7.1 Introduction

In the previous chapter, we presented a bottom-up analysis of the major communicative functions across the turns and discourse units of the spoken BNC 2014. This followed on from a similar study of the TLC L1 and TLC corpora earlier in the book. These analyses together bring us to the point where we can explore to what extent those functions are shared across different contexts of production—that is, our corpora. Our focus is on macro-structures in discourse, hence we will explore the discourse unit view. However, as the analysis proceeds, units below that level will, once again, be considered.

Table 7.1 shows, at the discourse unit level, distinct functions from the analyses presented in Chapters 2–6. It then shows which corpora have discourse units with that function attested by the short-text MDA. The dimension (D) and polarity (+/−) which we deem to identify that function is shown after the mention of each corpus. This allows us to see where, across the dimensions of the different corpora, produced in different contexts, the same function occurs. The third column records how many corpora the discourse function was shared across, with the corresponding types of language use displayed in the final column.

To what extent do the productions of the L2 speakers in the data approximate, in terms of discourse macro-structures, the productions of L1 speakers? Given that the language elicited in the TLC and TLC L1 was gathered in a context where specific tasks were set, we might expect to find the language of the L1 speakers in the exam was a better fit to our L2 data rather than to the general conversational English in the BNC. This is the case. There are three core functions which are shared across all three corpora – Information Seeking, Informative and Instructive as well as Seeking and Encoding Stance. There are a further three functions shared across the L1 and L2 corpora gathered from the GESE exam – Informational Narrative, Irrealis and Realis. Finally, there are only two functions which are shared by the L1 speaker corpora: Discourse Management and Situation-Dependent Commentary. We can view these intersections in different ways – for example, if we consider the set of functions in the TLC corpus, of those ten functions, three are shared across both L1 corpora and three are shared with the TLC L1 – so most of the discourse functions in the TLC corpus are shared by L1 speakers either in the same communicative context or in general conversation. Of the remaining functions, two relate to length, meaning that only two functions as such, Persuasion and Descriptive and Affective, are associated only with discourse units involving the productions of L2 speakers.

In the sections that follow, we will consider each category in the Type column in Table 7.1, considering to what extent, if at all, they cast light on differences in performance based on context (i.e. exam versus non-exam performance) or proficiency (i.e. L1 versus L2 production).

7.2 L2 Exam-Only Functions

With regard to the L2 exam-only functions, it is tempting to ask why these are only produced in the L2 exam and why L1 speakers taking the same exam or in everyday conversations do not use them. To ask that question would, however, be to misunderstand the technique used to reveal these functions. The functions observed are those which occur often enough to be identifiable using the short-text MDA approach. It is perfectly possible that there are low numbers of examples of discourse units both in the Spoken BNC 2014 and the TLC L1 which might, plausibly, be categorised as Persuasion, for example. Consider the following example from the Spoken BNC 2014, taken from file SJG5:Footnote ¹

This could easily have been coded as Persuasion if that function had been salient to the degree that it emerged as one of the functions in the BNC data. The first utterance alone shares many of the features of the Persuasion function in the TLC analysis – contracted forms and predicative adjectives abound. The predicative adjectives are accompanied by be copulas. The prediction modal going to is swallowed up in the contraction gonna, and there is an indefinite pronoun present (the one). These are most of the features whose presence defines Persuasion in positive Dimension 5 of the TLC data. The features of that function are very clear across the turns of Speaker A in this example, where the speaker tries to persuade the hearer that their behaviour is unacceptable.

This example also points to a possible reason why this function is not frequent enough in the conversational BNC to emerge in the short-text MDA of the discourse units. Persuasion, as can be seen in this example, can be distinctly face threatening and in the example given is clearly so; this is not a polite attempt to change the hearer’s views. The speaker views the hearer’s behaviour as ‘not okay’, they complain about both ‘having to text’ them and that they do not know whether the hearer is where they ‘are supposed to be’. They are clear about the impact on them, it leaves them ‘not interested’, feeling that they ‘can’t rely on’ the hearer, and that the hearer is ‘making me out to be like a massive cowbag’. They also note that their being together ‘was meant to be fun’. This is clearly a very different use of Persuasion than that which we discussed in the TLC data in Chapter 4. There the pretext for the persuasion was part of the task, the speakers were not known to one another and, consequently, there were no personal overtones to the Persuasion. Yet if Persuasion in L1 conversation is linked to a face-threatening act like conflict, then one might see why L1 speakers use it less frequently – and that may also explain why conflict is so rare across all three datasets. In the BNC data, if we consider the top-down coding introduced using the Egbert et al. (Reference Egbert, Wizner, Keller, Biber, McEnery and Baker2021) scheme, there are only 209 discourse units marked as relating to conflict from a total of 24,549 discourse units in the data. In the TLC L1, of the 6,687 discourse units in the corpus, only 17 are coded as having a conflict function. Likewise, only 69 out of 21,148 discourse units in the TLC data are coded for conflict. If conflict was a typical context for Persuasion to be employed in L1 data, then we might have a credible hypothesis regarding why Persuasion does not occur frequently enough in the L1 data, whether that be the Spoken BNC 2014 or TLC L1, to emerge during the short-text MDA. Yet why does Persuasion occur in the TLC where conflict is also rare? The answer to that question is apparent if the discussion of Persuasion in Chapter 4 is considered. Persuasion is closely linked to the Conversation task – the examiner is presenting the student with a topic and the student is commenting on that topic, often trying to persuade the examiner to take a specific action or attitude. The examiner has no particular investment in the topic (the topics are prepared by the examining board); likewise, they have no ongoing relationship with the student. The student – who they do not know and will probably not meet again – is engaging in Persuasion for the sake of practising the function of Persuasion. It is artificial and any face threat implicit in Persuasion in naturalistic L1 interactions, like the one in Example 66, is not present. In other words, the L1 speaker in the exam context has no face invested in the attempt to persuade. Hence their replies are, as noted in Chapter 4, largely phatic; their goal is not to be persuaded or not persuaded, to win or lose an argument. Rather, their goal is to test the performance of the student at the micro- and macro-discourse level, and a student who performs the function of Persuasion might well be deemed to have succeeded. By contrast, in our example from the BNC, it is reasonable to assume that the speaker who is trying to persuade will measure their success by whether or not they change the mind and behaviours of the hearer. So, Persuasion is both artificially inflated in the TLC and divorced from something that we hypothesise is likely to be linked to Persuasion in L1 conversation – conflict. This artificial inflation is not a deficiency in the data, since, as a way of testing the student’s ability to perform the Persuasion function, the Conversation task clearly performs well. In terms of interpersonal relations, it is also successful in allowing that function to be tested without conflict being engaged in, which is both pedagogically and ethically the right choice in a high-stakes testing context.

An account of why the Descriptive and Affective function occurs in the L2 data is rather different though, and it highlights another way in which the TLC varies from the Spoken BNC 2014 and TLC L1. The discourse units in the TLC are co-constructed by an L1 and an L2 speaker. While a discourse unit in the other corpora might be initiated by either interlocutor in principle, in the TLC the interlocutors vary in this crucial respect. So, the use of a discourse unit might not indicate that an L2 speaker necessarily initiates such discourse units as part of their communicative repertoire. We see this clearly in the Descriptive and Affective function, as was shown in passing in Chapter 3. The Descriptive and Affective function is sustained in part by requests from the examiner for information. The examiner’s goal in doing this is to allow the student to demonstrate their mastery of the macro-structure, and the micro-structures that allow it to be realised, as shown clearly in the discussion of the TLC Dimension 2 where it was noted that there was a positive association between the use of this function and high attainment. Yet the function is typically examiner-controlled and is a form of elicitation. This elicitation seems to be absent from the L1 data analysis – while it is possible that a hearer might repeatedly prompt a speaker to produce a Descriptive and Affective discourse unit in L1 interaction, the evidence that we have from the short-text MDA of the discourse units in the two L1 corpora examined shows that this does not happen enough for the pattern to be salient in these corpora. This is almost certainly a point where proficiency plays a role. In the TLC, the Descriptive and Affective function is needed to elicit evidence of proficiency. In the TLC L1, it is not.

So, once again, the needs of the examination, and its level, lead to a function becoming, in effect, over-represented in the L2 corpus. However, this is not a criticism of the exam or of the corpus – there are good pedagogical grounds for using this function to explore the student’s ability to respond to questioning with the purpose of revealing their ability to express opinions and produce descriptions. Also, we should note, of course, that there may be other conversational contexts in which both this function, and arguably Persuasion, might be more frequent and, in the case of Persuasion, less problematic; for example in a shop, where a salesperson might use Persuasion when trying to close a sale. That is not a hypothesis we will explore here, though it is a useful reminder that our observations across the three corpora, while well motivated by the purpose of the Trinity exam, do not necessarily represent the intended context of use for the conversational skills acquired by the learners. We are also reminded that the interlocutors in the TLC in particular are distinct, as L2 speakers, in important ways from those in the TLC L1 and the spoken BNC 2014.

7.3 Exam-Specific Functions Shared by L1 and L2 Speakers

In the category of discourse functions present in the exam data, but not conversational English, one might expect discourse functions which are specific to the exam itself – functions involved in introducing tasks, for example. However, that is not what is present in this category.

There are three functions that are shared between L1 and L2 speakers but which do not appear in conversational L1. These are: Informational Narrative, Irrealis and Realis. Realis and Irrealis form a pair on the same dimension and, in both datasets, exam takers favour Irrealis in the Conversation task and Realis in the Discussion task. Given that both sets of speakers use the functions for these tasks, we can reasonably conclude that the task determines the function – while it might be tempting to conclude that the L2 speakers are simply reproducing discourse functions that they have been trained to produce for these tasks, there is more than that happening here. The L1 speakers taking the exam have had no such training, yet they use the same functions – functions which we can see are not particularly salient in conversational British English. Hence, even if the L2 speakers have been trained to produce these functions, we can reasonably say that they are doing so appropriately if the yardstick by which we judge them is L1 speaker performance in the same context. Additionally, we may also note that the repertoire of discourse functions that we can observe for the L1 speakers in the BNC is not a complete reflection of the range of discourse functions available to such speakers. What we see in the analysis of the BNC are functions which occur frequently enough to be observed using the short-text MDA technique, which we may assume are appropriate to the context in which the data was gathered.Footnote ² The range of contexts of use for spoken language goes beyond casual conversational interaction and it is clearly the case that there are contexts where L1 speakers acquire and use the Realis and Irrealis functions, otherwise the L1 speakers would not have produced them in the TLC L1. So the use of functions such as Realis and Irrealis, used by both the L1 and L2 speakers in the exam context, is clearly appropriate to the context, even though the use of such functions in casual conversation is not salient enough to be observed using our method.

A similar argument clearly applies to Informational Narrative, though here we might note that in Chapter 4 we saw evidence that the L2 speakers develop competence in the Informational Narrative function across grades. Given that the function is also used by L1 speakers in the exam, but not in our corpus of casual conversations, we can draw similar conclusions as those drawn for Irrealis and Realis, while also noting that the path to proficiency for the students is directing them towards a discourse function which is used by L1 speakers in this context.

Before moving on, it is worth pausing to consider the Informational Narrative function further. Of all of the labels assigned to discourse units in the study of the three corpora, varieties of narrative are notably common – the variants are Extended Narrative, Informational Narrative, Narrative and Opinionated Narrative. Only discourse unit functions related to the giving and receiving of information (such as Information Seeking and Informational Recounts) are as frequent and one of those, Informational Narrative, encompasses both narrative and information. While the conveying of information might be seen as an obvious locus for the exam, it is less clear that narrative is – students are asked for information as part of the Conversation and Discussion tasks, but they are less obviously asked to relate narratives. For the moment we note that Informational Narrative is used by both L1 and L2 speakers in the exam, but we will return to consider the use of the other narrative functions as we encounter them in the sections that follow.

7.4 L1-Only Exam Functions

The functions performed only by L1 exam takers are, perhaps, the most important for the purpose of considering how L2 and L1 performance vary within the permitted confines of the social context of the exam. The contrast with L1 conversational English is instructive as it shows that these functions are not ones which are typical of casual conversations between L1 speakers. Likewise, the contrast with the L2 speakers shows that there are functions which seem to lie beyond the proficiency of the L2 speakers in the TLC corpus.Footnote ³ This last point needs to be nuanced – the question of task may intervene here. While it is tempting to see the L1 exam-only functions as being diagnostic, in a sense, of proficiency, we should be mindful of the possibility that one task which the L1 speakers performed which is not present in the TLC at grades 6–8 – the Presentation task – may give rise to the presence of these functions. In the analysis that follows, we will consider the L1 to L2 contrast in our discussion of the functions, and conclude with a reflection on the contrast of the L1 exam data and L1 casual conversation.

One function which is present in the L1 exam data only, Affective, is arguably present within another function in the L2 exam data – Descriptive and Affective. This may be viewed in at least one of two ways – firstly, one might conclude that, as some feature of proficiency, the affective separates from description and becomes a distinct function of its own, able to be deployed to provide affective commentary alongside a number of other functions without being tied to description. Secondly, we might conclude that affect is being used to perform a distinct function – while the language employed is affective, it is not being used to provide an affective commentary on a specific function, as appears to be happening in the L2 function Descriptive and Affective. Rather, the affective language has a distinct function of its own that does not, in fact, lead to the modification of another function. If this is true, what we would not see is an affective discourse unit following, for example, an Extended Narrative to provide affective context for that narrative,Footnote ⁴ such as ‘And it was the worst experience of my life. It was a miserable time. I am glad it is over.’ Rather, we would see affective language being used in some other way. When we look at the data, it is the second interpretation that fits what we observe. The discourse units marked as being most strongly associated with the Affective function in the L1 data are dominated by examiner speech and are replete with affective language, but are also oriented towards discourse management and discourse pragmatics. The overall goal of the turns is to perform a task in guiding the discourse to perform what may be called facework, a key part of the politeness theory of Brown and Levinson (Reference Brown and Levinson1987). Politeness theory, and the role of face within it, is predicated on two basic assumptions that Brown and Levinson claim ‘all competent adult members of a society have (and know each other to have)’ (Brown and Levinson, Reference Brown and Levinson1987: 61). These are:

1. 1. ‘Face’, the public self-image that every member wants to claim for himself [sic], consisting in two related aspects:

   1. (a) Negative face: the basic claim to territories, personal preserves, rights to non-distraction – i.e., to freedom of action and freedom from imposition

   2. (b) Positive face: the positive consistent self-image or ‘personality’ (crucially including the desire that this self-image be appreciated and approved of) claimed by interactants.

2. 2. Certain rational capacities, in particular consistent modes of reasoning from ends to the means that will achieve those ends.Footnote ⁵

Consider the following example, from TLC L1 file 69, where the examiner is using the Affective function to, in essence, provide positive backchannels to the examinee to encourage them by reinforcing their positive face. In turn, the examinee is doing the same to the examiner. This discourse unit was produced early in the interaction before the student had performed any tasks. At this point in the exam, they have just had the format of the exam explained to them:

The student is using the Affective function to signal understanding and consent – the description of the tasks the examiner has outlined are ‘perfect’, while the examiner is backchannelling as part of the Affective function, either by simply producing supportive laughter or by signalling that the student’s response is a good one. The operations are mutually reinforcing of face – the student is happy with the examiner and vice versa. In a way we might also see this use of the Affective function as having a key role in the exam itself – it is used to mutually reinforce positive face and to signal that both parties are ready to proceed and appear happy to do so. Hence, the Affective function may be used to gatekeep the interaction itself. Where the gate is not passed, the function shifts – following a similar sequence in TLC L1 file 94, the student switches to Information Seeking as there is, in fact, something that is not clear to them. Thus, when the examiner asks them ‘would you like to start’, the student does not agree and instead initiates Information Seeking by asking the question ‘can I put a timer on’. So, while the Affective function is important in gatekeeping progress through the exam, the possibility always exists that either speaker may invoke other functions to halt progress, even if the Affective function has been used to show that the interaction may proceed. Note also that this facework is cooperative – facework ‘is both personal but also negotiated’ (O’Brien, Reference O’Brien, Romanowski and Bandura2019: 83).

The Affective function is also used for another aspect of facework – repair. Consider the following example, from TLC L1 file 63. In the previous discourse unit, the examiner (E) has incorrectly assumed that the examinee (S) is an academic. The examinee has noted that they are a senior construction project manager, leading to this exchange:

This relates back to an earlier discussion of the lack of conflict in the exam data – here is a good example of the Affective function being deployed to minimise a threat to positive face for both interlocutors, while performing a repair and avoiding conflict. The face threat for them both is rooted in a mistaken assumption about the examinee. The examinee challenges this assumption in a repair. In the subsequent interaction, both signal, by staying in the Affective function and emphasising positive affect, that conflict will not be the result of the mistake. The facework to restore positive face for both is led by the examiner, who makes comments which allow them to evaluate the repair made by the examinee and, in so doing, acknowledging their error in a context where both speakers are engaged in the expression of positive affect. Both speakers produce laughter, a behaviour which can mark exactly the type of ambiguity and tension present in this exchange (Holmes, Reference Holmes2020). Here, the tension arises from the error, the ambiguity relates to whether this will, or will not, prove to be a source of conflict. Humour ‘generally creates and maintains solidarity … and it may hedge or attenuate face threatening acts such as directives and negatively affective speech acts’ (Holmes, Reference Holmes2020: 179). Humour plays that role in this example. The Affective function is used to avoid conflict, to show agreement on the repair, and thereby to minimise loss of positive face for both. This is key, as repair by the hearer in such contexts is seen as the dispreferred option as it is face threatening (Lerner, Reference Lerner1996) while ‘repairing oneself might be viewed as less face-threatening − and consequently treated as such interactionally’ (Evnitskaya, Reference Evnitskaya, Kunitz, Markee and Sert2021: 172). While there are contexts in which this may not be the case – for example, where disagreement can be seen as a strategy for forming group identity (e.g. Georgakopoulou, Reference Georgakopolou2001) or in adversarial contexts, such as courtrooms, where face threats may be a successful rhetorical strategy (e.g. Grainger, Reference Grainger2018) – the exam context is one which seems to avoid conflict, as noted earlier. However, in this case the possibility of self-repair is negated by the examiner believing the examinee had a job other than the one that they actually have. Rather than accept the tacit loss of face that ignoring the error would entail, the examinee opts to repair the examiner’s turn. The Affective function then helps the speaker and hearer cooperate to mitigate the mutual face threat.

In both examples, of course, there is also negative face at work – the examiner is making an imposition upon the examinee, that is, by requiring them to carry out specific tasks. In the case of the TLC L1 corpus, that imposition is not as great as it is in the L2 data. In the L1 corpus the examinees have volunteered to do the exam and the test is, essentially, consequence-free for them. However, in the L2 data it may be the case that the examinee is required to take the exam (perhaps to demonstrate language competence to gain a visa or proceed in their education) and the power relationship between the examiner and the examinee is more markedly asymmetrical as there may be consequences if the student does not pass the exam. So, we may view the TLC L1 corpus as a low-stakes, low-imposition test, while the L2 corpus is, relatively speaking, a high-stakes, high-imposition test. However, in both cases there is a degree of imposition as the examiner is directing the behaviour of the examinee. While the degree of engagement with negative face may vary across the two corpora, its shadow is cast across both. We will return to negative face shortly.

For now, let us turn to consider whether task, facework and the Affective function interact in TLC L1 in a way that explains why this function is unique to the TLC L1. To do this, we looked at the 100 prototypical discourse units related to the Affective function in this corpus. In each case the discourse unit, and its preceding and following context, was assessed to determine whether it constituted facework. Our finding is that there does seem to be a strong link between facework and the Affective function in the prototypical examples; 71 out of the 100 discourse units examined relate to facework. Why might facework be a feature of the TLC L1 that is so salient? The function is not apparent in the TLC corpus. We hypothesise that two factors together militate in favour of the appearance of the Affective function, and its link to facework, in the TLC L1: power and imposition. To begin with power, it is worth noting again at this point that the imbalance of power between the examinee and examiner is not as great in the L1 as in the L2 corpus – in the L1 corpus the examinees are volunteers, and the test might, as noted, best be described as a ‘low-stakes’ test – there are no negative consequences if the student fails. With volunteers, issues of facework must be more to the fore – facework tries to ‘decrease the threat to or repair damage to face’ (Turnbull, Reference Turnbull2003: 214) and asking L1 speakers to take a test which will rate, and possibly from their perspective negatively evaluate, their spoken language is certainly a face threat. This threat is amplified as, crucially, face threats increase as the degree of imposition for a speaker increases; for example, the face threat to a student performing a task increases as the examiner makes more impositions upon them by correcting them, asking for further information, switching task, inter alia. That increase in imposition leads to an increased threat to face and, hence, more facework. In the TLC L1 corpus, the degree of imposition is amplified relative to the TLC – the examiner has less power to make the imposition, yet must make it. The examiner cannot rely on any threat to the student’s face being offset against an outcome that the student desires—that is, a good exam result. Accordingly, the imposition intrinsic to the exam is amplified by the more equitable power balance between the interlocutors in the corpus, militating in favour of facework in the TLC L1. In that corpus, facework comes to the fore, as the imposition is at its greatest in the TLC L1 compared to the TLC corpus. That, we hypothesise, is why we see the Affective function carrying the burden of facework in the TLC L1. The function is absent from the TLC as the imposition there is relatively lower. It is absent from the BNC Spoken BNC 2014 as the imposition in that data is not all pervasive, as it is in the TLC L1, where volunteers are being required to perform tasks that they normally would not undertake—that is, there is consistent high imposition upon them throughout the exam. While we may find isolated examples of facework in the Spoken BNC 2014, it is not drawn upon consistently enough to become salient as a function of the data.

Before moving on, we should note that one response to the finding regarding the Affective function and facework could be to relabel this function as ‘Facework’. However, to do so would be to ignore the twenty-nine examples in our sample that did not constitute facework. Thus, while the Affective function may permit facework, it is not coterminous with it. It can also cover discourse units in which affect is coded, either linguistically or paralinguistically, as in the following example from TLC L1 file 74:

In this exchange, affect is expressed (e.g. through laughter), but there is no management of face in the context of some potential disagreement. Affect here is not a response to the type of face threat that we have explored in our previous examples. So while affect is clearly present in this example, face threats are less clearly present, if they are present at all.

What of the second feature unique to L1 exam performance, Extended Narrative? The discourse units falling into this category caused an issue during the annotation process. The discourse unit is an annotation designed to organise interactive dialogue. The discourse units in this function were identified as distinct when they were annotated – they are strongly monologic. They are also monofunctional – they can either be viewed as repeated sequences of discourse units with the same function or as one, unusually large, discourse unit. Given that these are sometimes simply a handful of turns, with one speaker dominating, we decided to treat them as one discourse unit and marked them as monologues. To give an idea of the scale of the units in question, if we look at the 100 prototypical discourse units for Extended Narrative, their average length is 772.34 words. We can dismiss swiftly the hypothesis that these monologic outputs are typical of narratives – if we look at the 100 prototypical discourse units in the Narrative category of the Spoken BNC 2014, we find their average length is 61.58 words. So, in the L1 exam data, something quite distinct is happening with narrative, though only in certain circumstances – the Informational Narrative function in this corpus is not monologic and the average length of discourse unit in the 100 prototypical discourse units for this function is 79.17 words. So, what causes narrative to become extended within the L1 exam but not within conversational interaction between L1 speakers?

Our discussion of politeness so far is a useful framework to use to consider why Extended Narrative is absent as a function from the BNC data – holding the floor to produce an extended narrative is, in the context of interactive casual conversation, a substantial imposition. Accordingly, the conditions for the suspension of turn taking in conversation ‘(by other than sheer listener apathy) requires special techniques’ (Levinson, Reference Levinson1983: 323). One such device identified by Levinson is the story announcement – the speaker, in effect, signals a potential suspension of turn taking by announcing that they have a story to tell. Here is the beginning of such a sequence from BNC file S7HR:

This follows the classic sequence for seeking permission to suspend turn taking to produce narratives outlined by Levinson (Reference Levinson1983) – firstly the story announcement, secondly a response from the hearer licensing the suspension, and then thirdly the commencement of the story, which then unfolds over the following discourse units with the hearer’s role reduced largely to producing backchannels. While observable in the Spoken BNC 2014, this patterning is not the same as that which occurs in the Extended Narrative – this function in the L1 exam data produces lengthy narratives with minimal input from a hearer. The most obvious explanation for this is task and, more specifically, a task which has something functionally equivalent to a story announcement. If we look at the 100 prototypical Extended Narrative discourse units, can we see a task which is strongly related to this function? There are two − and both produce slightly different behaviours. The task which dominates the function of Extended Narrative is the Presentation task – 47 of the 100 examples examined are from this task. The next most frequent task in the sample is Discussion (twenty-three), Conversation (sixteen), Listening (eight) and the Interactive task (five).Footnote ⁶ It appears that it is the choice of narrative in the Presentation task which has made the link between long discourse units and narrative so salient that narrative has not merely registered as a function, but it appears as a function, Extended Narrative, in Dimension 1—the dimension which is the strongest in the data. This might be slightly more understandable if the Presentation task was framed in terms of storytelling. If that were the case, our exploration of examples would have hardly been necessary – the invitation from the examiner to tell a story could be viewed as a story announcement, which, as it is produced by the prospective hearer, also signals a consent to the suspension of turn taking. However, that is not what the Presentation task calls for – it is a topic-oriented presentation. The Trinity guidance explains to students that they will be given five minutes in which ‘the candidate delivers a formal presentation on a discursive topic of his/her choice which they have prepared before the exam’.Footnote ⁷ There is no clear entailment in the choice of a narrative function arising from the request that the student should produce a topic-led presentation. The guidance does not mention narratives or storytelling – the choice to use the affordance of narrative is one that the student makes. However, it is one of a number of ways that students may approach the task. Consider the following examples. In the first, a student (from file 11) begins a Presentation which is more clearly oriented to the Informative and Instructive function:

In the second, a student (from file 184) clearly signals the beginnings of the use of the Extended Narrative function:

What we might claim, looking at both examples, is that even though the first example is not a narrative, it does contain an announcement, akin to a story announcement, marking that the suspension of turn taking is about to take place. The examiner is waiting for this point, and, in both cases, they have clearly signalled this in prior discussion. Accordingly, the examiner becomes passive. So, while the two examples are different in terms of the functions that they use, they are equivalent in the sense that they represent an imposition on the hearer and require turn taking to be suspended. In both cases this is flagged to the hearer, though in each case hearer consent may be assumed as prior to the point the examiner had asked for such a sequence. So, there is a pragmatic equivalence between the two in terms of politeness and turn management.

In the Listening tasks, roles are reversed. In this task the examinee needs to sit and listen to passages read by the examiner. After the examiner has finished, the student is asked questions about what they have just heard. Some of these passages fall into the Extended Narrative function. In common with the examples in the Presentation, however, politeness still produces a sequence similar to the story announcement – the examiner explains that they are about to take the floor and produce an extended monologue and then seeks the student’s assent for that.

With the handful of examples from other tasks, we see that, in principle, an Extended Narrative discourse unit may appear anywhere, but with a low frequency. Indeed, it may possibly appear with low frequency in the two other corpora.Footnote ⁸ Yet it is in the Presentation, where speakers are given the floor, that they often choose narrative as a macro-structure for their presentation. This then causes the form-to-function relation revealed by the short-text MDA to become salient enough not only to make the feature observable, but also to propel it into being in the first dimension to be identified in the data.

The examples have also shown us how politeness, as expressed in discourse rather than at the lexical level, is central to the two functions which are unique, in terms of salience, to the L1 exam data.

7.5 L1 Exam- and Conversation-Only Functions

The discourse functions shared by both the L1 exam and Spoken BNC 2014 data point to functions which are part of the repertoire of L1 speakers engaged in conversation and of L1 speakers responding to tasks in the GESE exam. Two functions fall into this category – Discourse Management and Situation-Dependent Commentary.

Discourse Management is a salient function shared by both L1 corpora, occurring in the first dimension of the analysis for both. To explore the function, we will once again down-sample and look at the 100 prototypical discourse units for the relevant functions for each corpus.

In terms of length alone, Discourse Management could be seen as an evolution of the Short function from the L2 data. They are certainly similar as they both relate to short discourse units and appear in Dimension 1. If we look at the average length of the 100 prototypical examples of Discourse Management in the TLC L1 data, for example, it is just over 15 words. This is slightly longer than the Short function of the Trinity L2 data, where the average is just over twelve words. However, to sustain the argument that these two functions are linked, and that the shift from the Short function to the Discourse Management function is related to proficiency, a clearer view of the nature of such a shift is needed. To explore the functions, we looked at the 100 prototypical discourse units for the Short function in the L2 TLC data and the 100 prototypical discourse units for Discourse management in the L1 TLC data and the BNC. We consider what this revealed, next.

Considering the twenty most frequent words in our prototypical examples of the Discourse Management function in general L1 conversation and those in the L1 exam, we see similarities and differences. In terms of similarities, ten words are shared between the two including pronouns (I, you), backchannels (yeah), a filler (mm, er), articles and demonstratives (the, that), a conjunction (and), a paralinguistic feature (laughter), the third-person enciliticized form of the verb be (s) and a preposition (to). Many of these we would expect from what we know about the spoken register (see Biber et al., Reference Biber, Johansson, Leech, Conrad and Finegan2021), but it is interesting to see turn-management features present in the data, indicating the Discourse Management function of these short discourse units. These include backchannels which carry an evaluation (yeah, laughter), and another which may play that role or act as a way of holding the floor in discourse (er). When we look at n-grams, however, there is little overlap. In the Spoken BNC 2014 material we see n-grams which have a role in discourse, especially face management, for example signalling disagreement, such as it’s not (four examples) and face, like yeah, yeah, yeah (six examples). By contrast, n-gram sequences in the TLC L1 sample relate exclusively to politeness formulae – thank you very much (eleven examples), haha thank you (seven examples) and nice to meet you (six examples). There are also n-grams which seek to reinforce positive face, such as yeah yeah yeah and repeated sequences of the use of mm as a backchannel. So, while there are similarities between Discourse Management functions within the L1 corpora, there is also a clear difference pragmatically between them. The TLC L1 data is more preoccupied with the management of face while that seems to be a less salient concern of the Spoken BNC 2014 L1 conversational data. Again, imposition probably explains the difference. The social context in which the two is produced is an important factor. The TLC L1 context (an exam taken in a formal context) requires more management of examinee face than a conversational interaction such as that in the Spoken BNC 2014, where we have conversations between two people well known to one another talking in a domestic context. It is also notable that the Spoken BNC Discourse Management sequences are longer than those in the TLC L1 – while still short relative to the other functions evidenced in the BNC, at just over seventeen words per discourse unit, it is longer than the lengths of the discourse units in the Short function for the L2 TLC (fifteen) and the Discourse Management function of the TLC L1 (fifteen).

So, the two L1 corpora, while giving rise to Discourse Management functions which are similar, also show that such a function can flex to meet different conversational demands. Given this view of our Discourse Management functions, might there be an argument for relabelling the Short function of the L2 TLC data as Discourse Management? If we explore these discourse units and find that, when viewed through the optic of function and context, we see a function which is, in fact, similar to Discourse Management, we may hypothesise that, while variant to a degree, the Short function in the L2 data is Discourse Management too. This will be explored shortly.

Exploring the short discourse units in the Spoken BNC and the TLC L1 reveals the different pragmatic purposes of functionally equivalent discourse units. Although the discourse units are involved in Discourse Management, the nature of that management varies depending on the pragmatic demands of both contexts. In the Spoken BNC, the most common pragmatic purpose of Discourse Management is to introduce an evaluation as part of the conclusion of a topic of discussion which is then agreed upon by both parties in the conversation, as in the following example from file SNXL. Prior to this example, an argument from a church for a rent-free period has been presented. In this discourse unit it is evaluated, and agreement on the evaluation is achieved between speaker and hearer. In the following example, each turn is numbered to aid the discussion:

The speakers then go on to criticise the church more generally. In the example, Speaker A works through an evaluation of the church’s argument (turns 1, 3, 5 and 7), with agreement being signalled by Speaker B through the backchannel yeah (2, 4 and 6). So, within the macro-structure that the discourse unit represents, we have two other macro-structures which intertwine – one relating to evaluation, driven by Speaker A, and a second to concurrence, vocalised by Speaker B, but indicating a shared evaluation of the evaluation given by Speaker A. There is also a sense in which the discourse unit represents a point at which the discourse pivots – the Discourse Management is bringing one theme within the conversation to an end (the evaluation of the church) and establishing agreement as a prelude to proceeding to a more general discussion of people being dishonest to achieve their goals.

If we look at the Discourse Management function of the BNC and TLC L1, can we discern functions within it that intertwine? To explore this, we carried out a bottom-up coding of the top 100 prototypical discourse units for Discourse Management in the BNC and TLC L1. What we found in the two corpora were a series of macro-structures within the Discourse Management discourse units, many of which worked, either singly or together, to realise a pivot in discourse either from one general focus to another or, where they occurred at the end of a conversation (or task in the TLC L1) to signal that. We will call these substructures. Note that these substructures are coded to explain the flow of the discourse, not its propositional content. So, for example, an utterance such as I would like to thank you for that, occurring at the end of a conversation, would not be coded to reflect the transmission of thanks from speaker to hearer, it would be marked as concluding a conversational exchange. The Discourse Management substructures, which we may say represent a meso-structural level of organisation, found in the analysis of the BNC and TLC L1 prototypical discourse units for Discourse Management are:

Overall, these could be characterised as concluding a coherent stretch of discourse (Conclude), permitting discourse to proceed or otherwise (Concur, Disagree), enriching a stretch of discourse (Evaluate, Inform), initiating within the discourse or its context, entailing an imposition, which may be agreed to or be the focus of disagreement (Direct, Initiate), engaging in facework (Apology) or carrying out a meta-discoursal function (Repair).Footnote ⁹

Note that when undertaking the coding we were mindful that context and language interact – so, for example, concurrence with Direction may not be linguistic, but may be implied through both a failure to disagree and/or by the discourse proceeding in accordance with the direction. However, our coding focuses on linguistic realisations of the substructures identified. Also, when coding we found that Concur was pretty ubiquitous across the data. So, for ease of coding, we coded a maximum of two substructures per discourse unit, choosing only to mark Concur where only one other substructure appeared with it. This decision, we found, allowed us to mark all substructures that were not Concur, and to mark examples of Concur where it accompanied only one other substructure.

All substructures found in the Spoken BNC 2014 are present in the TLC L1. However, in the samples studied, the substructure Disagree was found in the TLC L1 but not the Spoken BNC 2014. Before looking at these substructures and their interaction in a little more detail let us return to the question of whether the Short function in the L2 data is, in fact, related to Discourse Management. To do this, we can consider the degree to which these substructures are present in the Short TLC corpus discourse units also. Carrying out the same coding on the top 100 Short TLC prototype utterances, we found Apology, Conclude, Concur, Direct, Evaluate, Inform, Initiate and Repair. In terms of meso-structures in Short, the TLC bears a striking resemblance to the Spoken BNC 2014 and TLC L1 corpora. No new meso-structures emerged from the TLC discourse units examined. Given the small-scale, qualitative nature of this investigation, we should be cautious about drawing substantial conclusions from our study, of course. However, we can say that, as far as the analysis goes, it is supportive of the argument that the Short function in the TLC is, in fact, Discourse Management. If we view Short through the lens of the meso-structures identified in the Discourse Management function of the TLC L1 and Spoken BNC 2014, then an analysis which views them as the same function may proceed smoothly.

But before we conclude our exploration of the Discourse Management function, and the Short function in the TLC, some consideration should be given to any apparent differences in the frequency of the use of the meso-structures and in their intertwining within a discourse unit. Table 7.2 shows, for each corpus, the frequency of each meso-structure and also the relative ranking of each substructure.

Once again, we need to be duly cautious about what a down-sampled set of examples like this can reveal. Nonetheless there are some comments worth making. The Conclude substructure is the most common meso-structure in Discourse Management across each of the corpora, suggesting that it is a common feature of interaction whether that be task-oriented or not.

The most glaring anomaly in the table is the ranking of Initiate in the Spoken BNC 2014 on the one hand (2nd) versus the exam-based corpora (5th and 6th). The rank and frequency of Initiate in the Spoken BNC 2014 is much higher than in the other two corpora. This is likely to be context-bound – in the context of casual conversation like that represented by the Spoken BNC 2014, any speaker within the conversation is, in principle, free to initiate. This is not the case in the other corpora, where a power imbalance gives the examiner much more scope to initiate than the examinee – in particular by initiating a new task or a shift within a task. If we look at the Initiate substructures within the sample studied from the TLC corpus and explore who begins an Initiate, we see that it is, in all eleven examples, the examiner, as in the following example, which presents a discourse unit where a Direct meso-structure occurs with Initiate. This is from the Conversation task of a grade 7 exam, undertaken by a student in Argentina, who was awarded an A for the task (file 2_7_AR_38):

The student in this example concurs with a direction from the examiner to stop and initiate a new discussion (utterance 1) and to direct it (3). The discourse unit also contains concurrence (utterances 2, 4 and 6), though this is not part of the coding of this utterance in our scheme, for reasons noted earlier.

In this case, the meso-structures are concurrent turns from a speaker and are realised at the uppermost micro-structural level (the turn). The discourse unit occurs at the very start of the Conversation task and clearly signals a shift in the discourse, controlled by the examiner, with the student showing compliance. In terms of power structures, unlike casual conversation amongst friends, the student never really had the license in the interaction to select this, or any other, Initiate meso-structure which would have ended one task and begun another. So, pragmatics here is constrained by the roles of the speakers in social context and that constraint, we would argue, is what we are seeing in Table 7.2 in the two examination corpora, relative to the conversational English corpus. While one speaker in the examination corpus is constrained in accessing Initiate, this is not true of the speakers in the Spoken BNC 2014, of necessity. This alone, we hypothesise, explains the difference in the rank of the Initiate function between these two types of corpora.

With regard to the intertwining of meso-structures, Table 7.2 also shows another difference. There are fewer meso-structures coded on the L2 data than on the L1 speaker data. As noted, in principle two meso-structures may appear per discourse unit in our coding. Hence the upper limit for the totals in Table 7.2 in terms of meso-structures per corpus is 200. For the L1 corpora, that is nearly the case – the appearance of a single meso-structure within a discourse unit seems rare. Yet in the L2 data there are forty-two such examples. If we explore these substructures which occur alone, what pattern emerges across the samples studied? In the Spoken BNC 2014, there are nine such meso-structures – Conclude (two), Concur (one), Direct (two), Evaluate (one) Inform (two) and Initiate (one). The patterning in the TLC L1 is similar, with fourteen single meso-structure discourse units – Conclude (three), Concur (one), Direct (two), Evaluate (one), Inform (five) and Initiate (two). However, with the TLC we see forty-two such units − Conclude (thirty-four), Direct (one), Evaluate (one), Initiate (one) and Inform (five). Unlike in the other two corpora, Conclude clearly dominates in the TLC. A close inspection of the discourse units in question shows that the Conclude examples are all discourse units which cover goodbyes at the end of the examination. Next is an example of such a discourse unit (from file 2_6_CH_20) where the student is an L1 Chinese speaker taking a grade 6 exam:Footnote ¹⁰

While it may be possible to argue that there is a degree of concurrence here (right in utterance 5) or that utterance 4 might be conceived as some other substructure, we would argue that this is a Conclude meso-structure. It starts with the discourse marker okay and then proceeds through a ritualised exchange of farewells that brings to a close the largest macro-structure in the interaction—the examination itself. While there is one such discourse unit in the TLC L1 (in file 25), and that too is from the end of the exam, the following is a much more typical example of the concluding sequence from the TLC L1 present in the sample of the TLC L1 studied. Prior to this discourse unit, the examiner has announced that the exam is over:

In this discourse unit, there is a Direction provided (1) and the rest of the unit represents mutual concurrence to the Direction and to bringing the exam to a close. The only vestige of the rather repetitive closing sequence from the TLC example is the sequence cheerio followed by thank you at the very end of the exam (6).

While it is tempting to explore the meso-structures in greater detail, we have explored them enough to establish that the interpretation of negative Dimension 1 in the analysis of the TLC presented in Chapter 3 needs revision. The function is not simply Short. Like the TLC L1 and Spoken BNC 2014 material, the short units on this part of the dimension are Discourse Management discourse units. Some of the realizations of Discourse Management may be atypical, relative to L1 performance, but they are Discourse Management nonetheless. Accordingly, Table 7.3 needs to be amended and is shown as follows, with Discourse Management now being a feature which all three corpora share across D1−.

Before concluding this discussion and considering the remaining L1-only category, Situation-Dependent Commentary, we should note that the initial coding of the TLC D1 discourse units as simply Short was brought about largely by four things. Firstly, without focusing on discourse management (the management of the flow of the conversation), as realised by micro- and macro-level structures and as opposed to propositional content or some other meaning-related interpretation of the data, identifying the function Discourse Management can be difficult. Only by shifting the focus from understanding what is said to understanding instead its impact on the flow of discourse, can Discourse Management be readily spotted. Secondly, the ubiquitous nature of Concur makes the data hard to analyse – without pushing past Concur to see the other Discourse Management meso-structures in the discourse units, the short discourse units can be hard to analyse. Thirdly, the appearance of the discourse units from the conclusion of the exam, which were devoid of Concur meso-structures, made the data look, initially, functionally incoherent – it was hard to see what made these brief discourse units functionally coherent when the examples ranged from some largely composed of mm, yeah and okay to those which liberally repeated sequences of thank you. The fourth point is an important one – the final key to revising the categorisation was principled down-sampling that permitted close, qualitative analysis. A point we have emphasised throughout this book is apparent again – while our short-text MDA approach is indispensable for organising data and permitting down-sampling, the process of interpretation of that data requires close qualitative analysis also, as has been acknowledged for MDA itself (Biber and Conrad, Reference Biber and Conrad2009).

The recoding of the D1− function of the TLC leaves only one discourse function which is unique to the L1 data in the corpus – Situation-Dependent Commentary. The link between context, form and function explains this absence quite concisely. If we look at the 100 prototypical discourse units most strongly coded for Situation-Dependent Commentary in the TLC L1, we find that they are almost exclusively produced within tasks either not fully present or not present at all in the bulk of the L2 exams (grades 6 and 7) in the TLC; namely, the Interactive (88 examples, task taken from grade 7 onwards only), Listening (3 examples, task not taken at grades 6 and 7) and Presentation (3 examples, task not taken at grades 6 and 7) tasks. The only examples from a task also undertaken by all students in the TLC relate to Conversation (five examples) and Discussion (one example). The Situation-Dependent Commentary task is clearly of use in the Interactive part of the exam in which the examiner introduces a topic and the examinee is encouraged both to ask questions about it and make comments upon it. This task, especially the commentary aspect of it, is distinct from what is required in the Conversation and Discourse tasks that we considered in the TLC in Chapters 3 and 4. So, while Situation-Dependent Commentary is not salient to the degree that it is observable through short-text MDA in the TLC, L2 language production cannot be inferred to be the cause of this. Rather, it is the task which is key and, as the L2 speakers do not perform a task which selects the Situation-Dependent Commentary function frequently enough in the TLC because the bulk of the data are from grade 6, where that task is not taken, the Situation-Dependent Commentary function is not salient in the L2 data.Footnote ¹¹ Yet it is clearly visible in both the TLC L1 and the Spoken BNC 2014. The following example shows one of the Situation-Dependent Commentary prototype discourse units in the Spoken BNC 2014, from file S37E, in which a speaker and hearer make comments upon a situation that Speaker A has observed:

So, the use of Situation-Dependent Commentary in situations like that introduced in the Interactive task is well attested in the Spoken BNC 2014. However, the exam itself is composed of a set of tasks which do not elicit the full range of discourse functions that we might find in conversational English frequently enough for these functions to be visible through short-text MDA, as will be discussed in the next section.

7.6 L1 Conversation-Only Functions

The discourse functions present in the Spoken BNC 2014 which are absent from both the L1 and L2 TLC corpora are, clearly, functions which are not elicited by the tasks in those corpora to the extent that they are visible using short-text MDA. In this case, we can clearly say it is the tasks of the GESE exam – not L2 proficiency – that is the cause of these functions not being salient in the TLC data, as they are absent for both L1 and L2 speakers. The functions in question are: Attitudinal Descriptions, Elaborated Speech, Informational Recounts, Narrative, Non-Narrative, Opinionated Narrative and Reveal.Footnote ¹²

Of interest in the list of features is the presence of Narrative and it is on that function that we will focus here. Note that Narrative pervades both the TLC material and the Spoken BNC 2014. As noted, it provides affordances which can be drawn on to perform a task and, in the combination of discourse macro-structure and task in the TLC data, we see narrative take on a number of varying forms – Informational Narrative (both TLC corpora), Extended Narrative (TLC L1), Narrative (Spoken BNC 2014) and Opinionated Narrative (Spoken BNC 2014). The TLC narratives are clearly task-bound, as discussed. What we have not considered, so far, is to what extent, if at all, there is variation in the narratives between L1 usage and L2 usage. While we have seen that functions using narrative are common to L1 and L2 speakers approaching the same task, we have not considered the degree to which the form of the macro-structure is the same across the groups. In Chapter 9 we will, once again, move from a mainly quantitatively driven exploration of form and function, as pursued in this chapter, to a qualitative, close-reading-driven focus on narrative. In doing so, we are returning, as promised, to a formal consideration of narrative structure, and asking whether, for both L1 and L2 speakers in the TLC data, the functions underpinned by the macro-structure of narrative have identical realisations of narrative structure within them or whether, at the discourse unit level, we may be able to discern difference in the use of this macro-structure by the two groups of speakers. But before doing so we will, in Chapter 8, refine and contextualise our approach by considering research on narrative.

7.7 Conclusion

We began our exploration of the discourse unit functions across our three corpora with one main question in mind – does the GESE exam itself actually reflect the demands of conversational interactions in British English? The evidence presented in this chapter allows us to conclude that, in so far as the tasks in the corpus require the functions, yes it does. We have seen that in the TLC there are functions shared with the L1 corpora examined. We have also seen in the TLC L1, a higher grade examination with additional tasks, that both a further discourse function is called upon that appears in conversational English (Situation-Dependent Commentary) and that pragmatic competence is further tested in the demands placed on the examinee to gain the floor for a lengthy monologue in the Presentation task. At the same time, there are functions in conversational English that are not required by the exam, irrespective of whether it is being taken by an L1 or L2 speaker of English. By the same token, however, there are functions in the exam data that are not salient in everyday conversation. The mismatches between the functions demanded by the exam and those which are present in everyday conversation are, we believe, broadly linked to context. A formal exam for L2 speakers, often preparing students for formal interactions in educational settings, for example, is always going to be distinct from everyday conversation in a target L1. Likewise, the power imbalance and the impact that this has on face management is also likely to cause differences between the exam and the L1 conversational data. So, the differences observed, while clear, are also ones we might attribute to context of use. There is no clear evidence that, in terms of discourse functions themselves, there is a yawning gulf in performance between L2 and L1 speakers that can only be explained by proficiency, at least at the levels we are exploring.