Non-informative assertions: The case of non-optional wh-in-situ*

This paper focuses on inquisitive information-seeking utterances with non-fronted wh-words and declarative word order, which I dub wh-declaratives (WhDec), e.g., the party is where? Though they have not received much attention in the literature, they present an interesting theoretical puzzle: despite looking like declaratives syntactically, they receive an inquisitive interpretation, and yet they are not always interchangeable with (canonical) wh-interrogatives (WhQs). In this paper I use WhDecs as a window to explore how, by taking into account the interaction between semantics, discourse structure and dynamic updates, we can derive subtle interpretational differences while keeping the overall interpretation true to form. The paper also addresses the interaction of sentential force and prosody, extending the discussion of WhDecs to rising declaratives and examining the contribution of the nuclear contour in the construction of meaning. Along the way, I build on previous literature to provide a semantics for WhQs and WhDecs that bridges to pragmatics explaining the contextual dependence of the so-called existential presupposition in


Ways of requesting information: The taxonomy of inquisitive utterances
This paper focuses on inquisitive information-seeking utterances with non-fronted wh-words and declarative word order. I dub these wh-declaratives (WhDec). In this section I provide an overview of the data and situate WhDecs with respect to other inquisitive structures.

early access
Interrogatives in English and in other languages differ from declaratives and other clause-types with respect to several structural and prosodic features. The contrast between canonical utterances of declaratives, polar interrogatives and whinterrogatives is illustrated in (1)-(3) where '↓' is used as a shortcut for a final falling contour (FF) corresponding stereotypically to H* L-L% in the ToBI annotation system, and '↑' is used to signal a final rising contour (FR), stereotypically L* H-H%). 1,2 (1) Canonical Declaratives a. The boyfriend is at home↓. b. You know that via John↓.
(2) Canonical Polar Interrogative (PolQs) a. Is the boyfriend at home↑? b. Do you know that via John↑?
Canonical examples of wh-interrogatives (WhQs) involve a wh-word at the front and, compared to declaratives, reverse the relative order of subject and auxiliary. This latter property is also shared by polar interrogatives (PolQs). WhQs and PolQs present do-support when there is no form of the verb to be. Like declaratives, WhQs canonically exhibit an FF although, as in the case of declaratives, there are other possibilities (see Bartels 1999 for an overview and Hedberg et al. 2010 for corpus work on WhQs). 3 Polar interrogatives exhibit stereotypically an FR.
1 ToBI is a system of prosodic annotation that assumes that the underlying phonological representation of intonation consists of sequences of phonemic tones. The absolute pitch level, pitch range, the relative excursion size or properties of the pitch movement connecting tonal targets are not taken to be part of the underlying representation although they may have a paralinguistic function (such as the expression of emotions). In the ToBI system intonation is represented by means of L(ow) and H(igh) tonal events. Amongst them we can identify in English pitch accents (H*, L* as well as their combinations with other tones), phrase Tones (H-, L-) and boundary tones (H%, L%). Following the tradition of the semantics/pragmatics literature on the topic at hand, this paper uses this system in the annotation of the tunes and no further distinctions are made. 2 I take the final contour to include the final pitch accent and phrase and boundary tones (what corresponds to the nuclear tune in the British School). 3 While H* L-L% is taken to be the prototypical final contour for declaratives, other contours are possible too. Pierrehumbert & Hirschberg (1990) take H* L-L% and H* L-H% both to correspond to "neutral declarative intonation". See Bartels (1999) for discussion and other combinations such as L* L-H%. It is quite well known that a final rise is also possible in so called 'uptalk' dialects. This is currently the object of much heated debate and I leave it aside here. As for wh-interrogatives, Hedberg et al. (2010) report the analysis of 200 wh-fronted-questions collected from the CallHome Corpus of 2 early access There are other constructions that mix features of declaratives and interrogatives. The most studied ones are so called rising declaratives (henceforth RDecs), utterances exhibiting declarative word order but a final rising contour (FR) (canonically L* H-H%, but other realizations are possible). RDecs can be interpreted as "requesting information" (see Jeong 2018 for a more recent and refined analysis of declaratives with a final rise). Bartels 1999 (pg. 228) provides the schema below representing the prosodic differences between utterances with canonical declarative (surface) word-order that are used to request information and those that are used to provide it: According to Bartels (1999) (a) and (b) trigger a "question" interpretation whereas (c) is borderline. Finally, by uttering (d) the speaker claims that it is not the case that Aunt Mary reads the newspaper. The meaning of RDecs like (a) and (b) above, while inquisitive, is not the same as that of PolQs. Gunlogson (2003Gunlogson ( , 2008, e.g., argues that RDecs involve the speaker's tentativeness to commit to the content proposition (that Aunt Mary doesn't read the newspaper) but signal the need of confirmation that this is true. Less studied are English utterances with declarative word order and a non-fronted wh-word that are not echo questions (see Bartels 1999 and references therein). These are the focus of this paper, WhDecs:   Bolinger (1957: p.142) dubs these hybrids "reminder questions", utterances that somehow evoke a previous question. Examples like (5), (7) and (8), however, show that these are more than just questions that repeat a question uttered in the discourse long before. Bartels (1999) points out that with these questions sometimes "the speaker merely pretends to rely on a discourse link while knowing that no such link exists." This relaxes Bolinger's characterization by not requiring a specific previous utterance of such question but, e.g., to accommodate that such question is somehow 'present' in discourse. In this vein, Bartels 1999 (pg. 175) points out that "we find rising questions that do not seem repetitions in any sense, immediate or not, but are merely endowed with an ostensible link to prior discourse or the situation at large." 4 In this paper I aim to explain and model these intuitions, as well as other differences between WhDecs and WhQs.
The examples below further illustrate the use of WhDecs. They are taken from an episode in a North American TV-series, Major Crimes (Season 5, Episode 1, "Present tense"). This allows us to retrieve the utterances' intonation as well as the full details of the context of utterance, allowing us to keep track of what came before, what followed, and their discourse effect. Ltn. Provenza below is the highest ranking officer in the conversation and examples sound very natural to native speakers. 4 Bartels (1999) (pg. 192) notices that it is also possible to have final falling contour H* L-L% with non-fronted wh-questions. Bartels states that this is found when "calling for specification of the agent in the original statement" and are equivalent to narrow focus WhQs that she labels "reference questions". (Bartels 1999: ex. (57)): Notice however that WhDecs with such a contour are not limited to the types of contexts described by Bartels (we return to this in §3.2). The following example is from a TV series, The West Wing, S07 E13, "The Cold", portraying the same contour in a different context:

early access
The examples above illustrate that WhDecs do not need to be preceded by a previous utterance stating the equivalent WhQ in order for the WhDec to be felicitous. There is no such previous utterance in any of the examples (9)-(11) and discourse participants are aware that such utterances have not taken place. What is common to all the examples above is that WhDecs can be taken to request information that is very much related to an ongoing conversation, making WhDecs very much discourse-linked (see also Pires & Taylor 2007 for similar observations). 5 In addition, speakers report a difference between WhDecs and WhQs in the contexts above: while it is true that every time a WhDec is used a WhQ would also be felicitous (but not the other way around, see below), the use of WhDecs helps to move the discourse forward with a sense of urgency lacking when WhQs are used instead (unless extra-prosodic support is used).
There are further differences between WhDecs and WhQs. One is that WhDecs are not possible out of the blue (see also Ginzburg & Sag 2001 While WhDecs are not possible out of the blue, a preceding linguistic utterance is not strictly necessary to license a WhDec. Extra-linguistic context can sometimes offer the necessary support and WhDecs are felicitous as long as we can reasonably assume that the addressee is willing to take up the question: 6 (13) B is helping to tidy up after dinner at her friend's house and enters the kitchen carrying the dishes. B: These go where?
In addition, WhDecs are different from WhQs in that the speaker seems to assume that the addressee either knows the answer or knows whether the answer is known. (14) [Assuming the same preceding context as in in (9)] a. And the boyfriend is where? #Have you checked? b. And where is the boyfriend? Have you checked? 5 Notice that WhDecs can always be preceded by and or but. These are in such cases discourse markers (see Asher & Lascarides 2003 a.o.), indicating that the question is integrated in a larger discourse structure. WhQs can also be preceded by these markers when they are used in similar circumstances, but not in other contexts, showing that the distribution of WhQs is broader than that of WhDecs. 6 "Willing to take up the question" involves factors regarding the power dynamics, i.e., the speaker has to have authority over the speaker or else be in an "familiarity" situation and, as we will see below, be certain that the addressee knows the information.

early access
When it is clear that the addressee is very likely ignorant about the whereabouts of the boyfriend, the WhDec results in infelicity, (14a). 7 Here is another example illustrating the same point: (15) I want to repair the sink but I can't find your mother's toolbox. I don't seriously expect you to know this, but just in case, a. #. . . your mother's toolbox is where? (Do you know?) b. . . . where is your mother's toolbox? (Do you know?) This sets WhDecs apart from WhQs. In Searle's terms, the assumption that the addressee "may" know the answer is part of the preparatory condition of asking a question. With WhDecs, the speaker seems to make the stronger assumption that the addressee is in a position to solve it or to establish whether it can be solved. The WhDec has a metalinguistic flavor: it indicates information that Det. Sykes should have already inquired about. I set aside metalinguistic issues here. 8 Notice that this doesn't discard the possibility of the answer not being known. In (14) Ltn. Provenza can very well entertain that the whereabouts of the boyfriend are not known but, at any rate, he expects his detectives to have inquired about it. Knowing whether the answer is known is also a step forward to dealing with the issue. At least, it leads to agreeing that the issue cannot be solved at that point. This is not the same as the addressee not knowing anything at all: knowing that the answer is not known yet helps in deciding what steps to follow to address the question next.

early access
An account of WhDecs has to explain why (i) WhDecs have a more restricted distribution than WhQs, (ii) the different overall effect that we have observed between the two and, (iii) restrictions on the sequence of questions.
Here is an intuitive outline of the proposal to be developed in this paper. The key to the analysis proposed below is the idea that WhDecs are declaratives and, as such, they propose to update the context in the same way declaratives do: they are proposals to update the common ground (CG) and, hence, the associated context set (cs). In this sense, they are assertions (see Stalnaker 1978, 1984, Farkas & Bruce 2010, Malamud & Stephenson 2015 for this definition of assertion as the (grammatical) dynamic update triggered by the utterance of a declarative, which one could also call 'declarative (sentence) force'). 9 WhDecs differ from canonical declaratives in that the proposition they denote trivially updates the cs (it only establishes that cs is compatible with different possibilities). 10,11 The question is then how to derive the 'inquisitive flavor' of WhDecs. 12 I argue that we do not need any ad-hoc machinery to do this: the overall interpretation is derived from the dynamics of context update and well-known mechanisms of discourse anaphora triggered by focus (à la Rooth). The proposal, in a nutshell, is that via simple focus structure the WhDec the boyfriend is [where] F ? presupposes that there is a question open in discourse (that participants have to answer) of the form where is the boyfriend? (i.e., the immediate QUD, IQUD, in Roberts' 1996 terms). This is the same question presupposed via focus anaphora by The boyfriend is in [Joshua Tree] F . However, while the plain declarative answers the presupposed question, and hence it is dispelled, the WhDec doesn't provide an answer, and the question is left to be addressed by subsequent discourse moves. Hence, cooperative responses to WhDecs 9 This concept of 'assertion' (declarative force) refers to the dynamic update triggered by the utterance of a declarative and is grammatically encoded. It is not to be confused with how the term 'assertion' is used in the broader speech act literature. In fact, it doesn't provide by itself an account for how indirect speech acts performed by utterances of declaratives are derived. The same applies to our use of 'question' below, which is tied to the (grammatically encoded) dynamic update triggered by the utterance of interrogatives (which we could also term 'interrogative force'). Nevertheless, it is expected that an explanation of how indirect speech acts are derived will take into consideration the dynamic update triggered by the utterances. 10 See ftn. 31 for differences between WhDecs and the utterance of a declarative with an indefinite. 11 This analysis of WhDecs has parallelisms with Farkas & Roelofsen's (2017) claims regarding nonwh-interrogatives and their contrast with RDecs. Working within inquisitive semantics, in Farkas & Roelofsen's (2017) model declarative sentences are claimed to have a (typically) trivial inquisitive content while, conversely, interrogatives have a (typically) trivial informative content. While the analysis in this paper could potentially be framed within the Inquisitive Semantics framework, the proposal will instead be developed in a Hamblin framework, providing us with an easy link to Rooth's focus semantics and to work with discourse presuppositions. 12 While preparing the final version of this paper it has been brought to my attention that Bobaljik & Wurmbrand (2015) argue that WhDecs are syntactically declaratives.

early access
are also cooperative responses to its presupposed WhQ counterpart. In contrast to WhDecs, utterances of WhQs are proposals to pursue a new question, either as part of a strategy to answer a more 'general' question or a new question altogether (see Biezma & Rawlins 2017a,b). In this proposal, discourse differences between WhDecs and canonical questions stem from their different semantics and their different dynamic update: WhQs propose to pursue the inquiry they denote, while WhDecs presuppose that participants have already accepted to pursue such inquiry (the WhDec acts in a sense as a "reminder"). 13 If the analysis offered here is right, it is superior to potential alternative analyses requiring ad-hoc syntax-semantics for WhDecs, ad-hoc semantics for the wh-words in WhDecs or ad-hoc dynamic updates. The paper is organized as follows. In §2 I introduce the formal assumptions. In §2.1 I present the dynamic framework and in §2.2 the semantics of wh-interrogatives (I show how the system works in the case of WhQs). In §2.3 I address WhDecs and show how the behavior of WhDecs, and their differences with canonical WhQs, can be explained once we take into account all the elements involved in the construction of meaning of natural language utterances.
The case of WhDecs is certainly related to the other hybrid, rising declaratives (RDecs) (as we have seen, inquisitive utterances that differ from canonical declarative assertions in the final rise). In §3 I sketch how the proposal made for WhDecs can be extended to RDecs and tackle the contribution of the final contour in the construction of meaning. In §4 I compare the proposal made for WhDecs with previous analyses in the literature.
Before going ahead, let us make two disclaimers. First, let us point out that the analysis proposed in this paper aims to explain the North American English data, where WhDecs and WhQs differ in meaning (fronting is not optional). I have nothing to say about languages in which fronting is optional. Notice, however, that similar phenomena to English WhDecs can be observed in other languages (see e.g., Biezma 2018 for Spanish and Pires & Taylor 2007 for claims about similarities between English and Brazilian Portuguese). Given that WhDecs can be observed in other languages, one should be able to derive their properties in a principled manner, and not by appealing to a mere lexicalization of the construction. The key to the analysis in this paper is that English WhDecs are declaratives despite containing a wh-word and, hence, they update dynamically as such. The challenge then is how to early access explain their 'inquisitivity'. I compare the analysis offered here to other analyses in the literature in §4. 14 Second, in this paper I do not address echo questions (EchQs). The term "echoquestion" refers to utterances inquiring about what has been just said. A stereotypical echo-question is one in which the wh-word is also in situ and much of the literature proposes a special syntax to derive this word order (see e.g., Sudo 2007, Beck & Reis 2018. However, EchQs come in various shapes, and can also be WhQs or RDecs, (17). Prosodically, EchQs are characterized by a complex accent (L+H*, expressed by SMALL CAPS below) on the contingent element (while the rest is commonly deaccented) and final rising intonation (see e.g., Bartels 1999, Artstein 2002 An analysis of the puzzles presented by EchQs requires far more attention than we can devote to it in this paper (see Ginzburg & Sag 2001, Iwata 2003, Sudo 2007, Beck & Reis 2018 for discussion). I leave for the future an attempt to extend the proposed analysis for WhDecs to stereotypical EchQs. If this were possible, we would end up with a picture in which the term echo questions is an umbrella term for a variety of constructions that are also found in inquisitive non-echo environments (i.e., there are WhDec-EchQs, polar-EchQs, RDec-EchQs and WhQ-EchQs). The research question would be then how to formally derive the "echo"-effect.

Deriving the differences between WhDecs and canonical interrogatives
To model the differences between WhDecs and WhQs we need to consider the dynamic update of utterances. I follow most directly Farkas & Bruce (2010), Malamud & Stephenson (2015), Starr (2016) and Biezma & Rawlins (2017b), who build on Stalnaker's (1978) suggestion that utterances are proposals to update the context. A declarative, for instance, is a proposal to update the (Stalnakerian) CG 14 In fact, Pires & Taylor (2007), one of the few instances in which WhDecs have been addressed (see also e.g., Ginzburg & Sag 2001: chp. 7), take them to be interrogatives (and therefore questions) because, in their words, "wh-in-situ questions (including echo-Qs) do request information, thus a non-interrogative analysis (even for echo-questions) is inaccurate." In their analysis, Pires & Taylor (2007)

early access
and its associated cs with the propositional content of the utterance (see Farkas & Bruce 2010, Malamud & Stephenson 2015. 15 The addressee has then the option of accepting such proposal, rejecting the proposal or resisting the proposal until some considerations are pondered (see Bledin & Rawlins 2016 for resistance moves).
In the case of declaratives, the addressee's acceptance of the proposal triggers the actual addition of the proposition to the CG (and the shrinking of its associated cs). Acceptance is the default and hence, in absence of a reaction to the contrary, the propositional content makes its way to the CG. In the same way, imperatives are proposals to add a preference (the content proposition of the imperative) and its acceptance amounts to updating the ranking of alternatives according to such preference (see Starr 2016). Finally, the utterance of an interrogative is a proposal to update the context's questions stack (the set of unresolved questions in discourse that participants agree on answering, i.e., the Q(uestion)U(nder)D(iscussion) stack in Roberts's 1996), and its acceptance makes the question denoted by the interrogative the current/immediate question to be addressed, i.e., the IQUD (see Biezma & Rawlins 2017b). In what follows I borrow the formal details mostly from Biezma & Rawlins (2017b) (henceforth B&R). In §2.1 I introduce the formal apparatus in B&R with small adaptations for Hamblin-semantics. In §2.2 we see how the system works with canonical wh-questions. In §2.3 I lay out the results of the system for the case of WhDecs characterized as declaratives.

Dynamic assumptions
In order to capture the 'proposal' component of utterances, B&R distinguish what they call the 'local context' from the proposed update within a context. The local context of a given context c (l c ) is composed by the actual context set of c (cs c ) and the QUD stack of c (Q c , implemented using a simple alternative-semantics representation) [In the remainder of the text I use subscripts as needed, to avoid confusion.] The proposed update is the projected (future) context (F ). Utterances are proposals to update the context and are first 'recorded' in F , where they await evaluation (i.e., to be accepted or rejected). F is a 'copy' of the local context with the modification proposed. Once evaluated, F is emptied (F = / 0) and a new proposal can be made.

(18)
A local context l is a tuple cs, Q such that: a. cs is a context set. b. Q is a stack of sets of propositions.
early access (19) A context c is a tuple cs, Q, F where its elements are characterized as: The empty stack is notated by . The system assumes the standard push, pop and top operations on stacks (see Kaufmann 2000, Isaacs & Rawlins 2008. 17 In this model, the immediate QUD in the local context l c will always be at the top of the stack, i.e., given the stack of questions in a context c, Q c , the current QUD is top(Q c ). The updates triggered by declarative and interrogative utterances are defined in B&R as follows (' s,t ' is the semantic type corresponding to propositions): The purpose of conversation is taken to be a communal inquiry and, hence, it is organized around questions that participants commit to addressing. These questions are stored in Q. The structural-constraints governing Q (or QUD stack) are mainly relevance (of questions and assertions), and entailment. (Definitions below are adapted from the notion of relevance in Roberts 1996.) (21) Answerhood licensing: an assertion a is relevant to QUD c only if it entails, either positively or negatively, the resolution of at least one alternative in top(Q l ).

(22)
Question Entailment: A question q 1 entails another q 2 iff every proposition that completely answers q 1 completely answers q 2 as well.
, q is part of a strategy to answer top(Q l ); 16 Biezma & Rawlins (2017b) simplify by considering one possible future context or none (notated by / 0) accepting that in the general case, one may want to allow several future contexts. I adopt the same simplification in this paper. 17 Given a stack s, push(s, x) delivers the stack resulting from adding x to the top of s. Conversely, pop(s) delivers a stack in which the top element of s has been removed. Finally, top(s) just establishes what the top element on the stack is. 18 This will be modified in (46) to address some technical problems arising below. early access informally, a strategy is a sequence of subquestions that together answer a given question).
In an idealization of discourse, relevant declaratives are (partial or complete) answers to a (implicit) question, (21). 19 Implicit questions can be retrieved by conventional linguistic cues such as focus structure. Questions are licensed if they start a new discourse (when Q = ) or are entailed by top(Q l ). Lastly, we need a notion of accommodation. In B&R, presuppositions carried by utterances must be accommodated before the move is accepted. 20 (24) Accommodation. If ϕ presupposes ψ and ψ is not satisfied in l c , first update l c so that ψ is satisfied.
As said, full contexts are formed by the current context (l c = cs c , Q c ) and proposals to update it: the projected context F . The structure of F is the same as the structure of l c but also records what proposal is awaiting evaluation. The absence of a projected context is notated by / 0 (i.e., nothing is awaiting evaluation). Given the machinery above, B&R use the following basic definitions of context update making use of the operations defined in (20) for local contexts: The constraints in (25) ensure that (a) the projected context of the original c is empty and (b) that the projected context of the update is defined. The same applies to (26).
The definitions above capture the 'proposal' nature of uttering declaratives and questions: they modify the future context slot by proposing either to add the propositional content to the common ground, in the case of declaratives, or the question to the question stack in the case of interrogatives. Acceptance replaces the original context early access by the future context and leaves the proposal slot (the future context slot) empty. Utterances are felicitous if the update obeys the felicity constraints in (20). B&R's model also includes two 'maintenance' operations: (i) rejection of a move, i.e., rejecting a future context, and (ii) elimination of a QUD once it has been solved, which requires popping that QUD from the stack. 21 In the two sections below I turn to WhQs and WhDecs. In each case, I first present the details for their denotations and then examine their context update effects in terms of B&R's model as discussed above.

Canonical questions (WhQs)
In this section I provide the semantics and dynamic update of WhQs. The semantics offered builds on the Hamblin tradition and extends Biezma & Rawlins's (2012) proposal for non-wh-questions, bridging the semantics and the pragmatics to address the contextual dependence of the so called 'existential presupposition' in questions. Dynamically, WhQs are merely proposals to update the questions stack. This is achieved if the proposal for the update is accepted: the proposed question becomes the IQUD. With the semantics of WhQs in hand, it will be possible later to see how WhDecs differ from cannonical questions.
The semantics of WhQs Let us consider first canonical WhQs. Following current approaches to questions within the Hamblin tradition (Hamblin 1958(Hamblin , 1973, I assume that questions denote sets of propositions, the set of its possible answers. Alternative sets compose via pointwise function-argument application, 22 which in the case of singleton sets (e.g., the denotation of a declarative) behaves like standard function-argument application. I follow Kratzer & Shimoyama (2002) 21 B&R assume that when QUDs are not inquisitive / have been answered, they pop automatically (this mechanism to clear non-inquisitve QUDs follows most directly Groenendijk 1999). 22 From Kratzer & Shimoyama (2002): Pointwise Function-Argument Application If α M is a branching node with daughters β and γ, and β g,c ⊆ D σ and γ g,c ⊆ D σ ,τ , early access in disassociating the lexical operators introducing alternatives from the operators manipulating them: alternatives introduced lexically in the denotation (e.g., by wh-words) are collected by sentence-level operators such as '∃' (which provides existential force and returns a singleton set when several alternatives are introduced in declaratives, as in the case of disjunction; see e.g., Alonso-Ovalle 2006) (2002); (32) illustrates how the system works for the interrogative where is the boyfriend?: In (32) the set of alternatives manipulated by Q are the ones generated by the whword. The resulting semantics predicts that there is a place where the boyfriend is. However, the alternatives entertained by speakers when uttering WhQ like (32) may also involve the "null-alternative". Abusch (2009) illustrates it with (33): (33) I've alienated my colleagues completely. Who will vote for me? Probably nobody.
Abusch uses (33) to show that the so called existential presupposition in WhQs is a 'soft' presupposition: the context of utterance can cancel it. 23 The question is how to formally derive this observation. One way to do this is to bridge the semantics and the pragmatics so the semantics considers both the alternatives introduced lexically and those that are live contextually. This is in fact what the semantics of questions in Biezma & Rawlins (2012, 2017a achieves for non-wh-interrogatives. 24 The authors suggest that the proposal could be extended to wh-interrogatives. I show below how this could be done. The strategy to derive the difference between examples of questions with and without existential presupposition will be to allow the set of contextually salient alternatives to extend beyond lexically determined (semantic) alternatives. To formalize this we appeal to the notion of discourse 'move'. I assume that an explicit discourse move is a communication event tightly linked to a context of utterance and to a linguistic form. In what follows, given some move M, the context of utterance for that move at that point in discourse is notated c M .
ALT c M is the set of propositional alternatives that are salient in the context of interpretation c for move M. 25 (v1 preliminary, see (40)) The 'Q' operator in (36) does not affect the alternatives introduced semantically but acknowledges that there may be other alternatives in the context of utterance (clause (a); ALT= / 0 corresponds to the discourse initial situation). The difference between (30b) and (36) crucially is in definedness conditions for (36) that allow for non-semantic alternatives to be taken into account by appealing to alternatives available in the context of the discourse move. In Biezma & Rawlins (2012, 2017a, interrogatives present alternatives and request that the addressee choose between those and others present in the context. The definedness condition in (b) simply requires that there be more than one (live) alternative (i.e., questions are not superfluous). 26 Let us consider now the semantics of wh-words. Given that we will be 24 Biezma & Rawlins (2012) are primarily focused on how to derive differences between polar questions and alternative questions. The overall idea is that a polar question (Is the boyfriend in Joshua tree?) puts forward an alternative (that the boyfriend is in Joshua Tree) and requests that the speaker chooses between that alternative or others that are also contextually salient (see (36)). For alternative questions, (final falling) intonation signals that the spelled out alternatives are the only live alternatives. 25 Ultimately, this will be the possible answers to the QUD dominating M in Roberts's system, i.e., the IQUD. See discussion below leading to this result in (40). 26 See Biezma & Rawlins (2017a) for a milder view on (b) that is not important for our purposes. 16 early access considering below both the Roothian focus meaning ( · f ) and ordinary meaning ( · o ) of an utterance, I proceed to introduce them here for wh-words: 27 (37) a.
wh-f = D τ (where τ stands for the type of the given wh-word) b. wh In this approach, the alternatives introduced semantically by the wh-word do not fully determine the live alternatives in the context of utterance (ALT), which leaves room for the null-alternative to be entertained as a live alternative (see also Roberts 1996). That is, the utterance of a wh-interrogative puts up for consideration all the focus alternatives that are live and contextually salient alternatives, but it doesn't require that these be the only live alternatives. Crucially, the context of utterance may allow for the null-alternative to be part of ALT, e.g., "the boyfriend is not in a contextually salient location". In this latter case, we do not obtain the so called "existential presupposition". However, if the only live alternatives are the semantic alternatives, the "existential presupposition" arises simply as an entailment of the set of live alternatives under consideration (see Stalnaker 2014).
While this semantics derives the role of context in allowing for the null-alternative in a very smooth fashion, we still need to control what can be in ALT. The goal is (i) to leave space for the null-alternative (justifying this as being organic to questioning), and (ii) to constrain what alternatives are let in. Focus plays a crucial role in identifying the live alternatives.
Within the Roothian tradition, focus mediates the relation between utterances and discourse and plays a crucial role in identifying the alternatives that are entertained (a subset of the focus alternatives are live-alternatives). The focused element in 27 In Rooth's framework, a sentence is associated with both an ordinary semantic value ( · o ) and a focus semantic value · f . The focus semantic value of an utterance is the set of propositions obtained by substituting the focused element with expressions of the same type. (The English translation below is a shorthand for propositions.) Recall also that in Hamblin semantics, declaratives denote singleton sets: early access an utterance is the element evoking alternatives relevant for the interpretation. In the case of WhQs, this is the wh-word. Regarding the ordinary semantic value, according to (36) and (37), the semantic alternatives are already all the live salient alternatives that are focus alternatives. Is there any other contextual alternative? Notice that in this approach to questions, to utter an interrogative is to offer the addressee some alternatives to evaluate. One of the possibilities is that none of the alternatives offered is evaluated as being viable, and this is what the null-alternative is. This is captured in (39) and (40). (39) is an auxiliary definition that allows us to isolate the focus alternatives that are live alternatives and (40) establishes that ALT may be only that set (40a), or also contain the option that none of the alternatives in that set is viable (40b): In what follows I refer to it as top(Q c ) for short: the set of live alternatives in discourse, which is constrained but the focus value (it contains a subset of the focus alternatives) but also allows for the null-alternative. This proposal predicts that the null-alternative is not always available and this is borne out: If one is born, one is born somewhere (excluding metaphoric uses) and the nullalternative is not a live one, (41). However, one can entertain that someone is staying home during the non-teaching time and the null-alternative is not discarded, (42). Let us take stock. The semantics for WhQs offered above is very close to other proposals for wh-interrogatives found in the literature in the Hamblin tradition. The main difference is that, following Biezma & Rawlins (2012), the proposal above allows the semantics to bridge with the pragmatics (and discourse) by including in 18 early access the denotation an appeal to the set of alternatives in contention, top(Q c ) (/IQUD(M)), which is crucially shaped by focus. This move allows us to account for the contextual dependence of the existential presupposition: top(Q c ) may contain not only focus alternatives (which crucially shape it), but also, depending on context, may contain the alternative that none of the live focus alternatives is true. This appeal to contextual alternatives will play a role again in §2.3 in explaining the meaning of WhDecs, for which the existential presupposition is also contextually dependent. 28 The dynamic update of WhQs The dynamic update in a context c, following the definitions in (19), (20b), (26) and (27) is as follows: Uttering an interrogative amounts to proposing that the question it denotes be added on the top of Q l (i.e., it's a proposal that participants commit to addressing that question in the current discourse, making it the IQUD in Roberts's terms). By accepting the proposal, the future context F becomes the local context and the future context slot is emptied. Participants are thus committed to pursuing this inquiry next in discourse (see acceptance in (27) above). Hence, subsequent moves are either answers to the question (see answerhood licensing in (21)), or further questions relevant to answering such question (see question licensing in (23)). However, the addressee can also reject the proposal (i.e., reject the future context and hence empty the future context slot) and leave the context as it was before the proposal was made (see (28)).

WhDecs
The semantics of WhDecs builds on the ingredients already laid out above. For the dynamic update, WhDecs proceed as any other declarative ( §2.3.1). Declaratives

early access
address an IQUD that (if not explicit) is recoverable from the utterance focus structure, top(Q c ). At the end of the update with the declarative, that question is dispelled because it has been solved by the proposal made, and we only observe a change in cs c . WhDecs proceed in the same way ( §2.3.2). However, the question identified by focus structure that they presuppose is not solved, because their information update is trivial. The question presupposed is left there to be addressed by participants in subsequent moves. At the end of the update, WhDecs can only modify the questions stack and the final update is equivalent to that of wh-s, although we arrived to it through a different route, explaining the differences ( §2.3.3).
The semantics of WhDecs and first steps to their dynamic update We know that WhDecs and WhQs can receive the same responses which, ultimately, clear up an issue and let the discourse move forward. The fact that responses to WhDecs seem to be roughly similar to its parallel WhQ (but see rejection moves in §2.3.3) might lead to the impression that WhDecs are also semantically questions. However, this fails to capture the observed contrasts. What I propose instead is that WhDecs are declaratives (as indicated by their word order) and, hence, denote singleton sets in Hamblin semantics. Dynamically, they also propose to update cs. Given that alternatives are introduced semantically by the wh-word, in Hamblin semantics alternatives in the declarative are collected by '∃' (just as in the case of, e.g., disjunction). In the spirit of the semantics of WhQs laid out above, I assume the following semantics for WhDecs, adjusting the '∃' operator in (30a) along the lines of 'Q' in (36) above to bridge the semantics with the pragmatics.
The resulting semantics for our running WhDec example is as follows: (45) Let M be the move made by uttering the WhDec the boyfriend is where?: The semantics of WhDecs, hence, simply says that the context set is compatible with different alternative locations for the boyfriend and further states that one of the live alternatives is true. 29 However, speakers interpret that WhDecs request information and, hence, we still need to derive their inquisitivity. I argue that inquisitivity naturally results from the utterance's dynamic update.
To account for the dynamic update of WhDecs it is important to take into account that WhDecs are declaratives and, hence, they are proposals to update cs. The crucial difference with (regular) declaratives is that WhDecs do not resolve any (open) issue. Let us examine this difference in more detail. In order to explore the declarative update in Hamblin-semantics, where declaratives are singleton sets, we need to adjust the definition of ' ' in (20a), where the update proceeded by merely intersecting cs with the denotation of the declarative (a set of possible worlds). A quick fix to (20a) that delivers the right results is provided in (46), where I merely extract the set of possible worlds within the set denoted by the (declarative) WhDec and then conjoin it with the cs in the update: So, assertability for a WhDec will require compatibility between cs and its "content" proposition as well as ensuring that the proposition is relevant to top(Q l ) (we will revisit the notion of relevance in (56) below in light of what we will learn from the study of WhDecs).
As we saw in §1, WhDecs are linked to discourse and I argue that this is the result of simple focus-presupposition (à la Rooth). While WhDecs do not provide an answer, they do require, as any other declarative, to be embedded in a particular discourse to be felicitous and, just as with any other declarative, this link is mediated by focus structure: the focus structure of an utterance indicates anaphorically what is the question in discourse that the utterance is addressing. This relation between questions and answers in discourse is cashed out via a constraint on question-answer congruence. Informally, the Roothian congruence constraint can be summarized as in (47) . . . and presupposes that the context contain an antecedent C such that: ( In the Roothian system, the work of the '∼' operator is mostly specified in (c): the utterance presupposes that (i) there is a discourse antecedent formed by a subset of the focus value, (ii) that this antecedent is a set containing more than one proposition (i.e., it is a question) and, (iii) that the proffered proposition is one of the possible answers to that question. The Roothian system allows us to determine congruence by providing the felicitous conditions of an utterance (that there be a particular question open in discourse identified by the focus value of the utterance). Let us see how this applies to WhDecs. Following the discussion above for wh-interrogatives, the focused element in WhDecs is the wh-word itself. This provides the meaning in (49) for our running example, where we find the constraints imposed by the '∃' operator and the presuppositions imposed by '∼': . . . and presupposes that the context contain an antecedent C such that:

early access
To identify C (and, hence, understand the felicity conditions) we need to take into account the focus meaning of the utterance: a set of propositions varying in the location where the boyfriend is. This is the denotation of the natural language interrogative where is the boyfriend? when the null alternative is not considered; C is that question and C ⊆ top(Q c ) (top(Q c ) is crucially shaped by the live focus alternatives of the utterance, C). At the end, the WhDec the boyfriend is where? is only felicitously uttered in a discourse in which there is an open question equivalent to the denotation of where is the boyfriend? Regarding '∼', in the case of WhDecs its constraint that the ordinary meaning be a subset of C (which is itself a subset of the focus meaning of the utterance) is fulfilled by default because of the meaning of whwords (see (37)) and of '∃' (see (44)), which appeals to the alternatives in top(Q c ) (which, in turn, is crucially shaped but not solely determined by the focus meaning of the utterance). In declaratives without semantically introduced alternatives, however, the shape of top(Q c ) is only going to be determined by '∼' and the focus value (see §2.3.1). It is important here to show that the system proposed derives the right results in all cases.
To sum up, given the assertion conditions, the utterance of the WhDec the boyfriend is where? establishes that there is a question in discourse of the form where is the boyfriend?, the wh-interrogative counterpart of the uttered WhDec, and indicates that there are different possibilities. 31 How do WhDecs update? In order to understand the dynamic update of WhDecs it is useful to observe the update 31 I do not address in this paper how utterances of WhDecs are different from utterances of declaratives containing indefinites and in particular so called epistemic indefinites (see Alonso-Ovalle & Menéndez-Benito 2015 for a recent collection of papers on the topic). There are obvious relations between indefinites and question-wh-words as illustrated by the fact that in some languages such as e.g., Japanese and Tlingit they have the same (surface) form (see e.g., Kratzer & Shimoyama 2002, Cable 2010, Kotek 2014 and references therein for discussion). It is beyond the scope of this paper to address this issue but, at any rate, notice that while the semantics of WhDecs and that of Hamblin indefinites used in the analysis of epistemic indefinites is similar, focus structure is different: the wh-word in WhDecs is the focused constituent and determines the question in discourse that the WhDec is addressing, which is crucial in our analysis of WhDecs, while this is not the case with indefinites. Consider the question-answer pair A: Who is dating someone?; B: John is dating someone. Given that the dialogue is felicitous and we understand that B's response is an answer to A's question, we can conclude that the question and the answer are congruent and, hence, John is focused but someone isn't. Furthermore, B's utterance is not a complete answer to the question who is John dating? It may be a felicitous response, more so if one hedges it, I'm not sure..., but he is dating someone, discarding the null answer as a live-alternative, i.e., it can be a partial answer, but not a complete answer. Notice in fact that utterances with indefinites do not include the null-alternative as a live alternative (unlike WhDecs).
(i) John hardly ever eats anything.
(50) o = {λ w. the boyfriend is in Joshua Tree in w} defined only if there is a question in discourse of the form where is the boyfriend?, one of whose possible answers is that the boyfriend is in J. Tree.
To proceed with the dynamic update of the declarative, we need to accept its presuppositions (see (24) above). In the case of declaratives, besides other possible presuppositions such as, e.g., that there is a boyfriend (in (50)), the presupposition we need to accommodate (if it is not explicitly conveyed) is the discourse-presupposition triggered by focus anaphora (see (48c)) that there is a question open in discourse that the utterance is addressing, i.e., the top of the question-stack (top(Q c )/IQUD), (52ai). In the case of (50), this would be equivalent to the denotation of where is the boyfriend? Once this is accommodated, the proposal made by the move takes place, (52aii). The proposal made by the declarative is to update cs c in the (newly accommodated) context (l c ) with the new information regarding the location of the boyfriend. Once the move is accepted (see (27)), top(Q c ) is resolved and hence the question is popped (pop(Q c ) (see (29)), leaving the question stack as it was before the move, (52b) (Q c 2 = Q c ). It seems that indefinites require that alternatives under consideration be only semantic alternatives, resembling in this way the closure effects displayed by intonation in alternative questions or declaratives with disjunction (see e.g., Zimmermann 2000, Biezma & Rawlins 2012. early access b. Acceptance and resolution: Accept the proposed future context (see (27)) and pop(Q c ), since it is now resolved (see (29)): At the end of the process the only thing that we can see that has changed from the original context c is cs c , but in fact Q c also suffered some temporary changes, if only to end up the same it was after the presupposed question was resolved. This is going to be a crucial difference with WhDecs: WhDecs do not resolve the question they presuppose and hence it remains as top(Q c ) awaiting resolution.
Before proceeding to show how the same mechanism allows us to derive that WhDecs have an inquisitive overall meaning in the next section, let us finish with a note regarding the benefits of this model and the predictions it makes regarding the felicity of declarative utterances. This model allows us to explain why out of the blue declaratives are sometimes infelicitous. We already know that assertions are in general infelicitous if the proposed content is not compatible with cs c or is not relevant to a spelled-out question on top(Q c ) (see (25) above). What about out-ofthe-blue assertions (when Q c = )? How do we explain that they are sometimes but not always fine?: Three students are quietly working in their office. The door is open and another student shows up and says: Student: Hey, good morning! My uncle makes cheese in Vermont.
The student's utterance in (53) leaves everybody perplexed. Sure, there may be no problem in accepting that the speaker has an uncle, or that the uncle makes cheese in Vermont (it can all be compatible with cs). Why is it infelicitous? Let's assume that (53) has broad focus and hence that the question presupposed is of the form what is the case?, the least specific question. Still, presupposing the question amounts to presupposing a set of possible answers (the salient contextual alternatives that make up the question meaning). (53) is infelicitous because participants will not accommodate that the proposition that the speaker's uncle makes cheese in Vermont is one of the salient contextual alternatives making up the meaning of top(Q c ) (a question of the form what's the case?), or that it helps identify which amongst the live alternatives is the actual answer. It is, therefore, irrelevant (in the sense of (21)): (53) is infelicitous because participants cannot accommodate the presupposed question that would make the utterance relevant. Notice, in contrast, that the utterance in (54) is not infelicitous: (54) [Same context as in (53)] Student: Hey, good morning! There is free cake in the common-room. 25 early access In this case participants happily accept that there is a question open in discourse for which one of the contextually salient answers is that there is free cake in the common-room and, hence, (54) is relevant and the utterance felicitous. The discussion above shows that infelicity with assertions is not only related to incompatibility between the propositional content and the cs, or relevance regarding a spelled out question. Often times (as in out of the blue utterances) we are forced to accommodate that there is already a question open in discourse and infelicity arises when we are not willing to do that. That we have to accommodate questions and that we sometimes refuse to do so is at the core of our explanation for WhDecs and their discursive behavior. I return to the case of out-of-the-blue utterances below.

WhDec update
As in the case of regular declaratives above, WhDecs are proposals to update cs. As with regular declaratives, before proceeding with the update we need to accommodate the discourse-presupposition triggered by focus anaphora regarding the question in discourse that is addressed by the utterance, i.e., top(Q c ), (55ai) (if it has not been explicitly conveyed before). Given its focus meaning (see (49)) the question in discourse addressed by a WhDec is its wh-interrogative counterpart. Once this is accommodated (possibly in addition to other presuppositions such as, e.g., that there is a unique boyfriend), the proposal made by the move takes place, (55aii). Given that the WhDec doesn't provide the answer to the question in discourse, cs doesn't change, (55b), but what has changed is Q, since the accommodated question was not resolved and, hence, it was not popped.
The dynamic update of WhDecs: Following the definition in (46) for update with declaratives (together with accommodation, (24), and acceptance, (27) Notice that the final context obtained after the update via the WhDec is the same as the one obtained after the update with its wh-interogative counterpart (see (43)). However, we got to the final result through a different route: in WhDecs the context has become inquisitive without the speaker having proposed to pursue a question and thus without having given participants the chance to reject it. With WhDecs the modification of Q is done indirectly, i.e., the speaker has instead demanded that participants assume that the question is already in Q. In §2.3.3 we will see how the path to the final update matters and how discourse differences between WhDecs and WhQs can be derived from differences in this journey. 32 From the dynamic point of view, WhDecs do not change cs, but are devices to signal the content of Q (via (informative) presupposition triggered by focus anaphora). Certainly, for the accommodation to take place, the question that the speaker is taking to be on the top of the question stack has to maintain a well-formed question stack and be relevant (see (23)). The obvious way in which the WhDec can be relevant is by being (/possibly taken to be) (part of) a strategy to answer a question already in Q. I discuss below cases when WhDecs are not felicitous, but let us address first what WhDecs teach us about the notion of relevance.
Up to now, our dynamic system considers that assertions are only licensed as answers to (implicit) questions, see (21) above. WhDecs, however, do not provide any answer, they provide other kinds of information. WhDecs show that we need a broader notion of relevance that takes into account information about discourse structure: (56) Relevance for assertions: An assertion move is relevant in discourse iff a. it is relevant to Q, i.e., if it provides the resolution of at least one alternative in top(Q) either positively or negatively. b. it constrains Q (hence, it 'provides information' about Q).
WhDecs provide information about c by providing information regarding Q. 33 The claim is that providing information about what the structure of the discourse is (e.g., 32 WhDecs are not possible immediately following the utterance of its wh-interrogative counterpart. We can explain this in the current system: if we have just accepted that top(Q) is φ , it serves no purpose to point out that top(Q) is φ , which is the only effect that WhDecs have in our system. 33 A clarification is in order here. WhDecs constrain Q by presupposing what top(Q) is and, hence, provide information about c when this needs to be accommodated. This is, however, not the at issue information (the main point of the utterance, see Potts 2005) provided by the WhDec, whose denotation merely states that one of the answers to the presupposed question holds. The constraint on Q indicated by the WhDec is made via a presupposition.

27
early access what Q looks like) is to be considered on par to other kind of information: providing information about the context is as informative as narrowing down cs. 34 The discussion so far leads to the licensing conditions for WhDecs below (which summarizes what we have derived from independent assumptions regarding the dynamic instructions provided by the form, their semantics, and focus structure).

(57)
WhDec licensing: Let c be a context, Φ a WhDec, Ψ the set of propositions denoted by its WhQ counterpart and l c a local context: I turn now to examining how differences in the update process lead to overall differences in interpretation.

Comparing WhDecs with wh-interrogatives
After accepting the move made by uttering the WhQ where is the boyfriend? or the WhDec the boyfriend is where? the resulting contexts preserve the original cs c in both cases and have the same resulting top(Q c ) (i.e., top(Q c ) = Where is the boyfriend? o , see (52) and (55) above). However, as we have seen, we get there through different routes. In the case of WhDecs we get there indirectly, the speaker presupposes that the question is already there. This presupposition is often an informative presupposition (Stalnaker 1998), i.e., it is triggered in contexts in which it is not satisfied but is satisfiable: when the presupposition is not satisfied, the speaker 'requests' that the addressee accommodate it. In this way, the speaker 'imposes', without putting it up for discussion, what the next question to be addressed is. Accommodation is possible as long as the question can be considered a strategy for an ongoing inquiry and the addressee is willing to accept the mandate. Power dynamics may affect the speaker's choice of utterance: WhDecs impose what the next 34 Assertions can also provide information regarding cs. This would be instances in which the assertion does not provide new information but is merely used to remind participants about the properties of the worlds in cs.
(i) It's 6:00 a.m. and A and B are on the phone arranging how to meet to pick up Sarah at the airport. They know Sarah's plane is arriving at 12:30. A: Sarah is arriving at 12:30. What time will you meet me there? B: I'll be there at 12:45. A: She is arriving at 12:30!! B: Yes, but I can't make it any earlier.
A's second utterance is not providing new information, it is only reminding the addressee of something they already know and they know that they both know. A reviewer indicates a possible connection of this point with views of epistemic modals according to which they do not provide an information update but rather test properties of an information state.

28
early access issue to be addressed is instead of proposing to address the particular issue (as WhQs do), and not all participants in discourse have that authority. As a consequence, when WhDecs are uttered by a participant lacking authority, WhDecs come across as out of place manners-wise. For WhDecs to be felicitous either the speaker has particular authority or all participants are taken to be in a 'familiar' relation where impositions of this sort can be tolerated (see the contrast between (13) and (12) above). This is what one would expect when there is an "imposition" regarding what we should do next, i.e., dictating what our next discourse move should be (to address the question that is presupposed and left un-answered). Given this commanding nature, it would be expected that, in the wrong context, accommodation would naturally be more 'aggressively' resisted than regular accommodation of information in cs (which does not dictate what the addressee's following discourse move has to be in the same way). Factual presuppositions regarding cs are easily accommodated unless the information conflicts with other facts and are crucial for what is at issue at that time in the converstation. This expected asymmetry, however, has not been quantitatively investigated yet.
The overall 'urgency effect' in WhDecs: By uttering a WhDec, a speaker is able to cut corners: instead of proposing that a question be accepted (going through the process of awaiting evaluation of that proposal), the speaker presupposes that participants have already committed to addressing that particular question. This use of (informative) presuppositions is parallel to what we find in more familiar cases such as I can't meet tonight. I have to pick up my sister at the airport in which the speaker conveys that s/he has a sister without putting it up for discussion. 35 WhDecs do not put up for discussion what the question to be addressed next is. The speaker uttering a WhDec takes for granted that the question is already accepted and merely points at it. The sense of urgency associated with WhDecs, not replicated by plain WhQs, results from this presupposition combined with the fact that the peaker is highlighting that the question is (already) awaiting to be addressed.
Reactions to WhDecs and wh-interrogatives: After acceptance, both in the case of WhDecs and WhQs cooperative responses are geared to solving the same discourse-question. This gives the impression that WhDecs are questions: answers to the WhQ are also felicitous responses to its WhDec counterpart. WhDecs and WhQs differ, however, in the case of rejections. Rejecting a WhQ amounts to rejecting the proposal to add a question to the top of the question stack, i.e., refusing to pursue a line of inquiry. Rejecting the WhDec, on the other hand, does not amount to rejecting shrinking the cs, since WhDecs are not informative that way. Rejecting the WhDec 35 The alternative would be I have a sister. I have to pick up my sister at the airport.

29
early access is rejecting its presupposition and hence rejections to WhDecs are indications of presupposition failure. This is observed in out-of-the-blue scenarios, in which the addressee cannot/will not accommodate that s/he has already accepted a question.
(12) A stops a random pedestrian on the street and says. A: Excuse me, where can I buy an Italian newspaper? A': Excuse me, # I can buy an Italian newspaper where?
A' is infelicitous out of the blue because the addressee can't accommodate why/how s/he should have agreed to address a particular question. Cases in which the context of utterance seems to provide the cue for the question, and the power-dynamics are right, can be perfectly fine, as in (13). But again, when the power dynamics is not right, WhDecs are rather nosy or rude, (58): (13) B is helping to tidy up after dinner at her friend's house and enters the kitchen carrying the dishes. B: These go where?
(58) Passengers in a plane are getting settled before departure. Passenger A starts a conversation with passenger B, whom s/he assumes to be American. A: I hope this plane gets to NY on time. B: (With a strong non-English accent) Yes, otherwise I will be missing my connection. A: [Obviously surprised to hear a non-English accent] Oh! Yes, there is always that risk..., (and) you are from where?
The contrast between (13) and (58) depends on the willingness to accept the 'imposition' of the question to be addressed: we are usually less willing to accept that a stranger is entitled to impose where the discourse is going next and assume that we are willing to comply, let alone when this involves requesting personal information.
The bottom line is that WhDecs behave just as utterances with other kinds of presuppositions do. They offer a shortcut in the same way many other presupposition triggers do (as when a speaker utters I can't meet you later. I'm picking up my brother at the airport taking a shortcut regarding the information that s/he has a brother). The difference is that in the case of WhDecs the shortcut concerns what is the question that participants have committed to addressing next and, when accepted, WhDecs help to move the discourse along quickly. Because of their nature, WhDecs are related to declaratives and interrogatives in their licensing conditions. Infelicity in out of the blue WhDecs is related to infelicity in out of the blue declaratives (see discussion above in §2.3.1): in both cases there is a tension between what the speaker presupposes the addressee has committed to and how much the addressee is 30 early access ready to accommodate. Additionally, given that WhDecs request that a question be accommodated, the felicity of WhDecs is also related to the felicity of their WhQ counterpart.
WhDecs are predicted to be felicitous when the power dynamics are right and participants have no problem accepting that they can be told what their next move is and that the question the WhDec presupposes to be on the top of the stack is highly relevant to the ongoing inquiry, as in (9) above. WhDecs are also possible when the presupposed question has in fact been openly agreed upon before (i.e., having being openly accepted), as in (59) If the WhQ question has not been rejected, we can assume in the context above that it is top(Q c ) (the default). The WhDec can be used as a reminder that the question is still open awaiting on the top of the stack. Of course, given that the final dynamic update of a WhDec is identical to that of a WhQ, the WhQ-WhDec sequence is not good without an excuse to reiterate what top(Q c ) is: we can only remind the addressee of what is awaiting on top(Q c ) when we have indications that the addressee has not rejected the question but is not really addressing it. In these cases, we can use the WhDec as a reminder, as in (59).
Sequences of WhDecs, wh-interrogatives and both: Recall the contrast in (16) Given that the sequences above ultimately raise issues that are not in an entailment relation (see (22) above), they can only be felicitous when interpreted as spelling out subquestions that together form a strategy to answer a more general question, e.g., "what are the whereabouts of the persons of interest in the case?" (see (23)). This is fine in the case of two WhQs or WhDecs but not if we have a mixed sequence.

early access
The sequences in (16c) and (16d) are understood as the spell out of the subquestions forming a strategy and, hence, as presupposing (with WhDecs, (16c)) or proposing (with WhQs, (16d)) a question, the more general question that is addressed, e.g., "what are the whereabouts of the interested parties?". 37 The problem with (16a) is that it is contradictory. In uttering the WhQ, the speaker does not take for granted that the addressee will accept to address the question regarding the whereabouts of the boyfriend: with the WhQ s/he is proposing to pursue the inquiry, leaving room for rejection. With the WhDec, the speaker presupposes that the question they are already addressing is about the whereabouts of the parents. However, given that the two issues form a strategy, the inquiry that ultimately is being addressed is about the whereabouts of all the interested parties, and there is an inconsistency as to whether this (more general) question is proposed or presupposed.
Finally, judgements for (16b) are less clear: it is not as bad as (16a) (speakers find it better than (16a)) but not perfect). The addressee could reach the same conclusion as with (16a), and then it would be infelicitous. But here the addressee has the option of understanding it differently. (16b) could be understood as the speaker 'reminding' the addressee that top(Q) is a question about the whereabouts of the boyfriend and then proposing to also address the question regarding the whereabouts of the parents. At the end, the two issues still form a strategy and the proposal is to address a more general question regarding the whereabouts of the interested parties, but in this interpretation we ask to 'extend' a presupposed inquiry to a more general question. 38 Of course, the speaker could just have uttered two WhQs and, depending on how much participants are willing to accept the mix of shapes, the overall result could be more or less acceptable, with variability predicted.
Questioner's assumptions about the addressee's knowledge: There are other situations in which WhDecs are worse than WhQs (contrast (15) and (60)): (15) I want to repair the sink but I can't find your mother's toolbox. I don't seriously expect you to know this, but just in case, a. #. . . your mother's toolbox is where? (Do you know?) early access b. . . . where is your mother's toolbox? (Do you know?) (60) I want to repair the sink. You told me where the toolbox is but I forgot, a. . . . your mother's toolbox is where? b. . . . where is your mother's toolbox?
In the examples above the inquiry is not out of the blue and we can assume that the power dynamics are right. The only difference are the speaker's assumptions regarding the addressee's knowledge: WhDecs are not possible when it is likely that the addressee doesn't know the answer. This can be easily explained. We assume by default that a questioner proposing a question doesn't posses the information to resolve it and, hence, committing to pursuing the question involves being able to help towards resolving the issue, i.e., to provide some kind of information.
Hence, presupposing that a question is in Q is presupposing that the addressee has already accepted it and that s/he has information to solve it (or, in the limit case, has information regarding whether the answer can be found). The infelicity of the WhDecs in (15a) results from admitting that the addressee may lack the information and then uttering the WhDec presupposing that s/he has information leading to the answer. This is effectively a contradiction. It's the dynamic equivalent to #you may not have a sister but I just saw your sister in the cafeteria although, arguably, a contradiction at the level of facts is more disturbing than at the level of other discourse assumptions. 39 The canonical wh-interrogative in (15b) does not 39 Two reviewers raise questions regarding what it entails to accept a question. A reviewer points out that an addressee can always accept the question by saying That's an interesting question while admitting ignorance regarding the answer. Notice that such response is not an 'acceptance' of the question in the relevant sense: That's an interesting question, let's find out' indicates that the addressee has no information to solve it and hence the 'discourse' moves to a difference sequence: the speaker rejects pursuing the question (it cannot be solved within that discourse-sequence). Accepting a question in the relevant sense means that one commits to directing subsequent discourse moves towards answering it (see definitions above), not that one is willing to go ask someone else. A reviewer points out that one can accept a question and yet be not-knowledgeable about the answer.
The following example illustrates it: (i) A: Where is Susan? B: I do not know, isn't she in her office?/ # Is she in her office?
Notice that B's response is still cooperative: while B is not directly providing a (partial) answer, s/he is introducing a strategy to address the question. In this case to evaluate the alternatives one by one while indicating that s/he expected that Susan is in her office to be true. The plain (canonical) PolQ is marked across speakers because if A knew that Susan is in her office s/he wouldn't be (plainly) asking where Susan is. For a response to be felicitous across speakers it has to be at least a partial answer (where a partial answer is one that favors one alternative over others, see Büring 2003): B's felicitous response (a high-negation question) indicates that s/he would have expected Susan to be in her office, hence marking that alternative as more probable, while wondering whether this is indeed 33 early access presuppose that any question is already accepted but rather is a proposal that it be accepted and placed on Q. Since it is a mere proposal to place the question on the top of the stack, it gives the addressee the chance to reject the question on the basis of lack of knowledge. In such cases, the continuation offered by the speaker in which it is questioned whether the addressee has any information provides a possible reason to reject placing the question on the top of Q. In such cases, as expected, the continuation is fine.
In what follows I sketch how the system proposed here for WhDecs could be extended to another hybrid, RDecs, cashing out insights in the literature, §3. I also address the role of intonation in the interpretation.

Rising declaratives and WhDecs
Declaratives with a final rise (FR), RDecs, are the non-wh-counterpart of WhDecs. RDecs' prototypical final contour is H* H-H% or L* H-H% (but H*/L* H-L% is also possible; see Bartels 1999), the same as in polar questions (PolQs) Like WhDecs, RDecs are also hybrids: while they have a declarative word order, they have an inquisitive interpretation comparable but different from that of PolQs (which exhibit subject-auxiliary inversion and are interrogatives). In both cases the speaker seems to request that the addressee reveals whether the content proposition (e.g., that the fight is over a boyfriend or that it is raining in the examples above) is true (see, e.g., Gunlogson 2003, Malamud & Stephenson 2015, Farkas & Roelofsen 2017, Jeong 2018, Westera 2018 for discussion). However, unlike with PolQs, when uttering the RDec the speaker indicates that s/he is inclined to believe that the content proposition is true. Regarding their intonation, the final contours in true or, alternatively confirming that this is indeed not the case (see e.g., Ladd 1981 for the available readings). A single speaker can, however, utter the sequence where is Susan? Is she in her office? This contrast is not formalized within the QUD model and formalizing it is beyond the scope of this paper. At any rate, the take home message is that once the question is accepted, the questioner expects the addressee's cooperative move to be one in which a (partial) answer is provided. Consequently, WhDecs signal that the speaker expects the addressee to know the answer to the presupposed question or at least to be able to provide a partial answer.

early access
RDecs and PolQs are also comparable (see also figure 1, pg. 39). This means that meaning differences between PolQs and RDecs arise from differences in word order, just as between WhDecs and WhQs. Given the parallelism between WhDecs/ WhQs and RDecs/PolQs, the obvious question is whether the proposal to explain the contrast between WhDecs/WhQs can explain RDecs/PolQs. If this were the case, differences between PolQs and RDecs would also be derived from differences in the update process, i.e., from PolQs (as interrogatives) being proposals to update Q and RDecs (as declaratives) being proposals to update the context set, cs. 40 In what follows I summarize Malamud & Stephenson's (2015) (henceforth M&S) proposal (it summarizes insights already found in other proposals within an update model) and sketch a way to apply the proposal for WhDecs to RDecs while cashing out their insights, §3.1. In §3.2 I turn to intonation. A lingering question in the semantics/pragmatics literature on PolQs and RDecs concerns how FR contributes to inquisitivity in declaratives. I address the interpretation of the final contour in RDecs and WhDecs. I show what role it plays in the interpretation of inquisitive utterances, arguing that the analysis proposed to explain the contrast between WhQs and WhDecs can be extended to PolQs and RDecs.

RDecs and WhDecs
Building on Farkas & Bruce 2010 and on insights from Gunlogson 2003Gunlogson , 2008, M&S offer an analysis of RDecs in contrast to PolQs: RDecs involve the indication that the speaker wants to commit to the content proposition while the utterance of a PolQ doesn't make any indication regarding any participant's commitment. Full speaker's commitment, however, requires the addressee's confirmation that the content proposition is true in the context of utterance.
In Farkas & Bruce's (2010) model, the proposal nature of utterances is cashed out by placing their (semantic) content "on the table" for the addressee's consideration (the table is modeled as a stack of issues to be resolved, where issues are sets of propositions, singletons in the case of declaratives; see Farkas & Bruce 2010, Farkas & Roelofsen 2015 for a more complex system that also deals with anaphora and ellipsis phenomena). Besides placing the propositional content on the table, regular declaratives introduce the same proposition into the speaker's commitment set (modeled as a set of propositions, the propositions the speaker commits to). In M&S's system, RDecs do not introduce the propositional content in the speaker's commitment set, but rather in their projected commitment set (the set of propositions that the speaker is willing to commit to but has not committed to yet because, for example, s/he lacks evidence fully supporting them). 41 The declarative content is placed on the table for the addressee to assess (just as in canonical declaratives) together with a metalinguistic issue (see Ginzburg 2012), modeled as a set. The metalinguistic issue can be, for example, the speaker's uncertainty regarding the truth of the content proposition. To be more concrete, the RDec in (61a), for example, indicates that the speaker is willing to commit to the fight being over a boyfriend by adding that propositional content to her/his projected commitment set. That propositional content is also added to the table for the addressee to consider. On the top of it, a metalinguistic issue regarding (possibly) the speaker's uncertainty about the actual truth of such proposition is added on the top of the table stack. The addressee is then expected to evaluate the metalinguistic issue and, in doing so, to accept or reject the propositional content of the utterance. This evaluation triggers the final update (i.e., the proposition is added to the CG or its negation is). In M&S's words (pg. 296), "[by making]a move that will have the same effect as a plain assertion [that p], but only with hearer's involvement and approval, the speaker is, roughly speaking, seeking approval to make an assertion that p. Thus [RDecs] are predicted to be possible whenever the speaker isn't sure if a plain assertion is appropriate." PolQs, on the other hand, simply add an issue to the table (in M&S's system the set containing the PolQs content proposition and its negation, both of which the addressee needs to evaluate) without adding any content to the speaker's commitment or projected commitment set.
The question is whether the insights in M&S can be replicated in the model presented for WhDecs while reproducing the parallelisms and differences between WhDecs and WhQs explored above. In what follows I merely sketch a way M&S's insights translate in the system with the aim to show that this is possible and that the proposal presented in this paper provides a very elegant understanding of the construction of meaning of these hybrids and their contrast with the interrogative parallels. However, I leave exploring the details for future research.
In M&S, RDecs are treated already as other declaratives regarding what is added to the table, although RDecs also add a metalinguistic issue on top. In the model in this paper there is no "table", but RDecs are still proposals to update cs. As any other declarative, they should be relevant to a contextual question determined by focus structure. For example, (61a) presupposes that top(Q c ) is a question of the form what was the fight over? The declarative provides a tentative answer to this question while a "sure" answer would be provided by the canonical declarative exhibiting a final fall (FF). There are several ways to model the interpretational contrast between FF and FR in declaratives. One could, for example, argue that FR marks the utterance as contingent (see Gunlogson 2008), as adding the propositional 41 PolQs do not add anything to any commitment set. 36 early access content to the projected commitment (see M&S), or as suspending the maxim of quality (see Westera 2018). 42 I return to this below but, however we model the contribution of FF/FR, the consensus is that FR in declaratives signals that the resulting context is inquisitive (which means that there is more than one viable alternative in cs). How can we cash this out within the model proposed for WhDecs?
A declarative with a canonical FF signals that the proposition denoted is the answer to the contextual question identified via focus structure (no other alternative is an answer). With a RDec, however, the speaker signals that there are other alternatives to the content proposition that are also viable in cs (i.e., while proposing to update the cs with the information that it is raining in (61), the speaker acknowledges that there are other live alternatives, whether this is modelled via, e.g., metalinguistic issues, marking it as contingent or as suspending Quality). By proposing to add the proposition to cs c the speaker proposes it as the answer to the contextual question and we infer that the speaker is willing to commit to such proposition (rather than to any other non-mentioned alternative), and by signaling that s/he still considers that other alternatives are viable s/he signals her/his uncertainty. Given that the speaker only makes a proposal to solve the question while admitting that s/he lacks complete certainty, the question is not actually solved 43 and cannot be popped. The addressee then needs to either publicly accept the proposal made, discarding in this way the competing alternatives (the other answers to the contextual question) and finally solving the question, or reject the proposal. 44 At the end the same system used to explain WhDecs can capture the insights in M&S's proposal.
While the system proposed for WhDecs can be extended to explain RDecs, accounting for both declarative hybrids in the same way, we have not discussed whether the meaning of the final contour in RDecs is the same in WhDecs. If the construction of meaning is alike in both cases, the contribution of the final contour should be similar. However, while WhDecs and RDecs share the same canonical realization (H*/L* H-H%), WhDecs also allow H* L-L% (see ftn. 4 on pg. 4).

The role of the final contour in RDecs and WhDecs
In the proposal offered for the construction of meaning in WhDecs the final contour doesn't play any role. Differences between WhDecs and WhQs result from WhDecs being declaratives and updating as such (they are proposals to update cs) and WhQs being interrogatives (they are proposals to update Q). If we aim for a unified analysis of RDecs and WhDecs, this seems at odds with what we observe in RDecs, where 42 We are abstracting away from details on the final contour studied in Jeong (2018). 43 It's at best a partial answer in Büring's (2003) terms (see ftn. 39). 44 As it is, the framework presented here doesn't consider Ginzburg's (2012) metalinguistic issues as M&S do, but there is nothing in the model preventing us from incorporating them. Fig. 1 reveals the taxonomy that results from taking into consideration the conventional cues leading to the interpretation of declaratives and interrogatives, starting with word-order and finishing by considering the final contour. The result shows that the choice between FF and FR is used to mark a contrast in all the different realms, and that this contrast doesn't need to mark the same opposition across the board. In fact, looking at wh-interrogatives, it seems it can't.
We have already seen above that the contrast between FF and FR in declaratives allows us to distinguish between canonical declaratives and declaratives that are inquisitive. (As discussed, there are several ways to model this, and a proposal modeling the FR as indicating that there are several live alternatives in the context early access The proposal in this paper is related to Biezma's (2018) proposal for Spanish utterances with non-fronted wh-words. Biezma (2018) treats these utterances as questions and argues that the difference between wh-fronted and non-fronted questions is that the latter are strategies to indicate that the question is a "follow up question", i.e., a question that is a response to an assertion and presupposes that the utterance is part of a larger strategy to answer a higher question or that requests further information about the event predicated of in the previous assertions. The overall claim in this paper is similar to that in Biezma (2018): i.e., that WhDecs are linguistic devices whose role is to presuppose discourse structure, in particular the structure of Q. However, in Biezma's (2018)'s proposal, the fact that non-frontedwh-questions have to follow an assertion and that they are follow-up moves is stipulated and not derived from the properties of the utterance itself. In contrast, the proposal in this paper derives those properties from independently motivated ingredients and explains the data in Biezma (2018).

Conclusion
The proposal in this paper offers an analysis of the semantics and pragmatics of WhDecs making use of independently motivated ingredients. The analysis ends up deriving why an utterance with declarative word order can be inquisitive without appealing to ad-hoc solutions at LF and is in line with analyses of related phenomena such as RDecs that try to keep the interpretation of the utterance true to form.
WhDecs illustrate how different linguistic factors interact in the construction of meaning in natural language utterances. The overall interpretation of WhDecs is the result of the interaction between syntax, semantics and pragmatics. Taking into account the dynamic update proposed by the utterance as indicated by word order together with the link to discourse signaled by focus structure, we can derive differences between WhDecs and WhQs without ad-hoc stipulations.
WhDecs are the flip side of the coin of rhetorical questions (RQs) in B&R's analysis. B&R's objective was to propose an analysis of RQs that explains why they 'feel' like assertions and how they actually differ from them. In B&R's proposal, rhetorical questions are interrogatives conventionally indicating (e.g., via prosody or lexical items) that the answer is entailed in the context of utterance (and hence that all participants have access to it). That is, RQs are interrogatives triggering the speaker's presupposition that the answer is known by all participants. B&R propose that, dynamically, RQs behave like other interrogatives and are proposals to update Q but, given the presupposition they trigger, their acceptance leads to a non-inquisitive update: accepting the proposal to update Q means accepting the presupposition it triggers and, hence, the final update is trivial, i.e., it takes place in a context in which the (presupposed) answer is accepted and the resulting updated 42 early access context is, hence, non-inquisitive (the answer to the question is entailed by the context). Differences between rhetorical questions and assertions are explained with respect to their differences at the proposal stage: unlike rhetorical questions, which propose to update Q, assertions are proposals to update the context set (the reader is referred to B&R for details). WhDecs can be thought of as the flip side of the coin: WhDecs are assertions that presuppose what the IQUD is. Accepting the assertion amounts to accepting the presupposition, i.e., that all speakers commit to answering a particular question and, hence, subsequent responses ought to be answers to such QUD. This leads to a picture in which WhDecs differ from WhQs in that WhDecs do not propose that participants pursue a question, they indicate that participants have agreed on pursuing a particular question. Accepting the WhDec leads to accepting such presupposition and, hence, responses to the WhDec are understood as answers to the presupposed question. Indirectly, WhDecs add a question in discourse, the same way that, indirectly, RQs add information to the common ground.
In the broader picture, WhDecs are a window into the study of presupposition. WhDecs are conventional (linguistic) devices signaling speaker's assumptions about the structure of discourse, i.e., they trigger a discourse-structural presupposition. In addition, the study of WhDecs allows us to revise the notion of relevance in discourse: an assertion is relevant (and hence felicitous) if it provides information about the context, either about cs or about Q.
Once we take into account the morpho-syntactic features of declaratives and interrogatives we arrive at a taxonomy of utterances that raises questions about the role of the final prosodic contour in the construction of meaning and, in general, about the overall contribution of prosody in the interpretation. This opens different hypotheses that should be pursued in future research.