Rhetorical questions and epistemic stance in an Italian Facebook corpus during the COVID-19 pandemic

2.1 Corpus

The corpus analysed, split into two sub-corpora, A and B, consists of 106 comments containing 139 questions, selected from a larger corpus of 627 comments responding to two posts published, respectively, on 16 and 18 March 2021, on the public Facebook page of the Italian newspaper Il Corriere della Sera, both concerning the potential side effects of the AstraZeneca vaccine. Comments constituting this larger corpus were manually extracted the day after the publication of the two posts so that only comments directed to the posts themselves were included (thus excluding the replies to these comments), to obtain a homogeneous dataset and to avoid confusion between users’ reactions to the content of the articles – on which our interest was focussed when we collected the corpus – and the interactions between the different users.

The post published on 16 March is headlined ‘Cooke (EMA): The benefits outweigh the risks. Thursday [18 March] the decision on AstraZeneca’, carries the short text ‘The European Agency: “Thursday the final decision, to date the benefits of the vaccine outweigh the risks”’ and includes a link to the open-access article with the declaration by Emer Cooke, director of the EMA. The post published on 18 March reports on the EMA press conference at which the vaccine was declared ‘safe and effective’. In this case, only part of the article, readable in full only by subscribers, is quoted in the text of the post. The headline claims, ‘The EMA: The AstraZeneca vaccine is not associated with an increased overall risk of thrombosis’ (translated by the authors).

The selection of the 106 comments, including questions, was jointly carried out by the three authors, manually analysing the entire dataset of 627 comments (by considering the lexical, syntactic, semantic, and pragmatic dimensions, i.e. not merely detecting the presence of question marks). All the comments, including questions, were copied onto Excel spreadsheets to be coded and analysed.

Sub-corpus A, referring to the post of 16 March, consists of 40 comments, including questions, extracted from a larger set of 189 comments. In the 40 comments, 61 questions were identified (24 comments contain only one question; 11 two; 1 three; and 3 four).

Sub-corpus B, referring to the post of 18 March, consists of 66 comments, including questions, extracted from a larger set of 438 comments. In the 66 comments, 78 questions were identified (57 comments containing only one question; 7 two; 1 three; and 1 four).

The quantitative and qualitative analysis presented in this study, therefore, concerned 139 questions, 61 from sub-corpus A and 78 from sub-corpus B.

The term ‘corpus’ used alone means ‘sub-corpus A plus sub-corpus B’. Studying the linguistic differences, if any, between the RQs in A and those in B is beyond the scope of this study. From the point of view of their rhetoricalness and epistemic stance, RQs of both A and B are regarded as belonging to the same corpus.

2.2 Procedures

The research was carried out in two phases. In the first phase, each of the three authors independently analysed the corpus of 139 questions following a shared ‘reading grid’ with four main items: plain questions, rhetorical questions, candidate rhetorical questions, and neither plain nor rhetorical questions. In the second phase, all three authors compared their individual analyses and reflected on the results. The interobserver agreement was calculated using Jamovi (version 2.3.21), a free and open statistical tool built on top of the R statistical language. The overall K Cohen score was highly satisfactory, resulting in 0.959. After discussing their previous coding decisions, a 100% agreement was reached.

During the individual analyses, particular attention was paid to two procedures aimed at clarifying as many methodological problems.

2.3 Theoretical framework

In this section, the KUB model is described as briefly as possible but in such a way as to provide sufficient information to the reader unfamiliar with it; it is then applied to the main types of plain (i.e. non-rhetorical) questions.

2.3.1 The KUB epistemic model

KUB is a theoretical model of epistemic stance (Bongelli et al. 2018, 2020, 2023, Riccioni et al. 2014, 2018, 2022, Zuczkowski et al. 2017, 2021), linguistically and empirically grounded, according to which a speaker/writer (S/W) can assume the positions illustrated in Figure 1, each having two sides: one evidential (concerning S/W’s access to the information, S/W’s source of information while speaking/writing), the other epistemic (concerning S/W’s commitment towards the truth of the information conveyed):

• Knowing (K)/Certain Position;

• Not Knowing Whether (NKW) – Believing (B)/Uncertain Position;

• Unknowing (U)/Neither Certain nor Uncertain Position.

Figure 1

The structure of an S/W’s epistemic positions (K, B, NKW, U), each of which may be imagined as a coin having an evidential (left) and an epistemic (right) side. For simplicity, their abbreviated names derive from the former aspects.

Information communicated from a K position, i.e. as known to S/W, is simultaneously communicated as certain, and vice versa, i.e. as something S/W says they know to be true (e.g. The benefits of the vaccine outweigh the risks).

Information communicated as uncertain is simultaneously communicated either as something S/W says they do not know whether (NKW) it is true or not (I don’t remember if the benefits outweigh the risks) or as something they say they believe (B) to be true (I think that the benefits outweigh the risks). From the NKW position, both alternatives p (the benefits outweigh the risks) and non-p (the benefits do not outweigh the risks) are communicated as having the same probability of being true: I do not know whether p or non-p ( = I am equally uncertain whether p is true or false). It is thus a matter of doubt for S/W, in the sense that they communicate to be in doubt between p and non-p. From the B position, the explicit, lexicalised alternative (p in this case) is communicated as having more probability of being true than non-p: I believe that p ( = I am inclined to believe that p is true).

NKW and B are the two sub-positions of uncertainty corresponding to as many poles in the uncertain epistemic continuum between which the different degrees of uncertainty range: NKW represents the maximum uncertainty (S/W’s doubt), B the minimum (S/W’s belief, supposition, opinion, etc.).

The U position concerns all types of information S/W communicates that they do not possess and have no idea about (I don’t know why the benefits outweigh the risks): a piece of information p is communicated as missing; S/W communicates that they have no evidential access to it. Therefore, they can have no kind of epistemic commitment (neither certain nor uncertain) towards the truth of p just because p is lacking or absent. The information gap (caused by the absence of the source) characterising this position corresponds to a commitment void: S/W cannot commit to what they do not know.

2.3.2 Questions: Unknowing or uncertain

Following Stivers and Enfield’s (2010) terminology and definitions, plain questions are of the following primary types: wh-questions (also called content questions), alternative questions, and polar questions (also called yes-no questions). These last are divided into three subtypes: polar interrogatives (PIs), tag questions, and declarative questions.

When the KUB model is applied to these question types (Zuczkowski et al. 2021), they originate from two distinct epistemic positions of the questioner Q: the Unknowing and the Uncertain (Figure 2):

Figure 2

Questioner’s Unknowing and Uncertain positions, response types, and social actions of the five primary types of plain questions.

Wh-questions (WHs) come from the Unknowing position and are thus referred to as unknowing questions. The Unknowing position can be hence specified by adding the seven English wh-words introducing WHs: specifically, it is an Unknowing what–which–who–where–when–why–how position.

Instead, alternative questions and the three sub-types of polar questions come from the Uncertain position and are thus referred to as uncertain questions. They convey different degrees of uncertainty, consequently occupying different points along the epistemic continuum of the Uncertain position: alternative questions and neutral PIs are information-seeking; therefore, they are closer to the NKW pole of maximum uncertainty, while non-neutral PIs, tag, and declarative questions are confirmation-seeking of a belief, supposition, opinion, and the like, so they are closer to the B pole of minimum uncertainty.

Following the two different but not mutually exclusive readings of PIs given in the linguistic literature since Coleman (1914) and Bolinger (1978), respectively, the KUB model distinguishes between neutral and non-neutral PIs: in the former, Q is indifferent (neutral) to a yes or no answer since they are seeking information, whereas in the latter, Q is not indifferent (non-neutral) to a yes or no answer, since they are searching for confirmation of a (positive or negative) belief, etc. (Zuczkowski et al. 2021: Chapter 5).

3.1 Common RQs

We refer to this type of RQ as common because they are similar to and representative of those most studied in the literature. Typically, its IAs are declarative sentences in the indicative mood, containing neither modal nor epistemic verbs, unlike the other three types found in the corpus.

Example 1 is paradigmatic of this type of RQ. We saw (Section 2.3.4) that the WH ‘Benefits of the vaccine, what are they?’ read as plain: originates from Q’s U position, targets R’s Knowing position, presupposes (that it is true that) the vaccine has some benefits, and can be epistemically represented as I do not know + what the benefits of the vaccine are.

Read as rhetorical, the IA becomes something like the vaccine has no benefit, which negates the presupposition of the plain question, ‘the vaccine has some benefits’. This feature of reverse polarity questions, where the IA inverts not only the surface question but also its presuppositions, is central, though often underemphasised in the literature. The IA ‘the vaccine has no benefit’, being a declarative sentence in the indicative without any uncertain or unknowing marker, originates from Q’s K position. Thus, Q’s epistemic stance representation would be: I know that the vaccine has no benefit. To represent the whole structure of this RQ, we can write the epistemic stance representations of both the plain WH and the IA one after the other in the following way: I (Q) explicitly tell you (R) that I do not know what the benefits of the vaccine are [= plain question representation], but I implicitly tell you that I know that the vaccine has no benefit [= IA representation].

This epistemic contrast – from Q’s U to K position – appears contradictory and, from a strict epistemic perspective, unacceptable. Nevertheless, it is characteristic of rhetorical WHs in the traditional view (Figure 3, Section 3.1.1, arrows 1 and 2), where rhetoricalness involves both polarity inversion and an epistemic shift: on the one hand, Q uses the question to say that they do not know p (what the vaccine benefits are); on the other hand, they use the IA to say that they know non-p (the vaccine has no benefits).

Figure 3

Origin and destination of the plain WH (continuous arrow 1) and two possible IAs (dotted arrows 2 and 3) delivering Knowledge or Belief, respectively. The epistemic destination is supposed to be U.

3.2 Adynatic RQs

The adjective adynàtic derives from the rhetorical figure Adynaton, which in ancient Greek (ἀδύνατον) means ‘impossible thing’. Adynatic RQs convey impossibility in their IAs. Therefore, these questions could be referred to as RQs of impossibility, but in assonance with the adjectives deontic and epistemic, we prefer to use the term adynatic, even because so far in the literature, to the best of our knowledge, such a type of RQ is not indicated, if ever, by a standard term.

In our corpus, adynatic RQs include in the question: the verb potere [can] in the affirmative, explicitly (8 questions) or implicitly (2 questions); the adjective possibile [possible] in the expression come è possibile [how is it possible] (1 question); the expression come faccio/fate a + infinito [lit. how do I/you make to + infinitive, corresponding to how can I/you + infinitive] (2 questions); the expression chi mi dice che… [who tells me that…] (1 question).

Of the 14 adynatic RQs, 8 are WHs and 6 PIs. The following Example 4 is a PI, of which we will give (i) presuppositions, (ii) Q’s epistemic stance representations of its reading as neutral, non-neutral, and rhetorical, and (iii) RQ structure representation.

Example (4) – B 176

Potevano dire diversamente quando in ballo non c'è la salute ma miliardi di euro?

Could they say otherwise when at stake is not health but billions of euros?

The presence of the angry emoji (repeated four times) not only signals Q’s mood but also serves as a clear clue that leads us to interpret this PI as rhetorical and to acknowledge its critical, provocative intent.

Read as neutral (could they say otherwise [or not]?), this PI would be information-seeking, addressing R’s K position and expecting either a yes or no answer.

As for presuppositions, they are different in PIs (and in alternative, tag, and declarative questions as well) than in WHs. Following Heritage (2010, 47), a PI like the one in our example would not presuppose that the proposition p they could say otherwise is true, but rather that it is possibly true, meaning it is also possibly false. Thus, a PI leaves two possibilities open: that p is true, and that p is false. If p is false, then non-p is true, and vice versa. Accordingly, the presuppositions in Example 4 would be (it is possibly true that) they could say otherwise and they could not say otherwise. Q’s epistemic stance representation would be: I DO NOT KNOW WHETHER they could say otherwise (or not). The expression I do not know whether…or… makes explicit the uncertainty (NKW position) that Q communicates with a neutral PI.

If read as non-neutral, the question would be confirmation-seeking of the opinion expressed by Q (they could say otherwise), which is of the same positive polarity as the question. Q’s epistemic stance representation would be: I DO NOT KNOW WHETHER they could say otherwise (or not), but I BELIEVE they could say otherwise. The expression I believe that… indicates that with a non-neutral PI, Q is advancing a supposition, etc., coming from the Believing pole of the Uncertain position.

However, in the actual context (the EMA declaration), the rhetorical interpretation, where Q’s opinion is opposite in polarity to that of the question, is more plausible: Q implies that they COULD NOT say otherwise, given the economic stakes. The IA (they could not say otherwise), paraphrasable with IT WAS IMPOSSIBLE for them to say otherwise, exemplifies what we term an RQ of impossibility. This IA matches one of the plain PI’s presuppositions (the negative one), and is of inverse polarity to the positive one expressed in the question.

According to the traditional epistemic reading, the IA would be placed in a Knowing frame: I DO NOT KNOW WHETHER they could say otherwise (or not), but I KNOW they COULD NOT say otherwise. Here, the epistemic contrast is between NKW and K. In our view, such a representation appears to be epistemically inconsistent: it is contradictory to state, on the one hand, I do not know whether p or non-p and on the other hand, simultaneously, I know that non-p. Unless this epistemic contradiction is viewed as hyperbolic, as written in the previous section: I ask you something Uncertain (NKW), but I assert its Known opposite (K). This exaggerated contrast between not knowing whether and knowing would give strength to the assertion, from a hyperbolic perspective.

An alternative epistemic stance representation, replacing the verb knowing with believing, would be: I DO NOT KNOW WHETHER they COULD say otherwise (or not) but I BELIEVE they COULD NOT say otherwise. This removes the epistemic contrast. Q remains within the Uncertain continuum, shifting from NKW (plain PI) to B (IA), without stepping into the Knowing position (Figure 1). Such a shift is coherent, since even rhetorical PIs are structurally uncertain questions and cannot communicate knowledge, only belief. Belief is the pole of minimum uncertainty, bordering the K position. The difference between NKW and B in PIs is not a contrast like that one between U and K in WHs; it is a displacement within the same uncertain epistemic continuum. Nonetheless, the question retains its rhetorical status.

Thus, as in the previous section, the epistemic stance representations of both the plain PI and the IA can be expressed as: I (Q) explicitly tell you (R) that I do not know whether they COULD say otherwise (or not) [ = plain question representation], but I implicitly tell you that I KNOW/BELIEVE they COULD NOT say otherwise [= IA representation].

Let us recap what has been described so far in one single figure (Figure 4):

Figure 4

The continuous arrows 1 and 2 refer to the epistemic origin and destination of the plain PI read as neutral and non-neutral, respectively; the dotted arrows 3 and 4 signal two different epistemic origins of the IA. The epistemic destination is supposed to be U.

The social action performed by a neutral PI is to seek information (arrow 1); that of a non-neutral is to express an opinion and seek confirmation (arrow 2): the opinion is of the same polarity as the question. In contrast, a rhetorical PI has the function of expressing an opinion (mild epistemic reading of the IA, B position) or knowledge (strong epistemic reading of the IA, K position) of opposite polarity to that of the question and to one of the two assertions making up its presupposition.

In Example 4, the polarity of the IA they COULD NOT say otherwise is inverse to both the question COULD they say otherwise? and one of the two assertions (the positive one) forming its presupposition: they COULD say otherwise and they COULD NOT say otherwise. From this point of view, the IA is identical to the negative presupposition they COULD NOT say otherwise. This continuity among the four rings composing the rhetorical chain (presuppositionsplain questionRQIA) is outlined in Figure 5.

Figure 5

The epistemic continuity of neutral, non-neutral, and rhetorical PI in relation to polarity inversion, presuppositions, and social actions.

Being a PI, Example 4 can be considered representative of all rhetorical PIs and shows that, for a PI to be rhetorical, at least four shifts are required: (1) a shift from question to assertion, i.e. a change in the linguistic action performed by Q; (2) a shift from positive to negative polarity (or vice versa), as far as question and IA are concerned; (3) a shift from positive to negative polarity (or vice versa), as far as presupposition and IA are concerned; (4) a shift in Q’s epistemic position: from NKW to K or B.

3.3 Deontic RQs

In the corpus analysed, deontic RQs are those whose IAs include a deontic verb either in the affirmative or negative, which may or may not be part of the questions. Of the 15 deontic RQs (12 WHs and three PIs), 3 contain a deontic verb in the question (2 dovremmo [we should], 1 bisogna [one must]), the remaining 12 only in the IA: 2 of them include the expression come mai (lit. how ever = how come) in the question, 2 others come mai non… [how come… do not…], as in the following

Example (5) – B 98

Nn diranno maiiiiii che [il vaccino] è la causa [delle morti].

Hanno speso milioni e milioni per questo pessimo vaccino potranno buttarlo??

Allora ecco il crimine dicendo che è sicuro.

BUTTATELO

A Napoli è appena purtroppo deceduta un’altra donna (Nome) (Cognome) come mai nn ne

parlate?????

They will never say that [the vaccine] is the cause [of the deaths].

They spent millions and millions on this worst vaccine will they be able to throw it

away?

So here’s the crime by saying it’s safe.

THROW IT AWAY

In Naples just unfortunately died another woman (Name) (Surname) how come

you don’t talk about it?????

The whole content of this comment can be paraphrased as follows: Q urges the disposal of this ineffective vaccine (throw it away!) because, despite the EMA’s claims of its safety, Q believes it is the cause of the deaths. The EMA will continue to administer it because its production has been costly. The final sentence (the guilty silence about the latest vaccine victim in Naples) sounds like an accusation, probably directed at journalists.

The comment contains two RQs. The former potranno buttarlo?? [will they be able to throw it away??] (closed by two question marks) is an adynatic PI, i.e. an RQ of impossibility (Section 3.2). Its IA is they WILL NOT be able ( = IT IS IMPOSSIBLE for them) to throw it away (polar inversion: positive→negative). The reason for the impossibility is specified in the preceding sentence: they spent millions and millions on this ineffective vaccine. The second question come mai non ne parlate????? [how come you don’t talk about it?????] (note the five question marks) is the one we are most interested in because it is a deontic rhetorical WH introduced by a particular expression, both in Italian and English, come mai (lit. how ever = how come). In Italian, the peculiarity is that the interrogative adverb come [how] is followed by the time adverb mai [ever], and in English by the impersonal verb to come. In both languages, the expression can be replaced by perché [why].

In our rhetorical reading, the IA assigned to the question includes a deontic verb in the indicative or conditional mood, although such a verb does not explicitly appear in the question: voi dovreste/dovete parlarne [you should/must talk about it]

The first IA that we attributed to the RQ is the one with the deontic verb in the conditional (should), and the second is the one in the indicative (must); the former seems more in keeping with this example. The difference between the two moods of the deontic verb can be traced to two slightly epistemically different IAs as follows:

1. conditional mood = mitigated IA, I BELIEVE you should talk about it = mitigated RQ;

2. indicative mood = strong IA, no epistemic mitigation, I KNOW you must, you have to talk about it = strong RQ. The indirect message conveyed by the RQ would be something like: it is absolutely necessary, dutiful, obligatory to talk about it.

   Besides the combinations believe + conditional and know + indicative, at least in Italian, two other representations are possible, ‘mixed’ so to speak:

3. I BELIEVE + present subjunctive or indicative: CREDO che dobbiate/dovete parlarne [I BELIEVE you must talk about it]. This representation is less mitigated than (1).

4. I KNOW + present conditional: SO che dovreste parlarne [I KNOW you should talk about it]. This representation is less strong than (2).

Therefore, if the above four representations are listed in order of decreasing strength (or increasing mitigation), the order is 2, 4, 1, 3. In 2 and 4, the IA is treated as Q’s piece of knowledge; in 1 and 3, as Q’s belief.

Regardless of whether the IA is an instance of KNOWLEDGE or BELIEF, it has an inverse polarity (1) to that lexicalised in the question (negative →positive) and (2) to the presupposition of the plain WH as well. So we can now give Q’s epistemic stance representation of the RQ: I explicitly tell you that I DO NOT KNOW how come you do not talk about it, but I implicitly tell you that I BELIEVE/KNOW you SHOULD/MUST. We omit a further figure, as it would closely resemble Figure 3.

3.4 Epistemic RQs

In the KUB model, plain questions, either unknowing or uncertain, can be addressed not only to R’s K but also to the B position, i.e. to somebody who is not expected to know the answer but to be able to advance an assumption, supposition, opinion, and the like (Section 2.3.3). Such questions usually embed an epistemic marker of uncertainty, either lexical or morphosyntactic (Zuczkowski et al. 2021: Chapter 9). For this reason, they can be called epistemic.

The 20 epistemic RQs (14 PIs, five WHs, and one alternative question) found in the corpus contain both in the question and IA: an epistemic verb (think, seem…, 13 questions) or an epistemic expression (in your opinion…, two questions) or an epistemic future (i.e. will it be a casualty that…, two questions) or the verb volere [to want] used epistemically ( = meaning think, believe or pretend, expect, presume, suppose, etc.) in the expression cosa vuoi che [what do you want that] + verb (three questions).

Q addresses 17 out of the 20 questions to the potential readers of the comment: half of them are addressed to VOI [you plural], two to a generic TU [you singular], two to NOI [we], three to an INDEFINITE PRONOUN (two qualcuno [anyone], one chi [who]); the remaining three questions employ two epistemic futures and a hypothetical construction.

Example (6) – A 19

e vi sembra poco!!!? intanto per il vaccino quelle povere persone non ci sono più e nessuno

risarcisce i parenti di quelle persone ….

and does it seem little to you!!!? meanwhile because of the vaccine those poor people

are no longer there and no one refunds the relatives of those people ….

Starting with the IA helps clarify what is new in this type of RQ compared with previous ones. The IA is: [the EMA Declaration ‘the benefits outweigh the risks’] DOES NOT seem little to me, and I BELIEVE it does NOT seem little to you either. The novelty is that these RQs involve in their IAs not only the potential readers of the comment (as they explicitly do in the questions) but also Q implicitly.

During the second inter-subjective phase of our analysis (Section 4.3), we realised that the spontaneous IA assigned to all epistemic RQs communicates an opinion of Q (the EMA declaration DOES NOT SEEM little to me […]) and a belief about the potential readers’ opinion on the same topic ([…] and I BELIEVE it DOES NOT SEEM little to you either). The polarity of both opinions is inverse to that of the question (DOES it seem little to you?). Let us see this in detail.

The question is a PI embedding the epistemic verb seem. You refers to the potential readers of the comment. It refers to the EMA Declaration, ‘The benefits outweigh the risks.’ Context, co-text (in this case, what follows the question), content of the question, and punctuation (three exclamation marks before the question mark) support a rhetorical reading. As usual, the PI will be analysed progressively, first as if it were neutral, then non-neutral, and, finally, rhetorical, as it is, to gain a deeper understanding of the latter’s structure.

Read as neutral, the question presupposes two opposite possibilities: it is possibly true that [the EMA Declaration ‘the benefits outweigh the risks’] seems little to you, and it is possibly true that it does not seem little to you. Q does not know whether the former or the latter is true and seeks information; the expected answer would indifferently be yes or no. Q’s epistemic stance representation would be: I DO NOT KNOW WHETHER it seems little to you (or not).

In a non-neutral reading, the question would communicate Q’s belief about the potential readers’ opinion (I BELIEVE it seems little to you), and that opinion would be present explicitly in the question, that is, it would be the one lexicalised in the question (it seems little to you). Q’s epistemic stance representation would be: I DO NOT KNOW WHETHER it seems little (or not) to you, but I BELIEVE it does.

In its actual rhetorical reading, the question communicates a Q’s belief (I BELIEVE it DOES NOT SEEM little to you). This belief: (1) is implicit (expressed through the IA), (2) of inverse polarity to the lexicalised question and to one of the two presuppositions (it seems little to you), and (3) it concerns both the readers and Q themselves (it DOES NOT SEEM little to me either). Since Q cannot KNOW WHETHER the readers also think as they do, WHETHER they share their opinion, and much less can Q KNOW THAT they also think as they do, in the following Q’s epistemic stance representation, we prefer I BELIEVE instead of I KNOW.

Q’s epistemic stance representation: I DO NOT KNOW WHETHER it seems little to you (or not), but it DOES NOT SEEM little to me, and I BELIEVE it DOES NOT SEEM little to you either.

RQ structure representation: I explicitly tell you that I DO NOT KNOW WHETHER it seems little to you, but I implicitly tell you that it DOES NOT SEEM little TO ME, and I BELIEVE that it DOES NOT SEEM little TO YOU either (Figure 6).

Figure 6

Epistemic origin (NKW) and destination (B) of the question read as neutral PI (arrow 1). Epistemic origin (B) and destination (B) of the question read as non-neutral PI (arrow 2). Epistemic origin (B) and possible destination (U) of the IA of the question read as rhetorical (arrow 3).

At the very moment when Q asks the potential readers their opinion (I ask you if it seems little to you), in the IA Q expresses their own opinion (it doesn’t seem little to me) and belief that the other readers share their opinion (I believe that it does not seem little to you either).

4.1 Quantitative results

Out of 139 questions in the corpus, 75 are RQs (research question 1), and almost all of these 75 are WHs (47) and PIs (26) (research question 2).

4.2 Four types of RQs

During the research, the qualitative analysis of the 75 RQs revealed the presence of recurring linguistic structures in their IAs allowing the identification of four specific types of RQs: common (the most studied in the literature: usually their IAs are made up of declarative sentences in the indicative or conditional without any uncertainty or unknowledge markers), adynatic (or RQs of impossibility), deontic (or RQs of duty/obligation), and epistemic (or RQs of belief). The main difference between them is that, in their IA, the adynatic and deontic include a modal verb, the epistemic an epistemic expression of uncertainty, while the common neither. Concerning epistemic RQs, as far as we know, the only one who has dealt with something similar is Lee-Goldman (2006), calling them think-RQs, i.e. RQs including do you think/imagine/suppose. He only investigated WH questions and, mainly, their pragmatic function. Han (2002) analyses RQs containing deontic modals from a predominantly logical-semantic perspective, without addressing epistemic stance. Finally, to the best of our knowledge, no studies appear to have explicitly addressed RQs containing verbs or expressions of possibility. As the more detailed quantitative results reveal (Tables 2 and 3), there are clear differences in the amount of RQ types between the two sub-corpora. In sub-corpus A, common RQs are the most frequently used, accounting for 57.1% of the total. In sub-corpus B, on the other hand, the most used types of RQs are epistemic (35%), followed by adynatic and deontic (each at 25%); common are only 15%. We can assume that these marked differences are related to the content of the two posts commented on by Facebook users. Specifically, the first post (16 March), which emphasises that the benefits outweigh the risks and anticipates a final EMA decision on Thursday 18 March, elicits widespread use of common RQs, most of which are WHs (48.6% of all the RQs in sub-corpus A) focussing precisely on questioning ‘the benefits outweigh the risks’ statement (A 17: Ma quali benefici? [But what benefits?]; A 20: E che significa? [And what does this mean?]; A 64: Benefici per chi? [Benefits for whom?]; etc.). Conversely, the second post (18 March), which claims that the vaccine is ‘safe and effective’ and without an associated global risk of thrombosis, as shown by the tone of many comments, arouses disbelief, scepticism, anger, and sarcasm (sometimes even conveying radical anti-vaxxer or conspiratorial positions), and solicits comments (mostly in opposition to the communicated content) conveyed by epistemic, adynatic, and deontic RQs. Particularly: (a) epistemic RQs communicate users’ belief and/or certainty about the results of the tests and seem to be oriented towards seeking the consensus of other users, often with direct appeals to potential readers to confirm their own views through PIs (22.5% of RQs in sub-corpus B) (see Example 6, but also: B 41: Qualcuno aveva dubbi? [Did anyone have doubts?]; B 341: Avevate qualche dubbio che Ema desse il consenso per procedere alla somministrazione del vaccino? [Did you have any doubt that Ema would give consent to proceed with the vaccine?]; etc.); (b) adynatic RQs communicate users’ belief and/or certainty regarding the inadequacy of the tests carried out by EMA in only a couple of days and the impossibility of an alternative decision, mainly for economic reasons and interests (see Example 4, but also: B 14: Si poteva affermare il contrario? [Could the contrary be claimed?]; B 215: E cos’altro poteva dire un comitato che fa tutti gli interessi di Big Pharma? [And what else could a committee that serves all the interests of Big Pharma have to say?]); (c) deontic RQs communicate users’ belief and/or certainty that the information provided by the post (but also more generally by the mass media, journalists, health and political authorities, etc.), in particular with regard to the risks of the vaccine, is incomplete or insufficient, contrary to what it should be (to be convincing) (see Example 5, but also: B 210: come mai così tanto illustri scienziati non hanno saputo studiare un protocollo di cura???? [why have such distinguished scientists failed to study a treatment protocol????]).

4.3 Epistemic positions and hyperbole

Through the qualitative analysis of four examples of questions (two WHs and two PIs), Q’s epistemic position from the KUB perspective was shown in both plain and rhetorical readings of the above questions through plausible semantic representations and figures with circles and arrows (research questions 3 and 4). So far in the literature, the covert message of an RQ is decoded as an assertion, i.e. a declarative sentence usually in the indicative or conditional mood, with no uncertainty or unknowledge markers, thus inevitably decoded as communicating a K position (e.g. Koshik 2005), as it happens with common RQs. However, any assertion can be mitigated just by adding a lexical or morphosyntactic uncertainty marker. In the KUB model, the theoretical macro-marker for mitigation is I BELIEVE. We thus showed that (i) any IA can be inserted into an epistemic frame including either I KNOW or I BELIEVE; (ii) mitigation (I BELIEVE) does not cancel the rhetorical status of the question, it only dilutes it; (iii) in the semantic representations of three examples (1, 4, and 6), I BELIEVE seems even more epistemically appropriate than I KNOW; in the remaining example 5, both I BELIEVE and I KNOW seem acceptable. What, then, is the ultimate epistemic structure of an RQ resulting from applying the KUB model to the analysis of the above examples? An RQ is a hybrid not only of two different linguistic actions, as is well known in the literature, but as many (or more) epistemic positions as well: on the one (explicit) hand, a plain question seeks a piece of information p (or non-p) and delivers either a U (in WHs) or an NKW position (in PIs); on the other (implicit) hand, the IA gives an opposite information non-p (or p) and delivers either a K or a B position in both WHs and PIs, i.e. either a piece of knowledge (strong assertion) or belief (mild assertion). Therefore, from the epistemic view, this hybrid consists of four possible combinations of Q’s epistemic positions: (i) two epistemically unacceptable contrasts U versus K (in WHs) and NKW versus K (in PIs), which can only be accepted if understood as hyperbolic and functional to an IA intended as strong; (ii) two epistemically acceptable, i.e. non-contrasting, non-contradictory, combinations U–B in WHs and NKW-B in PIs, functional to a mitigated IA. The notion of hyperbole is thus central to solving the two epistemic contrasts outlined above in (i): they are viewed as an epistemic exaggeration, a trespass from B to K, acceptable only if the IA is considered a hyperbolic overstatement empowering itself. There is no epistemic contrast in the other two epistemic combinations outlined above in (ii), U–B in WHs and NKW-B in PIs: when Q DOES NOT KNOW p (in plain WHs) or DOES NOT KNOW WHETHER p (in plain PIs), it is epistemically acceptable to BELIEVE that non-p (in IAs), i.e. to deliver a mild assertion. In the former combination (U–B), Q moves from the U to the B position, not the K; in the latter (NKW-B), Q remains within a single position, the Uncertain one, moving from the NKW to the B pole, a displacement along the same epistemic continuum.

4.4 Presuppositions

In addition to the obvious links between RQs, IAs, and the corresponding plain questions, we felt it relevant to emphasise the less obvious links with the fourth ring of the rhetorical chain, namely the presuppositions of plain questions. We recap WHs first, then PIs. The presupposition of the plain WH what are the vaccine benefits? is (it is true that) the vaccine has some benefits; the IA (sans the epistemic markers I KNOW or I BELIEVE) is the vaccine has no benefits: its polarity is inverse to that of the RQ, as is widely argued in the literature, but the actual negation (of both the verb and indefinite pronoun, at least in Italian) concerns the presupposition, a feature not as widely emphasised in the literature (Ene 1983): the vaccine has some benefits becomes the vaccine has no benefits. Similarly, the presupposition of the plain PI could they say otherwise…? is (it is possibly true that) they could say otherwise and (it is possibly true that) they could not say otherwise; the IA is they could not say otherwise, which is identical to the negative presupposition and is of inverse polarity to the positive presupposition lexicalised in the question.

The fact that this study is corpus-based and, therefore, that the findings result from a quantitative and qualitative bottom-up linguistic analysis is one of its strengths and originality. However, this very feature could be a limitation but also a stimulus for further applications of the KUB model to RQs. The corpus analysed has some peculiarities: it is of limited size, it is written (or rather, it is an expression of the spoken-written style typical of social networks), and it is in Italian. It might be, therefore, interesting to extend the study to: (a) wider corpora, which would not only enable a bottom-up analysis of a larger number of occurrences of RQs, referring to different contexts (discursive genres, topics, etc.), but could also allow verification of the possible existence of other types of RQs based on the linguistic features of their IAs. It is reasonable to assume that the four types of RQs identified in this study do not exhaust all possibilities. For instance, since adynatic and deontic are concerned with the modal verbs can and must, respectively, one might assume that another potential type of RQ could involve the third modal verb want. From a typological perspective, the present study opens up a new research perspective; (b) spoken corpora, to be able to adopt a conversational and sequential perspective (and, therefore, to consider the management of RQs by the receiver(s), but also by Q themselves). Our next research goal will be to study R’s possible replies to Q’s RQs in a corpus that identifies sequences where a next turn-proof procedure can be applied. A sequential analysis, including replies to RQs, especially in the presence of epistemic disagreement or conflict between R and Q (Linares-Bernabéu 2023), would make explicit the different epistemic positions (strong or mitigated) that R and Q assign to the same IA; (c) corpora in other languages, to test possible similarities and variations.

Future research may also concern those questions (which we have only mentioned) featuring IAs but without polarity inversion, which were classified as neither plain nor rhetorical here.

In summary, this study integrates existing theoretical approaches and deepens our understanding of RQs in discourse. Its findings offer original insights by highlighting the nuanced epistemic positions encoded in RQs and by proposing a novel, corpus-driven typology within a clearly defined theoretical framework. This contribution not only enriches the understanding of RQ usage but also opens new avenues for research at the intersection of linguistic form, speaker stance, and pragmatic effect.

A further element of novelty lies in the application of the notion of hyperbole to explain epistemically conflicting configurations as acceptable exaggerations that enhance rhetorical force.

While the empirical, qualitative, and quantitative bottom-up methodology is a major strength of the study, it may also represent a limitation but also a stimulus for further applications of the KUB model to RQs. Further research should aim to test, refine, and expand the findings across larger and multilingual corpora, as well as in spoken interaction, particularly by exploring sequential dynamics and responses to RQs in dialogic settings. Additional types of RQs may emerge, and the model itself could support broader applications in discourse analysis, particularly in contexts involving epistemic asymmetry and conflict in digital and persuasive communication.

By offering both empirical and theory-driven systematic approaches to RQs, this study contributes to linguistic pragmatics, discourse analysis, and epistemic modelling. It also provides tools for analysing implicit communication in complex social contexts, including misinformation, institutional communication, and public opinion formation.