The interrogative flip with illocutionary evidentials

1 Introduction

The interrogative flip refers to the phenomenon of a marker that references the speaker’s perspective/point of view in declarative clauses “flipping” to the addressee in interrogatives (San Roque et al. 2017; Tenny and Speas 2003; Tenny 2006, and many others). This paper discusses this flip for illocutionary evidentials from the perspective of speech act theory. It is argued that the flip can be accounted for by analyzing such evidentials as contributing an evidential preparatory condition, requiring that the relevant evidence holder have the specified type of evidence. Who the relevant evidence holder is, is determined by sentence type: in declarative sentences the evidence holder is by default identified with the speaker S and in interrogatives with the addressee A. The discussion will focus on the reportative evidential in Cuzco Quechua (Quechuan: Peru). (1a) illustrates its use in a declarative clause, and (1b) its flipped reading in an interrogative. This evidential is of particular interest because it also has a non-flip reading in interrogatives in which the speaker asks the question on someone else’s behalf, as illustrated in (1c). The evidential condition EC is displayed on a separate line from the English translation to highlight the different interpretations.^[1]

The paper’s main argument is that evidentials in interrogatives put a constraint on the context in which the utterance is made just as they do in declaratives. In both sentence types, evidentials mark the speaker’s assumptions about the type of evidence that is available to the evidence holder and that is relevant to the issue under discussion at the time of speaking. This differs from previous analyses of the interrogative flip in which an evidential’s contribution is captured as a constraint on the possible answers (Bhadra 2017, 2020; Eckardt 2020; Faller 2002; Garrett 2000; Korotkova 2016; Lim 2015; Littell et al. 2010; Zimmermann 2004). The second main argument of the paper is that, despite constraining the utterance context, the evidential condition cannot be analyzed as a presupposition that is assumed to hold in the common ground CG (Stalnaker 1978, 2002). Instead, it should be considered a speaker presupposition, that is, a proposition that is taken for granted by the speaker, but which may be new information for the addressee. I argue that Gauker’s (1998) conception of presupposition is well suited to account for the evidential condition and preparatory conditions more generally. The discussion of these issues contributes to our understanding of perspective-sensitive linguistic expressions and types of presuppositions more widely.

The remainder of this introduction provides some basic background on Cuzco Quechua and its three main evidentials. Section §2 discusses how the contribution of illocutionary evidentials in declaratives can be analyzed as a preparatory condition in Searlian speech act theory. The meaning of the Cuzco Quechua evidentials in interrogatives is described in Section §3; Section §4 argues against analyzing evidentials in interrogatives as constraining the answers and instead proposes to analyze them as a preparatory condition also in this use. Finally, Section §5 discusses the status of the evidential preparatory condition as a speaker presupposition and Section §6 concludes.

Cuzco Quechua belongs to the Southern Peruvian variety of the Quechuan language family, which comprises 44 varieties in total (Eberhard et al. 2022). It has agglutinative morphology, and is mostly suffixing. It possesses a large number of enclitics, including three evidentials. In declarative sentences, these mark the speaker’s type of source of information for the proposition expressed (Cusihuaman 2001; Faller 2002; Muysken 1995; Weber 1986), as illustrated in (1a) for the reportative =si (allomorph =s), and in (2a, b) for the best possible grounds evidential =mi (allomorph =n)^[2] and the conjectural =chá (allomorph =cha).

2 The Cuzco Quechua reportative in declarative sentences

The Cuzco Quechua evidentials are best analyzed as illocutionary modifiers, that is, they constrain the illocutionary force with which the sentence they occur in can be used, not the truth conditions of the proposition expressed. I’ve argued for this analysis extensively in previous work, and will not repeat the arguments here, but the main observations are that the Cuzco Quechua evidentials cannot be embedded in conditional antecedents, scope under negation, or be agreed with or challenged (Faller 2002, 2007, 2011, 2014, 2019). The strongest argument for analysing the reportative as an illocutionary modifier is its ability to ask a question on someone else’s behalf, illustrated in (1c). Such examples can be relatively straightforwardly accounted for on the assumption that the reportative specifies the speaker’s type of grounds for making a particular speech act (see Section §4.3). However, there is considerable cross-linguistic variation in the empirical properties of evidentials, and different types of accounts will be appropriate for different types of evidentials. For example, some evidentials are best analyzed as modals, that is, quantifiers over possible worlds, which contribute to the truth conditions of the proposition expressed, and which have an evidential presupposition that restricts the set of possible worlds being quantified over to those in which the indicated type of evidence holds (Faller 2017; Izvorski 1997; Matthewson et al. 2007; Rullmann et al. 2008 and many others).^[3] Others are best analyzed as contributing not-at-issue propositions that directly update the CG, that is, without the addressee being expected to assent to it being added (Murray 2010, 2014). Evidentials also vary as to whether they can appear in questions and if so how they are then interpreted (San Roque et al. 2017). In sum, the different accounts proposed in the literature may reflect genuine cross-linguistic differences,^[4] and are therefore not necessarily in competition with other. The account developed in the following for the Cuzco Quechua reportative is only intended to extend to evidentials that have a similar empirical profile.

The discussion in this paper is couched within the Searle-Vanderveken framework (Searle 1969; Searle and Vanderveken 1985; Vanderveken 1990) for illocutionary force.^[5] In this framework, a speech act F(p) consists of an illocutionary force F, which combines with a propositional content p. F itself is complex, consisting of seven components which determine under what conditions a speech act is successful and non-defective. The illocutionary point is the essential purpose of a speech act which must be achieved in order for the speech act to be considered successful (Searle and Vanderveken 1985: 13–14). Of the other components, only sincerity and preparatory conditions are relevant to this paper. Sincerity conditions express the speaker’s mental state with respect to p, and are associated with a degree of strength, whereas preparatory conditions are propositions that are taken for granted in the utterance context by the speaker. These three components for simple, canonical assertion are given in (3) (adapted from Searle and Vanderveken 1985: 54–55).^[6]

Note that an assertion can still be successful even if the conditions in (3b) or (3c) are not satisfied. Such an assertion would be defective (insincere or unwarranted), but as long as it successfully represents as actual a state of affairs (3a), it counts as an assertion.^[8]

One of the main linguistic questions for this approach is to what extent the illocutionary force of an utterance can be derived from its linguistic form. I assume in this paper that while linguistic form typically does not fully determine illocutionary force, there are linguistic expressions which specify illocutionary components as part of their conventional meaning and in doing so constrain the illocutionary force that the resulting sentence can be used with.^[9] For example, the adverb alas adds the sincerity condition that the speaker S laments that p as shown in (4) (Vanderveken 1990: 150), while epistemic possibility adverbs arguably reduce the degree with which the speaker believes p, as shown in (5) for possibly (Wolf 2015: 77).^[10]

Sentence type does not conventionally code illocutionary force, though it correlates to some extent with it.^[11] Thus, declarative sentences are typically used for making assertions, but can also be used for asking questions when uttered with rising intonation, e.g., It is raining? Following Faller (2002, 2019), I therefore assume that declarative sentence type is conventionally only associated with the illocutionary force of presenting p as a proposition that is relevant to the question under discussion (QUD). However, in the absence of any illocutionary markers to the contrary, the assertive conditions in (3b, c) apply by default.

The various illocutionary components are not independent of each other in the sense that specifying one might imply another. For example, boasting has the sincerity condition “that the speaker takes pride in the existence of the state of affairs which is represented” (Vanderveken 1990: 138), which determines the preparatory condition that this state of affairs is good. With simple assertions, the preparatory condition that the speaker has evidence in support of p (3b) arguably implies that they believe p (3c).

On this approach then, illocutionary evidentials can be analyzed as adding a preparatory condition that specifies the type of evidence the speaker has in support of p. This preparatory condition in turn also implies a sincerity condition. Thus, the Cuzco Quechua evidential =mi specifies the preparatory condition that the speaker has the best possible grounds for p, which in turn implies that they believe p. Similarly, =chá specifies that the speaker can conjecture or infer p, which in turn implies that they believe p to a low degree. The reportative =si specifies that the speaker has reportative evidence for p, as shown in (6c) for (6a), but unlike the other two evidentials, it does not imply anything about the speaker’s belief regarding p. They may believe it or not, and even believe it to be false (Faller 2002, 2019). Instead, the reportative condition in (6c) implies that someone else believes p. Following Faller (2019), I refer to this third party as principal P, that is, “someone whose position is established by the words that are spoken, someone whose beliefs have been told, someone who has committed himself to what the words say” (Goffman 1979: 17).^[12] Note that the sincerity condition is still formulated in terms of the current speaker’s beliefs because it is their sincerity that is at stake in the current speech act of presenting p as relevant to the QUD, not the principal’s.^[13]

Thus, while the best possible grounds and the conjectural evidentials give rise to assertive illocutionary force in declarative sentences (weakened in the case of the conjectural), the illocutionary force associated with the reportative remains that of simply presenting p as relevant to, and possibly resolving, the QUD. Because the speaker does not express that they believe p even weakly, reportative presentations of p are compatible with the speaker asserting a contradiction of p. For example, (6a) was immediately followed up with (7).^[15]

The main question addressed in this paper is whether the illocutionary effects of evidentials when used in interrogatives can be accounted for in similar terms. The next section describes the interpretations of the Cuzco Quechua evidentials in interrogatives before moving on to the proposed analysis in Section §4.

3 The Cuzco Quechua evidentials in interrogatives

In interrogatives, the Cuzco Quechua evidentials may participate in the flip, that is, they can specify the addressee’s grounds for their answer as anticipated by the speaker. In wh-questions, evidential enclitics usually attach to the wh-phrase.^[16] (1b) above and (8a–c) illustrate the flipped reading of the reportative.

All three examples in (8) convey that the speaker expects the addressee to base their answer on what someone else said. In (8a, b), this is likely the referent of the subject, whereas in (8c) it is someone unspecified who may be identifiable from the context, for example, someone in charge of organizing the faena, a type of collective work event for the community.

With the best possible grounds evidential, the speaker expects the addressee to base their answer on best possible grounds, typically direct, first-hand evidence. This can, in context, result in an overall stronger force of the question, given that an addressee who has best possible grounds can be expected to give an answer.

Using the conjectural in an interrogative produces so-called conjectural questions, questions that do not necessarily expect the addressee to provide an answer. These questions still involve the flip reading, because, if the addressee nevertheless does provide an answer, the speaker expects them to only be able to conjecture it. The overall effect is one of the speaker “wondering”. Conjectural questions in Cuzco Quechua are analyzed in Faller (2023).

While I assume that the Cuzco Quechua evidentials are illocutionary modifiers, and this will play a role in their analysis, it should be pointed out that evidentials which contribute to the proposition expressed can also have flipped readings. This is shown for the reportative interpretation of the German (Germanic) modal sollen ‘shall’ in (11).^[17] The addressee is expected to provide an answer based on the penalty notice they received.

The Cuzco Quechua reportative however has an additional, speaker-oriented, reading that is not found with the German reportative, namely to ask a question on someone else’s behalf, as in (12a) (repeated from 1c). In this use, the reportative indicates that there is a third party source for the question speech act itself, not the proposition expressed.^[18]

The attempt of reproducing this with the German reportative sollen in (12b) fails, that is, it can only mean that the speaker is asking about what someone else says about how the addressee is. The analysis proposed in Section §4 is therefore only intended to apply to illocutionary evidentials. How the flip can be accounted for with non-illocutionary evidentials while preventing the on-behalf reading will need to be addressed on a future occasion.

Finally, it needs to be noted that there are evidentials in other languages that allow a speaker-oriented reading in which the speaker has evidence of their own in support of the proposition expressed, but asks the addressee for confirmation (Bhadra 2020; San Roque et al. 2017).^[19] This is not a possible reading with any of the evidentials in Cuzco Quechua, as far as I have been able to ascertain. It is an interesting question what factors determine which evidentials allow which readings, and I refer the reader to Bhadra (2020) for a discussion of this issue.

4 The pragmatics of the evidential flip

The interrogative flip with evidentials has received a substantial amount of attention in the previous literature, both from syntactic and from semantic-pragmatic perspectives. It is a type of perspective shift which other perspective-dependent linguistic items are subject to (Garrett 2000; Korotkova 2016; Tenny and Speas 2003; Tenny 2006). The type of perspective that is relevant varies for the different types of expressions. Thus, evidentials reference an evidence holder (Murray 2017), while epistemic modals reference a judge, that is, someone who evaluates the truth of a proposition (Bhadra 2017, 2020; Stephenson 2007). While these roles often coincide, and therefore tend to be subsumed under a general term such as perspectival center (Korotkova 2016: 237) or origo (Garrett 2000), they may also come apart. Reportatives are prime examples of the evidence holder being distinct from the judge/prinicipal, as illustrated in (6) (Faller 2019).

The illocutionary flip with evidentials concerns the evidence holder: it is the speaker in reportative presentations and assertions, and the addressee in questions. It is quite intuitive that this should be so, given that the point of a standard information-seeking question is to request the addressee to provide the speaker with an answer that resolves the QUD, while at the same time the speaker portrays themselves as not having the knowledge or evidence to resolve the issue themselves (Garrett 2000; Korotkova 2016). For Garrett (2000: 238), the role of evidence holder (origo in his terminology) is intrinsically linked to assertions, and it is therefore a pragmatic necessity that it is identified with the addressee in questions as the participant who is expected to assert an answer. Korotkova (2016) similarly argues that because sincere information-seeking questions are about the addressee’s information state, any element which makes reference to an origo’s perspective will be anchored to the addressee in questions. Such a purely pragmatic approach predicts that the shift might not occur with pragmatically non-canonical questions and Korotkova (2016: 252–254) shows that this prediction is borne out in, for example, quiz or biased questions. The pragmatic interaction of evidentials and questions may be further constrained by their syntax and semantics (e.g., according to Bhadra 2020, evidentials that obligatorily do not shift are ungrammatical in interrogatives, that is, they only occur in declarative sentences used as biased questions), but this does not mean that the shift itself is not pragmatic in nature. I therefore assume in the following that the identification of the evidence holder with the addressee in questions is a result of the pragmatics of information-seeking questions. My focus is however on how to treat the evidential condition in the pragmatics of questions once the evidence holder is resolved to the addressee.

5 The status of preparatory conditions

As mentioned in the preceding section, it seems fairly uncontroversial that a speaker who uses an evidential in a question presupposes the addressee’s type of evidence, even though most analyses of the evidential flip do not in fact implement it as a presupposition that needs to hold in the current context. However, considering only their use in declarative sentences, Faller (2002) argues explicitly against analyzing evidential conditions as preparatory conditions, which, recall, are felicity conditions that need to be presupposed or taken for granted by the speaker in order for speech act to be non-defective (Vanderveken 1990: 113). If preparatory conditions are understood as propositions that are taken by the speaker to be in the CG at utterance time, that is, constitute background information that the addressee as well as the speaker have accepted as true for the purpose of the current discourse (Stalnaker 2002), then evidentials cannot be presuppositions because the type of evidence the speaker has in support of a proposition is typically new information for the addressee (Faller 2002: 117; Murray 2010: 84). This however contrasts with their contribution in interrogatives, where, “the evidential content is presupposed: the speaker expects the addressee’s evidence to be of a certain type given previously introduced information. However, here it is clear that the contribution of the evidential must be in the common ground before the question, otherwise the question is infelicitous” (Murray 2010: 85). Consequently, Murray considers the evidential condition in questions a presupposition in the sense of Stalnaker (1973) (Murray 2010: 144).^[30]

In Murray’s analysis, evidentials contribute a not-at-issue proposition. She captures the difference in informativity of this proposition in the two sentence types as a difference in how it interacts with the CG. In declaratives, the evidential condition is directly added to the CG and eliminates any worlds from the context set (= the set of worlds in which all propositions in CG are true) in which the speaker does not have the indicated type of evidence. In interrogatives, however, the evidential condition does not eliminate any worlds, because it is required to be already in the CG at the point of utterance. It is however not clear what in this system ensures that the evidential condition in interrogatives is already in the CG at utterance time. Presuppositionality is usually treated as a lexical property, but as far as I understand, Murray does not treat the evidentials themselves as presupposition triggers. Doing so would amount to claiming a systematic polysemy such that every evidential that can occur in both declaratives and interrogatives has a non-presuppositional as well as a presuppositional sense. But even if one claimed such a polysemy, it is not clear what would prevent the non-presuppositional sense from occurring in interrogatives.

Without an account of how the evidential condition becomes presuppositional in interrogatives, there are two options. We can either treat evidentials as non-presuppositional in both sentence types (e.g., by analyzing them as a sincerity condition instead as in Faller 2002), but then lose the insight that interrogatives with evidentials are presuppositional, or develop an account in which they are presuppositional in both.

By analyzing the evidential condition as a preparatory condition, this paper explores the second option. The presuppositional status of the evidential condition in interrogatives is automatically accounted for, though not the fact that it is not necessarily a shared presupposition. For declaratives, what needs to be resolved on this approach is the fact that evidentials typically provide new information. This can be achieved by not analyzing preparatory conditions as presuppositions in Stalnaker’s sense but as propositions the speaker takes for granted, but which need not be shared by the addressee. This way of thinking about preparatory conditions is needed also for other types, for example, the preparatory condition of advice in (29).

Consider a situation in which someone made a mistake at work and intends to keep this from their boss. I might instead advise them that they should own up to it because I believe that this will be in their best interest. The addressee, at that point, does clearly not share this belief. Indeed, it seems that typically, advice is issued on the assumption that the addressee does not already accept that the advised course of action is good for them. If they did, then there would be no need to issue the advice. Similarly, consider the preparatory condition (i) that S does not know the answer in (23b) for canonical information-seeking questions. Usually, addressees only become aware that the speaker does not know the answer upon being asked. For example, if a stranger asks me for directions to the university, I would normally not have had any assumptions about what they knew or did not know about how to get there.

In each of these cases, one could appeal to accommodation, that is, the process by which the addressee comes to accept an informative presupposition into the CG as a result of the speaker manifestly having presupposed it (Stalnaker 2002: 710). However, as Murray (2010: 84) observes, presuppositions being informative “is a context-dependent effect, and not well suited to analyze the primary [emphasis added] effect of an entire class of morphemes.” This accords with Stalnaker’s (2002) view that accommodation exploits the standard way of conveying presuppositions: “If it is mutually recognized that a certain utterance type is standardly used, in some conventional linguistic practice, only when some proposition is (or is not) common belief, it will be possible to exploit this recognition, sometimes to bring it about that something is (or is not) common belief ….” That is, in order for the addressee to recognize the speaker’s intention for them to accept the presupposition into the CG, the utterance used must be of the type that is normally used in a context in which it is in the CG. To take Stalnaker’s running example, Alice can rely on Bob accepting that Alice has a sister when she utters “I have to pick up my sister from the airport” even if she knows that Bob does not know that she has a sister, because the possessive phrase my sister is standardly used in contexts in which the proposition “the speaker has a sister” is in the CG. That the informative use of a presupposition is an exploitation of its normal, shared use becomes clearer when considering examples that are not so easily accommodated. For example, explaining that one was late because “my fire engine broke down” (Levinson 1983: 205) would be inappropriate in most contexts in which the speaker is not part of the fire brigade, if the proposition that the speaker has a fire engine is not already in the CG.

However, as discussed above, the evidential condition in declaratives, the preparatory condition for advice, and the first preparatory condition of questions that the speaker does not know the answer, are standardly informative, not shared.^[31] Accommodation is therefore not the right tool to account for them. Instead of analyzing them as shared assumptions, the notion of presupposition proposed by Gauker (1998) is better suited for informative preparatory conditions. Gauker considers presuppositions to be the speaker’s take on those facts in the objective propositional context “that are particularly relevant to the conversational aims of the interlocutors, whether they are aware of these facts or not” (Gauker 1998: 150). This definition allows a speaker to take a proposition for granted without the addressee also being expected to share it, that is, presuppositions may be new information for the addressee. While Gauker aims to account for all types of informative presupposition, my aim here is more modest. As just discussed, the Stalnakerian accommodation approach works well for instances of informative uses of presuppositions that are normally used in contexts in which they are in the CG. But the Gaukerian conception is needed to account for the evidential condition and other preparatory conditions that are typically informative. While there is no CG in Gauker’s approach, he nevertheless seems to implicitly assume that the goal of discourse is for the interlocutors to come to a common understanding of what the actual context is. In cases where a speaker’s presupposition is new information for the addressee, they “may acquiesce in the speaker’s evident take”, especially when the speaker is viewed as having special authority (Gauker 1998: 168). Given that the speaker has special authority over their own source of information, the addressee will normally accept the evidential preparatory condition. While each participant has special authority over their own sources of information, this does however not mean that these are inscrutable. In fact, the fact that a speaker can use an evidential anchored to the addressee in a question requires that their source of information is accessible to the speaker as well. However, in a case in which the participants’ takes on the context diverge, it is the participant that the evidential is anchored to who takes precedence. Thus, as discussed in Section §4.1, if the speaker knows that the addressee spoke to someone about an issue but does not know that they also have direct evidence bearing on that same issue, then it is not infelicitous for the addressee to use the direct evidential in an answer to a question containing the reportative. Rather, it is the question that is defective (though not unsuccessful).

Given that preparatory conditions on this view do not need to be shared, the relevant context against which they are evaluated cannot be the CG. Instead, we might require that presuppositions are satisfied if the speaker believes that they are true (or is committed to them in a commitment-based framework). For example, we might say that the preparatory condition of having reportative evidence is met if the speaker believes that they have reportative evidence. However, Gauker argues instead that the right type of context contains facts, that is, propositions that are true in the objective context, not the participants’ beliefs about those facts. Sbisà (2002) argues in a similar vein that “cognitive” contexts (involving the representations of the intentional states of participants) are not sufficient to account for speech acts as social actions because they “demand too little of the agent” (p. 428). If only cognitive contexts were relevant, speakers would only be required to hold the relevant intentional state, but there would be no requirement to make sure that their intentional state correctly reflects reality. She gives the example of a speaker who mistakenly believes the door to be open and orders the addressee to close it. But if the addressee knows that the door is in fact already closed they would clearly be under no obligation to close the door as a result of the order.^[32] Thus, when evaluating whether or not a speech act is actually appropriate and not merely appears to be appropriate (p. 428), it is the actual circumstances that matter, not the speaker’s beliefs about them. This is even clearer for declarative speech acts, where a preparatory condition not holding results in an overall unsuccessful speech act. For example, if a low-level manager mistakenly believes that they have the authority to fire an employee, then sincerely uttering the words You are fired does nevertheless not constitute an act of firing.

On such an objective conception of context, speech acts with an evidential would be considered inappropriate in contexts in which the evidential condition does not hold. If someone was told that p but misremembers that they themselves witnessed that p, then using the direct evidential would be considered inappropriate even though the speaker believes that they have direct evidence. That this is the right approach is supported by an observation by Weber (1989: 424) about the use of evidentials in Huallaga Quechua. The neighbours of a man do not believe anything he says because he is known to constantly use the direct evidential and therefore “always speaks as though he had witnessed what he is telling about.” [quotation marks in the original]. The implication here is, though Weber does not say so explicitly, that the man in question did in fact not witness most of what he is telling about.^[33] That is, the evidential condition in this case does not hold, and the other community members therefore consider this man’s assertions inappropriate, regardless of what he himself believes. Thus, Gauker’s conception of the context seems to be applicable also to evidentials, though a more in-depth discussion of what role exactly it plays in communication is needed. Since it is not necessary to make the move to objective contexts to account for the interrogative flip with evidentials, I will leave this discussion here. The main aspect of Gauker’s approach to presuppositions that we need in the analysis of illocutionary evidentials is that the evidential condition is not necessarily a shared assumption.

6 Conclusions

The main aim of this paper was to argue that the evidential condition should be analyzed as a constraint on the current context also in interrogative clauses, instead of analyzing it as a constraint on the possible answers. This was implemented as a preparatory condition in the speech act theory of Searle and Vanderveken (1985) and Vanderveken (1990). The paper focussed on the reportative evidential of Cuzco Quechua, which was analyzed as contributing the preparatory condition that the evidence holder has reportative evidence that supports S raising the issue I denoted by the sentence it occurs in. In declarative clauses, this amounts to the condition that the speaker is in the possession of reports that support the proposition expressed p. In interrogative clauses, there are two readings: on the flipped reading, the addressee A is expected to base their answer on reportative evidence, on the on-behalf reading, the speaker has reportative evidence for someone else asking the question. In the two cases where the evidence holder is the speaker, the reportative moreover implies a modification of the corresponding sincerity condition. In declaratives, the speaker believes that a third-party principal P believes the proposition expressed, on the on-behalf reading of interrogatives, the speaker believes that P wants to know the answer. There is no effect on the sincerity condition on the flipped reading because the splitting off of the principal from the speaker role takes effect in the answer, and this is not recorded as a felicity condition for the current speech act. This is summarized in Table 1.

An anonymous reviewer has asked whether the current account is really that different from accounts that make evidentials part of the answers, given that placing a constraint on the current context crucially also affects the answers to a question on the flipped reading. To this, I would reply that the difference lies in what we take to be encoded by an evidential. In accounts that make the evidential semantically part of the answers, it is left to pragmatic inference that the speaker must have knowledge in the current context of what type of evidence the addressee is likely to have. On the account proposed in this paper, it is the latter aspect that is encoded, and it is left to pragmatics to explain why the default is for the addressee to use the same evidential in the answer as the one in the question. For languages that do not require answer parallelism, such as Cuzco Quechua, the current account is more appropriate. For languages such as Cheyenne in which non-parallel answers are considered infelicitous or non-compliant, the former type of account would be more appropriate.

The second aim was to discuss what type of constraint the evidential condition is. Because it is typically new information to the addressee in declarative sentences, it cannot be a shared assumption in the CG. Instead, Gauker’s definition of presupposition as the speaker’s take on the objective context is a better fit for the evidential condition, and indeed preparatory conditions more generally. Gauker’s notion of presupposition has also been appealed to by Matthewson (2006) to account for typical presuppositions triggers like hu7 ‘more’ and múta7 ‘again’ in St’át’imcets (Northern Interior Salish, Canada), which, unlike their English counterparts, do not “place constraints on the common ground of discourse.” (p. 63) The current paper therefore contributes to making the case for distinguishing between presuppositions that must be in the CG, and others that only require the speaker to take them for granted.

This paper has taken the perspective of classical speech act theory. In other work (Faller 2019, 2023), I have developed accounts of evidentials in Cuzco Quechua in the commitment-based framework of Farkas and Bruce (2010). My aim in using these different frameworks is to understand what (dis)advantages each of them has. As I show in Faller (2023) in particular, the commitment-based approach to discourse replicates several notions from speech act theory, the main difference being that commitments are public commitments which do not necessarily map onto the participants’ true beliefs. In making a discourse commitment a speaker takes on certain responsibilities (Alston 2000). For example, in committing to having a certain type of evidence the speaker assumes responsibility for producing their evidence if requested (Brandom 1994; Krifka 2014). However, the framework of Farkas and Bruce (2010) and work that builds on it, does not readily accommodate speaker presuppositions of the type discussed in this paper, which are not shared with the addressee. Faller (2023) therefore treats the evidential preparatory condition like a regular discourse commitment, which does however not quite capture the idea that the speaker takes it for granted. It is therefore an interesting question for future research how such presuppositions can be modelled.