2.3.1 Face and Advice

In Section 1.1, we briefly outlined why face is an intrinsic aspect of any kind of interaction; here, we focus on its role in advice exchanges specifically. As the overview of previous research on advice showed, advice-giving and advice uptake are a form of interaction of a ‘delicate nature’ (Reference Locher, Herring, Stein and VirtanenLocher, 2013:352) with regard to face considerations. They can damage the self-image of the advisee and can risk changing or unbalancing the existing relationship between the interlocutors. People use a range of discursive strategies to mitigate this risk. The complex relationship between face and advice is evident at every stage of interactions involving advice, as each stage carries identity implications with a potential for damage to face for both interlocutors (Reference GoldsmithGoldsmith, 2000:3). Asking for advice entails devaluing one’s own face by admitting need, as well as enriching the adviser’s face by identifying them as a valued source of assistance. The adviser needs to consider the advisee’s face vulnerability in seeking assistance, while not harming their own face by minimising the display of their own competence. If the advice is rejected, the advisee risks devaluing the adviser’s face by disregarding their expertise, while if it is accepted, they risk devaluing their own face by confirming their own lack of competence, although there is also space for enhancing the adviser’s face when the advice is appreciated. In the case of unsolicited advice, the potential harm to both interlocutors’ face is greater than in situations where the advice is solicited, as the advisee is positioned as someone who not only needs help, but is unaware of it, while the adviser positions themselves as someone who can be seen as intruding uninvited into another person’s affairs. These challenges contribute to our analysis of advice solicitation and uptake in Section 4.2.

Notably, this view of face and facework involves an understanding of the socio-cognitive frame for advice, specifically, the relationship between the interlocutors, the context of their interaction, and the identities that they claim for themselves in that context (e.g., competent co-worker, knowledgeable tutor, self-reliant daughter, wise friend). Such social factors need to be considered alongside the specific linguistic constructions used to give advice for a more comprehensive understanding of its effectiveness, which perhaps explains MacGeorge et al.’s comment that ‘attempts to identify precise linguistic strategies that consistently reduce perceived face-threat have been inconclusive’ (Reference MacGeorge, Feng, Butler and BudarzMacGeorge et al., 2004:44). The same utterance can be perceived very differently depending on, for example, the age, gender, and relationship of the interlocutors, that is, the social factors considered in this study. To support the interpretation of these utterances, the role of face, and the combination of linguistic, dialogic, and social factors they embody, we turn to politeness theory. Broadly speaking, politeness helps explain interlocutors’ linguistic (and non-linguistic) behaviours. In the next section, we describe the frame-based approach to politeness, and how it can provide us with an appropriate framework to account for all the factors at work.

2.3.2 Frame-Based Approach to Politeness

In this Element, politeness is understood as a technical cover term for the choices interlocutors make to preserve other people’s self-image as well as their own. Important areas of research include impoliteness, rituals and conventions, cross-cultural differences and similarities, among others. It covers a range of theoretical perspectives in pragmatics and linguistics more broadly, from the earlier approaches focusing on linguistic strategies (Reference Brown and LevinsonBrown & Levinson, 1987; Reference LakoffLakoff, 1973; Reference LeechLeech, 1983) to more recent developments that encompass sociocultural considerations and behavioural norms beyond linguistic forms, and which place more emphasis on the participants’ perspective of what is polite rather than the researcher’s (e.g., Reference EelenEelen, 2001; Reference LocherLocher, 2006; Reference MillsMills, 2003; Reference WattsWatts, 2003). Other scholars have pointed out perceived weaknesses of some of these approaches and sought to bring together the different factors at play; one of the most detailed presentations of these criticisms is given by Reference TerkourafiTerkourafi (2005, Reference Terkourafi2015). Her view is that theories centred on linguistic forms, such as Reference Brown and LevinsonBrown and Levinson (1987), make undue assumptions about the universal applicability of linguistic strategies without closer consideration of what might be appropriate in specific cultures or social situations (Reference TerkourafiTerkourafi, 2005:238). To account for both linguistic strategies and sociocultural norms in her conceptualisation of politeness, Terkourafi proposes a frame-based approach.Footnote ⁴

For this approach to politeness, Reference TerkourafiTerkourafi (2002, Reference Terkourafi2005) develops the notion of Fillmore’s frame (cf. Section 1.1) by extending its application to interactional contexts, using it as an explanatory tool to understand how interlocutors might know when something is appropriate to the communicative situation. She argues that speakers’ repeated experiences of interactions are abstracted into frames to be called upon again when a relevant situation arises. Specifically, Terkourafi translates this perspective to politeness by positing that regular co-occurrence of forms and contexts over time comes to be perceived as polite, leading to politeness as ‘a knowledge of which expressions to use in which situations’ (Reference TerkourafiTerkourafi, 2002:197). In other words, politeness includes awareness of the expected linguistic formulae to be used in a given situation, with the correct use passing unnoticed and marking the speaker as polite. Face considerations remain an important part of the framework, but with the broader focus on one’s sense of (public) self, rather than specific distinctions between face threat and face enhancement.

Terkourafi’s frames are structures of co-occurring components, where the components include both the actual linguistic forms used and ‘social categories such as the sex, age, and social class of the participants, the relationship between them, the setting of the exchange, and whether an act is occurring for the first time or is repeated’ (Reference TerkourafiTerkourafi, 2005:247). They are central to her view of politeness-as-conventionalisation (see, e.g., Reference TerkourafiTerkourafi, 2015 for a recent presentation of the theory), which moves away from the abstract categorisation of particular linguistic strategies as polite or impolite, focusing instead on their association with specific contexts. She argues that it is conventionalisation, rather than varying degrees of indirectness, that determines perceptions of politeness, such that it is not automatically the case that a more indirect formulation will be considered more polite, especially where this is not a conventionally indirect form backed by repeated use (Reference TerkourafiTerkourafi, 2015:15). In this view, politeness is not seen ‘as tentativeness, but as providing outward displays of one’s familiarity with the norms governing the current exchange’ (Reference TerkourafiTerkourafi, 2015:15), and receiving a positive evaluation from the addressee as an acknowledgement. For example, consider the imperative Break a leg!. Out of context, it is a rather aggressive and malicious suggestion, but in the appropriate frame – when someone is about to go on stage – it is the most welcome way to utter a wish of good luck (see also Reference Quirk, Greenbaum, Leech and SvartvikQuirk et al., 1985:831–832 for their treatment of imperatives as being used for a wide range of illocutionary acts, where information about the interlocutors as well as the relative benefit of the action is needed). We see politeness as a theoretical framework through which we can interpret the interplay of linguistic choices and contextual factors involved in advice-giving and advice uptake. The frame-based approach is well placed to support these objectives.

The frame-based approach is also suited for a diachronic study of advice-giving. Reference TerkourafiTerkourafi (2005:251) recognises that not every interaction will take place within a pre-established frame, and that novel combinations of linguistic choices and contextual factors will always arise (what she terms nonce contexts). In these cases, interlocutors will still be motivated by their own and the other’s face needs as well as the underlying assumption of politeness (rather than aggression), and they will rely on these for the choice of linguistic forms and their interpretation. Of course, every frame once started as a nonce context that subsequently was repeated and entrenched in the speech community, just like what happens to the use of words (Reference Paradis, Benczes, Barcelona and de Mendoza IbáñezParadis, 2011). Terkourafi’s observation that ‘[s]ocio-historical conditions determine what is regular’ and ‘[w]hat is regular then gets interpreted as polite’ (Reference TerkourafiTerkourafi, 2005:250) points to the importance of the diachronic perspective alongside the synchronic for a more comprehensive understanding of the politeness norms at work in a given situation. The design of our study, with two comparable corpora spanning a period of approximately fifty years, makes us uniquely placed to trace the development of new regularities in advice-giving as social norms change and develop over time. Next, we give a brief overview of the main lines of enquiry in diachronic politeness studies, and how these inform our study.

2.3.3 Diachronic Perspectives on Politeness

As previously mentioned, diachronic politeness studies can play an important role in our understanding of present-day politeness. One of the objectives of this study is to investigate how, if at all, advice-giving may have changed over the past half a century in spoken British English, and what this might tell us, indirectly, about changing politeness norms. We take a similar line to Reference Culpeper and ArcherCulpeper and Archer (2008), who looked at requests in earlier centuries and concluded that the changes in frequency of different forms of requesting are not a sign of speakers becoming more polite, but rather of the conventional modes of performing different speech acts changing in step with broader sociocultural changes. We also use the term ‘diachronic studies’ in a restricted sense. Although there is a large and growing body of research on historical politeness, that is, how politeness was conveyed at particular points in time (for detailed overviews, see, e.g., Reference Fitzmaurice and TaavitsainenFitzmaurice & Taavitsainen, 2007; Reference JuckerJucker, 2020; Reference Taavitsainen, Jucker, Schneider and BublitzTaavitsainen, 2018), we are chiefly interested in the development of politeness phenomena over a period of time. Jucker and Kopaczyk note that ‘an overall account of the development of politeness or impoliteness across several centuries is still lacking’ (Reference Jucker, Kopaczyk, Culpeper, Haugh and KádárJucker & Kopaczyk, 2017:433); while we cannot cover centuries of dialogue in this Element, our findings can at least contribute to filling some of the gaps regarding politeness in spoken dialogue in recent decades. In fact, the recent past is not well represented in diachronic politeness studies. A useful starting point is Reference Culpeper and DemmenCulpeper and Demmen (2011), who focus on requests and politeness in nineteenth century British English. They suggest that historical and social changes in Britain around the mid-1800s led to a change from collectivism to individualism, with the latter being seen as a positive notion. As a result, politeness culture also shifted to one privileging indirectness and freedom from imposition, to show respect for an individual’s needs and abilities over those of the collective. Similarly, Reference JuckerJucker (2020) finds an increase in the use of conventional indirect requests in American English as the twentieth century advances. However, Reference Culpeper and DemmenCulpeper and Demmen (2011:76) hypothesise that this type of Victorian individualistic politeness might have been waning in more recent decades.

A different line of enquiry that can yield fruitful insights comes from research on change in modality in Present-Day English (see, e.g., Reference Mair, Leech, Aarts, McMahon and HinrichMair & Leech, 2020 for a detailed overview). Modality – especially deontic modality – is often governed by politeness. In advice-giving, for example, the adviser might opt for a deontically weaker format out of considerations for the advisee’s face (e.g., you could change your shirt versus you must change your shirt). Using a variety of corpora, different researchers have observed the demise of deontically stronger modal verbs such as must and the concomitant rise of semi-modals such as have to and need to in recent British English, covering a time span overlapping with that of the present study. For example, an analysis of the Diachronic Corpus of Present-Day Spoken English (DCPSE), covering the period between the 1960s and early 1990s, finds must to have been significantly supplanted by have (got) to in deontic use (Reference Close, Aarts, Lenker, Huber and MailhammerClose & Aarts, 2010), and a substantial decline of must, may, and shall, particularly in informal face-to-face conversation (Reference Bowie, Wallis and AartsBowie et al., 2013). However, the authors conclude that there is no neat explanation of underlying factors for these changes (Reference Bowie, Wallis and AartsBowie et al., 2013:91). A similar time period is covered by Reference Leech, Hundt, Mair and SmithLeech et al. (2009), who look at both written British English (the Lancaster-Oslo/Bergen and Freiburg-Lancaster-Oslo/Bergen corpora; LOB and FLOB, respectively) and spoken British English (subsets of DCPSE taken from the years 1961 and 1991). They, too, find a particularly strong decline in the use of must and may, and suggest that there are various factors at play: not just an increase in semi-modals, particularly need to and want to, but also a shift away from must towards less dictatorial forms such as should (Reference Leech, Hundt, Mair and SmithLeech et al., 2009:116). An explanation that is offered in support of this change is the democratisation hypothesis, a general advance towards forms of communication that emphasise ‘equality of power’ and ‘express obligation less directly’ (Reference SmithSmith, 2003:259). Expanding the time frame to 1931–2006 and using further additions to the LOB family of corpora, later research has confirmed the general trend of should increasing in its use as a deontic modal (Reference Smith, Leech, Aarts, Close, Leech and WallisSmith & Leech, 2013), although this study looked at written English only. Reference Love and CurryLove and Curry (2021) bring us to the present day by comparing changes in modality in spoken British English using the Spoken British National Corpus 1994 and the recent Spoken British National Corpus 2014. Among the relevant findings of their detailed study, they find evidence of a decreased use of must, shall, and will between the 1990s and 2010s, as well as, interestingly, a decreased use of have to and want to. Conversely, they observe an increase in the use of weak modals such as could, might, and would. The study is unclear with respect to changes in the distribution of modal functions (e.g., deontic versus epistemic), since none of the modals showed a significant trend.

This said, the overall pattern that has emerged from the research briefly summarised in this section on change in recent English sees a movement towards greater indirectness, less imposition, and linguistic forms that shift emphasis away from deontic obligation. It is also evident that language change can be tracked even across a relatively short period of time, especially in spoken dialogue, the most common use of language (Reference ClarkClark, 1996). The diachronic component of our study is a test case for these claims, offering an up-to-date comparison of spoken data about a communicative event which necessarily requires some combination of indirectness, imposition, and deonticity. Furthermore, by focusing on one speech act only, we move beyond the broad categories of deontic and epistemic modality addressed by previous research. This makes it more likely that any changes observed are attributable to changes in politeness norms over time, rather than effects of different communicative events.

3.3.1 The Extraction and Identification of Advice

The extraction and identification of relevant data is both the core strength and core challenge of corpus pragmatics. Corpora provide a sizeable amount of data that allows for robust findings, but the lack of simple form–function relationships makes the task of extracting the data much harder than part-of-speech tagging or simple lexical searches. Specifically, there is no single way to perform a given communicative function (e.g., an apology can take the forms sorry, I apologise, it was bad luck), and the same form can fulfil different functions depending on the context (sorry can be an apology, a request for repetition, an invitation to move out of the way). Research in corpus pragmatics follows either a form-first or function-first approach (e.g., Reference RühlemannRühlemann, 2019:8). In the former, all instances of relevant (known) forms are extracted using corpus software tools. This approach can be applied to corpora of any size with little effort, but it will necessarily overlook more unusual or creative examples that are not captured by the search terms, thus potentially leaving some areas of the topic unexplored. A function-first approach, on the other hand, prioritises coverage of the pragmatic phenomenon of interest by selecting all its instantiations, regardless of form. This can only be achieved by significant human intervention in the form of pragmatic annotation, which requires manual identification of each instance and is therefore only possible with small-scale corpora, though efforts at automated pragmatic annotation are in early stages of development (see, e.g., Reference De Felice, Darby, Fisher and PeplowDe Felice et al., 2013 and Reference WeisserWeisser, 2019; for a comprehensive overview of pragmatic annotation, cf. Reference Archer, Culpeper, Jucker, Schneider and BublitzArcher & Culpeper, 2018).

Our study is double-barrelled in that it first adopts a form-first approach and then a function-first approach. The reason for starting with the form-first approach is that the nature of advice-giving is such that there is broad agreement on a set of linguistic forms canonically used for this function, as evidenced by the significant overlap and similarity among the phrases studied in previous work on advice-giving (cf. Section 2.2). These are easily searchable due to their formal predictability. Of course, they do not always yield accurate results, but at least they provide an initial list of candidate examples based on which further manual work can take place (see below). The full list of search terms used in the study is given in Part 1 of the Annotation Manual (available as part of the supplementary material of the Element); here, we provide an overview of the main categories of search terms in Table 4 and describe the main principles that guided our process of constructing the full list.

We began by considering the phrases listed by Reference Couper-Kuhlen, Thompson, Deppermann and HaughCouper-Kuhlen and Thompson (2022), as they provide one of the most comprehensive overviews of advice-giving constructions, all of which have formal predictability: imperatives, interrogatives, and different forms of modals. This was motivated by the belief that relying on a previously established set of forms would facilitate comparisons between previous studies and our own. However, a pilot study involving manual read-throughs of a subset of the conversations in the corpora revealed a number of additions to the set of search terms, of two main types. The first consists of the natural grammatical and lexical expansion of the initial set of forms by including either a wider range of subjects (e.g., from second person modal and semi-modal constructions to all subjects), verb choices (e.g., from have you tried to have you + any verb), or interrogatives (e.g., from what about to how about). The second stems from a broader understanding, arising from our pilot study, of the different parts of speech and constructions that can provide advice, such as nouns (suggestion, idea), adjectives and adverbs (advisable, worth(while)), and performative verbs (I recommend). The pilot study also pointed us towards the inclusion of mitigators and stance markers (e.g., I think, just) as a further signal of an advice-giving utterance, ensuring that we took into account all the different components of this kind of speech act, formal and attitudinal. Through this iterative process of combining insights from previous literature, our own understanding of the texts, and simple substitutions and expansions of the terms licensed by the grammar, we have arrived at a comprehensive list of search terms giving us broad coverage of advice-giving in our data. Indeed, often different search terms yielded the same utterance (cf. the conditional with just in Table 4), a redundancy which reinforces our confidence in the exhaustiveness of the coverage of the search terms. Naturally, there will be ad hoc or idiosyncratic formulations which will not be captured by our terms. These formulations are highly individual and specific to, and comprehensible within, only a particular conversation. Although such formulations might be interesting examples of linguistic creativity, their idiosyncratic nature makes them of less relevance for our study, which aims at uncovering frames and generalisable patterns of advice-giving in language.

The procedure of extracting candidate examples of advice based on the search terms above was as follows. The texts were prepared by first removing all XML markup (including descriptions of events such as <event desc= “phone rings”/>) and passages of untranscribable text. Almost all the searches were done using simple command line queries with wildcards allowing for intervening modifiers or false starts (e.g., you really should, what what about driving). These search terms are syntactically straightforward and belong to a closed set of forms. Imperatives and evaluations (comparative and superlative forms), however, belong to an open set and therefore required part-of-speech tagging and syntactic parsing for their extraction. This was done with a Python script using the Pattern package (Reference De Smedt and DaelemansDe Smedt & Daelemans, 2012), which includes a suite of natural language processing tools including part-of-speech tagging and full syntactic parsing. For the evaluations, the former was sufficient: the script identified all instances with words tagged as comparative and superlative adjectives. For imperatives, we had to take into account the lack of a specific part-of-speech tag for imperative forms, which in this model was labelled bare verb (VB), a category which also includes items such as read and go in constructions such as I can read and he wants to go. The script thus relied on full parsing of each instance, returning only those where verbs tagged as VB were also lacking in subject, to model the characteristics of imperatives. The nature of spoken language is such that there were many utterances incorrectly identified as imperatives, as in this example: probably boil them with the spikes. For the simpler searches, the problems were false starts and repetitions, which returned examples such as you are you are going to try this as a result for are you going to. The automatically extracted data, then, still required manual intervention and cleaning. The sizes of our corpora allowed us to reap the benefits of both automated and manual approaches: they are large enough to provide a substantial number of examples for robust bias-free generalisations, while just about manageable for human annotators to review and annotate the data, as we go on to describe.

Following the form-first approach, we turned to function by comparing the candidate examples to the felicity conditions of advice established for this study (see Section 2.1). The procedure we followed is outlined in Part 2 of the Annotation Manual. The felicity conditions cover a broad network of instantiations of directive–commissive speech acts. A more detailed breakdown of what the felicity conditions look like in discourse is given in (a)–(f); we consider an utterance to be an instance of advice-giving only if it exhibits all the properties below.

1. (a) The focus of the utterance had to be on the advisee’s action; the adviser had to elicit an action from the advisee, not just communicate their attitudes or values.

2. (b) The proposed action had to include the advisee as one of its undertakers; the advisee could act alone or jointly with the adviser, but the action could never be carried out by the adviser alone.

3. (c) The advisee – either solely or jointly with the adviser – was the benefactor of the action.

4. (d) The action had to take place in the future; references to past actions were not considered.

5. (e) The utterance had to be directed at people who were present in the conversation at the time of the utterance; reported speech and general statements without a clear addressee were excluded.

6. (f) When the utterance was ambiguous between advice and another type of speech act such as a general statement, the addressee’s response was consulted; the utterance was considered to convey advice if the addressee either agreed or disagreed to carry out the proposed action, rather than simply providing information.

To illustrate the presence or absence of the properties above, the utterances in italics in (3)–(5) were considered to convey advice, while the utterances in (6)–(8) were not.Footnote ⁷ Example (3) is a clear example of advice-giving. It focuses on a future action intended to be carried out by the advisee to his benefit. Both the adviser and the advisee are present in the conversation, and the advisee’s response (yes) indicates that he accepts the advice. Example (4), a proposal, also has all the necessary properties. The major difference with (3), however, is that, in (4), both the adviser and the advisee are expected to carry out and benefit from the action. The utterances in (5) are very indirect ways of making suggestions in English; however, the focus is still on the proposed action, that is, to have a Christmas get-together.

1. (3) (B is an architect who advises A on a renovation project)

   1. A: I mean what I’m imagine that we’re doing in pulling back to ensure that we can pay for things that we really want is at least getting a good job done on the things that we do get done rather than shaving all over the place …

   2. B: yeah … I think that well my advice would certainly be to you to do structural and constructional alterations

   3. A: yes

1. 1. A: oh we should do something for Halloween

   2. B: we should

1. 1. A: Christmas is a good excuse maybe to just have an impromptu coffee and mince pie …

   2. B: that’s a good idea maybe just after the end of the term but not too long after because otherwise people start disappearing …

   3. A: yeah definitely after term

These properties are not present in (6)–(8). In (6), what is in focus is B’s assessment of what might or might not work, rather than a call for action. The offer in (7a) and the request in (7b) did not qualify because of the addressee’s lack of involvement in the action and the addressee’s lack of benefit, respectively. Finally, in (8) the interlocutors make a series of statements about what constitutes a good lecture by using the second-person pronoun you. However, the pronouns seem to be used in a general sense without necessarily being directed at the addressee.

1. (6) (The speakers are drawing up a floor plan)

   1. A: that division doesn’t need to be there anymore

   2. B: oh that doesn’t need to be there oh yeah that could work

   3. A: or the door could even open like wham and open like that

1. 1. a. I’ll cover your answer in my letter

   2. b. I wonder if I might see you very briefly

1. 1. A: I’ve never given the same lecture twice yet it’s very wearing … you can’t do the same thing twice really

   2. B: the beauty of an excellent lecture is you’re supposed to be able to give it again that’s what compensation for the sweat of preparing it

   3. A: that’s right but it’s not like that with media … we haven’t got a textbook you see

In addition, we excluded what looked like instructions (e.g., what have you put on your thing -> you just stick that on the back). While instructions seem to have all the properties in (a)–(f), according to Leech, they ‘tend to have a collaborative illocutionary function, and therefore they lie outside the realm of (im)polite behavior’ (Reference LeechLeech, 2014:137). Face and politeness are important aspects of this study, so we decided not to include instructions after all.

The Annotation Manual helped us to be consistent in our annotation decisions and it ensured that the annotators followed the same guidelines. To assess the reliability and replicability of the guidelines, a series of inter-rater reliability tests were carried out based on the annotations of the first author of this Element and the annotations of ~10 per cent of the data by the second author. The inter-rater reliability tests were conducted over three rounds of three corpus texts each, separated by discussion sessions where disagreements between the annotators were discussed and resolved together. The intermediate sessions served to progressively refine the annotation guidelines and address unanticipated problems. All choices made during those sessions were recorded and added to the Annotation Manual, which was then used to revise the rest of the annotations. The results from each annotation round (R) are given in Table 5. We report the values of two measures: observed agreement (in percentages) and Reference CohenCohen’s (1960) chance-corrected kappa.

Table 5 shows that the kappa scores obtained for the identification of advice reached at least substantial levels of agreement across all three rounds, according to the scale of Reference Landis and KochLandis and Koch (1977). The third and final round, in particular, showed the best result (k = 0.922, ‘almost perfect agreement’), which indicates that the annotation task had become clearer and better defined by the end.

3.3.2 The Annotation of Advice

Following the extraction and identification of the advice-giving utterances in the corpus sample, we performed a close manual annotation of the utterances and their context for linguistic, dialogic, and social factors. The most important of these were presented in Section 2.2 (e.g., form of advice, (non-)solicitation of advice, advisee’s response to advice), but there are many other factors typically not dealt with in the literature that might affect how advice is given and received in conversation (e.g., social factors such as the age and gender of the interlocutors). Table 6 summarises the annotation factors and values. The factors can roughly be divided into three groups. The first group (in light grey) comprises linguistic factors that we used to develop a systematic overview of the constructions used to give advice along varying degrees of deontic and epistemic authority and, by extension, communicative (in)directness, which in turn was the basis for systematic comparisons across time and discourse contexts. The combined values of the linguistic factors also make up the first factor based on which we predict the advisee’s response to the advice. This allowed us to determine how, if at all, constructional choice affects advice uptake. The other predictors of advice uptake, dialogic and social factors, are in the second group of factors in Table 6 (in darker grey). The third group (in darkest grey) comprises just one factor, the advisee’s response to the advice, that is, the factor to be predicted. It is important to keep in mind that the factors in Table 6 are limited to the types of factors that the corpus data allowed us to consider. Advice uptake may depend on a range of other factors (e.g., comprehensibility, relevance, feasibility, absence of limitations; see Reference MacGeorge, Feng, Butler and BudarzMacGeorge et al., 2004); however, these factors are better suited to be tackled by survey and experimental methods than corpus methods. In the following, we provide brief descriptions of each factor and their values. A more detailed overview is given in Part 3 of the Annotation Manual.Footnote ⁸

We start with linguistic factors, all of which relate to the advice-giving utterance. The first factor, form of advice, stands out because it is open-ended and therefore defies strict classification. This is due to the diverse range of linguistic forms used to give advice, a feature which would have made it difficult for us to impose strict categories on the data. Instead, we identified the part of the advice-giving utterance which contained the main illocutionary message of the utterance. Among the items included are references to the participants involved in the action (typically pronouns), the verb phrase conveying the action, noun phrases and adjective phrases (e.g., comparatives) in the subject and subject predicative positions, question words, negative markers, and so forth. If the advice-giving utterance did not contain any such items, the whole utterance was included (excluding modifiers; see below). Examples (9)–(13) illustrate some of the items included (in italics).

1. (9) slow down your pace a little bit [verb phrase]

2. (10) you should get into a job as quickly as possible [pronoun, verb phrase]

3. (11) I think your paraphrasing could be better [noun phrase, verb phrase, adjective phrase, comparative adjective]

4. (12) darling why don’t you bribe Jo to lend you her Cold Comfort Farm [question word, negative marker, pronoun, verb phrase]

5. (13) I think twice a week [whole utterance]

The next linguistic factor, utterance type, has four values – imperatives, interrogatives, declaratives, conditionals – and concerns the grammatical form of the advice-giving utterance. While imperatives and interrogatives are straightforward, the distinction between declaratives and conditionals requires some explanation. The distinction lies in the location of the proposed action in the utterance in that, when the action was in the main clause, the utterance was considered to be a declarative (e.g., if the weather allows you can mow the lawn), but when the action was in the conditional clause, the utterance was a conditional (e.g., if we could go to York a few days before that would be really nice). The advice-giving utterances were also annotated for the presence or absence of modifiers, which were divided into hedges (e.g., I think, well, tag questions), emphasisers (e.g., obviously, always, intensifier do), and other stance markers that mark, for instance, viewpoint, attitude, or focus (e.g., ideally, honestly, actually). Importantly, the modifiers had to take scope over the whole advice-giving utterance. Taken together, the linguistic factors provide us with enough information about the degree of deontic and epistemic authority of the advice-giving utterance, and they also help us to test the prediction that there is a relation between the format of the advice-giving utterance and its uptake.

The dialogic and social predictors of advice uptake in Table 6 go beyond the advice-giving utterance and refer to the broader discourse and sociocultural context. The first two dialogic factors are type of event and deictic time. Type of event refers to the spatio-temporal structure of the proposed action. It has two values: dynamic and stative. While dynamic events include change of state (e.g., you must start to rearrange your estimates of Lawrence’s novels) and state of change (e.g., go to the planning and buildings committee), stative events include state of no change (e.g., I think that’s about the minimum of what would make it worthwhile). Deictic time is concerned with when the advisee is expected, or is in the position, to carry out the action relative to the time of the utterance. The cut-off point is the present conversation, in which case the deictic time is immediate if the advisee is to carry out the action during the course of the conversation (e.g., I think weekdays would be preferred in a context where the advisee has to propose a suitable weekday immediately) and it is distant if the action goes beyond the present conversation (e.g., this could be held alongside a teaching job in a context where the decision about taking the job does not have to be made immediately). We associate both factors, type of event and deictic time, with the effort required to carry out the action. Based on the assumption that it is more difficult to change someone’s mind than to elicit physical action, and that distant actions are more demanding than immediate actions, we associate dynamic and immediate action with low effort and stative and distant action with high effort. Assuming also that there is a relationship between perception of effort and advice uptake (cf. Reference MacGeorge, Lichtman and PresseyMacGeorge et al., 2002), we expect dynamic and immediate action to lead to higher levels of acceptance of advice than stative and distant action.

The last dialogic factor, (non-)solicitation of advice, is concerned with the way in which advice is initiated in the preceding context. The values are based on Reference GoldsmithGoldsmith’s (2000) typology of six patterns of advice initiation, ordered from the clearest case for advice solicitation to the clearest case for when the advice is not solicited. The first four values are initiated by the advisee, meaning that the person who is expected to respond to the advice is also its initiator. First, when advisees ask for advice, they do so in a relatively explicit way, through clear statements (e.g., I need your advice), explicit requests (e.g., what should I do), or somewhat narrower questions (e.g., should we try and get tickets for Duna Jam).Footnote ⁹ Second, when advisees ask for an opinion on some action (e.g., what do you think) or simply for information (e.g., which was the troubled pear), the advisability of the action is implied rather than explicitly stated. Third, when advisees disclose a problem, their announcement of the problem may make advice solicitation plausible, as in when they make confessions of ignorance or uncertainty (e.g., my teaching evaluations were terrible again I really don’t know what to do) or pose rhetorical questions (e.g., how can I possibly finish all my work before break). Fourth, when advisees announce a plan of action, they may suggest that the plan is in some way problematic, as in (14).

1. 1. A: I’m inserting a new stylus into the record player [advisee announces plan of action]

   2. B: why not put it into the head [advice-giving utterance]

The last two values are initiated by the adviser, meaning that the person who gives the advice is also its initiator. The first one of these is when advisers identify a problem, in which case the problem and the advice are given in separate turns. Consider (15), where the advisee’s acknowledgement of the problem provides a basis for the advice in the next turn. By contrast, the constructed example in (16) illustrates a situation where the adviser volunteers the advice and where the advice and the problem are in the same turn.Footnote ¹⁰

1. 1. A: do you have a Young Persons Railcard [adviser identifies problem]

   2. B: no I don’t

   3. A: you should get one [advice-giving utterance]

1. 1. A: do you have a Young Persons Railcard # you should get one [both in the same turn]

Following previous research on the topic, we expect there to be a decrease in the level of acceptance of advice as we move from explicitly solicited advice to unsolicited advice.

The social factors are represented by factors such as the relationship between the interlocutors, and their age and gender. The interlocutors’ relationship was determined based on the distinction made in LLC between conversations among equals and disparates. All combinations of the latter were considered: when the addressee was of lower rank, higher rank, and when there were addressees of different ranks. When the interlocutors in conversations between disparates were on the same level, they were considered to be equals instead. A similar principle was followed in the annotation of the age and gender of the interlocutors. The combinations of age included conversations where the advisee was younger, older, when the interlocutors were roughly the same age (± ten years), and when there were advisees of different ages. The combinations of gender included same-gender conversations (male–male, female–female), mixed conversations (male–female, female–male), and when there were advisees of different genders. The relative lack of research on the effects of these three factors on advice uptake makes us hesitant to make any predictions at this point.

Finally, we annotated the advice sequences for the advisee’s response to the advice. The factor has four main values – acceptance, resistance, rejection, other response – of which the first three were further divided into sub-values. The choice of the (sub-)values draws on research from a number of studies, most notably, Reference Hepburn and PotterHepburn and Potter (2011), Reference Heritage, Sefi, Drew and HeritageHeritage and Sefi (1992), Reference PudlinskiPudlinski (2002), and Reference Shaw and HepburnShaw and Hepburn (2013). First, when advisees accept the advice, they do so in one of three ways: (i) commitment to act (17), (ii) positive evaluation (18), and (iii) marked acknowledgements, which treat the advice as informative and helpful (e.g., yes, okay, oh right, partial repeats, as well as yeah and mm with a rising–falling and rising intonation).Footnote ¹¹

1. 1. A: maybe you should just check to be sure

   2. B: I will tomorrow morning I’ll be checking [commitment to act]

1. 1. A: like Lost in the City or something

   2. B: oh that’s a good one [positive evaluation]

Second, advice resistance involves cases where the advisee avoids treating the advice as informative and helpful, but where the proposed action is not necessarily rejected. The two types of resistance are: (i) unmarked acknowledgements (e.g., right, as well as yeah and mm with a falling intonation)Footnote ¹² and (ii) assertions of prior knowledge and/or competence (19).

1. 1. A: it’d make it easier if you could give them an example of someone else who is doing it right

   2. B: that’s what I’m saying [assertion of prior knowledge]

Third, rejections include (i) elaborated rejections in which case the advisee provides a justification for the rejection (20), and (ii) straightforward rejections where no such justification is provided (21).

1. 1. A: we can go over and do mine

   2. B: mm but you wouldn’t know how the system works [elaborated]

1. 1. A: you could use your Club Card uh no Colonel Card thing

   2. B: but I won’t [straightforward]

Finally, other responses include, for instance, silence, laughter, and topic shifts.

To assess the reliability and replicability of the guidelines in this part of the analysis, we carried out two types of inter-rater reliability tests – observed frequency and Cohen’s kappa – over three annotation rounds (see Section 3.3.1 for the same procedure). The results of the three annotation rounds are given in Table 7. Note that no measures were obtained for the open-class factor of form of advice, which due to its nature would have required a different approach. Also, the disagreements associated with this factor were so minor that they had little effect on the development of the constructional overview in Section 4.1.1.

Overall, the results in Table 7 suggest that the guidelines for annotating advice in this study were clear and well defined. The kappa scores were especially high in the case of utterance type, type of event, and deictic time (‘almost perfect’ or ‘perfect’ agreement across all three rounds). Somewhat lower levels of agreement were obtained for the presence or absence of modifiers, although the good result obtained in the last round shows that the problems encountered earlier in the annotation task had been resolved by the end of the rounds. Perhaps unsurprisingly, the kappa scores were lowest for (non-)solicitation of advice and advisee response to advice, both of which are factors that are inherently subjective and context dependent. Those factors were also most sensitive to the types of texts that were analysed in each round. Texts with many instances of advice were more challenging than texts with only a couple of instances of advice, because it was difficult to determine in each case whether or not the advice was solicited, or which response should be considered. Conducting the annotations over several rounds helped us to become more aware of the influence of such confounding biases (see also Reference Põldvere, Fuoli and ParadisPõldvere et al., 2016).

3.3.3 Statistical Analysis

The multifactorial nature of our data requires the use of advanced statistical analyses. So far, the norm in pragmatics research has been to rely on simple monofactorial analyses, which work well in cases with only one potential cause-effect relationship but are less suited in situations where the cause–effect relationship is expected to be moderated by multiple factors, which simultaneously amplify and downplay each other’s effects. In fact, it is the latter scenario that is believed to be the situation in linguistics (Reference Gries, Gilquin, Granger and MeunierGries, 2015a:175), as is also evident from the growing number of multifactorial analyses in corpus pragmatics (e.g., Reference Oben and BrôneOben & Brône, 2016; Reference Rühlemann and GriesRühlemann & Gries, 2020, Reference Rühlemann and Gries2021; Reference Tantucci and WangTantucci & Wang, 2018, Reference Tantucci and Wang2021, Reference Tantucci and Wang2022; Reference Van Olmen and TantucciVan Olmen & Tantucci, 2022). We contribute to this line of research by using different types of regression models including generalised linear models and random forests. The models are at varying levels of complexity, ranging from simpler models with one independent variable and a binary dependent outcome to more complex models with several independent variables and multiple levels of outcome. The goal is to show the potential of the models to answer different questions in pragmatics. In some places, we have chosen a model with fewer predictors over a model with more predictors in order to ease the reader into multifactorial corpus pragmatic analyses and to keep the level of statistical complexity manageable to novices. We used the software environment RStudio (version 1.3.1093; RStudio Reference TeamTeam, 2020) for all analyses.

To compare the occurrences of the advice constructions first across LLC–1 and LLC–2 and then across casual and institutional conversation, we fitted a number of binary logistic regression models to our data using the glm function of the stats package in base R. An advantage of regression models over traditional monofactorial tests (e.g., chi-squared tests) is that all the levels of the independent variable, that is, the advice constructions, are considered together. The plots comparing the occurrence of the constructions across the corpora and the conversational contexts were generated using the ggplot2 package. To predict the outcome of advice based on the range of linguistic, dialogic, and social factors, we modelled the data using Conditional Random Forests (CRFs), implemented in the cforest function of the party package (Reference Strobl, Boulesteix, Kneib, Augustin and ZeileisStrobl et al., 2008). CRFs have been widely used in syntactic analyses of language to predict particular linguistic variants (e.g., Reference Klavan, Pilvik and UiboaedKlavan et al., 2015; Reference Larsson, Callies and HasselgårdLarsson et al., 2020; Reference Szmrecsanyi, Grafmiller, Heller and RöthlisbergerSzmrecsanyi et al., 2016; Reference Tagliamonte and BaayenTagliamonte & Baayen, 2012); however, they have not been used very often in corpus pragmatics (but see Reference Tantucci and WangTantucci & Wang, 2018, Reference Tantucci and Wang2022; Reference Van Olmen and TantucciVan Olmen & Tantucci, 2022) despite the fact that CRFs are particularly useful in situations with highly correlated predictors, and where the sample size is small but the number of predictors is large – a common scenario in pragmatics research, and also the case in the present study. CRFs are based on Conditional Inference Trees (function: ctree), which predict outcomes by recursively partitioning the data into smaller and smaller branches and leaves according to those predictors that co-vary most strongly with the outcome.Footnote ¹³ CRFs extend that by using ‘ensemble methods in a forest of trees built on randomly sampled data subsets to arrive at an aggregated estimate of a particular outcome’s probability’ (Reference Szmrecsanyi, Grafmiller, Heller and RöthlisbergerSzmrecsanyi et al., 2016:114), and they also rank the predictors based on their explanatory importance. In the final model, we grew 1,000 trees (ntree = 1000) and selected three random predictors for each tree (mtry = 3; square root of the total number of predictors; see Reference LevshinaLevshina, 2015:297). As before, we used the ggplot2 package to visualise the proportions of the levels of the most important predictors.

As previously mentioned, the output of the statistical models was intended as a basis for further qualitative analysis. Also, due to data sparseness we were not able to include all levels of all the factors in the statistical models, requiring some simplification. However, the fine-grained analysis proposed in this section provided us with useful entry points for further qualitative investigations of specific advice situations. The next section presents the results of the qualitative and quantitative approaches.

4.1.1 Classification of Advice Constructions

In this section, we start by proposing a classification of the constructions used to give advice in LLC relative to their degree of speaker authority, and we then relate the types of constructions to the degree of communicative (in)directness and face considerations. As previously mentioned, the development of the classification of the advice constructions was based on three linguistic factors: (i) the form of advice, (ii) utterance type, and (iii) the presence or absence of modifiers. These factors were manually examined and correlations between them were identified for the classification of the types of unsubstantiated as well as substantiated constructions that speakers of English draw on in advice frames. Our classification was aided by previous work on advice constructions, which has proposed similar classifications, albeit only partial ones in terms of the range of the constructions (see below). It follows that the classification is not strictly data driven but rather the result of subjective, yet explicit, decisions made by the annotators with support from existing research. On the basis of the three factors, we classified all 1,234 utterances in the sample relative to their degree of speaker authority. The notion of speaker authority was operationalised in terms of the degree of the deontic force of the advice, and the strength of epistemic commitment and confidence conveyed by the speaker or adviser. The constructions form a continuum ranging from constructions that are expressive of the strongest degree of deontic force coupled with the highest level of the speaker’s epistemic commitment and confidence about the benefits of the advice for the advisee at the one end, to constructions that are only weakly deontic with low speaker commitment and confidence at the other end. The order of the continuum of constructions follows from our analysis of the constructions in their various contexts in the dialogic sequences. We then consider the speaker authority continuum with reference to the notion of communicative (in)directness, which also forms a continuum that to some extent goes hand in hand with the degree of speaker authority. The concurrence is likely to be a natural consequence of the fact that speakers’ choice of (in)direct expressions is a matter that relates to their intersubjective social and psychological considerations of the advisee. The choice of form is utilised by advisers to make it easier or more difficult to express alternatives, that is, to expand or contract the communicative space vis-à-vis the advisee (cf. Section 1.1).

Table 8 shows the types of constructions and their instantiations in LLC, ordered in terms of strength of speaker authority. The left-most column features five general-level types of constructions and their subtypes; we refer to them as unsubstantiated as they are not substantiated by language forms. The introduction of subtypes allows us to make comparisons at different levels of constructional granularity across the corpora and the discourse contexts (see Section 4.1.2). The column in the middle displays representative substantiated examples of the constructions from the corpora. The most important parts of the examples are underlined. Absence of constructions in Table 8 does not mean that such constructions cannot be used to give advice, but it simply means that no examples were found in our data. The column to the right presents the ordering of the constructions as a function of the degree of speaker authority as well as the extent to which we assume the advisee is expected to follow the advice. The ordering applies both to the general-level constructions and the specific examples of the constructions in the middle column. As shown in Table 8, the principal order of speaker authority is: imperatives → performatives → interrogatives → declaratives → conditionals. This order partly mirrors Reference Couper-Kuhlen, Thompson, Deppermann and HaughCouper-Kuhlen’s and Thompson’s (2022) order – imperatives → interrogatives → declaratives – but it also differs from theirs since they did not consider performatives and conditionals, at least not as separate categories.

Table 8 The constructions and their instantiations in LLC, ordered on a continuum of deontic and epistemic speaker authority (Strong = strongly directive, committed, confident; Weak = weakly directive, committed, confident); the most important parts of the examples are underlined

The investigation of the occurrence of the different constructions in LLC shows that utterances in the imperative mood, the highest level of the continuum, make up 22 per cent of all the advice-giving utterances in the sample (276/1,234). They represent the most straightforward way of trying to make somebody else do something for their own benefit. Giving advice in the imperative puts clear constraints on the advisee’s uptake in most situations – an uptake of acceptance (or not) is expected by the adviser. In affirmations, such utterances convey an obligation, and in negated form, a prohibition (don’t do that), which do not leave much room for debate or negotiation. The exhortative let’s utterance is a special and a slightly softer type of imperative construction and less binding from the advisee’s point of view since it includes the adviser and conveys a piece of advice where the interlocutors share part of the burden. In our data, both types of imperatives are often modified by hedges (26 per cent of the time), but rarely by emphasisers (5 per cent), which speaks to their forcefulness and need for softening.

Next on the continuum of speaker authority in Table 8 are performatives, which make up a very small part of the sample (1 per cent or 12/1,234). They are expressions that explicitly state that an act of advice-giving is being performed. The performative expressions may be verbs (I would strongly suggest that …) or nouns (well my advice would be to …), with verbs coming across as more forceful. Unlike directives in the imperative, mention of the advisee is optional in performatives, softening the impact of the advice on the recipient; the advisers, on the other hand, are always explicitly identified, adding personal involvement and highlighting the higher authority assumed by the adviser. Due to the limited number of performatives in the sample, we cannot say much about patterns of modification. However, as the examples in Table 8 indicate, performatives can be modified by both hedges and emphasisers.

Next in terms of deontic force and epistemic commitment and confidence are different types of interrogatives, which occur in the sample 7 per cent of the time (82/1,234). They necessarily appeal directly to the advisee, from whom a response is expected. We consider negated interrogatives to be more deontically forceful and epistemically committed than non-negated interrogatives, regardless of whether reference is given to the advisee, both the advisee and the adviser, or to the action instead. This is because only negated interrogatives convey that ‘the solution being recommended is so obvious or natural that there must be good reason not to embrace it’ (Reference Couper-Kuhlen, Thompson, Deppermann and HaughCouper-Kuhlen & Thompson, 2022). Among both the negated and non-negated interrogatives, the more forceful are constructions with an explicit mention of the advisee followed by an explicit mention of both parties, in which case the burden of the action is shared. Utterances without any mention of the participants but with reference to action instead are the least compelling ones, because the information about who should carry out the action is held back. The negated and non-negated interrogatives also have in common that questions introduced by, for example, wh-words (why don’t you/what about) are stronger than their modal counterparts expressing possibility (would it not/could it). This is because the former are more explicit in their call for action. Interestingly, in our data only a small proportion of all the interrogatives (21 per cent) are modified in one way or another, and if they are, they are typically hedged. The negated interrogatives are more likely to be hedged than their non-negated counterparts, presumably due to their strength. We consider interrogatives to rank higher on the continuum of speaker authority than the next category, namely, declaratives, because the former are more explicit in their demand for an answer and so are more impositive. Leech makes a note about the relative strength of interrogatives and declaratives in requests, which we believe applies to advice as well.

Next on the continuum of speaker authority are declaratives, which are the most frequent type of construction in the sample (65 per cent or 806/1,234). As shown in Table 8, they come in several different flavours, ranging from more forceful formats of advice as conveyed by modal auxiliaries of necessity to less forceful formats with modal auxiliaries of possibility. The necessity modals are exemplified by stronger core modals (must), various types of semi-modals (have to, need to, want to), and weaker core modals (should), in that order, and the possibility modals by can (stronger), and could, might, and would (weaker). The order of declaratives in Table 8 partly mirrors Reference Couper-Kuhlen, Thompson, Deppermann and HaughCouper-Kuhlen’s and Thompson’s (2022) order – you should/ought to → I’d/I would/I wouldn’t → you can/could – however, they remain agnostic about mixed constructions of necessity and possibility without explicit mention of the participants (there ought to be a textbook, it would be great to kind of push it down the line), as well as short fragments (maybe just after the end of the term). Both are placed on the lowest end of the continuum in Table 8 because they leave open the source of the advice as well as who is expected to carry out the action. As with interrogatives, explicit mention of the advisee in declaratives is considered more forceful than explicit mention of both parties. Other features that may make a difference are the polarity of the modal auxiliaries, different clause patterns, and the presence or absence of contractions, but as indicated in Table 8, this is a level of detail that we are not concerned with here. Advice in declarative form combines with the most diverse range of modifiers of all the construction types. Hedges are the most common (43 per cent), but other stance markers such as honestly and actually are also used with higher frequency than in other constructions (5 per cent). Interestingly, in our sample we did not find any instances of fragments modified by emphasisers, which supports their interpretation as the least forceful type of declarative construction due to their brevity and undemanding tone. As previously mentioned, we consider declaratives to be less impositive than interrogatives due to their lack of demand for an answer. However, we acknowledge the possible overlap between the more forceful types of declaratives (e.g., necessity modals with reference to the advisee) and the less forceful interrogatives (e.g., non-negated interrogatives with reference to the action), which needs to be taken into account in quantitative analyses based on the finer-grained classification.

At the lowest level of deontic force and epistemic commitment and confidence in Table 8 are conditionals, making up 5 per cent of the sample (58/1,234). They are divided into two types: complete and insubordinate. In terms of strength, the stronger ones are complete conditionals with both the matrix clause and the subordinate clause expressing a predicted outcome or a hypothetical situation. Insubordinate conditionals are independent conditional clauses where the resulting outcome, given the condition, is unexpressed. They are the least forceful constructions and come with weak speaker authority. Added to this, nearly half of all the conditionals in our data (48 per cent) are modified, typically by hedges and rarely by emphasisers, thus mirroring the pattern for fragments above. As was the case with constructions at the juncture of interrogatives and declaratives, conditionals may also overlap with weaker forms of declaratives in terms of directive strength (e.g., fragments). Therefore, the finer-grained classification might be a more useful reflection of the scale of directive strength than the general-level classification (see Section 4.1.2).

To some extent parallel to the scale of speaker/adviser authority is the scale of communicative (in)directness, which concerns advisers’ openness towards the advisee to alternative options and viewpoints. This scale is however not a simple reflex, or one-to-one match, of the modality scale of deontic and epistemic strength of speaker authority but overlaps with it at the top and bottom of the continuum. Unlike speaker authority, the scale of (in)directness is not primarily about deontic force and speaker commitment and confidence, but instead pertains to the intersubjective relationship and face concerns between the participants. In slightly different terms, we may say that (in)directness is at the juncture between the discursive strategies and the social and psychological processes related to the intersubjective relationship between the participants (cf. Section 1.1). As described in Section 2.3.1, damage to both the adviser’s and the advisee’s face are clearly at stake in situations where advice is given.

With respect to communicative (in)directness, imperatives are emblematic examples of the top of the scale. They are the most direct utterances that express advice directly and bald-on-record to the advisee. Also, at the strong end of the scale we find utterances that explicitly mention you as is the case with some performatives (I would strongly suggest that you …), the stronger examples of interrogatives (why don’t you …), and the stronger examples of declaratives (you must/need to/should). In these cases, the advisee is expected to take a stand for or against the piece of advice. This contrasts with conditionals or more indirect declaratives referring to situations and entities, where the absence of a response would not be strange. In communicatively direct utterances, the staging is simple in that the act of advice takes place in the actual speech situation as a bald-on-record directive message from the adviser to the advisee, and the latter can accept or resist the piece of advice by saying yes or no. In such cases, potential harm to the interlocutors’ face derives primarily from the unambiguous advisory nature of the utterances. The adviser cannot plausibly claim deniability, should the advice be unwelcome, and the advisee has fewer (linguistic and practical) opportunities to avoid engaging with the advice, thus risking damage to their own as well as the adviser’s self-image.

Now, indirectness may already come into the picture in these sequences in the form of various kinds of hedging strategies since it concerns the portrayal of a piece of advice where the speaker sets up an alternative mental space in order to move away from the here and now to a space that recognises that the advisee may have different thoughts about the situation (e.g., what about, I would, you can; Reference FauconnierFauconnier, 1994). This is an important strategy of intersubjective consideration of face to promote successful mutual coordination of mental states in communication. With respect to indirectness in the communicative sense of the term that we use here, speakers link the here-and-now space with an irrealis space (Reference ParadisParadis, 2009, Reference Paradis, Kloudová, Šemelik, Racochová and Koptik2020), which in turn has the effect of expanding the workspace of the common ground for the uptake/perlocution. By using alternative mental spaces, the adviser also saves their own face. As previously observed, the adviser’s self-esteem is at risk if the advisee is unwilling to accept the advice. The alternative mental space has the effect of playing down the level of epistemic control on the part of the adviser and encourages the advisee to assess other options (cf. Reference Couper-Kuhlen, Thompson, Deppermann and HaughCouper-Kuhlen & Thompson, 2022). Therefore, by giving the advisee a way to feel included in the process, advisers boost the advisee’s face at the same time as they reduce the damage to their own face in the case the advice is turned down.

The above types of indirectness are all spatial construals that have a mitigating effect on the intersubjective bonding between the adviser and the advisee. They portray the piece of advice in a more modest and considerate way than the strongly assertive direct forms where the speaker’s viewpoint is the non-debatable viewpoint. Such space builders establish mental spaces that are distinct from the reality space of the communicative situation. In the case of imperatives and deontic modal utterances with expressions such as you must/need to/should, the advisee has the option of accepting or rejecting the advice by saying yes or no, while in the case of speakers’ portrayal of the advice in a communicatively indirect way, the advisee is treated with intersubjective consideration as a thoughtful person that is capable of assessing a piece of advice and its consequences. In other words, directness in our sense has the effect of contracting the communicative space, while indirectness has the opposite effect of expanding it (Reference Põldvere, Fuoli and ParadisPõldvere et al., 2016). A simple response in the form of a yes or no in the uptake by the advisee is infelicitous in the case of advice presented in the indirect way. Some kind of elaboration on the part of the advisee is expected in such sequences.

4.1.2 Comparison of the Constructions across Corpora and Contexts

In this section, we compare the frequencies of the types of advice constructions in Table 8 across LLC–1 and LLC–2 and the conversational contexts, between equals and disparates, which by and large correspond to casual and institutional conversation. The goal is to track the development of the constructions over the past fifty years in spoken British English with respect to the degree of deontic and epistemic speaker authority as well as communicative (in)directness and the interlocutors’ intersubjective consideration of each other’s face concerns. The same principle will be applied to the conversational contexts, where the goal is to determine the role of context in constructional choice and thereby to complement existing conversation analytic investigations of advice with a more comprehensive understanding of how, if at all, advice-giving differs across casual and institutional contexts and among different constellations of people. We start with the diachronic investigation.

As previously mentioned, there are many more instances of advice in the sample from LLC–2 than LLC–1 (775 and 459 occurrences, respectively). While this difference is interesting, we refrain from drawing any conclusions as to the reason for the more frequent use of advice-giving in contemporary conversation compared to fifty years ago. Instead, we are interested in the constructions used in each period. The corpora follow the same pattern in terms of the proportions of the general-level constructions in Table 8 (imperatives, performatives, interrogatives, declaratives, conditionals). This is illustrated in Figure 2, which shows that the most common way to give advice in both LLC–1 and LLC–2 is through declaratives, followed by imperatives, interrogatives, conditionals, and performatives, in that order. However, the corpora differ from each other in terms of the more frequent use of imperatives and declaratives in LLC–2 compared to LLC–1, and the less frequent use of all other forms of advice-giving. Specifically, the regression model identified a significant association between LLC–2 and imperatives (ß = 0.8438, SE = 0.1313, z = 6.429, p < 0.001), and between LLC–1 and interrogatives (ß = −0.8438, SE = 0.2569, z = −3.284, p = 0.001), and conditionals (ß = −0.9820, SE = 0.2942, z = −3.338, p < 0.001).Footnote ¹⁴ Declaratives did not reach significance (ß = −0.2945, SE = 0.1503, z = −1.960, p = 0.05), and we excluded performatives due to their low number in the data.Footnote ¹⁵ In Table 8, we placed interrogatives and conditionals at opposite ends of the scale of directive strength, with imperatives at the very top. Therefore, these findings call for a closer investigation of what might be behind the seemingly contradictory results. On the one hand, it seems clear that the more recent dataset includes more imperatives and fewer conditionals; on the other hand, the older dataset also comprises some of the strongly directive, and unambiguously advice-giving, interrogatives. In other words, the shift towards more democratic and less deontic forms of advice that we hypothesised based on previous research appears not to have materialised.

Figure 2 Comparison of general-level advice constructions across LLC–1 and LLC–2

A more nuanced picture emerges when we consider the finer-grained constructions in Table 8. Some constructions were conflated due to low numbers, resulting in eight different types of constructions: affirmative/negative and exhortative imperatives; negated and non-negated interrogatives; declaratives expressing necessity, possibility, and declaratives with reference to the action (both mixed declaratives and fragments); and, finally, conditionals (both complete and insubordinate). The regression model revealed a significant association between LLC–2 and affirmative/negative imperatives (ß = 0.8829, SE = 0.1453, z = 6.078, p < 0.001), and between LLC–1 and negated interrogatives (ß = −1.4135, SE = 0.4242, z = −3.332, p < 0.001), non-negated interrogatives (ß = −0.6270, SE = 0.3083, z = −2.034, p = 0.042), declaratives with reference to the action (ß = −0.5569, SE = 0.1976, z = −2.819, p = 0.005), and conditionals (ß = −1.0211, SE = 0.3007, z = −3.396, p < 0.001). The rest of the construction types did not reach significance (see the supplementary material for details). The results support our decision to propose the fine-grained classification of advice in Table 8, without which we would have been unable to isolate the precise differences in construction types across the corpora. Specifically, we find that, while declaratives did not reach overall significance, those with reference to the action (it would be great to kind of push it down the line, maybe just after the end of the term) are more common in the older data. Compared to declarative constructions that place the burden of the action on specific individuals, declaratives with reference to the action have a more ambiguous and indirect advisory nature whereby the advisee is presented as a person who is capable of making their own decision.

On the one hand, then, the decline of less forceful types of declaratives and conditionals (if you feel like a film tomorrow night Mike) seems to point to a move towards stronger forms of advice-giving today. Indeed, it is the stronger forms of imperatives (affirmative/negative rather than exhortative) that are more common in LLC–2 (compare well don’t do that and let’s begin with your own department). On the other hand, it is clear that negated interrogatives, in particular, carry a similar force to imperatives with regard to directness in conversation and their potential impact on the interlocutors’ relationship. We suggest that what we are seeing over time is a change from interrogatives to imperatives in what is considered the default form for the most explicit and face-harming instances of advising. However, the rise of imperatives does not necessarily mean that advice today is overly indicative of unequal interaction. Instead, we argue that elaborate forms of advice-giving such as conditionals, despite being low on the face damage scale, can still be perceived as socially distancing due to excessive indirectness and elaborateness. They are more indicative of the kind of Victorian individualistic politeness observed by Reference Culpeper and DemmenCulpeper and Demmen (2011) in the nineteenth century. Democratisation, then, is not just a question of strong or weak modality, but of a more general shift towards more informal and less elaborate and convoluted ways of giving advice, which can include more direct ones, too. From the perspective of politeness-as-conventionalisation, we consider direct forms of advising as becoming, over time, more acceptable and socially appropriate, a result which ties in with Culpeper’s and Demmen’s observation about the use of requests in recent decades. The social acceptance of more direct forms of advising can be fruitfully explored in other aspects of interaction, too, which leads us to differences between conversations between equals and disparates.

For comparisons between conversations between equals and disparates, we considered the whole dataset to be able to draw meaningful conclusions about the two types of conversational context. As previously mentioned, advice-giving utterances in conversations between disparates in LLC are more than twice as frequent as in conversations between equals; they contain forty-nine utterances, while conversations between equals contain nineteen utterances (per 10,000 words). This is not surprising considering that, in the former, the conversations mainly take place in institutional contexts where advice is implicitly solicited through the roles given to the participants, while in the latter, advice is given in casual contexts and therefore not necessarily solicited or expected. In terms of the proportion of the general-level advice constructions in each conversational context, the results seem to mirror the patterns above, but with one major exception: in conversations between disparates, the third most common construction type is no longer interrogatives, but conditionals (see Figure 3). This was confirmed in the regression model, which showed a significant association between conversations between disparates and conditionals (ß = 0.82750, SE = 0.30830, z = 2.684, p = 0.007); for conversations between equals, a significant association was found for interrogatives (ß = −0.57401, SE = 0.26047, z = −2.204, p = 0.028). No significant differences between the conversational contexts were found for imperatives (ß = −0.02899, SE = 0.12040, z = −0.241, p = 0.81) or declaratives (ß = −0.08030, SE = 0.13955, z = −0.575, p = 0.565). Performatives were again excluded due to low numbers. As before, the fine-grained constructions revealed a more nuanced picture. In addition to the significant association between conversations between disparates and conditionals (ß = 0.789774, SE = 0.313086, z = 2.523, p = 0.012), the regression model also revealed a significant association between conversations between equals and negated interrogatives (ß = −1.058556, SE = 0.458611, z = −2.308, p = 0.021). The remaining differences were not significant (see the supplementary material).

Figure 3 Comparison of general-level advice constructions across conversations between equals and disparates

Taken together, the results seem to support the interpretation that, in casual conversation where advice is not typically the default aim of the conversation, it is given in a way that is more recognisable to the advisee, that is, by using communicatively direct forms of the type why don’t you (cf. Reference Couper-Kuhlen, Thompson, Deppermann and HaughCouper-Kuhlen & Thompson, 2022). Interlocutors who know each other well may also feel more comfortable giving advice in a direct way due to reduced face concerns. Constructions where the advisee is explicitly mentioned could therefore be seen as supportive moves between equals working together to address the problem immediately. This is different from conversations between disparates, which favour conditionals. The indirectness of conditionals could be perceived as unnecessarily roundabout and formal in casual conversation, but in institutional contexts where the participants typically are merely acquaintances, conditionals are useful because they soften the potential blow of the advice and draw the advisee in by opening up the space for alternative viewpoints. We observe further qualitative differences between the social settings when the conversations between disparates are broken down to show the direction of the advice-giving: to or from a person of higher status. When advice is given by a person of higher status, for example, by an employer to an employee or a supervisor to a supervisee, it seems to be more direct, particularly through the use of affirmative/negative imperatives. Such imperatives carry a high risk of damage both to the adviser’s and the advisee’s face as they carry the expectation of compliance from the advisee and position them as not being capable of considering alternatives, and they also position the adviser as being entitled to issue such uncontroversial advice. However, if the imperatives are of a positive kind, they could in fact enhance face by presenting the adviser as someone who cares about the interlocutor. We came across several such imperatives particularly in the LLC–2 data. Consider (22), a study abroad advisory session where adviser A’s negative imperatives (italicised) are meant to encourage student B to apply for a year abroad.Footnote ¹⁶

1. 1. A: but that’s it’s okay don’t worry but I would just say don’t worry don’t worry too much I know that’s easier said than done

   2. B: yeah <vocal desc=“laughs”/>

   3. A: cause you know it’s just a big it’s such an integral part of the next year <pause/> but I really wouldn’t worry too much … yeah oh also don’t worry I can’t stress that enough oh I feel really worried now that I’ve like <pause/> I’ve made things way more complicated yeah don’t worry

We also found other kinds of strong imperatives in the conversation, which concern the student’s choice of destinations (stick with those regions, don’t go too wild card, put California first). The adviser’s viewpoint is presented as the viewpoint with little space for the student to participate in the problem-solving process. However, the social setting is such that the use of these imperatives is unlikely to be affected by face concerns because the adviser’s knowledge and experience give her the authority to issue the advice, and in most cases the advice has also been made relevant by the advisee (e.g., I’ve made myself sick [worrying] about [it]). Moreover, this pattern seems to be more common in LLC–2 than in LLC–1. Consider the example in (23) from LLC–1, which is comparable to the examples from LLC–2; they all involve advice given by a person of higher status and are about a life-changing decision for the advisee. However, the difference is that, in (23), adviser A uses the conditional construction if you were living in London to guide advisee B’s future action.

1. 1. A: how far is it from Huddersfield to Coventry <pause/>

   2. B: uhm <pause/> about uhm a hundred miles <pause/>

   3. A: so in fact if you were living in London during that period <pause/> you would be closer

This apparent difference between the corpora supports the interpretation above, namely, that direct forms of advising have become more acceptable and socially appropriate today. However, more research is needed to unpack the changes in each social setting using more complex models, as well as with regard to changes within the construction types observed here. The fact that declaratives expressing necessity, for example, did not show any differences might be because this category is rather broad and includes necessity modals of varying strengths (compare must and should); their development may therefore be more complex than our analysis might have revealed (see Reference PõldverePõldvere, forthcoming).

Equally important to the advice frames is the outcome of the advice or how the advisee actually responds to the advice constructions in different discourse contexts, and the interaction of these factors with other relevant aspects of advice exchanges. For instance, the fact that advice is either explicitly or implicitly solicited in conversation, as in (22)–(23), does not necessarily mean that it is well received by the advisee, so the full dialogic sequence needs to be taken into account. In the following, we expand the current investigation by identifying the full frames of advice exchanges in LLC.