Attitudes in discourse : Italian polar questions and the particle mica *

This paper explores ways in which discourse participants convey an attitude about another discourse participant’s conversational move. We examine the semantics/pragmatics of Italian positive and negative polar questions (building on the literature on biased questions) and propose the first fully compositional analysis of the Italian particle ‘mica’, appearing in negative polar questions and negative assertions. The core is that ‘mica’ is member of a family of presuppositional, epistemic ‘common ground management’ operators, leading to a new account of epistemic inferences in biased polar questions that relies on the presuppositional nature of these operators. We argue that ‘mica’ is a high-left-periphery particle that indicates a presupposed bias against a proposition being added to the common ground, anchored uniformly to the speaker and therefore not showing ‘interrogative flip’. The paper develops connections between common-ground management operators and evidentials, arguing that interrogative flip (and lack thereof) is a phenomenon that should be studied for a wide variety of discourse particles.


Introduction
This paper explores modal displacement at the level of discourse, where an agent conveys a mental attitude toward another agent's conversational move. We focus on polar questions (henceforth, PQs) in Italian, and the ways in which such questions convey 'biased' attitudes. For example, by uttering the PQ in (1), B conveys a prior bias in favor of the positive answer to the question: that at some point before A's utterance, B thought/assumed that A visited the South American country Peru.
1 There is some controversy on how to characterize biases in PQs. Two types of 'biases' have been discussed: what Domaneschi et al. (2017) term the original speaker bias or original belief prior to the utterance, and the newly acquired belief or contextual evidence bias emerging from contextual evidence (see also Sudo (2013)'s distinction between epistemic and evidential bias). Both types of bias have been used to model the speaker's expectation for the most likely answer to the PQ. Ladd (1981) and  focus on the original speaker bias; Büring & Gunlogson (2000) focus instead on the newly formed bias emerging from contextual evidence; van Rooy & Safarova (2003) merge the two types of biases into one; see also Gunlogson (2001), Asher & Reese (2005), AnderBois (2011), Gutzmann & Miró (2011), Sudo (2013), Krifka (2017), Roelofsen & Farkas (2015), Farkas & Roelofsen (2017), Domaneschi et al. (2017), Goodhue (2018b) for further discussion of bias and types of biasing effects we do not consider here. 2 As noted by Cinque (1976), mica can also be used in negative imperatives, which we will set aside.
The reverse holds for (3). Here, B's negative question with mica conveys that B had a prior expectation for the negative answer (that A would not be invited to the party); omitting mica in this case, would lead to infelicity. 3 This question is also motivated by epistemic conflict: the content of A's claim provides evidence against B's pre-existing negative expectation.
(2) A: Tomorrow I am finally staying in. B: Non eri stato (#mica) invitato alla festa per Alex? 'Weren't you invited to the party for Alex?' (3) A: I know that Alex hates me and tomorrow there's gonna be big drama. B: Non sei #(mica) stato invitato alla sua festa a sorpresa? 'You aren't invited to his surprise party, are you?' (roughly) The second part of the paper develops an analysis of Italian biased questions by extending  /Romero (2014)'s analysis of biases in English PQs to Italian, though we will make substantial revisions to the core theory. With the toolbox for PQs in hand, we return to the account of mica, proposing mica is uniformly an epistemic common ground management operator that indicates an agent's beliefs about whether some proposition should be part of the common ground. This analysis therefore unifies mica with other common ground management operators that have been proposed to account for question bias and denials across languages; in particular, the operators VERUM and FALSUM , Repp 2013, Romero 2014. We propose that mica is special in that it is an inherently subjective operator. VERUM and FALSUM mirror many documented evidential operators by reflecting the perspective of the speaker in unembedded assertions and of the hearer in questions, but mica does not shift its perspective in questions; thus it does not undergo what in the evidential literature is referred to as interrogative flip (Speas & Tenny 2003). We argue that this feature of mica is responsible for its bias reversal effect in PQs. More broadly, our proposal contributes to a larger understanding of the typology of perspectival anchoring in epistemic operators.
3 Mica is subject to dialectal/regional variation. As discussed in Squartini (2017), there is a secondary use of mica in some dialects with a distinct intonation, which does not trigger 'bias reversal'; in these dialects, with the relevant intonation pattern, (2) is acceptable with mica. This secondary use is not accepted by all speakers of Italian: "It is in fact a regional phenomenon restricted to varieties of Italian, whose geographical boundaries are not clear yet, but in a preliminary delimitation, can be located in an area in the North-West of Italy, possibly centered in Lombardy and Northern Emilia" (Squartini 2017:12). In this paper, we will not consider this variety. For other aspects of variation in the use of mica across Italian dialects see also Garzonio & Poletto (2009), Penello & Pescarini (2008), and Visconti (2009)  'It's not cold in Rome.' The particle mica can appear as a discontinuous element of non, as in S's response in (5), or as an autonomous negative element (Cinque 1976) as in S , without any difference in interpretation. (What that interpretation is exactly, we return to in section 2.2). Following Cinque (1976), Zanuttini (1997), we assume that autonomous mica is derived from discontinuous mica via movement: mica moves and takes the place of non, incorporating its negative meaning. As a discontinuous element of non, mica follows verbal elements (auxiliaries, modals, participles), but cannot precede the first one of them and it cannot occur after non-verbal elements that follow the verbal group (Cinque 1976). The spaces in the sentence below show where mica can occur in a sentence, the stars the places where it cannot occur:  (Cinque 1976 ex. 6) We will adopt a version of Zanuttini's syntax for mica. Zanuttini (based on a range of evidence about negation in Romance) proposes an extended sequence of functional projections that includes two positions for negation. (Standard) Italian negation non is base-generated in the higher position, NegP-1, and mica is generated in the specifier of the lower position, NegP-2; see (b) in Figure 1. The surface structure of autonomous mica, illustrated in (a) in Figure 1, involves mica moving to Spec,NegP-1, and as a result, the head of that projection not being spelled out independently. The featural annotations in these trees are our own augmentation to Zanuttini's proposal. While these syntactic proposals do not settle what the LF of a mica sentence might be, they do suggest that, consistent with the interpretive evidence so far, autonomous and discontinuous mica should be treated in a similar fashion.  Unlike plain negation non, mica in declaratives requires a prior claim or salient expectation to deny (Cinque 1976). Compare the minimally different dialogues below, where the negated sentence with mica is marked in the first dialogue, but acceptable in the second. The difference is that in (8), A asks a question that signals he is wrongly assuming that S's sister is in a position of giving them a ride, i.e., she can drive, she has a car, etc. The implicit assumption is what is licensing mica here (same judgments for autonomous mica). 4 Unlike non, which can be used to assert new negative propositions, mica in declaratives seems to be limited to denials. It is known since Horn (1985) that the essential function of a denial is to object to a previous utterance on any grounds whatever, including its conventional or conversational implicata, its morphology, its style or register, or even its phonetic realization. We identify three types of context that license the use of mica in assertions: rectification of an utterance -where mica is used to object to a previous utterance on any ground whatever; rectification of implied content, where mica is used to deny a proposition that the speaker is attributing to the addressee, even when the addressee has not uttered anything to that effect; and self-rectification -where mica is used by the speaker to deny one of her own prior expectations. 5 Some examples are included below, where p refers to the propositional content of α in "(non...) mica α"; we will call this the prejacent. There are two take-home points from this data. First, mica in declaratives occurs in a subset of the environments where non could or does occur -in all of the above examples, plain negation would also be acceptable. Second, mica is not just for surface denials: it can deny a proposition that has never been expressed linguistically, nor presupposed or implied by a previous utterance, including one that has been simply inferred as a belief of another participant.

Previous proposal: mica as presuppositional in assertions
Cinque (1976) (see also Zanuttini 1997, Penello & Pescarini 2008 suggests that mica in declaratives is a presupposition trigger: a sentence of the form (non) mica p asserts that ¬p and presupposes that p was expected. We sketch a particular version of this in (12): This directly captures the distributional facts for declaratives from section 2.1. First, it straightforwardly predicts that mica-sentences will be good in a subset of the cases where regular negative sentences would be good, because the presupposition introduced in (12b) as a definedness condition leads to mica-declaratives being defined in a subset of the contexts where negative sentences are defined, and having the same truth-conditions when defined. The proposal also captures the contextual licensing conditions for mica: the presupposition is intended to cover cases where any participant (including the speaker) said or implied or acted as if they believed α c . A speaker being aware of their own prior assumptions, even if they haven't communicated them, is just another special case. Despite capturing many of the distributional facts we have shown so far, there are some facts that do not obviously follow from this analysis (at least as we have stated it). First, as we will discuss in section 2.3, mica must scope higher than other operators in the sentence (e.g. modals). Second, in section 3 we show that mica in polar questions induces what we label bias reversal, a pattern of data that will occupy much of the rest of the paper. 7 early access Frana & Rawlins

Mica must scope high
As Penello & Pescarini (2008) discuss, the presence of mica interacts with other scopal elements. Where regular negation is ambiguously scoped with respect to the deontic universal modal dovere, as shown in (13), mica-negation must scope above, as shown by the fact that (14) (2008)'s observation can be reproduced with existential deontics, as well. In this case though, the scope of the modal with respect to negation is fixed. Specifically, when negation is placed above the modal, the only possible interpretation is the one following surface scope (NEG CAN: "It is not the case that you can..."). Conversely, when regular negation is placed below the modal the only possible interpretation is CAN NEG ("It is allowed to not..." = "you don't have to..."). The interesting fact is that mica can only be be used in the first sentence type, where negation takes wide scope over the modal, as illustrated below. This observation is not about felicity in particular contexts, but rather is general: there is no possible scenario that would rescue the acceptability of mica in the second example (same judgments for autonomous mica). An analogous contrast can be replicated with epistemic modals. As in the case of the existential deontics, existential epistemic modals have fixed scope with respect to negation in Italian. When the modal potrebbe (roughly, "could" or "might") precedes negation, the only available reading is the one following from surface scope (MIGHT NEG). This is shown in the context below, where Sherlock's first reply to Watson can only mean that the murderer might not be (= does not have to be) Mr. Purple. Replacing ordinary negation with mica in this particular configuration, leads to ungrammaticality (same judgments with discontinuous mica).
On the other hand, if Sherlock intends to express a strong denial of Watson's claim, namely that the murderer can't be Mr. Purple (NEG MAY), this can be rendered by placing negation above the possibility modal può (roughly, "can" or "may"). This configuration is grammatical with both regular negation and mica and both versions of the sentence obligatorily have the reading following from surface scope (which in this case is NEG MODAL The last piece of scopal evidence comes from conditionals. As shown below, mica cannot occur in the antecedent of a conditional (17), while it can occur in the consequent (18). We take this to be evidence that mica at LF must be high enough in the left periphery that it does not embed in (conditional) adjuncts headed by se ('if'; see Iatridou 1991, Haegeman 2003 et seq for discussions of the ways in which the internal syntax of conditional adjuncts can be truncated relative to root CPs). 8 (17) A: It'll rain tomorrow.
7 The sentences L'assassino non/mica potrebbe essere (/mica) Mr. Purple are also grammatical, but dispreferred to the stronger denial made by using può instead of potrebbe. 8 An alternative explanation for why mica is ruled out in the antecedent of conditionals follows from the restrictor analysis of conditionals (Lewis 1975, Kratzer 1986, Heim 1982. Under this view, if-clauses restrict the domain of a (possibly covert) modal, thus effectively trapping mica under the scope of a modal; a configuration which, as we already observed, leads to ungrammaticality.

early access
Frana & Rawlins S: Se domani non/*mica piove (/*mica), possiamo andare al mare. 'If it does not rain tomorrow, we can go to the beach.' (18) A: Tomorrow we'll go to the beach. S: Se domani piove, non/mica possiamo (/mica) andare al mare. 'If it rains tomorrow, we can't go to the beach.' All together the data suggest that mica-negation must scope high at LF, even higher than epistemic modals; if it must appear at LF high in the left periphery, both the scopal data and the conditional data fall out straightforwardly. This contrasts with the scopal possibilities for non, which can be substantially lower. These facts aren't incompatible with Cinque's presuppositional proposal, but they also aren't explained by it: why should negation with mica scope so much higher than regular negation? In summary, mica in assertions has three key properties. First, it is felicitous in a subset of cases where non is: those where the complement proposition of mica (the prejacent) is assumed or taken for granted by some discourse participant. Second, this isn't just a 'surface' denial: mica can be used to deny not just linguistic claims, but propositions that have not been previously expressed, nor presupposed or implicated. Third, mica in declarative clauses scopes high, at least high enough to necessarily outscope epistemic modals. 9 In the next section, we will characterize the felicity conditions of mica in polar questions and introduce the phenomenon of bias reversal.

Biases in English polar questions
A polar question (PQ; sometimes called a 'Yes/No' question) is standardly treated as a question that asks whether a given proposition is true or not (Hamblin 1973, Karttunen 1977, Groenendijk & Stokhof 1984, Roberts 1996. Despite this simple idea, PQs have proven to be quite complicated in their pragmatic details (Borkin 1971, Bolinger 1978, Ladusaw 1979, Ladd 1981. The examples below illustrate five distinct ways of raising a polar 'issue' in English. 10 These alternatives vary on the basis of the presence/absence and position of negation (High vs. Low negation) as well as extra emphatic marking; the 'V' in 'V-PPQ' stands for what has been termed 'verum' marking or 'verum focus' (Höhle 1992, Gutzmann & Miró 2011 rendered by intonational marking on the early access Italian polar questions and 'mica' main verb or an epistemic adverb like really. 11 Although they raise the same issue (i.e., whether or not the addressee is married), the PQs below each have different felicity conditions and cannot be used interchangeably by native speakers. and what we will call evidential bias -that is, a proposition implied by contextual evidence. In the remaining of this section, we will briefly review the felicity conditions of PPQs and HiNPQs (we return to LowNPQs in 4.1). PPQs without polarity focus are seemingly 'neutral' with respect to the context, as they can be felicitously used in interview scenarios, where the speaker can have no prior expectations regarding the resolution and the contextual evidence may as well be absent: (20) Do you have a college degree?
However, as Büring & Gunlogson (2000) first pointed out, although unbiased in a certain sense, English PPQs are not fully neutral with respect to contextual evidence, and they are subject to the relatively weak evidence condition (21).
(21) Evidence condition on PPQs: A PPQ of the form "p?" is felicitous only if there is no compelling evidence against p, i.e. either there is no evidence or there is evidence for p.
The two scenarios below from Büring & Gunlogson (2000) illustrate this point. Instead, V-PPQs and HiNPQs are felicitous in contradiction scenarios, i.e., scenarios in which contextual evidence seemingly contradicts the speaker's prior expectation. More specifically, V-PPQs are felicitous in cases where the speaker had a prior expectation for the negative answer to the question (negative prior bias) and the contextual evidence seemingly contradicts that expectation. HiNPQs, on the other hand, are felicitous in cases where the speaker had a prior expectation for the positive answer to the question (positive prior bias) and the contextual evidence seemingly contradicts that expectation. The contrast is brought up by the two contexts below, which manipulate the polarity of the expectation/evidence. Outside of contradiction scenarios, HiNPQs can also be used in the absence of contextual evidence in order to make a suggestion, as first pointed out by Ladd (1981) with the famous example in (27) The key types of biased polar questions are summarized in Table 1. 12 With this picture in hand, we can turn to positive and negative PQs in Italian and ask a similar question: what expectations involving the speaker and the hearer can be conveyed by different types of PQs?

Biases in Italian Polar Questions
Italian PQs differ from English in at least two respects: there is no marking beyond intonation for PQs (which we signal with punctuation), and there is only one possible surface position for regular negation. 13 However, we will show that there are sub-12 This table is not intended as exhaustive: we have not dealt with either LowNPQs, or with other biasing mechanisms such as tag questions or minimizers. Moreover, as it will emerge, simply describing the polarity (positive vs. negative) of the biases does not give a complete characterization of the felicity conditions of biased questions, even from a descriptive point of view. The felicitous use of PQs may be further restricted by constraints regarding the type of available evidence (direct/indirect), which we are not addressing here. 13 The exact intonation that marks questions in Italian is complicated and dialect/region-specific; see D'Imperio (2002), Grice et al. (2005) (2016) illustrate the contrast between Italian PPQs and NPQs. In Good Manners v. 1, the PPQ is infelicitous, because the context violates the evidence condition (by asking for food, Miles provides evidence against the proposition that he has eaten already), whereas the NPQ is felicitous because the scenario meets the contextual requirements: (prior) positive bias (Clara expected Miles to have eaten already) and negative contextual evidence (Miles' behavior provides evidence against Clara's pre-existing expectation). In Good Manners v. 2, the opposite holds. Here, the PPQ is felicitous because the context satisfies the evidence condition (M.'s behavior provides evidence for the proposition that he has eaten already), whereas the NPQ is marked because the context makes it clear that the speaker did not have a (prior) positive bias (C. invited M. for dinner, thus she expected him to not have eaten already).

(29)
Contradiction: Good Manners v. 1 (S expected p, evidence for ¬p) Clara invites Miles for drinks and tells him to come after dinner. When he gets there, Miles asks if she has any food. Clara asks: a. #Hai have.2sg già already mangiato? eaten?
'Did you eat already?' rise, and in Southern dialects, it is a rise leading to the nuclear pitch accent. In the Southern Italian variety that this paper focuses on, given our preliminary fieldwork, mica receives the nuclear accent in most circumstances, so there may be a regular interaction in this variety between the position of mica and the question contour. We will not investigate the intonational properties of mica here, leaving it for future work. 14 In Italian, the third person is used to address interlocutors in formal situations. 14 early access So far, despite syntactic differences, the Italian PQ system appears to parallel the parts of the English PQ system under consideration. In the next section, we will return to mica and introduce the phenomenon of bias reversal, something quite different to anything seen in English.

Mica in Polar Questions induces Bias Reversal
As anticipated, mica can also occur in PQs; though it cannot appear in constituent questions (Cinque 1976). In PQs, mica has the same syntactic distribution as in declaratives, as shown in (32) Frana & Rawlins (2016) demonstrate that Italian NPQs are felicitous in scenarios where the speaker holds a prior positive bias, i.e. a prior bias for the positive answer to the question (with conflicting contextual evidence, as in contradiction scenarios, or neutral evidence, as in suggestion scenarios). However, when mica is added to an NPQ, it does not simply reinforce that positive bias (which we might expect from the assertion case), but reverses it -a mica-PQ is compatible with a negative prior bias, i.e. it is felicitous in contexts where the speaker holds a prior expectation for the negative answer to the question (and the context provides counter-evidence). Thus, unlike in assertions, regular negation and mica-negation in PQs are in complementary distribution. The mica examples in our 'good manners' scenarios illustrate this bias reversal effect. As before, the alternative word order (discontinuous mica) has equivalent acceptability conditions. 15 (33) Context: good manners v. In these scenarios Mica-PQs pattern with PPQs: they are both compatible with the speaker expectation of the negative resolution, and evidence for the positive resolution. However, Mica-PQs are not just compatible with this bias, but require it. Thus, in contrast to PPQs, they are inappropriate in neutral interview contexts. This completes the paradigm introduced earlier in (28) Mica-PQs are also distinct from V-PPQs, in both English and Italian. Just like V-PPQs, Mica-PQs mandatorily convey the speaker's prior negative bias and are 15 From now on, we will stop bringing up this alternative word order, which has has equivalent meaning as the autonomous mica order in the variety we focus on. 16 English has no general translational equivalent of mica, and we therefore will not always provide a full translation. Paraphrases that come quite close for this particular example would be: oh no, you didn't eat, did you? and you were not supposed to eat...did you?, but these do not generalize.

Italian polar questions and 'mica'
infelicitous in interview-type contexts. However, these two PQs do not have the same felicity conditions. The Italian V-PPQ Veramente hai giá mangiato? (just like its English counterpart Did you really eat already?) is infelicitous in Good Manners v. 2, despite the fact that the scenario satisfies both biases (see Table 1) -we will return to an explanation of this contrast in §6. Moreover, while V-PPQs are limited to contradiction scenarios, a Mica-PQ -just like an NPQ -can be used in absence of counter-evidence. For instance, nothing in the context described in (36) counts as evidence in favor of the positive proposition ("that you told your mother that you were coming over"), yet mica is felicitous. qui? here 'You didn't tell your mother that you were coming over, right?' Another typical use of Mica-PQs in the absence of contextual evidence is found in polite requests, where intuitively the speaker signals their prior bias for a negative answer in order to take some weight off the hearer's shoulders, by freeing them from fulfilling the request (Cinque 1976, Manzotti & Rigamonti 1991). In these cases, mica can be paraphrased with Italian per caso or its English equivalent by any chance. From these examples we conclude that mica PQs can be used in scenarios with neutral contextual evidence, as long as the context satisfies the other contextual parameter, namely that the speaker had a prior expectation for the negative answer to the question (or pretends to have it, as in the case of polite requests). Table 2 summarizes the phenomenon of bias reversal.
One last data clarification point before leaving our data section on PQs: while many of the mica examples so far have expressed the speaker's disappointment (or some kind of bouletic modality), and this is a common function, this is not necessary. For example, it is absent from the incredulous reply in (38). early access Frana & Rawlins Type of polar question speaker's contextual evidence prior bias (H's implied bias) NPQ: Non hai mangiato? p ¬p (or absent) Mica-PQ: Mica hai mangiato? ¬p p (or absent) Non hai mica mangiato? Table 2 Bias Reversal with mica in contradiction scenarios: Similarly, while one common use is to express surprise, this is again not necessary, as shown by polite requests such as (37) above.

Summary: Mica in declaratives vs. mica in PQs
To conclude the description, we want an analysis of Italian PQs and mica that derives: i. The distributional facts: mica in assertions is felicitous in a subset of the cases where non is. Mica-PQs and NPQs are in complementary distribution.
ii. The scopal facts: a modal cannot outscope mica at Logical Form and mica does not embed in the antecedent of conditionals.
iii. The felicity conditions: assertions of the form (non) mica α. are felicitous in contexts in which the proposition expressed by α is assumed by some participants of the context. Question of the form (non) mica α? are felicitous when the speaker had a prior (to the exchange) bias for the negative answer to the question.
iv. The phenomenon of bias reversal: With respect to the speaker's prior bias, Mica-PQs and NPQs have opposite polarity.
In order to address these points we need to first understand how biases work in PQs. This will be the topic of the next several sections, starting with English.

The scope of negation and Negative Polar Questions
We will review evidence in §4.1 that English NPQs with high/preposed negation (HiNPQs) and NPQs with low/non-preposed negation (LowNPQs) have slightly different felicity conditions (Büring & Gunlogson (2000), Romero & Han (2004), a.o.). Given that Italian, as we have discussed, has only one surface position for non, this raises the question of whether Italian should receive a different analysis of NPQs than English. To address this, we introduce the data on inner vs. outer negation readings, first discussed in Ladd (1981), and then examine comparable data in Italian. The main finding of this section is that Italian NPQs also have inner and outer negation readings.

LowNPQs vs. HiNPQs
The now-classic puzzle of positive vs. negative questions is that on standard accounts, the positive and negative forms of the questions are denotationally identical. For example, on Hamblin's (1973) proposal, the semantic content of a question is described as the set of propositions corresponding to possible (complete) answers, which for PQs are constructed from a proposition by simply taking the positive and negative version of that proposition: whether ¬p = {¬p, ¬¬p} = {p, ¬p} c. Therefore, whether p = whether ¬p Setting aside V-PPQs, where the difference in felicity conditions may follow from the presence of (VERUM) focus and/or (possibly covert) epistemic particles such as really, the truly puzzling contrast is the difference in felicity conditions between PPQs and NPQs, with only the latter leading to mandatorily biased interpretations.
However, a closer look at the data reveals that the standard semantic analysis of PQs may salvageable (from this puzzle at least), and that the observed difference in felicity conditions between PPQs and NPQs may rely on a special type of negation. We initially observed that, in English, NPQs can be realized with 'preposed'/high negation and non-preposed/low negation: a. Doesn't this bus stop at the mall? (HiNPQ) b. Does this bus not stop at the mall?
So far, we have ignored this distinction in the position of negation and focused only on HiNPQs. However, the position of negation in English PQs correlates with certain interpretive differences, including the 'strength' of the speaker's prior bias. As shown by , HiNPQs obligatorily convey the prior positive 19 early access Frana & Rawlins Type of polar question speaker's contextual evidence prior bias (H's implied bias) PPQ: Are you married? can be absent p (or absent) V-PPQ: Are you MARRIED? ¬p p Are you really married? HiNPQ: Aren't you hungry? p ¬p (absent in suggestions) LowNPQ: Is she not coming either? can be absent ¬p (or absent) Table 3 Summary of English polar question and bias types bias, whereas LowNPQs do not necessarily convey such bias. Hence, the latter, but not the former, are felicitous in contexts where the speaker had no prior expectation regarding the answer to the PQ. The contrast is illustrated by the example below from  This data leads to the updated picture of English PQs in Table 3. PPQs and LowNPQs do not differ with respect to the first parameter (speaker's prior bias), and both of them are subject to an evidence condition matching the respective polarity (Büring & Gunlogson 2000). The cases that stand out are V-PPQs and HiNPQs. One could easily take it that polarity focus, or the presence of epistemic particles such as really, would have an impact on the compositional semantics of the PQ, but what is special about preposed negation? Outside of questions, the syntactic position of negation (high vs. low) has been shown to have several consequences at the semantics/pragmatics of the sentences in which it occurs (Ladusaw 1979, Szabolcsi 2004, Schwarz & Bhatt 2006, Romero 2014, most famously licensing of polarity items. Polarity items have been described as having a disambiguating effect that correlates with the scope of negation, between what have been termed inner and outer readings; this will be the topic of the next section.

Inner vs. Outer negation readings
As already noted, NPQs sound very natural in contradiction scenarios. These are cases in which the speaker had a previous expectation for the answer to the question (e.g. bias for p) and the context, or the addressee, is providing partial evidence against it. When faced with this type of epistemic conflict, the speaker might decide to ask an NPQ with one of these two intentions in mind: he or she may intend to confirm, or "double-check", their (positive) prior expectation for p (outer negation reading) or to double-check the (new) implied proposition that ¬p (inner negation reading; see Ladd 1981, Büring & Gunlogson 2000. The example below brings out the two readings intuitively. In both cases, the polarity of the prior bias is positive. It is just the proposition that gets 'double-checked' which changes: a prior bias for p vs. a (new) implied inference that ¬p (where p = the bus stops at the Hampshire Mall).

(42)
Context: Hampshire Mall (S expected p, evidence for ¬p) S wants to go to the Hampshire Mall and has been told that the B43 stops there. While on route, the bus goes past what the speaker thought was the stop. S asks the driver: a. (What're you doing?) Doesn't this bus stop at the Hampshire Mall? S had prior expectation that the bus stopped here (p) and thinks driver may have skipped the stop, so is double-checking the prior expectation that p.
(Outer NPQ) b. (Oh no!) Does this bus not stop at the Hampshire Mall?
S had a prior expectation that the bus stops there (p) and now thinks she may have been wrong, so is double-checking the implied inference that ¬p.
(Inner NPQ) Our example employs a HiNPQ for the outer reading and a LowNPQ for the inner reading. The HiNPQ from Ladd's suggestion scenario (27) is also an OuterNPQ because it expresses the speaker's request for confirmation of her prior positive expectation that there is a vegetarian restaurant. One may legitimately wonder whether inner and outer negation readings correlate strictly to the (surface) placement of negation (HiNPQs/outer vs. LowNPQs/inner). The answer to this question, at least 21 early access Frana & Rawlins for the English data, is not so straightforward due to cross-speaker variation. Ladd reports that English HiNPQs can have both outer and inner readings (though this judgment of ambiguity doesn't hold across all English speakers, and some accept only the outer reading with high negation). 18 The example below is Ladd's own variant to (27) This example, according to Ladd, is an example of inner negation reading: Bob assumed the truth of the proposition that there is a vegetarian restaurant around here (p), but he had to infer from Kathleen's response (contextual evidence) that this proposition might actually be false, and is using the HiNPQ to double check on the newly implied inference that ¬p (inner negation reading) due to epistemic conflict.
Ladd then argues that, at least for English, the difference between inner and outer readings is a genuine syntactic/semantic ambiguity, involving a difference in the scope of negation and an operator whose nature remains to be determined. In the case of outer negation readings, Ladd takes a sentential negation to be outside the scope of the operator, which intervenes for licensing purposes -thus the proposition being double-checked has positive polarity. On the other hand, in the case of inner negation readings, sentential negation is below the operator, and the proposition being doublechecked has negative polarity. Ladd uses data from polarity items as a probe into a scope-based analysis; see  for much further discussion of this diagnostic. In particular, these authors show that a PPI disambiguates an NPQ in favor of the outer reading and that an NPI (choosing NPIs that do not already have questions as a licensing environment) disambiguates an NPQ in favor of the inner 18 Based on consultation with many native speakers, there is substantial variation across speakers in whether inner readings are available (by all diagnostics) with preposed negation. The situation is summarized below.
(i) a. Returning now to Italian, we have already seen that Italian NPQs, just like English NPQs, are only felicitous in contexts compatible with the speaker having a prior positive bias. We now show that Italian, despite surface appearances, has Ladd's inner vs. outer distinction. (See Höhle 1992, Büring & Gunlogson 2000 19 The additive particle too has classically been thought to be anti-licensed in negative contexts since Klima (1964), including in the polar questions literature, and for this reason we present it in the main text. However, Rullmann (2003) has argued that its distribution can be explained by a combination of its presupposition, a preference for low attachment, and competition with either, which Rullmann argues should be treated as an NPI. In Italian, we can construct a similar diagnostic using certain polarity items and their positive counterparts. We will employ the polarity items anche ('too/also') and neanche ('neither') to distinguish between the two readings. 20 The data show that Italian NPQs can have both inner and outer readings, despite having only one surface position for negation. For instance, in the contradiction scenario in (46), S is double-checking the positive proposition p "that A is (also) going" (outer negation reading). The motivation behind the double-checking move is an apparent epistemic conflict: contextual evidence contradicts S's prior expectation for p. NPQs also license NPIs; moreover, when neanche ('neither') is used in an NPQ, it disambiguates in favor of the inner reading (double-checking implied contextual inference for ¬p). In (47) below, S is double-checking the implied proposition ¬p, i.e. "that A is not going"; as before the motivation behind the double-checking move is epistemic conflict: contextual evidence contradicts S's prior expectation for p.  (49) is ungrammatical with the NPI neanche, while the variant with the PPI anche is grammatical and would be felicitous in a scenario in which for e.g. the speaker told her younger sister to not go out with her and her friends, but is now faced with counter-evidence.

Analyzing Biases
In this section, we develop a novel account of biased PQs in Italian, and show how to extend it to Mica-PQs. We take as our starting point the , Romero (2014) account of biases in English PQs via common ground management operators, such as VERUM and FALSUM; however, we modify this account substantially to treat 'CG-management content' as presuppositional (providing contextual bias conditions), and to drastically simplify (and, we hope, clarify) the pragmatic reasoning that leads to evidential bias. We then develop an account of mica in this framework. We argue for a treatment of mica as a CG-management operator, FALSUM, but one that differs from the NPQ uses of that operator in terms of its perspectival anchoring. While VERUM and FALSUM mirror evidential operators in that they show interrogative flip, or what Garrett (2001) termed 'origo shift', mica remains stably anchored to the speaker. The analysis has the further advantage of providing a unified account of mica in PQs and declaratives, as well accounting for the scopal facts from 2.4 and the polarity licensing facts from the previous section. Stalnaker (1978) famously proposed that in modeling assertions in a context, we need to think of that context as a common ground (CG): a set of propositions that discourse participants have publicly and mutually agreed to. This can be simplified to a context set, the intersection of propositions in the CG, which represents the shared information state of agents in a discourse. An assertion, on this view, proposes to add new information to the CG, which results (if everything goes well) in a monotonic shrinking of the context set over the course of discourse. 25 early access Frana & Rawlins (50) Let a common ground CG w for some context be the set of propositions that discourse participants for that context have publicly accepted (abstracting away from time) in a world w .

Common Ground Management
The conversational move of assertion is the instruction or request to add the asserted proposition p to the CG (see e.g. Roberts 1996Roberts , 2012 and it is governed by the maxim of Quality. One could accept a move, in which case the proposition asserted is added to the CG. (51) Maxim of Quality: Try to make your contribution one that is true, specifically, (i) do not say what you believe to be false and (ii) do not say that for which you lack adequate evidence. (2010) we assume that a default assertion (i) commits its author to the propositional content of the asserted sentence, (ii) raises an issue (the denotation of the sentence radical) and (iii) directs the conversation towards a unique resolution of that issue, namely confirmation of the assertion. Thus, an assertion yields an output context that is categorically biased in favor of confirmation of the asserted proposition. Default (= unbiased) Polar Questions, on the other hand, raise an issue and indicate two possible ways of updating the CG, corresponding to the two answers: "The context state after a default polar question is inquisitive with respect to the denotation of the sentence radical because the Projected State contains both future common grounds in which p is added and future common grounds in which ¬p is added" (Farkas & Bruce 2010: 10). While Stalnaker was primarily concerned with the role and result of conversational moves such as assertion, a substantial body of recent work suggests that natural languages contain rich sets of morphology for indicating meta-conversational information of various kinds (see e.g. Krifka 2008, Farkas & Bruce 2010, Zimmermann 2011, Ginzburg 2012, Repp 2013. Here we build most directly on Repp (2013), who refers to such items as common ground management operators.

Following Farkas & Bruce
An example of a meta-conversational move would be to challenge an interlocutor's assertion instead of accepting it. This sort of move is typically marked; for example, in English this could be done using a V-PPQ, as shown below. In Repp's terms, therefore, it is plausible to take really to be a CG-management operator. CG-management operators indicate the status of a proposition relative to the CG and may play at least one of the following functions: (i) they indicate whether the proposition is part of the CG, or whether it is new; (ii) they indicate the interlocutor's current stance towards the proposition: whether the proposition is e.g.  Höhle's (1992) Verum focus; but see Gutzmann & Miró 2011, Goodhue 2018a. This operator has the semantics of an epistemic modal, though it intuitively operates at a 'meta-level' with respect to discourse. The following is Romero & Han's entry for VERUM (their ex. 43) . In what follows, x is a free variable whose value is contextually identified with the speaker in declaratives and the addressee in questions; Epi x (w) the set of worlds conforming to x's knowledge in w; Conv x (w ) the set of worlds where all the conversational goals of x in w are fulfilled (e.g attain maximal information while preserving truth); CG w is the Stalnakerian common ground at a world w, i.e. the set of propositions that the speakers assume to be true at w (c.f. Stalnaker 1978).
Paraphrase: x is sure that, in all the worlds satisfying x's conversational goals, p is added to the CG.
VERUM on this view is a perspectival operator: it introduces entailments about the state of the CG from the perspective of one of the participants in the discourse. The perspective taken by the operator is bound by the speech-act operator. That is, in assertions VERUM is speaker-oriented, and in questions, it is hearer-oriented. This behavior closely resembles the phenomenon of 'interrogative flip' in many languages with evidential systems (see Bhadra 2017 for a recent overview). As an example of the role of VERUM in assertions, consider (55). This scenario begins with S putting forward the unmarked assertion of the proposition that Tom is tired. A responds by arguing with this claim (a meta-conversational move itself), so the claim doesn't immediately become CG. S then responds with a meta-conversational reiteration: establishing over and above the initial claim their high degree of certainty in it, by indicating that they are certain that the proposition should be added to the CG. A then gives in and accepts the claim. 27 early access Frana & Rawlins (55) Context: the hike S: Tom is tired. ASSERT S [Tom is tired] A: I am surprised, we have only walked for 10 minutes. S: He really is tired.
ASSERT S [VERUM S [Tom is tired]] (S is certain about adding [Tom is tired] to the CG) A: Ok, fine.
(⇒ 'Tom is tired' is added to CG.) Repp additionally proposes that illocutionary negation corresponds to the metaconversational operator FALSUM, which is a VERUM with embedded negation: x is sure that, in all the worlds satisfying x's conversational goals, p is not added to the CG.
FALSUM is in the general family of illocutionary or metalinguistic negative operators (see also Krifka 2017). An example can be found in denials. Depending on the context and on intonation, a negative sentence can be read in two ways: (i) as an assertion of a negative proposition, as in (57) or (ii) as a denial of a positive proposition, as in (58) -a common characteristic of illocutionary negation is that it carries stress (Horn 1985, van Leusen 2004 A major question at this stage is how, exactly, the CG-management content of a meta-conversational move relates to the prejacent. We will approach this by asking: what is at issue in a CG-managing move? There are two main answers.  as well as Repp (2013) and Frana & Rawlins (2016) take the CGmanagement proposition itself to be at-issue. That is, S's response in (55) asserts that in all accessible worlds, the proposition that Tom is tired is added to the CG. The alternative possibility (Romero 2014) is that all that is at-issue in such a move is the prejacent of the CG-management operator; the CG-management content would then operate on another dimension, e.g. as a conventional implicature in the sense of Potts (2005) or as a presupposition (Yuan 2015). We briefly review two arguments for treating the CG-management content as not-at-issue, and then revise the current denotations of VERUM and FALSUM. The arguments point towards a multi-dimensional analysis of the differences between CG-managing operators and epistemic modal auxiliaries.
The first argument comes from a commonly used diagnostic for (non) atissueness in the evidential literature, the challengeability test, according to which only the at-issue content of a speaker's utterance can be accepted or rejected by an addressee (Faller 2002, Papafragou 2006, Murray 2017; but see Korotkova 2016a for a different account). With an epistemic modal, a denial can target the modal claim (in addition to the prejacent in some contexts), whereas in the case of VERUM/really, just like in the evidential case, the denial must target the embedded claim, not the meta-conversational claim about certainty: (59) A: The keys have to be in the car. (Epistemic modal) B: No, they don't have to be, they might be somewhere else. The second argument comes from the interaction of such operators with question meanings and answerhood licensing. As noted by Gutzmann & Miró (2011), Romero (2014, a negative answer to a question containing an epistemic modal can be taken as negating the modalized epistemic proposition (thus, expressing lack of certainty). In contrast, the corresponding negative answer to a question containing epistemic really, or VERUM focus, can only mean that the negative alternative is true, and cannot express a lack of (un)certainty. This is expected if CG-managing operators, such as VERUM, are not part of the at-issue content.  (2011) give a rather different account of the content of VERUM (they deal only with declaratives, not PQs), but also move this content out of the at-issue content into the 'expressive' dimension. While we think that the basic move to not-at-issue content is necessary, the evidence for choosing between categories of not-at-issue meaning is somewhat more equivocal.
Once again the situation resembles that of evidential marking, where the so-called 'family-of-sentences' tests (see e.g. Chierchia & McConnell-Ginet 1990, Tonhauser et al. 2013 are challenging to apply, because VERUM and FALSUM can't be embedded; embedded NPQs with whether seem to involve regular negation, e.g. do not indicate a speaker or attitude holder's prior beliefs or evidence. 22 Though Gutzmann & Miró (2011) take a stand in a particular direction, they also did not provide arguments for choosing expressive content over other options.
In what follows we will develop a new version of this proposal that treats CG-managing content as presuppositional, but we do not rule out the possibility of translating the analysis to other categories of not-at-issue meaning. The core intuition behind choosing a presuppositional account is that negation and mica in denials and PQs impose appropriateness conditions on input contexts, which lead to judgments of infelicity when these conditions fail. Since one of our primary desiderata is to capture these conditions, presuppositionality is a natural tool. To the extent that biased questions indicate new information (e.g. about the speaker's bias), we will derive it as an implicature, via Romero & Han-style reasoning (to be developed in §5.3), but we suggest that the bias content that we encode lexically is not generally used to convey new information. We will conclude this discussion by presenting an argument towards a presuppositional account of FALSUM in PQs, involving the interaction of HiNPQs and conditionals. One case where NPQs can embed with a biased meaning is in the consequent of a conditional question. Following Isaacs & Rawlins (2008) we take it that the if -clause has to scope over a question operator, because it is providing a restriction to that operator, while FALSUM (on Repp's syntax at least, see (53)) must syntactically scope in the consequent below the if -clause. The question is if the not-at-issue implication of FALSUM, which for purposes of discussion we will assume is the CG-management content in (56), is interpreted locally or globally. On a presuppositional story we expect local interpretation to be possible in the consequent of conditionals (based on well-known filtering data; see Karttunen 1973 et seq). On e.g. a Pottsian conventional implicature analysis (Potts 2005), we would instead expect the implication to project globally. In (64) we see that the biased contribution of an NPQ in a conditional question gets conditionalized. That is, in B in contrast to B, the speaker indicates a prior expectation that the party will be indoors if it rains. To make sense of this, the implication of FALSUM must be interpreted in the local context where the temporary assumption that it will rain is already in play, rather than escaping the conditional.

VERUM and FALSUM in Italian polar questions
We give an account of the pragmatics of biased PQs that builds on 's framework, keeping their syntax as well as the core of their pragmatic reasoning -that the use of a meta-conversational operator triggers reasoning about the choice of a complex discourse move over a simpler competitor. We depart from  in treating the contribution of CG-management operators as not at-issue (Gutzmann & Miró 2011, Romero 2014). More specifically, we propose an analysis in which the meta-conversational content is modeled as a presupposition. As a result, our analysis avoids complex theoretical tools like biased partitions, (as in , and non-at-issue inquisitive content (as in Romero 2014). 23

The LFs of Polar Questions and an explanation of the polarity facts
We adopt 's LFs for PQs (with the innovation introduced by Romero (2014) of using FALSUM instead of ¬VERUM), together with their account of inner/outer readings in correlation with polarity items (section 4.2) through an interplay of the CG-management operators VERUM/FALSUM and negation. In a nutshell, outer negation in NPQs does not introduce standard negation at LF. Instead, it corresponds to the negative operator FALSUM which, unlike logical negation, fails to license NPIs (or anti-license PPIs). Thus, it follows that NPIs are only licensed in early access Frana & Rawlins Inner-NPQs, where regular negation takes scope over the prejacent proposition and is below the (optional) VERUM operator. We propose to extend this analysis to Italian PQs. As already noted, despite having only one surface position for negation, Italian NPQs also have inner and outer negation readings. We take this as evidence for postulating two types of negation; the outer one corresponding to the operator FALSUM, the inner one for regular negation. Thus, by the same logic, it follows that the PPI anche is licensed in Outer-NPQs because FALSUM, unlike regular negation, is not a PPI anti-licensor. Conversely, the NPI neanche is licensed in Inner-NPQs where we find standard negation.

Meta-conversational content as presuppositional
Our presuppositional entries for VERUM and FALSUM are presented in (67) and (68) respectively. The CG-management content of VERUM/FALSUM is rendered as a presupposition on the input context from the perspective of some individual in discourse. Adopting a term from the evidential literature, we will indicate the perspectival parameter of these operators with a contextual variable Origo c . In root assertions, Origo c is the speaker, and in root questions, Origo c is shifted to the hearer. Following much of the literature on evidentials, we assume that the Origo parameter is effectively bound by the illocutionary operator in ForceP, without intending to presuppose that the ultimate analysis must literally involve binding. 24 (67) VERUM c,w = λ p s,t . p Defined for p, c, w only if ∀w ∈ Epi Origo c (w) : (∀w ∈ Conv Origo c (w ) : (p ∈ CG w )) "Origo c is sure that, in all worlds satisfying their conv. goals, p is CG." ∀w ∈ Epi Origo c (w) : (∀w ∈ Conv Origo c (w ) : (p ∈ CG w )) "Origo c is sure that, in all worlds satisfying their conv. goals, p is not CG." As in , PQs with CG-management operators are metaconversational moves, which are subject to a discourse-economy constraint: 25 (69) Principle of Economy (after  Do not use a meta-conversational move unless necessary (to resolve a Quality dilemma).
A 'quality dilemma' (our term) is when some agent a in discourse is faced with one of these two possible situations: (70) Quality Dilemmas a. Epistemic Conflict: S has a pre-existing bias (for p or ¬p) and the context (e.g. the hearer) presents counter-evidence. b. Lack of Evidence: S has a pre-existing bias for p, but not enough evidence to assert it and the context does not support either p or ¬p.
The Principle of Economy is intended to explain why V-PPQs and HiNPQs are felicitous only in scenarios where the speaker held a prior expectation regarding the answer to their own question. In contradiction scenarios, where S believed p and H produced evidence against p, S is justified in questioning the appropriateness of adding p (or the newly implied ¬p) to the CG in order to resolve the apparent epistemic conflict. In suggestion scenarios such as (27), where S believed p, but doesn't have enough evidence to assert it, she can use the meta-question move to double-check this belief and potentially gather evidence, while indicating the belief. It follows that in scenarios where the speaker holds no prior expectation regarding the answer to her PQ, V-PPQs and HiNPQs are predicted to be infelicitous, because they involve the use of a meta-conversational move without a justified reason, in violation of the Principle of Economy. A corollary of this is that when the metaconversational move is optional, as in some NPQs (e.g. LowNPQs in (41)), the Principle of Economy won't trigger and such questions are felicitous in neutral contexts.
Let us now turn to the derivation of the biases, starting with V-PPQs; recall that these PQs license the inference that the speaker had a prior expectation for early access Frana & Rawlins the negative answer to the question, and they are felicitous only in contradiction scenarios, where the context (e.g. the hearer) provides counter-evidence. We take the polarity of the speaker's prior bias for a V-PPQs to be derived as shown below. Because this is a question, Origo c is resolved to the hearer, which we notate h c .
"h c is sure that in all the worlds satisfying h c 's conv. goals p is CG" The presupposition is met in the input context, given that H has just asserted p. By quality, if one asserts p, it follows that they believe p to be true and that they have adequate evidence for p. Moreover, since the default outcome of an assertion is confirmation (Farkas & Bruce 2010), it is reasonable to assume that, through their assertion, H conveys the expectation that p will become CG. Note that since the presupposed content of VERUM in questions requires that H has indicated certainty for adding p to the CG, suggestion scenarios (where contextual evidence is neutral) are automatically ruled out. S's epistemic bias can be derived via a competition effect, following the Principle of Economy: S chose a complex (meta-conversational) question ([Q [ VERUM [you go]]]) over a simpler competitor (same question without VERUM). She must have a reason to do so. Thus, by the Principle of Economy, S is facing a quality dilemma. Since lack of evidence is ruled out (as explained above), S must be facing epistemic conflict (contextual evidence contradicts their prior bias). Given that H gave evidence for p (h c goes), then S must have had a prior bias for ¬p, and is requesting confirmation for (double-checking) the newly implied proposition that p because of epistemic conflict. 26 Furthermore, the analysis explains why a V-PPQ is infelicitous in scenarios like our Good Manners v. 2 (34): the presupposition introduced by VERUM in '[Q [VERUM [you ate already]]]' would be satisfied in scenarios where H had indicated certainty about adding p (= H ate already) to the CG. Such a requirement is not met in our scenario, where the evidence provided by H (barely not touching food) is not strong enough to imply their certainty for wanting to add p to the CG. Summing up, the analysis derives the felicity conditions of V-PPQs of the form '[Q [VERUM [X]]]', namely that they can be felicitously uttered in contradiction scenarios only, where contextual evidence -in the form of H's commitment towards X -contradicts the speaker's pre-existing bias for the negative answer to the question.
Analogous reasoning can be applied to the derivation of the speaker's bias in Inner-NPQs, for instance, NPQs with low negation and the particle really: "h c is sure that in all the worlds satisfying h c 's conv. goals ¬ p is CG" The presupposition is met in this context since H asserted ¬p. As above, the speaker's epistemic bias can be derived via a competition effect, following the Principle of Economy: S chose a complex (meta-conversational) question over a simpler competitor (same question without VERUM). She must have a reason to do so. Thus, by the Principle of Economy, S is facing epistemic conflict (we know that "lack of evidence" is ruled out with VERUM). Given that H gave evidence for ¬p (H is not going), then S must have had a prior bias for p, and is requesting confirmation for (double-checking) the newly implied proposition that ¬p because of epistemic conflict. 27 Finally, Outer-NPQs contain FALSUM. These questions are felicitous both in contradiction scenarios and in suggestion scenarios. This follows from the fact that, unlike VERUM, FALSUM does not require the hearer to have expressed the intention of adding a given proposition (p or ¬p) to the CG. In fact, the non-at-issue content of FALSUM in questions is logically compatible with two possible states of affairs: (i) H is sure ¬p should be CG; for e.g. H uttered/implied ¬p, or (ii) H has not taken a stance towards p (e.g. H has professed discourse neutrality towards the p/¬p issue) and nobody else has taken a stance towards the issue (neutral evidence context). Case (i) corresponds to contradiction scenarios and case (ii) to suggestion scenarios. We illustrate with a contradiction scenario: "h c is sure that in all the w satisfying h c 's conv. goals, p is not CG." The presupposition is met in this context since H asserted q, which implies ¬p. As before, S's epistemic bias is derived via a competition effect: S chose a complex (meta-conversational) question over a simpler competitor (same question without early access Frana & Rawlins FALSUM). She must have a reason to do so. Thus, by the Principle of Economy, she is facing a quality dilemma. Given that H gave evidence for ¬p, then S must have had a prior bias for p. Another possibility, compatible with H being certain about p not being CG, would be that H did not take a stance with respect to p, as in suggestion scenarios. In both cases, the speaker wants to double-check their prior bias that p (outer negation reading) because of epistemic conflict (contradiction scenarios), or lack of evidence (suggestion scenarios).
Summing up, the analysis derives the felicity conditions of biased questions (V-PPQs, Outer and Inner NPQs), namely that they can be felicitously uttered in contradiction scenarios, where contextual evidence contradicts the speaker's prior bias for the answer to the question. Furthermore, it follows that only Outer-NPQs are felicitous in suggestion scenarios. This is because FALSUM stays for lack of certainty for adding p to the CG, which is compatible with evidence against p (contradiction scenarios) but also lack of compelling evidence (suggestion scenarios). The analysis also derives parallel licensing facts to English, despite surface differences.

Mica and bias reversal
The CG-managing operators VERUM and FALSUM are perspectival operators: they introduce information about the state of the CG explicitly anchored to the perspective of one of the discourse participants. The perspective taken by the operator is effectively 'bound' by the speech-act operator . That is, in assertions VERUM/FALSUM are speaker-oriented, and in questions, they are hearer-oriented. As we have noted, this situation resembles the common behavior of interrogative flip seen with evidential marking cross-linguistically (Garrett 2001, Speas & Tenny 2003, Murray 2017, Korotkova 2016b, Bhadra 2017. Our proposal for mica is that it is the same kind of perspectival operator, but without the flip: mica introduces a FALSUM operator that, rather than having bound perspective variables, has its perspective necessarily anchored to the speaker. This makes mica analogous to the type of evidential recently discussed by Bhadra (2017) (focusing on Bangla) that does not undergo interrogative flip. We will show that this perspectival stability inverts the pragmatic reasoning triggered by the use of a CG-management operator in regular NPQs, leading to a reversal in the polarity of the bias on the part of the participants in discourse. Under our proposal, the bias reversal phenomenon for mica amounts to lack of interrogative flip.
We earlier presented arguments that FALSUM should be treated as not-at-issue and one argument for a specifically presuppositional account; the arguments for not-at-issueness were straightforward and apply directly to mica as well, though the arguments for presuppositionality were somewhat equivocal. The situation for deciding among types of not-at-issue content is not much easier for mica: once again, the family-of-sentences largely do not apply, since mica is hard to embed; it can't occur in the antecedent of conditionals, with a higher negation ("It is not the case that (non)...mica p"), or in alternative questions. However, we do have one interesting case: mica can embed under certain attitude verbs. In (77) the projective content is the implication that Carla -not the speaker -is certain that p should not be CG (p = Gianni stole the printer); thus the contrast between the two possible continuations (importantly, the global assertion with just plain negation non would be acceptable). In these cases we see what Tonhauser et al. (2013) described as 'obligatory local effect' -the anchor/origo for mica must be the attitude holder, not the speaker, and the implication acts as if it is embedded in the belief context. This data is suggestive of a presuppositional analysis, in that the other English triggers showing obligatory local effect for Tonhauser et al. (2013)  Our entry for mica is given in (78). For the moment, we will take it to be a lexical property that mica is speaker-oriented (Origo c is resolved to the speaker, which we notate s c ), though we return to this below. 29 (78) mica c,w = λ p . ¬p Defined for p, c, w only if ∀w ∈ Epi s c (w) : (∀w ∈ Conv s c (w ) : (p ∈ CG w )) Definition is equivalent to: mica c,w = FALSUM c/Origo→s c ,w 28 We leave open the question of what the exact mechanism is for the local effect in this case. One possibility is that projection is blocked because 'Origo' is bound locally by a doxastic operator. If mica were maximally projective (e.g. Potts-style CI-content), but required local binding of the origo variable, we might expect overall infelicity when embedded under attitude verbs, contrary to fact. One may also wonder whether mica exhibit 'strong contextual felicity' effects, another test used by Tonhauser et al. (2013) to diagnose presuppositional content. However, as we have shown, although mica gives rise to contextual felicity effects -in that it is infelicitous in a neutral context, like e.g.(35), it can also be used discourse-initially (as in polite requests or in the paranoid scenario), as long as the inference can be easily accommodated. 29 Where x is a named slot in c, the notation c/x → y indicates a modified context where x is evaluated as y relative to that context. 37 early access Frana & Rawlins In assertions, perspectival stability has no effect: Origo c in regular FALSUM would be bound to the speaker regardless. For questions, the prediction is that mica in PQs does not undergo interrogative flip, remaining anchored to the speaker, even though regular FALSUM would have a flipped Origo c .
(79) [Q [mica [p]]] c,w = {p, ¬p} ('S asks H whether p or not p') Defined if ∀w ∈ Epi s c (w) : (∀w ∈ Conv s c (w ) : (p ∈ CG w )) "s c is sure that in all the w satisfying their conversational goals p is not CG" Just as in the case of hearer oriented FALSUM, this presupposition can be met in two ways: (i) in the input context, S is sure that p is not CG because she believes ¬p to be CG, or (ii) S is sure that p is not (yet) CG because nobody has taken a stance towards p (or ¬p) -the context is neutral with respect to the issue. To see this in example form, let us revisit the contradiction scenarios used before: The presupposition introduced by mica is met in the second scenario only, where the speaker -prior to her own question -believed ¬p (= Miles has not eaten already) to be CG. The mica-PQ presupposes certainty that p (H ate already) should not be common ground; this presupposition is satisfied in Good manners v. 2 because S believed this was an established expectation, so ¬p ought to have been common ground in the input context. However, this is a question and so the speaker isn't sure, relative to the current context. Moreover, it is a meta-question, so the speaker is trying to resolve some epistemic or evidential conflict in the current context. In this case, there is contextual evidence against their prior expectation, in the form of Miles' apparent lack of hunger, and so they do not have sufficient evidence to decide the question themselves. However, Miles should certainly be able to report on his own hunger, and so would be in possession of appropriate evidence to decide whether adding p to the CG is compatible with quality from the speaker's (or any) perspective. Consequently, the mica-PQ is felicitous in this context for indicating the speaker's prior expectation that Miles would not eat.
The opposite holds for the Outer-NPQ that is felicitous in good manners v. 1 only. The presupposition of hearer-oriented FALSUM is straightforwardly met in this context since H (Miles) implies -through his request for food -that he didn't eat already, indicating that ¬p should be common ground, from his perspective. As in the previous cases, the speaker's epistemic bias can be derived via a competition effect. By the Principle of Economy, the speaker is facing a quality dilemma. Given that H gave evidence for ¬p, then S must have had a prior bias for p.
Furthermore, the analysis makes the welcome prediction that mica-PQs and V-PPQs won't have the same felicity conditions: recall from section 3.3 that, unlike a mica-PQ, a V-PPQ is not acceptable in the Good Manners v. 2 scenario. In fact, despite sharing the same type of prior speaker bias, the content of the presupposition introduced by the two CG-management operators in PQs is not the same: while VERUM presupposes the hearer's certainty for p being CG, which is too strong to be satisfied in our scenario (where Miles's indication is very indirect), mica requires the speaker's certainty for p not being CG, which is instead satisfied in the scenario.
Finally, this analysis predicts the polarity licensing data for mica. Recall that mica in PQs does not license NPIs and does not anti-license PPIs (49). Given that mica introduces FALSUM and we have already established that FALSUM, unlike logical negation, does not license NPIs (or anti-license PPIs), we predict that mica-PQs will behave exactly like Outer-NPQs with respect to licensing of polarity items.
Before proceeding to assertions, we will briefly sketch three hypotheses about why it is that mica is perspectivally stable. The first is that, in line with our presentation so far, as well as Frana & Rawlins (2016), the reference to the speaker is just a lexical property of mica. We know of nothing that would rule out this possibility, but will consider two hypotheses from which speaker anchoring might follow. These build on recent work by Bhadra (2017) on variation in interrogative flip for evidentials across languages. Bhadra's core proposal is that interrogative flip happens when an evidential's perspectival anchor is determined by a speech-act operator. Thus, following Bhadra (2017) on Bangla, the second hypothesis is that mica occurs in the analogue of rising declaratives, not 'true' questions. In Bangla, Bhadra argues that there are no true PQs, which generally leads to a lack of interrogative flip in the language. However, in Italian we see interrogative flip in PQs with veramente and plain negation, so the hypothesis must be more limited relative to Bangla -at least some PQ-like utterances would need to involve genuine PQs. 30 The third hypothesis, also in line with (but not directly predicted by) Bhadra (2017), is that mica scopes high enough that it can never have its perspectival anchor shifted to the hearer, i.e. it 30 The intonational evidence we have examined in Italian is fairly inconclusive about differentiating 'true' PQs from something more declarative, where the word order is the same. That there is such a distinction is actually an intuition some speakers have, but one that requires more investigation than we can do here.

early access
Frana & Rawlins actually scopes above the question operator; we would also need to assume that the unmarked Origo is the speaker. 31 We have already provided a lower bound for mica at LF that is high in the left periphery, so this third hypothesis would be a natural extension of the scopal evidence examined earlier.
In summary, our proposal is that a question of the form "mica p?" presupposes that the speaker is certain about p not being CG, leading to the expression of a mismatch in expectation with the current context. This presupposition can be met in two ways. Either S's certainty follows from a prior bias for ¬p, in which case the reason for still posing the question would be epistemic conflict, i.e., that H implied p to be true, or some other type of contextual evidence against ¬p (contradiction scenarios). The other possibility, compatible with S being certain about p not being (yet) CG, would be that no one has yet taken an explicit stance towards the p/¬p issue. One reason for posing the question could be to double-check on their prior bias that ¬p is still the case, as in the paranoid scenario in (36). Another reason, as in the polite requests, S thinks that p is possible -but for politeness they want to be explicit about the fact that there is no (public) reason to expect p, giving H an easy path towards saying denying their request.

Back to 'mica' in assertions
Our proposal for mica in declaratives is that, just as in PQs, it expresses a form of meta-conversational negation, which we have treated as a subjective FALSUM operator. Since assertions are speaker-oriented already, the special perspectival properties of mica (vs. regular FALSUM) neutralize, and the operator remains speaker-oriented. This effectively turns previous accounts on their head, aside from Squartini (2017): Cinque (1976) (see also Zanuttini 1997, Penello & Pescarini 2008, Pescarini 2009 suggests that mica is a presupposition trigger, a sentence of the form (non) mica p asserts that ¬p and presupposes that p was expected. On our proposal, the inference that p was expected is derived via a competition effect, whereas the presupposition (just as in PQs) is that the prejacent p is not taken to be CG, from the speaker's perspective. We therefore capture the insight of the previous accounts of mica in assertions, while providing a proposal that can account for the wider range of data and unify the use of mica in assertions and PQs. Our proposed LF for a mica assertion as in (82) is illustrated in (83) Instead, the contrast above follows neatly from our analysis. Only in (84b), S is justified in taking for granted that it is not CG that S's building has an escalator. The opposite holds for (84a); here S would not be justified in assuming that the proposition that their building has an elevator should not be in the CG, given that it would be a very common thing to assume. Finally, the account of the scopal facts in section 2.3 follows from the nature of CG-managing operators, which scope high on the left periphery, like discourse particles (Zimmermann 2011); something that wouldn't be explained if mica were a regular negation that is presuppositional. 32

Conclusions
In this paper we have developed a new account of Italian polar questions as introducing presuppositional meta-conversational operators that are perspectival. Based on the presuppositional treatment of these operators, we have presented a new, and we hope, cleaner, version of the logic by which the biased 'epistemic inferences' of  can be inferred from both questions and declaratives with meta-conversational content. The star of the paper is the particle mica, which shows up in both assertions and polar questions, with a complex distribution and interaction with negation. Building on work on evidential operators, we proposed that meta-conversational operators undergo interrogative flip (something already implied by , but not under that name; our contribution here is 32 The analysis would also explain why mica leads to binding blocking effects. As shown below when mica intervenes between a quantifier and a bound variable, the sentence is marked. Related to that is the fact that mica is banned from restrictive relative clauses, as well as from constituent questions Cinque (1976). Following Kratzer (1999) and Zimmermann (2008) on discourse particles, we suggest that these three effects can be given a unified account under the assumption that mica selects a closed proposition (i.e., a proposition with no variables bound from above).
(i) Ogni candidato che ha partecipato al concorso non lo ha (*mica) superato. 'Every candidate x i : x i took the exam, NEG x i (*MICA) passed it.' to link this idea to the larger literature on interrogative flip), but that mica does not undergo interrogative flip -it is inherently speaker-oriented. Italian, though lacking a grammatical evidential system, therefore is an example of a language that mixes both standard interrogative flip behavior (Garrett 2001, etc.) with non-flip behavior that has been less commonly described (Bhadra 2017). This account of mica as a perspectivally stable metaconversational FALSUM operator fully explains the complex distribution of the particle, as well as many related observations, including scopal facts and polarity licensing. This project ties into the larger open question of why interrogative-flip-like or 'origo shift' effects can happen or not happen; previous work has approached this on a language-by-language basis (Bhadra 2017) or simply taken it to always occur (Korotkova 2016b). Italian contributes to this question by suggesting that a more fine-grained parameter is necessary, i.e. a mechanism for variation between particular constructions or items. We have suggested two possibilities that could account for the data in Italian: variation among question-like clause types (following Bhadra 2017) or variation in the scope of particles (following, roughly, syntactic work on the left periphery).
We end with a prediction: there might be many languages that show mixed flip behavior, when the right semantics-pragmatics interface elements (e.g. particles) are examined, and the scope of interrogative flip extends beyond strict definitions of evidentials -and understanding the conditions for anchoring behavior at a finegrained level is crucial to an account of force at the semantics-pragmatics interface.