Assessing alternatives: the case of the presumptive future in Italian

Abstract

In this paper, we study the distribution and interpretation of a non-temporal use of the future tense in Italian, called ‘presumptive’ or ‘epistemic’, which we label here PF. We first distinguish PF from its closest modal relatives, namely epistemic necessity/possibility/likelihood modals, as well as weak necessity modals. We then propose an account of PF in declaratives and interrogatives that treats it as a special comparative subjective likelihood modal, and test its empirical predictions. A theoretical lesson drawn from this detailed study of the semantics of PF is that semantics needs sharpened theoretical tools to be able to capture the fine-grained distinctions languages make when it comes to signaling modulated epistemic commitment to a proposition.

1 Introduction

Languages use a variety of expressions to signal nuanced commitment to a proposition, such as epistemic modals (Gianni may/must be home), weak necessity modals (Gianni ought to be home), and likelihood adverbs (Gianni is likely to be home). In this paper, we argue that the non-temporal use of the future tense in Italian is one of the ways commitment can be modulated in this language. In our account, the non-temporal use of the future tense in Italian is similar to, yet different from, the modals mentioned above.¹

The future tense in Italian has an ordinary temporal use, exemplified in (1).²

This morphological form also has a non-temporal use, called here the presumptive future (PF), which allows co-occurrence with non-future temporal adverbs, exemplified in (2) and (3).³

As illustrated in (4) and (5), PF also occurs in constituent and polar interrogatives.

In the cases exemplified above, the future acquires a non-temporal flavor that has been labeled ‘presumptive’, ‘epistemic’ or ‘evidential’. Because of the non-future adverbs, examples (2) and (3) can only receive a presumptive interpretation, according to which the speaker expresses, roughly speaking, her best guess with respect to Gianni’s whereabouts.⁴

Other languages within and outside the Romance family that have similar but not necessarily identical non-temporal uses of the future are French, Spanish, Romanian, Dutch, Greek, and English. For earlier work see Fălăuş (2014), Fălăuş and Laca (2014), Frana and Menéndez-Benito (2015, 2019), Giannakidou and Mari (2013, 2017), Irimia (2010), Mari (2012), Mihoc (2014), Mihoc et al. (2019), Winans (2016), among others. We concentrate here on Italian, leaving a cross-linguistic comparison for another occasion.

In what follows, we differentiate PF from epistemic modals, simple likelihood modals, variable-force modals, and weak necessity modals in Sect. 2. We give our proposal for the semantics of PF and explore its consequences in Sect. 3. In our account, PF is a comparative subjective likelihood modal that imposes a restriction on the question under discussion. In Sect. 4, we discuss PF in interrogatives, and Sect. 5 concludes.⁵

2 Distinguishing PF from other modals

In this section we show that PF must be distinguished from epistemic necessity, epistemic possibility, likelihood predicates, weak necessity modals, and that it cannot be treated as a variable force epistemic modal either. We end the discussion by summarizing the empirical properties that an account of PF has to capture.

2.1 PF is different from epistemic necessity

Giannakidou and Mari (2017) argue that PF in Italian, as well as in Greek, Dutch and German, is synonymous with the epistemic necessity modal must. In this account, both PF and must are necessity modals that have an epistemic conversational background and a stereotypical ordering source. Furthermore, the epistemic modal base is subject to a condition called ‘subjective nonveridicality’: with respect to a proposition p and an anchor i, i’s epistemic state must entail neither p nor $\lnot p$.

In this subsection we challenge the general claim that PF is synonymous with a universal epistemic modal. Any such account predicts that PF and the universal epistemic modal dovere in Italian have the same distribution and interpretation. We show below that this prediction is not verified, and therefore that PF and the universal necessity modal dovere must be distinguished analytically.

In what follows, we assume that PF sentences involve an operator, PF, that takes a proposition p in its scope. We call p the ‘prejacent’ of PF, and represent the semantic structure of such sentences as PF(p).

Note first that in (6), while a doctor could felicitously report her inference about the patient’s health on the basis of the available evidence by using the epistemic necessity modal must/dovere, an utterance with PF would sound odd in this context.

Intuitively, PF is inappropriate in this scenario because it suggests that the doctor is guessing rather than drawing an inference based on the (medical) evidence and information available.

Consider next the case in (7), modified from Mandelkern (2018). This example shows that when the speaker knows that the inference being drawn is true, must/dovere is acceptable, but PF is not.

Next, note that there are also situations in which PF is acceptable while must/dovere is not. This is the case in the scenario exemplified in (8), in which the speaker has asserted lack of knowledge concerning p:

Note that the Italian expression non avere la più pallida idea, ‘not have the faintest idea’, in (8) is idiomatic: by using it, the speaker in (8) emphatically asserts that she does not know the answer to the question she addresses. This is, of course, compatible with her having some ideas about Maria’s whereabouts. This is precisely what is conveyed by : the speaker does not know where Maria is, but she has some ideas (having to do with Maria’s daily routines, for instance) that support her guess that Maria is at home.⁶

Note next that PF and must/dovere behave differently in interrogative sentences too. We have seen above, in (4) and (5), that PF is appropriate in polar and constituent interrogatives. As illustrated in (9) and (10), must/dovere sounds odd in such sentences.

We have illustrated above both cases where epistemic necessity modals are acceptable but PF is not, and the reverse. On the basis of these data we conclude that PF should not be analyzed as an epistemic necessity operator.⁷

We turn below to a comparison of PF with possibility modals.

2.2 PF differs from epistemic possibility

We distinguish now PF from the epistemic possibility modal might/potere. Example (11) shows that a speaker can use an epistemic possibility modal but not PF to report a state in which she entertains multiple epistemic possibilities.

Conversely, example (12) illustrates that when the speaker expresses high credence in the prejacent, PF is appropriate but the possibility modal is odd.

Next, we show that PF differs from a likelihood predicate as well. An analysis that treats PF as synonymous with predicates like likely/probable can capture the fact that its force is somewhat stronger than epistemic possibility (as in the best alternative and multiple possibilities scenarios), as well as the unacceptability of PF in the math case. Crucially, however, such an analysis cannot explain why PF is not appropriate in the doctor scenario, a context in which, as (13) shows, a likelihood modal is appropriate.

The data discussed so far lead us to conclude that PF should be differentiated from epistemic necessity and epistemic possibility modals, and that it cannot be treated as synonymous with likelihood predicates either.

2.3 PF is not a variable-force modal

Pursuing the comparison with epistemic modals, Mari (2012) shows that, as illustrated in (14), PF exhibits a mixed behavior: PF can co-occur with strong adverbs like certamente, ‘certainly’, and have an interpretation close to that of a necessity modal, but it can also co-occur with weak adverbs like forse, ‘maybe’, and have an interpretation close to that of a possibility modal. (See Bertinetto (1979) and Pietrandrea (2005) for discussion.)

To address this puzzle, and following Matthewson et al. (2008), Mari (2012) proposes that PF is a variable-force modal. According to this view, PF existentially quantifies over a non empty set of worlds that can either be equal to or smaller than the set of accessible worlds, as shown in (15) from Mari (2012).

In addition, in Mari’s proposal, PF comes with a requirement that the speaker has indirect evidence in support of p.

Focusing only on the issue of the force of the modal and returning to the pair in (14), the lexical entry in (15) ensures that PF will be appropriate with an existential reading in case X is smaller than the set of accessible worlds (f(w)), as in ; it will also be appropriate with a universal reading in case X equals f(w), as in .

It is not clear, however, how this account addresses some of the questions we raised in the preceding two subsections. Abstracting away from the indirect evidence condition, if the context is what determines the force of the modal claim, then we would expect PF to be appropriate in the multiple possibilities scenario, where it should receive an existential interpretation, as well as in the doctor-type scenario, where it should receive a universal interpretation. Taking into account the indirect evidence condition does not solve the problem: PF is still infelicitous in the versions of the multiple possibilities and doctor-type scenarios in (16) and (17) below, where the indirect evidence condition is met.⁸

In (16) PF should be felicitous under its existential interpretation, while in (17) it should be felicitous under its universal interpretation. The infelicity of PF in these examples shows that the flexibility allowed by a variable-force analysis is not enough to capture the distribution of PF in Italian, even when coupled with an indirect evidence requirement.

We conclude that PF differs from epistemic necessity modals, epistemic possibility modals, and likelihood predicates such as likely/probable, and furthermore that its interpretation cannot be captured by a variable force epistemic modal account either.

2.4 PF is not a weak necessity modal

In this subsection we show that the distribution of PF differs from that of weak necessity modals, namely should/ought to in English and dovrebbe in Italian, despite overlap in some cases.

Weak necessity in Italian is expressed by combining the strong necessity modal dovere, ‘has to/must’, with the morphology that occurs in the consequent of a counterfactual conditional (the conditional mood).⁹

Let’s begin with the similarities. Both PF and dovrebbe are infelicitous in the math and in the multiple possibilities scenarios, as shown in (18) and (19) respectively.

Yalcin (2016) treats English should and ought to as normality modals. According to his proposal, $\alpha$ should F means that it would be normal, all relevant things considered, for $\alpha$ to F. In this treatment, weak necessity modals are not epistemic. Following Veltman (1996), Yalcin takes the epistemic flavor of these modals to be the result of a default inference from what is normally the case to what is presumably the case.

There are, however, several contexts in which PF and weak necessity dovrebbe come apart. First, as shown in (20), PF differs from weak necessity in that only the latter is compatible with a counterfactual prejacent.

Next, (21) shows that dovrebbe, unlike PF, is appropriate in the doctor scenario.¹⁰

In (21) the use of dovrebbe suggests that the doctor is not sure about the diagnosis but has enough information to attempt one.

Note next that, as shown in (22), if the speaker has reliable evidence against the prejacent, a weak necessity modal is still appropriate, whereas PF is not. This is different from the non-counterfactuality requirement on PF since in the following example, the speaker is still open to the possibility that Maria is at home, as the expression riproviamo, ‘let’s try again’, indicates.

The best alternative scenario is another instance where PF and weak necessity modals differ: as shown in (23), PF is fine but dovrebbe is not.

Thus, PF but not dovrebbe, can be used to express the speaker’s ‘best guess’ concerning the answer to the question under discussion. These data suggest that despite similarities in distribution and interpretation, PF should be distinguished from weak necessity modals.

On the basis of these contrasts, and against our own earlier proposal in Ippolito and Farkas (2019), we argue that PF cannot be treated as a normality modal along the lines of the account proposed by Yalcin (2016) for English weak necessity modals. In our earlier proposal, PF statements involve an assessment of subjective likelihood based on what the evaluating agent considers to be normally the case. This account makes the wrong prediction relative to above. In that example, we see that in the presence of strong contextual evidence against the prejacent, PF is not felicitous. This runs against what a normality account predicts, since strong evidence against p is nevertheless compatible with p being normally the case.

The empirical contrasts between PF and its closest modal relatives are summarized in the chart below.¹¹

Our account of PF is developed in the next section.

3 Proposed analysis

In this section we first propose a semantics for PF in Sect. 3.1; in Sect. 3.2 we discuss its empirical coverage, and in Sect. 3.3 we investigate the status of the contribution of PF and discuss its interaction with negation before turning to PF in interrogatives in Sect. 4.

3.1 The semantics of PF

In the account we work out below, PF in Italian has the semantics of a special type of comparative subjective modal.

We propose that when a speaker uses a declarative sentence whose logical form is PF(S), where S denotes a proposition p, the speaker commits to p having the highest degree of subjective likelihood among a set of contextually salient alternatives C. More precisely, PF(S) presupposes that there is a highest ranked proposition in C and asserts that p is that proposition. PF(S) is true just in case, relative to her doxastic state, the speaker takes p to be subjectively more likely than any of the alternatives in C. We take the subjective likelihood of p for an agent i at a time t relative to a world of evaluation w and a doxastic base $D_{i}$ to be i’s degree of credence at t that w is an element of p, established on the basis of $D_{i}$, where $D_{i}$ is the set of worlds $w'$ such that for every propositions $p'$ such that i believes at t in w that $w\in p'$, $w'\in p'$.

Furthermore in our account, PF associates with focus in that the set of alternatives C is a subset of the focus semantic value of S. The constituent bearing a focused feature is marked by the diacritic F. Following Rooth (1992, 1996), we assume that this feature introduces the focus value of this constituent, i.e. the set of entities of the same type as the entity denoted by the phrase carrying the focus feature. The focus semantic value of the whole prejacent is the set of propositions obtained by substituting each of the elements in the focus value of the focused constituent for the semantic value of that constituent.

The truth conditions of PF(S), where S denotes p, are given in (25).

C is a salient, contextually restricted subset of , which includes the prejacent and at least some alternative in other than the prejacent. Following Roberts (1996), among others, we take prosodic focus (marked by F) to be connected to the question under discussion (qud), where the qud is conceived as a set of propositions: for an assertion $\alpha$ to be congruent to the qud, the set of its focal alternatives (or a contextually salient subset thereof) must be a subset of the qud. Therefore, asserting PF(S) is congruent to the qud iff C is a subset of the qud, i.e. only if $p\in$ qud.

The essential difference between the semantics of PF given here and the proposal in Ippolito and Farkas (2019) is that, in the latter, worlds in the doxastic modal base of the evaluating agent were ordered relative to that agent’s normality base while in the current account the relevant ordering is done based on the evaluative agent’s subjective credence.

The proposal given above has two consequences concerning the doxastic state of the evaluating agent i relative to the prejacent p. First, note that given this semantics, it follows that the doxastic state of the evaluating agent cannot entail $\lnot p$. An agent cannot have highest subjective credence in p while believing $\lnot p$.

Next, we argue that PF(S) gives rise to the mandatory implicature in (26):

The notion of p being positively settled relative to a set of worlds D is defined in :

The implicature in (26) arises because of the competition between asserting p and asserting PF(p) in a context in which p is one of the answers to the qud, as required by the semantics of the PF. The genesis of this implicature is as follows. First, we assume that a sentence S with semantic content p is simpler than PF(S) with semantic content PF(p). Next, note that when a speaker asserts p she publicly commits to p, which means that she presents herself as being in a doxastic state that entails p, i.e., $\forall w' \in \cap$D$_{i}$, $w' \in p$. Asserting a sentence whose semantic content is PF(p) merely commits the speaker to being in a doxastic state that leads her to place highest subjective credence in p. Thus, given the qud presupposed by PF(p), asserting p is always both relevant and more informative than asserting PF(p). It follows that if a speaker asserts PF(p) rather than p, she must have a reason for making a weaker and more complex statement, rather than a simpler and more informative one. This gives rise to the implicature that the speaker is not in a position to assert p, and therefore that $\exists w' \in \cap$D$_{i}$ such that $w' \in \lnot p$.

The npi implicature fits the profile of what Lauer (2014) calls mandatory implicatures, i.e., implicatures that are not cancellable. Lauer is primarily concerned with disjunction, but the reasoning behind the implicature is similar to what we have argued above: asserting A or B implicates, by mandatory implicature, that the speaker is not in a position to assert either A or B, each of which would be simpler and more informative.¹²

Evidence that npi is not cancellable is provided by :

The infelicity of is difficult to overcome. A parallel example involving disjunction is given in :

From what was said above, the evaluating agent’s doxastic state is not settled in favor of p (mandatory implicature), and it is not settled in favor of $\lnot p$ (entailment). Putting these pieces together, the evaluating agent’s doxastic state is not settled in favor of either p or $\lnot p$.¹³

To illustrate the proposal briefly, consider (2), repeated in (30):

In this case, the qud is the denotation of Where is Gianni now? The constituent carrying the focus feature is the locative prepositional phrase a casa, ‘at home’, and the alternatives to be considered are of the form $\ulcorner$Gianni will be at $x\urcorner$, where x is a member of . We have already argued that the set of contextually salient alternatives C is a subset of the qud. Let us now suppose that C = {{w: Gianni is at home}, {w: Gianni is at the office}, {w: Gianni is on the bus}}. The issue the speaker’s utterance is meant to address is which proposition in this set contains the evaluation world w.

According to our account, the PF sentence is defined just in case there is a unique alternative q in C such that q is subjectively more likely for the speaker than any other alternative in C relative to the speaker’s doxastic state. If defined, the sentence is true if and only if the subjectively most likely alternative for the speaker is that Gianni is at home. Furthermore, the sentence implicates that p is not settled in the speaker’s doxastic state.

Note now that npi targets only the prejacent, and that—since C is required to be a non-singleton set—for the PF statement not to be vacuously true, at least one alternative to the prejacent must be compatible with $D_i$. However, our semantics does not require all alternatives in C to be compatible with $D_{i}$. This is as it should be because it is perfectly appropriate to follow a PF(p) statement with denying one of the alternatives that is different from p.

Both Maria being at home and Maria being out shopping are assumed to be compatible with the speaker’s doxastic state, and this suffices to satisfy npi.

Before moving on to the empirical predictions of our proposal, we take a closer look at the role focus plays in our semantics. As pointed out above, following Rooth’s theory of focus, we take focus—marked by F—to allow the construction of a set of alternatives to the prejacent (the argument of PF). Moreover, following Roberts (1996), we take focus to mark a connection with the qud: for an assertion to be congruent to the qud, the set of its focal alternatives must be a subset of the qud. Therefore, focus on different constituents will generate different sets of alternatives and will be congruent with different quds. Consider (32).

The set of alternatives in (32) are all about Maria: for example, {Maria is sick, Maria is scared, Maria has not left her house in weeks}.

In (33), on the other hand, for Bea’s assertion to be a congruent answer to the question, the focus must be broad, i.e. at the sentential level.¹⁴

3.2 Empirical predictions

Given npi, the account predicts that PF will not be felicitous in case the speaker is assumed to be knowledgeable relative to the prejacent. This is confirmed by the infelicity of Bea’s response in (34).¹⁵

Under normal assumptions, people, including Bea, know what it is that bothers them. Bea’s answer in (34), however, claims that the subjectively most likely cause of her state is her being sick. Using the PF gives rise to the mandatory implicature that she is not in a position to simply assert the prejacent alone. In our account, Bea’s answer above is predicted to be odd because, even though she has first-person authority over the content of her psychological state, she presents herself as merely guessing about it. In (35), on the other hand, Aldo’s question targets the cause of Bea’s physical state.

Our account predicts the acceptability of Bea’s answer here because people are not assumed to always be knowledgeable about the causes of their own physical states, and, when they are not, guessing can be appropriate.¹⁶

The ban against a counterfactual prejacent, illustrated in (20) above, follows from our account because PF(p) entails that $D_{i}$ is compatible with p and therefore $\lnot p$ cannot be part of the common ground. The ban against a factual prejacent, exemplified in the math scenario, follows from npi.

We have seen in the discussion above that PF(p) statements come with the mandatory implicature that p is not settled in D$_{i}$. As we pointed out at the end of the last section, it follows from our account that some other alternative in C should be compatible with D$_{i}$. Consequently, a PF(p) statement is predicted to be fine in a context in which an alternative other than p is compatible with the speaker’s doxastic state. The example in (36) shows that this prediction is correct.

B’s statement is interpreted as claiming that Ezio is the most likely person to have invited Maria, while at the same time allowing for the possibility that in fact it was Carlo and not Ezio who did so.¹⁷

The account also predicts that a PF(p) statement should be felicitous in case the speaker asserts that all alternatives in C are compatible with her doxastic state. The felicity of PF in shows that this prediction is met.

The infelicity of PF in the doctor scenario is due, in our account, to the fact that PF(p) statements require credence to be assigned to p relative to a doxastic modal base, a base that is not reflexive. In the doctor scenario, however, the doctor is supposed to assess the alternatives in the qud based on what she knows, modelled as a reflexive modal base. Epistemic modals such as must or might are appropriate in this scenario precisely because these modals quantify over epistemic modal bases. To exemplify, in the analysis defended by von Fintel and Gillies (2011), epistemic modals such as must and might are universal/existential quantifiers over those worlds where all the propositions that constitute a kernel of information K are true. In (38) we give exemplify this proposal for must.

K has two properties. First, information in K is direct or privileged information. Because one cannot have direct evidence about a proposition q unless q is true, $\cap K$ is reflexive. Second, the prejacent, p, is presupposed not to be settled relative to K. Now, the doctor case illustrates a situation in which the speaker is expected to make an assessment based on knowledge. Note that a statement made relative to what the anchor knows (or a relevant subset thereof, as in von Fintel and Gillies’s analysis) entails the corresponding statement made relative to what the anchor believes (or a relevant subset thereof). Therefore, a speaker that bases her assessment of the truth of a proposition p on what she believes, implicates that she is not in a position to assess the truth of p based on what she knows and, in particular, based on a body of relevant and direct information K available to her in the context. This is so because K is either empty or irrelevant to the prejacent and the qud. In our account, the ‘guessing’ flavor of PF(p) statements is due to (i) the fact that the modal base is restricted to being doxastic, and (ii) the competition with a stronger knowledge-based statement. The PF statement is odd in a doctor–patient exchange because the doctor’s assessment is based merely on beliefs and implicates that the contextually relevant kernel of direct information available to the doctor cannot support the prejacent. In this situation and given the formal doctor–patient setting, one expects a doctor to refrain from making an assessment.¹⁸

The proposed account predicts that PF statements will be inappropriate in contexts in which the speaker is supposed to base her assessment of the truth of p on known facts rather than on her beliefs and assumptions. We therefore expect PF statements not to be appropriate as answers to questions asked in situations where the addressee is supposed to base her response on what she knows. This is the case not only in the doctor scenario but also in the teacher-student scenario in (39):

Finally, the contrast between knowledge and beliefs is also responsible for the infelicity of PF in the math scenario, since mathematical conclusions are based on knowledge and not beliefs.¹⁹

On the flip side, our account predicts PF statements to be appropriate in situations in which the speaker is not required to base her assessments of possibilities purely on known facts. Such a scenario is illustrated in (40), a variant of the doctor example above.

Addressing not a patient but a peer (and, therefore, without holding a position of authority over the addressee), the doctor is free to reveal that her knowledge is insufficient to draw any conclusion and to express her subjective opinion, which is not based on her medical knowledge.

Note that the implicature that the speaker is not in a position to make a knowledge-based inference can be explicitly suspended as in the following scenario.

In (41), a PF statement is followed by a necessity epistemic statement suspending the implicature that the speaker is not in a position to support the prejacent with what she knows.²⁰

The contrast between PF and epistemic modals exemplified in the ignorance scenario above is also accounted for. Epistemic modal claims must be based on knowledge. Therefore, in cases where the speaker has emphatically asserted her ignorance about the topic, as in our ignorance scenario, an epistemic must-claim will be infelicitous, while a PF statement is expected to be appropriate since such a statement is not based on what the speaker knows, and does not require any particular kind of evidence. The relevant example is repeated below.

Here the speaker is interpreted as guessing her answer to the question since her assessment is based on beliefs, rather than knowledge (which she emphatically claims not to have).

We now turn to justifying the claim that the subjective likelihood involved in the interpretation of PF is comparative rather than absolute.²¹ Under the hypothesis that the relevant notion is absolute likelihood, PF(p) statements would require the prejacent to receive a credence level that surpasses a particular threshold, while under the comparative hypothesis what is required is for the prejacent to be the most likely alternative, a situation which is compatible with scenarios in which the credence level the speaker assigns to p is nevertheless rather low. The example in (42) supports the comparative likelihood hypothesis: PF is acceptable here even though the speaker claims to be ignorant about Maria’s whereabouts. The sentence is interpreted as asserting that the prejacent is more likely than any other alternative in the qud. By asserting (42), the speaker is not assumed to believe that the prejacent is subjectively more likely than some threshold of likelihood. What is required is that there be no alternative the speaker gives higher credence to.

To further illustrate this point consider the scenario in , where multiple alternatives are in play, and where one alternative is ranked higher than the others but still below a likelihood threshold of 0.5. As predicted by our account, PF is felicitous in such a situation.

However, the assertion in does not commit A to the claim that it is likely that Bayern won, as shown by the following continuation.

The account also predicts that PF will be appropriate in the best alternative scenario in (45): the speaker selects Gianni’s being at home as the subjectively most likely answer given her doxastic state.

The fact that (45) is still interpreted as a guess is the result of an implicature arising from a competition with an epistemic modal sentence: the speaker chose not to evaluate the prejacent relative to a relevant kernel of direct (therefore, true) information, even though it would have been more informative than assessing the prejacent merely on the basis of what the speaker believes. As mentioned above, the implicature will be that the speaker was not in a position to do that because there is no relevant body of direct information that would support the prejacent. Note that the kernel of direct information that is part of the semantics of an epistemic modal is presupposed not to settle the prejacent, and thus it only allows the speaker to indirectly infer the prejacent. As result, the implicature will be that the kernel (that is, the body of direct evidence available to the speaker) does not even indirectly support the prejacent, i.e. it is either empty or it contains information irrelevant to the prejacent.

The contrast between likely/probabile and PF discussed in Sect. 2.2 can be explained along the same lines: unlike PF statements, we take likelihood predicates to evaluate the prejacent relative to a body of true propositions and, therefore, they are appropriate in contexts where knowledge-based assessments are requested, as in (46).

The fact that PF contrasts with possibility epistemic modals in not allowing multiple possibilities, illustrated in the multiple possibilities scenario above, follows from the uniqueness presupposition in its semantics.

The examples we discussed till now involved mutually exclusive possibilities in C, and we only considered simple PF statements, disregarding the possibility of conjunctions and disjunctions. We turn now to these two cases, beginning with disjunction.²²

First, staying with mutually exclusive possibilities, and taking an example inspired by a reviewer, let us assume that the qud is Where is Leo, and that the context has narrowed down the possibilities to the four given in (47):

Furthermore, assume that, given her subjective credence, the speaker ranks these possibilities as in (48):

Under these circumstances, the speaker can truthfully assert any of the sentences in (49):

Substituting other city names for the ones used in (49) would result in false statements. We give a sample in (50):

We show now that these facts follow from our account. In the case of , the prejacent proposition is m and the propositions in C are the focus semantic value of the prejacent sentence, namely {m, t, g, b}. Among these propositions, the speaker assigns highest credence to m, and therefore is true and felicitous, while substituting any other city name from our set to Milano results in a false sentence.

As for , first note that while this sentence is intuitively true in the given context, it carries the implicature that the speaker prefers not to distinguish between the credence level she assigns to m and the credence level she assigns to t. Thus, is interpreted as communicating that the speaker assigns higher credence to m and to t than to either g or b, and that she wishes to avoid ranking m and t relative to one another.

Turning now to our account, we assume an LF in which PF takes scope over the disjunction, as in (51).²³

The prejacent proposition is the ordinary semantic value of the prejacent sentence, namely $m \vee t$. The propositions in C are the set of propositions in the focus semantic value of the prejacent sentence, given in :

Among these, the speaker assigns the highest credence to $m \vee t$, and therefore is true in our original context. If the qud is assumed to contain not only the most informative propositions, namely m, t, g, and b, but also the less informative ones obtained by the binary disjunctions over this set, no problem arises. The implicature that the speaker is not willing to commit to ordering m relative to t arises given that, by uttering a disjunction, she chooses to order only the less informative disjunctive propositions in a situation in which ordering the more informative propositions is relevant. Thus, the speaker implicates that her credence in m is not significantly different from her credence in t and therefore she avoids committing to either m or t.

Under the assumption that the qud contains only m, t, g, b the condition that C be a subset of qud is violated in . Under our account, the speaker is not addressing the current qud but rather, an accommodated, less informative qud, one that includes the binary disjunctive propositions as well.

Under both assumptions, we capture the fact that is true in our original context, as well as the implicature that the speaker is not comfortable distinguishing between the credence level she assigns to m and the one she assigns to t. Were she willing to distinguish between the credence level she assigns to these two possibilities, asserting the one she has highest credence level in would be both simpler and more informative. Note that in the given context, replacing with any other binary disjunction results in a false sentence.

The account of is analogous. In this case, the LF of the sentence is as in .

The propositions in C are those in :

Of these, $m \vee t \vee g$ is highest ranked relative to the speaker’s credence level, which explains why is true in our original context. And just as in the case of , the speaker implicates that she is not in a position to distinguish between the credence levels she assigns to each disjunct. Her statement entails, however, that all three are ranked above b. By uttering a three-way disjunction the speaker implicates that she is not ready to commit to assigning highest credence level to m, t, or g or any binary disjunction over these propositions since if she could do so she could assert a sentence that would be both simpler and more informative. Our account also explains why, unlike the three sentences in (49), the following PF-disjunction is odd in the scenario in which the alternatives being considered are m, t, g, and b.

Since the prejacent is $[m \vee t \vee g \vee b]$, and since this disjunction exhausts all possibilities, there is no other alternative in C and therefore the focus requirement that C be a non-singleton set is not met.

Let us turn now to cases where the alternatives in C are not mutually exclusive by considering a scenario in which the qud is What did Leo eat?, and the possibilities in the context are that he ate at least the apple, at least the pear or both. The possible PF statements in this case are given in (56).

Let us denote by a, p and $a+p$ the propositions denoted by Leo ate the apple, Leo ate the pear and Leo ate the apple and the pear respectively, where the apple and the pear is interpreted as denoting the sum of the apple and the pear. These then are the propositions in C for all three examples in (56). By uttering the speaker claims that she places highest credence in a, by uttering she claims that she places highest credence in p, and by uttering she claims that she places highest credence in $a+p$. Uttering either or is compatible with the speaker being committed to the possibility $a+p$ being the case, though her credence in this possibility is lower. This prediction is confirmed by the felicity of (57).

Our account also predicts, correctly, that the conjunction in (58) is infelicitous:

This sentence is infelicitous because it violates the uniqueness presupposition of PF: the claim it makes is that a has the highest credence level, and that p has the highest credence level.

Before concluding this section, we briefly summarize the main features of PF in our account.

• 1. PF(p) involves an assessment rooted in a doxastic base, rather than in an epistemic or normality base.

• 2. The prejacent p has to be a member of the qud in the context of assessment.

• 3. The prejacent p can neither be factual nor counterfactual relative to the context, i.e., the context cannot have established either that p is true or that it isn’t.

• 4. PF is felicitous in cases in which the evaluating agent has low credence in the prejacent as well as in cases in which she has high credence in it, as long as the evaluating agent assigns to it a higher subjective credence than to all other alternatives in the C, a contextually salient subset of the qud.

In the next subsection we defend treating the contribution of PF as part of the compositional semantics of the sentence in which it occurs, and briefly discuss connections with evidentiality.

3.3 The status of the contribution of PF

The main issue we address in this subsection is whether the contribution of PF is part of the semantic content of the sentence in which it occurs, as in the account developed above, or not. We take non-semantic content to include expressive content as well as content that signals contribution to the conventional discourse effect of a sentence. Evidence in favor of treating the contribution of PF as part of the semantic content comes from three sources, polarity particle interpretation, embedding, and comparatives.

With respect to polarity particles, we assume, following Roelofsen and Farkas (2015) among others, that they are anaphoric to a discoursally prominent proposition provided by the previous utterance. In the case of a polar interrogative, this proposition is contributed by the semantic content of the sentence radical. If PF contributes to the semantic content of the proposition in which it occurs, we expect the whole proposition, PF(p), to be able to serve as an antecedent to a polarity particle. We see in (59) that this expectation is fulfilled: sì, ‘yes’, in B’s response takes PF(p) as its antecedent.

The answer in (59) makes clear that B does not know where Gianni is and cannot rule out the possibility that Gianni is somewhere other than home. Therefore, in answering ‘yes’ to A’s question, B is not assenting to the prejacent (that Gianni is at home) but to the weaker, modalized, proposition (that the subjectively most likely alternative is that Gianni is at home). The possibility of PF(p) to serve as antecedent to a polarity particle response, as in (59), supports an account that treats PF as contributing to the semantic content of the sentence in which it occurs.

There are also cases in which the antecedent of a polarity particle is a subpart of the semantic content of the utterance the particle reacts to, as exemplified in :

Since the embedded clause in the sentence uttered by Carol is directly relevant to the qud (i.e. where is Bob?), it can serve as antecedent to the polarity particle in Anna’s response. This situation can arise with PF as well, as illustrated in (61).

The qud at the beginning of the discourse in (61) is ‘where is Gianni?’ Let us assume that the propositions in this qud are {Gianni is at home, Gianni is at the office, Gianni is on the bus}. B’s answer is a PF statement which is pertinent to the unmodalized qud but does not directly answer it. In C’s reply, the antecedent of the negative polarity particle ‘no’ is the prejacent in B’s utterance. C’s statement reveals that he is in a different doxastic state than B and, because of this, he is in a position to rule out one of the alternatives in the qud. C could have also disagreed with B’s whole modalized proposition, as shown in (62), in which case C would also be indirectly addressing the qud. In (62), by uttering the negative polar particle no, C does not have to reject the possibility that Gianni is at home but merely rejects B’s assertion that the most likely alternative is that Gianni is at home.

We take the possibility of reacting to both the prejacent (as in (61)) and the whole PF claim (as in (59) and (62)) to show that, in a context where the qud is about the prejacent (whether p), asserting PF(p) indicates that the speaker’s knowledge does not support an answer to the qud.

Note that the same facts are true of epistemic modals as well: the polarity particle can either be anaphoric to the prejacent only (as shown in (63)) or to the whole modalized proposition (as shown in (64)).

Turning to the embedding potential of PF sentences, the picture that emerges is less clear mostly because constraints on embedding possibilities have a variety of sources, and therefore one cannot use embedding as an absolute test for diagnosing semantic content. Restrictions on the embedding potential of epistemic necessity and possibility modals have been noted by Hacquard and Wellwood (2012), Anand and Hacquard (2013), Ippolito (2017), among others. At least some of the proposals that have been advanced to explain such embedding restrictions have maintained the assumption that epistemic modals contribute to the truth-conditions of the sentences in which they occur.

First, as exemplified in (65), PF sentences can occur as complements of propositional attitude predicates, a fact that is at least consistent with the assumption that PF contributes to the semantic content of the sentence in which it occurs.²⁴

As discussed above, PF presupposes that the prejacent is a member of C, which in turn is a subset of the qud. Therefore, even when PF(p) is embedded, p must be directly relevant to the current qud. This aspect of the semantics of PF sentences can explain why embedded PF clauses are most natural under doxastic or epistemic predicates, preferably in the first person and in the present tense. However, as long as the prejacent is relevant to the resolution of the qud and the matrix predicate is of the right type, embedded PF clauses are fine with third person subjects as well, as in the previous example or in (66).²⁵

We want to stress, however, that embedding of PF sentence is not unconstrained. We already noticed that embedding under some attitude verbs (e.g. credere, ‘to believe’) affected acceptability. The example in (67) shows that a PF sentence also resists being embedded in the antecedent of a conditional.

We are not going to explore these constraints further in this paper since they would take us too far afield and would be beyond the scope of this paper, but see Mihoc et al. (2019) for relevant discussion.²⁶ The data we collected show that there is considerable variation across speakers with respect to which embeddings are acceptable and which are not. Most of our consultants found at least some kind of embedding or anchoring to an individual other than the speaker possible, but preferences varied across speakers.

One last piece of evidence in favor of a semantic account comes from the possibility of having PF in comparative sentences. The example in (68) illustrates this.

This example is interpreted as claiming that the likelihood that Gianni is in his office is higher than the likelihood that he is at home. Thus, the comparative clause più che a casa, ‘more than at home’, appears to realize explicitly the comparison term for the likelihood relation that was implicit in the semantics we gave in Sect. 3. Because ‘being in the office’ and ‘being at home’ are not gradable predicates in the relevant sense, we can rule out the possibility that the comparative is part of the prejacent since that would give rise to the proposition that Gianni is in his office to a degree that is greater than the degree to which he is at home.²⁷ Our proposal can account for (68) straightforwardly: the two alternatives that are being ranked are the proposition that Gianni is in the office, which we denote by o and the proposition that Gianni is at home, which we denote by h. C in this case is {o, h}. When uttering (68), the speaker asserts that the subjectively most likely alternative is o, not h. Formally, the truth conditions of (68) are as in (69).

Though a detailed analysis of these comparative structures is beyond the scope of this paper, we take the possibility of this construction to be further evidence that the comparative relation expressed by PF is part of the semantic content of a PF sentence, and that the second argument of this relation can be expressed overtly in the sentence.

So far in this subsection we have addressed the question of whether the contribution of PF is part of the semantic content of the sentence or not. In the literature, this question is often linked to the issue of whether PF is a marker of indirect evidentiality. It is to this question that we now turn. In the account proposed by Frana and Menéndez-Benito (2019), PF in Italian signals that the prejacent is a conjecture based at most on indirect evidence. In support of this proposal, they argue that PF sentences share the following three characteristic properties with evidentials: (i) the evidence supporting the prejacent is at most indirect, and therefore PF sentences are compatible with complete lack of evidence; (ii) PF is always anchored to the speaker in declaratives; (iii) the evidential component is not at-issue and therefore it is unchallengeable.²⁸ In the remaining part of this section, we offer some arguments that challenge the conclusion that PF must be treated as an evidential marker.

Frana and Menéndez-Benito discuss the example in (70) in support of the first property:

The claim is that in (70), PF is appropriate even though Carmela has no evidence whatsoever with respect to Elena’s husband’s situation because what PF requires is at most indirect evidence, which is consistent with having no evidence at all. We agree that the function of the PF statement in this example is to reassure and comfort Elena, and we also agree with the observation that the speaker in (70) does not need to have any factual evidence about the soldier’s situation. These properties of (70), however, are accounted for under the present analysis as well. Carmela asserts that the prejacent is the subjectively most likely alternative but the PF statement is silent as to what kind of evidence (if any) the speaker might have. However, as discussed above, the statement will implicate that her assessment is not supported by what she knows. Normally, since she has offered her PF statement in response to the (implicit) question about the soldier’s situation, she will be understood as having a doxastic state that supports, however tenuously and vaguely, her credence in the prejacent.

Our account departs from Frana and Menéndez-Benito’s in that it predicts that PF(p) is not acceptable in a situation that enforces the speaker’s ignorance relative to p and where the issue of subjective likelihood does not arise. To show that this prediction is correct, consider the coin-tossing example in (71).

In this scenario, since Bill knows that the machine is random and that this is the machine’s first coin-tossing, the presupposition that there is a subjectively most likely alternative based on what he believes is not satisfied. This explains the oddity of (71). Frana and Menéndez-Benito (2019) predict PF to be fine in (71) because in their proposal the speaker does not need to have any evidence in support of the prejacent.

The second property Frana and Menénedez-Benito ascribe to PF sentences, namely that they are necessarily anchored to the speaker in declaratives, does not always hold. We already saw in example (66) above that the anchor of the embedded PF can be someone other than the speaker, i.e. the police, while in (65) it is the people who know Carlo, though there appears to be variability across speakers’ judgements in cases in which the anchor is not the speaker.

In support of the claim that PF sentences must be anchored to the speaker, Frana and Menéndez-Benito give the example in (72) (modeled after an example in Kratzer (2009)):

As pointed out by Ippolito and Farkas (2019), the problem with the PF sentence in (72) is due to the fact that the anchor in (72) is non intentional, and therefore not able to assign subjective likelihood to a proposition. This explains why PF is acceptable in (66) but not in (72). We therefore conclude that PF is not obligatorily speaker-oriented.

Turning to non-challengeability, note first that, as argued by Korotkova (2019), non-challengeability is not unique to evidentials, but rather, it characterizes expressions that involve a certain type of subjective content. In cases like (73), (74) and (75), the speaker is the ultimate authority with respect to her inner states and therefore her reports on those cannot normally be legitimately challenged:²⁹

The oddity here is due to the fact that B is in the best position to decide what B considers subjectively most likely, and therefore challenging her claim is infelicitous.

Recall that in our account, the semantics of PF involves a subjective component: the anchor assesses p as being subjectively the most likely alternative based on her doxastic base. This subjectivity, we claim, is the reason why some denials of PF sentences are odd. Thus, the reaction in (76) is odd if interpreted as challenging the PF(p) statement without shifting the anchor, and thus asserting that it is not true that B assigns the prejacent the highest likelihood according to his doxastic state.³⁰

Note, however, that not all denials of PF claims are infelicitous: in particular, denying a PF claim by saying ti sbagli, ‘you are mistaken’, is acceptable, as shown in (77).

In our account, cases like (77) are instances of what has come to be known as faultless disagreement.³¹

PF sentences express an attitude towards the prejacent that other people may or may not share. This means that an interlocutor may agree with a PF statement if she too considers the prejacent to be the most likely alternative given what she believes, or may disagree with such a statement if she does not share this assessment. (For a discussion of the difference between assessing a claim as true or false, and agreeing with a claim, see Roberts (2017).) In (77), C does not share B’s state of mind relative to Gianni’s whereabouts, and expresses her own assessment. In B’s utterance, B is the anchor of the response, which means that she reports her own subjective likelihood assessment, while in C’s utterance, the anchor of the response is C. A case of agreement is exemplified in (78):

In this case too, the anchor of the agreeing response is the person who utters it, in this case B, rather than her interlocutor. We take the facts in (77) and (78) to show that a PF claim can be challenged as long as this is done in ways that allow the anchor to shift from one speaker to the other.

In sum, the addressee can challenge the speaker’s PF claim based on two reasons: (i) because she knows more about the qud, in which case she will deny the prejacent as in (61); or (ii) because she does not know more about the qud but her subjective assessment is different, in which case she will deny the whole PF statement as in (77). In the latter case, the anchor shifts from the original speaker to the challenger.

3.4 Interaction with negation

It is a robust generalization that PF cannot take scope below negation.

The lack of scope flexibility with respect to negation can shed doubt on the conclusion of the preceding discussion, i.e. that PF is part of the asserted content of a PF sentence. In this subsection, however, we argue that the apparent impossibility of negation to take scope over PF is the result of the existence presupposition triggered by PF.

According to our account, PF presupposes that there exists a unique alternative that is the subjectively most likely alternative in C. Because of this presupposition, $\lnot PF(p)$ is equivalent to $PF(\lnot p)$. To see why, consider first the case in which the set of alternatives is $\{p, \lnot p\}$ and assume one asserts that it is not the case that p is the subjectively most likely alternative. The presupposition that there is a subjectively most likely alternative entails that this most likely alternative is $\lnot p$. For cases where C contains more alternatives, the same result is reached, whether or not these alternatives are mutually exclusive. For example, suppose that the issue is Gianni’s whereabouts, and suppose that there are three possible mutually exclusive alternatives {Gianni is at home, Gianni is at the office, Gianni is at the gym} (or, in short, {home, office, gym}).

In (80), the speaker asserts that it is not the case that office is the most likely alternative. Given the presupposition contributed by PF (that there is a most likely alternative in C), it follows from $\lnot PF(\textit{office})$ that one of the other alternatives is the most likely one. Since the alternatives are mutually exclusive, it follows that the most likely alternative is a non-office alternative, i.e. that it is subjectively most likely that Gianni is not at the office. Given our semantics, $\lnot PF($office) and $PF(\lnot$office) are equivalent, and this equivalence is responsible for the illusion of fixed scope relative to negation.

This analysis works for non-mutually exclusive alternatives too. Suppose that C is {Leo ate the apple, Leo ate the pear, Leo ate the apple and the pear} (or, in short, {apple, pear, apple+pear}). The sentence in (81) has the logical form $\lnot$PF(apple) and it expresses the proposition that it is not the case that apple is the most subjectively likely alternative for the speaker.

This means that the most subjectively likely proposition is either pear or apple+pear. However, assuming that for any propositions p and q with subjective probabilities $P_S(p)$ and $P_S(q)$, $P_S(p\wedge q)\le P_S(p)$, it follows that apple+pear cannot have a higher credence than apple. Therefore, given the existence presupposition triggered by PF, the proposition that it is not the case that it is subjectively most likely that Leo ate the apple is strengthened to the proposition that it is subjectively most likely that Leo didn’t eat the apple.

This concludes our discussion of PF in declaratives. We turn to interrogatives in the next section.

4 PF in interrogatives

We assume a Hamblin-style semantics for interrogatives according to which the denotation of an interrogative is the set of propositions that constitute its least informative complete possible answers. We exemplify schematically with the denotation of a polar interrogative PF sentence in (82).

Each proposition in the denotation of the interrogative has the PF semantics given in (25). Due to the subjective nature of PF, interrogatives are subject to what Murray (2010) calls ‘interrogative flip’, which requires the contextually bound value for the anchor i in questions to be the addressee rather than the speaker. The positive answer to the interrogative commits its speaker to the proposition PF(p), whereas the negative answer commits her to $\lnot PF(p)$. This is illustrated in (83).

Note that the negative answer in (83) is understood obligatorily with narrow scope negation for the reasons explained in Sect. 3.4.

The account predicts that in a PF interrogative, the prejacent is required to be unsettled in $D_i$, where i is the addressee. In other words, the addressee is assumed to be agnostic about the answer to the qud. To see that this prediction is correct, compare the lack of competence case in (84) with the quiz case in (85).³²

PF is acceptable in (84) because the addressee is not assumed to know whether the mummy is a man or a woman. In (85), on the other hand, PF is inappropriate because in quiz contexts the addressee cannot be assumed not to know the answer to the question. Note, however, that in the more complex situation described in (86), where the contestant has already proven his ignorance about the question, the host’s PF interrogative becomes acceptable.

By remaining silent when asked the question for the first time, B reveals that he is ignorant about the answer to the question. A’s subsequent PF question is felicitous because it is inviting B to search for the answer that is subjectively most likely given what B believes.³³

Finally, (87) shows that in contexts in which the addressee is presupposed to know whether the prejacent is true, PF is infelicitous:

The question in (87) becomes felicitous in situations where the addressee is absent (as in a written communication). In this case, the question can be interpreted as self-addressed, and PF is felicitous.

Before concluding this section, we note that questions asked using a PF interrogative have been labeled conjectural questions, i.e., questions that signal that the speaker does not assume that the addressee will settle the issue raised by the question (Frana and Menéndez-Benito 2019; Eckardt and Beltrama 2019; Littell et al. 2009). In our account, the conjectural nature of PF questions follows from their semantics. Each alternative in the denotation of a PF interrogative is a proposition in C embedded under the scope of PF (i.e. {PF(p), PF(q), ...}). These are the responses such a question projects, in the sense of Farkas and Bruce (2010). Because of the interrogative flip, the anchor of the propositions in the denotation of the interrogative is the addressee. As a consequence, the questioner assumes that at least some alternative other than the prejacent is compatible with the addressee’s doxastic state, and therefore she assumes that the addressee is not in a position to settle the qud. The addressee is projected to respond by asserting which alternative she finds subjectively most likely. Note also that under default assumptions about asking a question, the participant who utters a PF interrogative is not assumed to know the answer to the qud either, and therefore PF interrogatives are predicted to be felicitous in situations in which neither the speaker nor the addressee know the answer to the qud. As a consequence, we expect PF interrogatives to be infelicitous when used as rhetorical questions whose answers are supposed to be obvious in the context. As shown in (88), this expectation is met.³⁴

The same is true for rhetorical questions with a positive answer, as illustrated in (89) where the answer is ‘the speaker’.

Under the assumption that it is common knowledge that the speaker provided the money for the party, (89) is a rhetorical question whose point is to stress the obviousness of the answer. As expected, PF is not felicitous in this case.

We conclude that the semantics of PF developed in Section 3 extends to interrogatives without further modification.

5 Conclusion and open issues

To summarize, we have argued that PF in Italian has the semantics of a subjective comparative modal: a sentence with the logical form PF(S) asserts that an evaluating agent finds p, the proposition expressed by S, subjectively more likely than any other alternative to p in a set of propositions C. C is a subset of the focus semantic value of the prejacent sentence. We have also argued that the choice of PF over an epistemic modal triggers the implicature that the speaker is not in a position to appeal to what she knows in order to support her credence in the prejacent. More specifically, PF implicates that there is no set of privileged, direct information relevant to the qud and available to the speaker in the context that would support the prejacent. Hence, the observation that PF is used to express a guess concerning an answer to the qud. The proposed account covers PF in both declarative and interrogative sentences, and predicts the possibility of its occurrence in embedded contexts, though the constraints on embedded PF have not been explored here. According to this account, PF is used to express a particular type of weakened commitment to the prejacent in the case of declaratives, and to elicit such a commitment in the case of interrogatives. The semantics of PF proposed above results in a special type of commitment nuance, one that involves an agent choosing among alternatives based on her beliefs and her credence.

There are two further non-temporal uses of the future in Italian that have been discussed in the literature (cf. Squartini 2012). The first is its use in echo sentences rejecting a previous statement, as shown in (90).

The second, more widely attested in other languages, is the future used in concessive clauses, exemplified in (91).³⁵

We leave to the future a discussion of whether and how these occurrences are connected to the ones we have discussed here.

Going beyond Italian, there are two interrelated questions that arise at this point. The first concerns the issue of why it is that the future tense has this type of non-temporal interpretation not only in Italian but in other languages as well. The second concerns identifying the cross-linguistic similarities and differences found in languages that have PF-like uses of the morphological future tense.

Answering the first question requires comparing various existing theories of the temporal future (cf. Copley 2009; Kaufmann 2005; Condoravdi 2003; Cariani and Santorio 2018, among others), a task that is beyond the scope of this paper. For the time being, we can only offer some initial remarks. Note first that the connection between temporal and modal-like uses of the morphological future tense cannot be so tight as to directly derive one from the other, given that not all languages exhibit such a connection. Note also that an account that derives one use of the future from the other would rule out the situation found in Romanian, where there are two morphologically distinct future forms that are synonymous in their temporal use but contrast in that only one of them is also compatible with a PF-like interpretation. (For discussion, see Irimia (2010) and Fălăuş (2014).)

The connection between PF, modal interpretations and the future is nevertheless real. Exactly how to capture it is, however, far from clear. First, note that the connection between future tense morphology and modality has been argued convincingly in the literature (Enç 1996; Abusch 1997; Copley 2009; among many others). Furthermore, such morphology is often historically derived from deontic or desiderative expressions, as noted in Bybee and Pagliuca (1987), for instance. Bybee and Pagliuca suggest that non-temporal interpretations of the future tense are often connected to the modal origin of future morphology. They note that PF-like interpretations of future morphology in Romance languages, Dutch, and in the case of the Korean be-future, arise when temporal future morphology originates in expressions denoting deontic modality. If this suggestion is correct, the semantic connection we should explain is the three-way connection between deontic modality, PF, and the future. For now, we can only say that using the future tense to express a particular nuance of weakened epistemic commitment has its roots in the robust connection between future tense morphology and modality.

Turning now to the second question, note that an explanation of the various ways in which future tense morphology can be used to express PF-like nuanced commitment would have to be based on solid semantic micro-variation work that does not exist yet. We hope that the detailed empirical discussion of Italian in Sect. 2 will prove useful in establishing the relevant cross-linguistic generalizations. For now, we can only offer some suggestions concerning cross-linguistic expectations that the account presented here leads us to. We expect that PF-like interpretations of future morphology will be compatible with weakened epistemic commitment to the prejacent. If this commitment is required to be based on the agent’s beliefs rather than known facts, we also predict the presence of the obligatory implicature concerning the prejacent being unsettled as far as the evaluating agent is concerned.

Variation may arise concerning the nature, and therefore the possible strength of the agent’s commitment. In French, for instance, future morphology has a PF-like interpretation similar to PF in Italian, as shown in (92), from Mari (2018).

Italian and French contrast, however, in that in French, but not in Italian, the non-temporal use of the future is compatible with the evaluating agent having decisive evidence in favor of the prejacent. This is shown in (93), where the French example is also taken from Mari (2018) (bold-face is ours).

A similar point can be made about the relation between PF in Italian and the non-temporal uses of will in English. Italian PF and will overlap in some cases, as shown in (94) from Winans (2016).

However, Winans argues that non-temporal will cannot be used when inferring a cause from an effect, while this constraint is not valid in Italian. For example, will is disallowed in (95) where—according to Winans—the speaker utters the future sentence to express her inference about the cause of the sculpture’s melting, while the Italian equivalent is fine.

Conversely, in some of the contexts discussed above the non-temporal use of will is fine in English but PF in Italian is not.

The last issue we mention and leave open is the question of how PF compares with modal particles such as wohl in German or darou in Japanese, which have similar though not identical effects. (See Eckardt (2020) and references therein for wohl, and Uegaki and Roelofsen (2018) and references therein for darou.) Common to all these forms is that they are used as means of nuancing commitment to the prejacent. To illustrate the kind of questions that one should explore further consider (96), illustrating a use of wohl where the truth of its prejacent is supported by factual evidence (Göbel 2017).

Here A concludes, based on reliable evidence, that her previous geographical beliefs were erroneous. Were A to speak Italian, acknowledging her mistake by using the PF statement in would not be appropriate:

Our account predicts the infelicity of in this context: since A’s evidence is reliable, it settles the issue whether p, and therefore such a scenario is incompatible with PF; after looking at the map, A’s epistemic state will entail that Athens is in Greece, and that her previous beliefs were wrong, and therefore npi will no longer be met. This and other differences between PF and evidential particles should be explored further.

Finally, we note that PF is related to but different from the ‘future verification’ use of the future. In the latter use, the future time reference is not that of the eventuality described in the prejacent but rather, the time of its verification (cf. Mari 2012). Here, we are merely going to point out that Italian PF is not a future of verification: if it were, PF would not be felicitous in cases where future verification of the prejacent is not possible. However, as shown in (98), PF is fine even when future verification is impossible: Mary can felicitously use PF knowing full well that there will be no time when she (or anybody else) will be able to verify whether her childhood pictures where in a box in the attic.

Accounting for the non-temporal use of the future in a particular language, as we have done above, provides, we hope, the basis for future explorations of the semantics of such non-temporal interpretations of the future across languages, as well as for understanding the similarities and differences between PF and the use of modal particles in languages like German and Japanese or evidential systems elsewhere.