorite food, leisure activities and travel experience) and were asked to talk about three of them. The second task was a dialogue about topics of general interest. The third task was to reconstruct a narrative on the basis of four sequential pictures. Teachers of English conducted the interviews and the learners participated voluntarily. Each interview lasted approximately 15 minutes.

The 183 learners were grouped into six fluency levels according to the aural evaluations of their sample performances by trained examiners. Learners’ performances under the headings of range, fluency, accuracy, phonological control and coherence were rated according to the descriptors of CEFR (Council of Europe 2018Council of Europe 2018 Common European Framework of Reference for Languages: Learning, Teaching, Assessment. Companion Volume with New Descriptors. Strasbourg Cedex: Council of Europe., 171–172). The 100 learners in LINDSEI-Czech and LINDSEI-Taiwanese were assessed independently by two qualified Cambridge IELTS examiners, who had previously been trained in CEFR rater standardization. The rating results from the two raters correlated closely with each other (ρ = .893), as did the rating of the 83 supplementary corpus items (ρ = .84).66. Huang et al. (2018)Huang, Lan-fen, Simon Kubelec, Nicole Keng, and Lung-hsun Hsu 2018 “Evaluating CEFR Rater Performance through the Analysis of Spoken Learner Corpora.” Language Testing in Asia 8 (14): 1–17. documented the details of rating LINDSEI-Czech and LINDSEI-Taiwanese. The rating of the supplementary data is reported in Huang and Gráf (2021)Huang, Lan-fen, and Tomáš Gráf (2021) “Expanding LINDSEI to Spoken Learner English from Several L1s across CEFR Levels.” Corpora 16 (2): 271–285. . Cases showing a discrepancy were sent to a third rater for adjudication. In the present study, the learners’ levels of fluency in CEFR were used for grouping because we were investigating the relationship between fluency and the use of DMs. The qualitative features of fluency operationalized in CEFR are presented in the Appendix.

As shown in Table 1, the post hoc assessment resulted in a division into six groups: A1 (n = 5), A2 (n = 23), B1 (n = 33), B2 (n = 69), C1 (n = 48) and C2 (n = 5). This study focused on the development of DMs from A2 to C1; the sample sizes of A1 and C2 were too small to represent these two levels and therefore they were not included in this study. To compare the use of DMs by speakers at the four different fluency levels (n = 173) with that of their native counterparts, 50 interviews with British university students from the Louvain Corpus of Native English Conversation (LOCNEC; De Cock 2004De Cock, Sylvie 2004 “Preferred Sequences of Words in NS and NNS Speech.” Belgian Journal of English Language and Literatures New Series 2: 225–246.) were also examined. The construction of the LOCNEC corpus followed the same structure as LINDSEI, making them directly comparable.

To investigate the relationship between the learners’ immersive experiences (whatever their form) and their use of DMs, the 173 learners were further grouped on the basis of experience: learners with no immersive experience in an English-speaking country (n = 101), and those with immersive experience (n = 72), ranging from 0.2 months to 167.8 months (SD = 22). Table 2 below presents the distribution of learners based on their stay-abroad experiences and fluency levels. The effect of this experience is analyzed and discussed in Section 5.3 below.

4.2Data analysis

The first part of the analysis examined the corpus data quantitatively to measure the overall frequencies of the three DMs (well, like and you know) across each fluency level of CEFR. The three DMs were retrieved with the Concord tool in WordSmith 7 (Scott 2016Scott, Mike 2016 WordSmith Tools (Version 7). Stroud: Lexical Analysis Software.) and one of the present authors manually disambiguated the instances between their discourse and non-discourse use, as defined in the Cambridge Grammar of English (Carter and McCarthy 2006Carter, Ronald, and Michael McCarthy 2006 Cambridge Grammar of English. Cambridge: Cambridge University Press.). The classification was then double-checked by a research assistant with a master’s degree in English language.

The frequencies of DMs were normalized as the number of instances per hundred words (phw). The resulting relative frequencies in the learner data were compared to those produced by their native counterparts. The relationships between the relative frequencies of DMs and learner fluency levels and between the speakers with/without immersive experiences and the duration of learners’ immersive experiences were then evaluated77.Spearman rank order correlation is used when one of the variables consists of non-parametric ranked data (e.g., CEFR fluency levels and speaker groups; Pallant 2011Pallant, Julie 2011 SPSS Survival Manual (4th ed.). Crows Nest NSW: Allen and Unwin.). using Spearman rank order correlation. A Kruskal-Wallis test88.A Shapiro-Wilk normality test (p < 0.05, except for well in the native-speaker group) showed that the frequencies were not normally distributed; therefore, a Kruskal-Wallis test, an alternative to ANOVA for non-parametric data, was conducted (Leech et al. 2005Leech, Nancy L., Karen C. Barrett, and George A. Morgan 2005 SPSS for Intermediate Statistics: Use and Interpretation (2nd ed.). Mahwah, NJ: Lawrence Erlbaum Associates.). and Dunn’s pair comparisons were then used to determine whether the differences were statistically significant.

The corpus methods allowed us to quantitatively analyze the distribution and developmental patterns of DMs across CEFR levels, but we were unable to explain adequately the observed use of the DMs in a particular speaker group unless we explored the co-texts. For this reason, we then conducted a largely qualitative immediate context analysis to exemplify and explain how DMs were produced by a speaker at a certain fluency level. All the instances of the three DMs were re-arranged to identify typical instances and their broader co-texts according to the immediate co-occurring items to the left and right of each DM.

5.1Overall frequencies of the three discourse markers

A total of 3,395 instances of three DMs was identified, comprising 1,280 instances of well, 853 of you know and 1,262 of like. The descriptive statistical information is presented in Table 3. On average, the learners from A2 to C1 levels produced 0.19, 0.27, 0.63 and 1.28 DMs phw. The 48 C1-level learners used them almost as often as their native counterparts did (1.3 DMs phw).

To yield a better visual display, the relative frequencies of the three DMs are presented in boxplots. Figure 1 shows that the use of DMs develops linearly with fluency levels. A Spearman’s correlation was run to determine the relationship between the relative frequencies of DMs and the fluency levels. Overall, there was a strong, positive correlation (r_s (223) = 0.65, p < 0.0001). The relationship between the relative frequencies of individual DM and fluency levels was also strongly positive for well (r_s (223) = 0.674, p < 0.0001) and you know (r_s (223) = 0.576, p < 0.0001), but weakly positive99.The strength of the correlation adopts the guide that Cohen (1988Cohen, Jacob 1988 Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Hillsdale, NJ: Lawrence Earlbaum Associates., 79–81) suggests for the absolute value of correlation coefficients: 0.10 to 0.29 ‘small’; 0.30 to 0.49 ‘medium’; 0.50 to 1.0 ‘large’. for like (r_s (223) = 0.243, p < 0.0001).

In Figure 1, the boxplots of the advanced C1 level are broadly similar to that of the native counterparts. The medians for C1 and native counterparts are 1.04 and 1.13 instances of DMs phw, respectively. In the native data, four outliers are shown. Since the native counterparts were taken as a benchmark and the frequencies of the DMs used by the high users show the phenomena of natural speech, we decided not to reject the outliers. In the learner groups, one outlier in C1, nine in B2 and two in B1 were identified. A closer look at these outliers revealed their preference for a particular DM. The only outlier at C1 (FI112) produced 2.55, 0 and 0.7 instances phw of well, you know and like respectively. Six of the nine outliers in the B2 group were frequent users of like, of whom one (TW003) employed like only. In the B1 group, one outlier (TW040) was a high user of well and the other (TW106) was a high user of you know. It is not easy from the current data to explain why a speaker uses a given DM more frequently than other DMs. Possible interpretations could be the speakers’ idiosyncrasies, or perhaps that the immediate contexts where the DMs occurred required their use.

To test if there was a statistically significant difference between speaker groups, the data from learners at four levels (A2, B1, B2 and C1) and those of their native counterparts were used to run a Kruskal-Wallis test. It showed that the frequencies of DMs had a significant, strong1010. Lomax and Hahs-Vaughn (2012)Lomax, Richard G., and Debbie L. Hahs-Vaughn 2012 Statistical Concepts – A Second Course (4th ed.). New York: Routledge. suggest that effect size values measured with epsilon squared can be interpreted similarly to those of eta squared; therefore, a value of 0.01 is considered a small effect, 0.06 a medium effect and 0.14 a large effect (Cohen 1988Cohen, Jacob 1988 Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Hillsdale, NJ: Lawrence Earlbaum Associates., 284–287). effect on the levels of perceived fluency, χ ²(4, n = 223) = 99.065, p < 0.0001, ε ² = 0.446. Dunn’s post hoc tests were then conducted on each pair of groups.1111.In order to preserve a family-wise 0.05 significance level, we applied the Bonferroni adjustment by dividing the alpha equally across the ten tests (Pallant 2011Pallant, Julie 2011 SPSS Survival Manual (4th ed.). Crows Nest NSW: Allen and Unwin.; Tabachnick and Fidell 2012Tabachnick, Barbara G., and Linda S. Fidell 2012 Using Multivariate Statistics (6th ed.). Boston: Pearson.). It was found that the A2 group was significantly different from the C1 (p < 0.0001) and native counterpart groups (p < 0.0001), but not from the B1 (p = 1) and B2 (p = 0.064) groups. A statistically significant relationship was found between the B1 and B2 groups (p < 0.043), between the B1 and C1 (p < 0.0001), between the B2 and C1 (p < 0.0001) and between the B2 and the native counterpart groups (p < 0.0001). As suggested earlier, the C1 learners performed similarly to their native counterparts. The statistically significant difference did not lie in the difference between these two groups (p = 1).

5.2Three focal discourse markers in learner and native groups

Almost all the 50 British native university students under investigation used DMs well and you know and only 9 (18%) of them did not use DM like (see Table 4). All of the C1 speakers were users of DMs, which means that their speech was closer to native norms. The percentages of non-users of the three DMs at A2 and B1 were higher (39% and 42% respectively) than those in the B2, C1 and native data (13%, 0% and 0% respectively), suggesting that these three common DMs may develop in line with speakers’ fluency levels. This trend was particularly marked in the use of DMs well and you know (see Figure 2). Their frequencies started to increase from B1 to B2. The use of DM like appeared distinct from the use made of the other two DMs. Nearly half of the speakers at A2 and B1 and four-fifths at B2 and C1 used like as a DM.

5.2.1Discourse marker well

Among the three DMs for analysis, well is the only one used by over 90% of the advanced C1 learners and native counterparts, while only approximately one third of the B2 learners (38%) use it. The high frequency of the use of well is thus a characteristic feature of the more fluent C1 speakers, who may have adopted it as one of their strategies for maintaining fluency.

In the extract, the mark-up <A> refers to the turn produced by the interviewer and <B> to that of the learner. In order to discuss possible reasons why well is used and demonstrate how the use of well relates to fluency, we selected a learner at C1. In Example (1) below, the first instance produced by Speaker A follows a disagreement with the other speaker. Speaker B, a C1 speaker, is one of the outliers (i.e., a user who used DMs relatively frequently) cited in Figure 1 above. The second example may act as a “frame” for introducing clarifications or for self-repairs (Svartvik 1980Svartvik, Jan 1980 “ Well in Conversation.” In Studies in English Linguistics for Randolph Quirk, ed. by Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik, 167–177. London: Longman., 175). The third instance, which follows a short silence, might indicate Speaker B’s attempt to compensate for disfluency.

(1)

Example of DM well at fluency level C1 (File: FI112, LINDSEI-supplemented)

<A> I really want to learn that’s my goal in spring is it difficult </A>

<B> I don’t think so </B>

<A> well ⁽¹⁾ you started at two </A>

<B> it’s (eh) yeah . well ⁽²⁾ it’s difficult if you want to be good at it . but </B>

<A> (mm) </A>

<B> I think you can just be an amateur and try and just . well ⁽³⁾ slide down you’re not gonna be good at it at first but </B>

Based on the qualitative features of C1, “only a conceptually difficult subject can hinder a natural, smooth flow of language” (Council of Europe 2018Council of Europe 2018 Common European Framework of Reference for Languages: Learning, Teaching, Assessment. Companion Volume with New Descriptors. Strasbourg Cedex: Council of Europe., 171–172). Speaker B is able to speak fluently and appropriately employs DM well – when reformulation is needed to perform pragmatic functions rather than producing long pauses, which would probably have resulted in an impression of disfluency. If the instances of well in Example (1) above had been taken out, the speakers would also have been able to get their messages across, but they would have produced (un)filled pauses and lacked the pragmatic functions that well serves.

5.2.2Discourse marker you know;

In the learner data a linear developmental pattern can be seen. The proportions of non-users declined from 87% at A2 to 43% at C1. In Example (2), Speaker B at B2 produced three instances of you know. The first one co-occurred with an intensifier, actually, signaling the introduction of key information. The second instance prefaced a re-start and the third one follows the vague item kind of. These two instances may indicate that the speaker is searching for content or appropriate words. This use of you know, not surprisingly, is common in learner data because learners need more time to formulate what they say in a foreign language (Polat 2011Polat, Brittany 2011 “Investigating Acquisition of Discourse Markers through a Developmental Learner Corpus.” Journal of Pragmatics 43: 3745–3756. ). Other evidence suggesting disfluency was seen in the hesitation markers (eh, mm, er and em, underlined) and silent pauses (transcribed into one, two and three periods). In the speech produced by a B2 speaker, “there are few noticeably long pauses” (Council of Europe 2018Council of Europe 2018 Common European Framework of Reference for Languages: Learning, Teaching, Assessment. Companion Volume with New Descriptors. Strasbourg Cedex: Council of Europe., 171–172, see Appendix). If Speaker B had not employed DMs, there would have been more evident pausing, which could possibly have downgraded Speaker B’s fluency level to B1.

(2)

Example of DM you know at fluency level B2 (File: TW035, LINDSEI-Taiwanese)

<B> in Korea . well . Korea (eh) re= (eh) really really . I mean . polite </B>

<A> the Korean people right </A>

<B> yeah . (eh) . (mm) well . (er) superficially they are polite but actually you know ⁽¹⁾ <overlap /> <starts laughing> it’s another way <stops laughing> it’s another </B>

<A> <overlap /> <laughs> oh okay </A>

<B> you know ⁽²⁾ another .. things . (em) </B>

<A> were they friendly . do you think they were friendly to you </A>

<B> they are friendly to me cos they are kind of you know ⁽³⁾ association . to: greet foreign students <overlap /> so: .. it’s the job </B>

Like DM well, you know can be employed as a compensatory filler. This function could be found in the speech of nearly half the B2 learners, while most of the learners below B2 could have used silent pauses and such filled pauses as er and mm to serve the same function. The latter group was more likely to be seen as disfluent, distinct from the B2 speakers. In CEFR, the fluency at B1 featured evident pausing (see Appendix).

5.2.3Discourse marker like

Of the three prominent DMs under investigation, well and you know were used by all 50 British native university students in LOCNEC, although 9 (18%) of them did not produce like as a DM. In terms of relative frequencies, like was least frequently used (0.33 instances phw), compared to well (0.5 instances phw) and you know (0.47 instances phw) by the native group. In contrast, the analysis of learner data shows that like was most often used by learners at all four levels. The proportion of its users in each learner group was also higher than for the other two DMs. 48% of A2 and 48% of B1, 78% of B2 and 79% of C1 used like as a DM. It was also found that the relationship between the relative frequencies and fluency levels was weak. In other words, compared to well and you know, like was the least likely to be a distinguishing feature between fluency levels. It is therefore interesting to examine how lower-level speakers use like.

Of the three DMs, like was the most popular among the 23 A2 learners, producing 0.14 instances phw of like (see Table 3 above). Like often co-occurred with (un)filled pauses and false starts (73% of the 22 instances), as demonstrated in Example (3) by an A2 learner, and was thus clearly due to hesitation, exemplifying the qualitative features of fluency at A2, where a speaker “can make him/herself understood in very short utterances, even though pauses, false starts and reformulation are very evident” (Council of Europe 2018Council of Europe 2018 Common European Framework of Reference for Languages: Learning, Teaching, Assessment. Companion Volume with New Descriptors. Strasbourg Cedex: Council of Europe., 171–172, see Appendix). Although the development of like starts early at A2 (see Figure 2 above) compared to well and you know, Example (3) shows A2-level learners’ excessive use of like as a hesitation marker, which may be evaluated to signify a lower fluency level.

(3)

Example of DM like at fluency level A2 (File: JP101, LINDSEI-supplemented)

<A> (mhm) okay (em) how is your Chinese .. ho= how is your Chinese </A>

<B> (erm) <laughs> .. (em) .. (er) I studied Chinese .. four year </B>

<A> (mhm) </A>

<B> (er) . (erm) . I . can speak .. (erm) . like like . conversation . and . I can listen the class Chinese class </B>

<A> okay . okay so it’s good </A>

The empirical corpus data show that speakers at higher fluency levels use more DMs to enhance fluency as well as for their pragmatic functions. The pragmatic functions of the DMs are discussed in the above examples in their immediate co-texts, demonstrating the contribution of DMs in the framework of fluency in CEFR.

5.3Use of discourse markers and learners’ immersive experience

In order to analyze the relationship between the use of DMs and learners’ immersive experience, the learners were divided into two groups: those with immersive experience in an English-speaking country (n = 72) and those without (n = 101; see Table 2 above for more detail). The native counterparts (n = 50) served as a benchmark. A Spearman’s correlation test indicated that the relationship between the relative frequencies of the three DMs as a grouped category and the three speaker groups was moderately positive (r_s (223) = 0.506, p < 0.0001). The same positive relationship was found individually between the three speaker groups and the use of well (r_s (223) = 0.54, p < 0.0001), you know (r_s (223) = 0.535, p < 0.0001) and like (r_s (223) = 0.196, p = 0.003).

Native students used more DMs overall and with greater within-group consistency than those with or without immersive experience (Figure 3). A Kruskal-Wallis test found that the relative frequency of DM use by any speaker group was significantly different from its use by the others (χ ²(2, n = 223) = 56.948, p < 0.0001, ε ² = 0.257). Dunn’s post hoc pairwise comparisons revealed significant differences between the three groups (the no-experience group and those who had had such experience, p < 0.0001; the no-experience and native groups, p < 0.0001; and the immersive experience and native groups, p = 0.006). These results suggest that immersive experience may positively influence the acquisition of DMs.

This study further analyzed each level individually between learners with and without immersive experience. The results revealed a significant difference at the B2 level, suggesting that B2 learners with immersive experience (Mdn = 0.55) produced significantly more DMs than those at the same level without immersive experience (Mdn = 0.18, U = 779.5, p = 0.004, r = 0.35).

6.1Developmental pattern of discourse markers across fluency levels in CEFR

Analysis of learner data reveals that the use of DMs positively correlated with fluency level. On the whole, it is reasonable to conclude that the use of DMs was indicative of perceived fluency. As mentioned earlier, instead of silent pauses and filled pauses (e.g., er and mm), DMs can be employed to increase fluency (Hedge 1993Hedge, Tricia 1993 “Key Concepts in ELT.” ELT Journal 47 (3): 275–277. ; Götz 2013Götz, Sandra 2013 Fluency in Native and Non-Native English Speech. Amsterdam: John Benjamins. ).

It is worth noting, however, that the use of DMs does not develop steadily with fluency levels. In the cases of well and you know, the breakthrough is made between B1 and B2, and then the maximum use is observed on attaining C1, when most learners adopt these two DMs as frequently as their native counterparts do. This phenomenon was previously reported in Neary-Sundquist (2014Neary-Sundquist, Colleen 2014 “The Use of Pragmatic Markers across Proficiency Levels in Second Language Speech.” Studies in Second Language Learning and Teaching 4 (4): 637–663. , 652), who identified a sudden rise in the frequency of a wider range of pragmatic markers “between Levels 5 and 6”, Level 6 being the highest. However, we cannot infer their equivalent levels in CEFR.

The use of DM like displays a different developmental pattern. It is least influenced by learners’ fluency levels, possibly because the frequency of like was found particularly high in A2 learners. They prefer to use like often with (un)filled pauses and false starts when a lexical gap or speech difficulty emerges, thus clearly indicating hesitation. This confirms the phenomenon that DM like might fulfill “potentially disfluent functions” by monitoring, punctuating and reformulating (Cribble 2017Crible, Ludivine 2017 “Discourse Markers and (Dis)fluency across Registers: A Contrastive Usage-based Study in English and French.” PhD diss., Université de Berne.).

While the fluency levels of the learner data in this study had been assigned on the basis of CEFR, previous studies (e.g., Neary-Sundquist 2014Neary-Sundquist, Colleen 2014 “The Use of Pragmatic Markers across Proficiency Levels in Second Language Speech.” Studies in Second Language Learning and Teaching 4 (4): 637–663. ) did not offer sufficient information on their learners’ proficiency or fluency levels. Although a few studies (e.g., Jones et al. 2018Jones, Christian, Shelley Byrne, and Nicola Halenko 2018 Successful Spoken English: Findings from Learner Corpora. Oxon: Routledge.) also adopted CEFR to describe this variable, the data collection methods and contexts differ from each other. These discrepancies make it problematic to compare the use of DMs across datasets.

6.2Effects of immersive experience on the use of discourse markers

This paper reports a strong and positive relationship between the frequencies of DMs in general and learners’ immersive experiences, suggesting that DMs can be acquired incidentally when learners are given the chance to stay in a country where the target language is used, a finding that is in line with previous studies (e.g., Polat 2011Polat, Brittany 2011 “Investigating Acquisition of Discourse Markers through a Developmental Learner Corpus.” Journal of Pragmatics 43: 3745–3756. ; Gilquin 2016Gilquin, Gaëtanelle 2016 “Discourse Markers in L2 English: From Classroom to Naturalistic Input.” In New Approaches to English Linguistics: Building Bridges, ed. by Olga Timofeeva, Anne-Christine Gardner, Alpo Honkapohja, and Sarah Chevalier, 213–249. Amsterdam: John Benjamins. ; Liu 2016Liu, Binmei 2016 “Effect of L2 Exposure: From a Perspective of Discourse Markers.” Applied Linguistics Review 7 (1): 73–98. ; Götz and Mukherjee 2018Götz, Sandra, and Joybrato Mukherjee 2018 “Investigating the Effect of the Study Abroad Variable on Learner Output: A Pseudo-Longitudinal Study on Spoken German Learner English.” In Learner Corpus Research, ed. by Vaclav Brezina, and Lynne Flowerdew, 47–65. London: Bloomsbury.). Such cultural and linguistic immersion may provide L2 learners with naturalistic input, increased opportunities for social interaction and repeated exposure to the DMs used in naturally-occurring contexts, which may serve as a driver of L2 DM development and further enhance oral fluency. Although Götz and Mukherjee (2018)Götz, Sandra, and Joybrato Mukherjee 2018 “Investigating the Effect of the Study Abroad Variable on Learner Output: A Pseudo-Longitudinal Study on Spoken German Learner English.” In Learner Corpus Research, ed. by Vaclav Brezina, and Lynne Flowerdew, 47–65. London: Bloomsbury. report that a significant positive effect on the use of DMs was found only when the immersive experience continued for more than one year, this study finds a positive relationship after a shorter interval (average 9.25 months) on learners across CEFR levels.

The DMs you know and well, in particular, were found to be influenced by learners’ immersive experience. This may perhaps be explained by the interpersonal nature of these DMs. For example, you know is commonly considered an interpersonal DM, signaling that speakers are sensitive to the needs of their listeners and are monitoring the state of shared knowledge in the conversation (O’Keeffe et al. 2007O’Keeffe, Anne, Michael McCarthy, and Ronald Carter 2007 From Corpus to Classroom: Language Use and Language Teaching. Cambridge: Cambridge University Press. ). Exposure to the target language in its natural environment encourages learners to understand and use the language for these interpersonal purposes, which spurs the development of native-like patterns of language use (Liu 2016Liu, Binmei 2016 “Effect of L2 Exposure: From a Perspective of Discourse Markers.” Applied Linguistics Review 7 (1): 73–98. ). Similarly, well as an interpersonal DM is often under-represented in EFL learner speech due to the lack of both exposure in authentic contexts and social interaction (Huang 2019Huang, Lan-fen 2019 “A Corpus-Based Exploration of the Discourse Marker Well in Spoken Interlanguage.” Language and Speech 62 (3): 570–593. ). The present study finds that well can be developed rather sooner in the natural exposure of L2 learners, which is consistent with Liu (2016)Liu, Binmei 2016 “Effect of L2 Exposure: From a Perspective of Discourse Markers.” Applied Linguistics Review 7 (1): 73–98. .