2.2.1 Speakers and addressees

Meta-why’s distribution in discourse is restricted in terms of the content it can access and the interlocutors who may deploy it. As noted by Ginzburg (2012: 313), only the addressee of the relevant question can use meta-why:

However, we argue that the addressee of the original question may be understood as slightly broader than the single person to whom the question is addressed. George Tsoulas (p.c.) suggests that the following scenario, minimally different from (13), is in fact felicitous:

As the defence lawyer has the right to speak on behalf of the defendant, she is effectively the addressee of the prosecution lawyer’s question too.

We further observe that the addressee of the original question need not be one unique individual, as meta-why is fine in contexts where the original question is addressed to a group, as long as the utterer of meta-why can plausibly be understood to be an addressee of the original question.

Further to this, meta-why also cannot be used by the original questioner to induce the addressee to make explicit the questioner’s own motivations (see also Ginzburg 2012: 314):

Compare (17) with the legitimate (18).

In (18), why appears to be requesting a reason for the teacher’s question. Though this seems to be precisely the meaning of meta-why, we have already seen that meta-why differs from reason- and purpose-why. We resketch the differences between reason- and purpose-why in (19) below:

While why in (19) can be replaced by For what purpose? to elicit response B or For what reason? to elicit response B′, meta-why cannot be replaced by either of these fragmentary paraphrases:

To understand why this is, we will now investigate how different types of why differ based on the information that informs their possible answers, ultimately drawing an analogy between different types of why and different modal flavours.

2.2.2 A brief ontology of why

Koura (1988) distinguishes between a why whose answer pertains to the achievement of an endpoint (purpose-why) and a why whose answer depends on natural laws and how they determine causal relationships to explain a connection between the explanandum and the explanation (reason-why). Bromberger (1992) also invokes natural law as a condition that defines possible answers to reason-why questions. This characterisation of different types of why recalls a Kratzerian approach to modality (Kratzer 1981; 1991; 2012), wherein modals are distinguished by two contextual factors. This is summarised by Katz et al. (2012: 489) as follows:

[…] the modal base, which determines the possibilities that are relevant in the current context, and the ordering source, which provides a set of propositions encoding the priorities among these possibilities. The ordering source induces a “goodness” ordering on worlds, and this ordering serves as the basis for interpreting [modal] sentences.

To flesh out the analogy: just as teleological modals and deontic modals are distinguished by their ordering sources (the performance of actions favourable to some outcome versus adherence to some set of laws), so answers to purpose- and reason-why are distinguished by similar sets of determining factors—the actions that are favourable to reaching a given outcome versus the laws that dictate whether an event or state will obtain in some context.

Interestingly, when these types of why query propositions that contain modals, the explanandum—that which requires explanation—shifts from the outcome/event/state to target the modal’s ordering source. This is illustrated in the reason-why⁹ examples in (21)–(22). Moreover, when the premises that underpin the modal ordering source are overt, as in (23), the explanandum shifts again to the legitimacy of the ordering source:

The behaviour of meta-why-fragments differs, however. Even if premises for the original question are provided, the interpretation of meta-why remains the same:

We can therefore observe that the explanandum of a meta-why question is always the premise(s) that motivate the previous speaker’s question, which implies that no matter how much context the previous speaker has given for their question, the utterer of meta-why believes that there is some information that still has not been shared with them. It also suggests that meta-why outscopes modality, unlike reason- and purpose-whys.

Moreover, answers to meta-why questions are not defined by epistemic factors in general, such as stereotypical expectations as to why any speaker would ask some given question, but they specifically target information that only that original questioner could possibly provide, as it pertains to their motivations and intentions within the discourse—information that Ginzburg (2012: 300–301) refers to as genre recognition. In other words, the ordering of answers to meta-why questions can only be determined by the speaker’s intentions, and never by circumstantial or stereotypical factors. Let us briefly investigate this using two apparent (constructed) counterexamples where a third person responds to the meta-why-fragment.¹⁰

In this case, the defence lawyer’s meta-why is considered insincere as the prosecution lawyer’s reasons for asking the question have already been established and accepted by all interlocutors (namely relationship of the shirt colour to the case and the role of the prosecution lawyer in the context), so the meta-why question itself is rejected rather than answered.

In this case, the parents are effectively acting as one speaker as they share the exact same intentions with respect to the question at issue.

We therefore conclude that meta-why targets the private knowledge and intentions of the original speaker and that that knowledge must not be already shared between the speaker and the addressee. As an extension of that, answers to meta-why cannot be solely defined by any other set of factors, such as circumstance, natural law or stereotypical expectation.

2.2.3 Lexical realisations of different whys

We also find empirical motivation for a distinction between reason- and purpose-why on the one hand, and meta-why on the other, in languages that have different lexical items for reason- and purpose-why. That said, we see some ways in which meta- and reason-why are similar and distinguished from purpose-why.

Take Russian as an example. Reason- and purpose-why are expressed by two different lexemes, počemu and začem respectively, which Stepanov & Tsai (2008) argue to be merged at different positions in syntactic structure.

However, neither can be used metacommunicatively:¹¹

Russian a chto is not the same type of element as meta-why as it can be used metacommunicatively in response to declaratives (just like English And (what)?/So (what)?) and does not appear to be sentential, as we will show in Section 5.2. We will otherwise leave elements like a chto for future work.

Ginzburg (2012: 313) also notes that in Hebrew and German, languages with multiple lexemes glossed as why in English, only one may be used metacommunicatively. In these cases, the lexeme used for meta-why must also be compatible with a reason-why reading:

This is intuitively unsurprising and is reflected in our working paraphrase for meta-why as asking for a reason for making a question speech act. However, we have already seen in Section 2.2.2, and will further see in the sections to follow, that meta-why should still be treated as a different lexical item from reason-why, irrespective of homophony.

In the rest of the article, starting with Section 3, we will provide evidence for our principal novel theoretical claim with respect to meta-why: that ellipsis is involved in meta-why-fragments.¹²

5.2.1 A brief excursus: languages without meta-why-stripping

Let us briefly consider again languages whose meta-why equivalent does not derive from the word for why, for example Russian. We have suggested that these items are truly non-sentential, so we would predict that the meta-why equivalent in these languages may not appear with an additional non-wh-remnant or a full recapitulation of the question. This is borne out in Russian, as shown in (74), and a sentential response is necessary instead:

Fenno-Swedish is another language in which bare varför (lit ‘why’) cannot be used metacommunicatively and hur så (lit. ‘how so’) is employed instead (Anders Holmberg, p.c.). In Fenno-Swedish, as in Russian, hur så cannot participate in why-stripping.

However, Holmberg also reports the following exchange to be licit, as it is in the Swedish spoken in Sweden:

Holmberg also notes that in the Fenno-Swedish equivalent of our cooking teacher example in (63), a varför-strip also behaves like a typical why-strip again. Holmberg suggests that “Sally” in B’s response in (76) could be a hanging topic, in which case the utterance would contain something like sluicing rather than stripping. However, this does not explain why the metacommunicative intention obtains in (76) when it does not in the absence of the non-why-remnant, leading us to conclude that (76) is a case of meta-why-stripping. We will return to why (76) is licit while bare meta-varför is illicit in Fenno-Swedish in Section 5.3.2.

To return to our main data, having established that ellipsis of an illocutionary question act is the process by which meta-why-fragments are derived, we now justify why we believe that the meta-why element is an SAP-adjoined modifier of how the utterance entity affects the discourse context.

5.3.1 Modelling meta-why as a discourse-level modifier

A core purpose of this section is to justify the claim made in Section 5.2 that meta-why is not simply the wh-word why in a high clausal position. Throughout this paper we have indicated using ⋍ that meta-why only approximates a request for a reason for making some speech act, and that the environments that permit meta-why are a subset of the possible environments that typical reason-why questions can be used in. We propose that a better paraphrase for meta-why is “There’s some reason for your question that I’m not understanding”, to really capture that the original questioner’s intentions alone determine the response, rather than any potentially plausible answer to the non-meta-why question “Why are you asking me that?”, as well as the intuition that these intentions are unknown to the user of meta-why. If we test this paraphrase on meta-why and “Why are you asking me that?” in their differing contexts, we see that it is licit where meta-why is licit ((77) and (80)) and not otherwise:

This paraphrase reflects the fact that meta-why is a non-canonical response to the question by the interlocutor to whom the question is addressed—in short, they do not answer the question. However, while it affects the next steps in the discourse, its use does not reject or represent a refusal to answer the question outright.

Moreover, as we have already shown, meta-why is also highly dependent on the prior, currently live discourse move being an open information-seeking question. It can never stand alone from, or precede or anticipate, that question act, whereas locutions like “Why might I be about to ask you X?” are perfectly acceptable. For these reasons, we conceive of meta-why as a modifier of a speech act or, more precisely, the modifier of a context-change potential. To progress in this section, we shall first present Krifka’s (2014) semantic proposal for illocutionary operators and context change potentials, then discuss how (and to what extent) that kind of proposal presents a plausible way ahead for formalising the meaning of SAP and meta-why. We will present some empirical evidence for our proposed meaning for SAP and for meta-why being an SAP-level modifier, before demonstrating how these proposed meanings play out in a model of discourse, specifically Farkas & Bruce’s (2010) Table model.

Krifka (2014) takes a dynamic approach to speech acts, based on work by Szabolcsi (1982), in which speech acts are an “index changing device” (Krifka 2014: 64). Krifka claims that in making a speech act, the speaker not only expresses an attitude but also takes on certain commitments with respect to the content of that attitude, in the case of an assertion, this is the “liability” that the proposition expressed is true and that evidence can be presented in support of that truth (Krifka 2014: 65). He claims that the production of the speech act results, instantaneously, in an update in the commitment state of the speaker to the effect just described. Commitment states can, however, be modified—Krifka describes cases such as repeated assertions, where the speaker’s commitment to the content of the assertion holds from its first production, but repeated instances of the assertion, rather than leading to tautology or redundancy, might instead increase the strength of the commitment of the speaker (Krifka 2014: 65–66).

Speech acts, then, are a different kind of semantic object than we have henceforth been dealing with—they are not truth values (propositions) or sets of truth values (interrogatives); nor are they entities. Instead, they are changes from one index to another, where at the first index a commitment state did not hold but at a second index the commitment state does hold. Krifka models this for assertions by proposing an assertion operator ASSERT (81) and an operation expressing changes of indices (82), such that the performance of an assertion consists of the process described in (83) (Krifka 2014: 68–69):

(82)	i′ i [F[i]]
	⟺ def	i′→i ∧¬F[i′] ∧ F[i] ∧
		for all formulas G such that F and G are logically
		independent G[i′]↔G[i]
	“This expresses a minimal change from i′ to i, consisting in the change of the truth value of the condition F[i].”

(83)	λi∃i′[i′ i [ASSERT(i)(p)(y)(x)]
	“This proposition is true for all indices i that differ from the immediately preceding i′ only insofar as at i, x is liable towards y for the truth of the proposition p at i.”

How does this relate to our proposed SAP-IAP structure? We claim syntactic equivalence between Krifka’s ASSERT and our QUESTION operator, which we analysed as located in the head of the illocutionary act phrase (IAP), as they both apply to the propositional content of the clause. However, we do not adopt the exact details of Krifka’s proposal, firstly because the output of our operator is an entity and the output of Krifka’s is a truth value, and secondly because we will claim that the relevant expression of liability for truth occurs at the next level up; at the SA head. This means that we would have to adapt (82) such that the condition F is defined as the speaker expressing the relevant liability for the truth of the propositional content contained within their utterance act. (We will return to the issue of who takes liability for truth in the case of a question shortly.)

This modified operation of changing indices with specific reference to liability for truth applies at the speech act head, such that an appropriately modified version of the process in (83) reflects the meaning of the SAP, which is where the new index i is introduced. Again, we endorse the core elements of this approach to understanding meta-why but the full formal details of these proposals are left for future work.

Based on our proposal that meta-why modifies how the original question act has tried to move the discourse forwards, it follows that it must outscope the QUESTION operator and the output of its composition with the propositional content of the question, thereby applying at the SAP level.

The proposal that meta-why modifies SAP is empirically attractive as meta-why appears to be in complementary distribution with other items that play similar roles and might otherwise appear in this utterance-peripheral position, such as speech act adverbs and exclamative vocatives; cf. Hill (2013: 208–9):

Note that the strings in both (84) and (85) are in fact licit, but only in the case that why is reason- or purpose-why rather than metacommunicative. This follows if we assume an analysis for reason- or purpose-why such as that in Stepanov & Tsai (2008) or Rizzi (2002), where they sit in ForceP or lower.

The division of labour across these two projections predicts that it should be possible to separate off the utterance act (IAP) from the event of updating the context with the speaker’s commitment (SAP). We can demonstrate this with evidence from Germanic syntax and Korean morphophonology.

In Germanic (and in many other language families), it is perfectly possible for a speaker to perform a root utterance act without committing to the truth of the content of that utterance—in other words, to utter an IAP that is not itself selected by SAP, as in the following examples:

The examples above show that this typically requires embedding in English, but can be achieved without embedding in languages like German. In (86) the speaker presents a proposition for discussion without necessarily committing to its truth, but that proposition must be embedded within some other context—here the debate context. In (87), in contrast, we have a root verb-final utterance—there is no embedding predicate or preceding discourse—with truth conditions that can be given, but it is not interpreted as an assertion by the speaker (see Truckenbrodt 2006 for more on such utterances). Moreover, in languages like Korean, morphemes that express the illocutionary force of an utterance and those which express who is responsible for committing to the content of the utterance can be realised separately (all examples below from Ceong 2017: 13):

In these examples, the clause-typing information, illocutionary force (reinforcement or echo question) and the commitment holder (speaker or addressee) are all marked independently of each other, the last of these being marked by the rising or falling intonation at the end of the utterance.

Finally, recall that overtly-marked illocutionary acts can be embedded, for example in cases of Germanic embedded verb second and English embedded interrogatives, but they do not commit the current interlocutors to their truth (Wiklund 2010; Krifka 2014; Woods 2016; Djärv 2019):

On the basis of these data we claim it is justified to separate off the utterance act from markers of commitment, overt or covert, and from any modification of that commitment, by the current speaker. We achieve this by distinguishing between the IAP and SAP projections, where the SA head selects for an IAP complement. Consequently, we claim that meta-why modifies the SAP, not the IAP.

Returning to our description of the semantics and pragmatics of these projections, Krifka (2014: 69) notes that the process in (83) results in an event (“as with index changes in general”). As should be evident from the discussion above, the level at which this happens—and at which we are claiming meta-why operates—is one at which we move from true ‘semantic’ composition to a form of discourse composition, where the nature of the objects being combined and modified moves from (sets of) propositions to discourse entities and events. As a result, we need to think more carefully about how discourse composition might work, and what discourse modification would then mean—and specifically for this paper, what meta-why modifies with respect to the discourse structure. We will not go deep into detail here but will use a model concerned with the effects on discourse continuations of basic assertions and polar questions (i.e. propositional content embedded under speech act operators), namely that of Farkas & Bruce (2010), to take a first, descriptive, pass at modelling the effect of a meta-why-utterance on the discourse continuation. We will also use this model to define the commitments involved in the uttering of a question act. We therefore caveat what follows as being the first step on the road to understanding discourse modification, with plenty of work that must follow up on this.

As alluded to at the start of this section, we believe that a meta-why utterance does not introduce a new question into the discourse. This might seem like a strange proposal, as meta-why utterances start with a “question” word, often have rising intonation and elicit an answer that (as we’ve discussed extensively) looks a lot like an answer to a reason-why question. However, as (77) and (80) illustrate, meta-why’s impact is not exactly the same as that of a question—it appears to ‘freeze’ the context update that the question would normally make at the point of the change where the commitments come to hold, for the specific purposes of clarification and expansion on the motivations behind the speaker’s question. We will show how meta-why and other wh-questions differ in detail in what follows.

We will model the effect of meta-why using Farkas & Bruce’s (2010) Table model for discourse. Similarly to Krifka, they treat speech acts as dynamic objects with the potential to update the discourse context. Focusing initially on assertions, they tabulate three changes that a single declarative sentence S[D] makes when asserted in an input context K_i:

Assertions of some proposition are for Farkas and Bruce, therefore, also a function from input context states to output context states, where in the output context states the effects of the three changes above hold. This process is visualised in Table 1 for the simple root assertion “Sam is home”, taken from Farkas & Bruce (2010: 91). In Table 1, the input context state K₀ consists of no prior discourse commitments or items on the Table, and the existing common ground (shared information) and projected sets both consist of s₁, the initial common ground at the start of the conversation.

Table 1

K₁: A asserts “Sam is home” relative to K₀.

Note that the projected set in Table 1 contains only the input projected set and its union with the singleton proposition expressed by S[D]. This is because Farkas and Bruce claim that there are canonical ways of responding to different types of speech acts, and in the case of assertions of a proposition in the form of a declarative, this is acceptance of that proposition by the addressee. Essentially, “an assertion yields an output context that is categorically biased in favour of confirmation of the asserted proposition” (Farkas & Bruce 2010: 92).

Farkas and Bruce also model the impact of a speech act containing a ‘default’, or a neutral information-seeking, polar question. Assuming that the denotation of a polar interrogative S[I] is the set {p, ¬p}, they claim that the update made to an input context K_iby such a speech act is not threefold, but twofold:

This differs from an assertion in that there is no addition of propositional content to the speaker’s discourse commitments in the case of a question. Note also that in neither assertions nor (polar) questions are speaker or addressee commitments added to the model, though Farkas and Bruce note in a footnote (fn. 12 on page 95) that their model is compatible with adding information such as who the source of the speech act is. We will therefore add the following context changes to (92) and (93) respectively:

In neither case is this propositional content added to the Table, as these propositions are not directly asserted and therefore cannot be directly challenged.

Returning to polar questions, the canonical projected set contains two members after a question is asked—output contexts containing p, or output contexts containing ¬p. This is schematised in Table 2, taken from Farkas & Bruce (2010: 95) with our addition from (95).

Table 2

K₂: A asks “Is Sam home?” relative to the initial input context K₀.

The canonical response to a polar question speech act as in Table 2 is that the addressee B accepts the liability for the true answer, whether p or ¬p, and asserts that answer in a similar way to that laid out in Table 1. This necessitates an assertion move of some kind which A must then accept, either overtly or tacitly. Only then will the common ground be equal to the relevant member of the projected set indicated in Table 2.

Note that this formulation of the discourse means that by answering a question, B commits to the truth of their answer twice; both in accepting the liability that A confers upon them through A’s question, and by expressing that answer as an assertion. This is recognised by Farkas and Bruce as they note that while assertions can be accepted tacitly, answers to questions must be overt. We suggest that this tacit-overt distinction falls out from the fact that commitment is expressed multiple times. Following Krifka (2014), multiple expressions of commitment do not result in redundancy but have the effect that the commitment is reinforced in some way, and this would be very difficult to do through silence—silence is not compatible with strong commitment. We also see that acceptance of and proffering of liability for the truth of an answer are separable over the course of a discourse. Consider the following context:

We take examples like (96) to suggest that building multiple expressions of commitment into the model for questions is in fact on the right track.

Moving on to meta-why specifically, it is should be immediately clear how a meta-why response to a question (we will stick here to polar questions) differs from a canonical response in that it does not assert p or ¬p. Neither does it place a syntactic form that denotes a full specific proposition or set of propositions on the Table. In this way, it is similar to constituent questions such as “Who is Sam?” (cf. Farkas & Bruce 2010: 106), which could also be used as a non-canonical response to the polar question “Is Sam here?” in the event that the addressee needs more information in order to fully answer. In a dialogue such as (97), such a constituent question appears at first glance to play a similar role to meta-why, in the sense that the resolution of the question “Is Sam here?”, via canonical or non-canonical means, is paused, as the question “Who is Sam?” takes its place as the main question currently under discussion:

However, while the question about Sam’s whereabouts is put on hold because it has been replaced on the Table by a question about Sam’s identity, this is not the same as ‘freezing’ the discourse, because B has not challenged A’s commitment to B’s liability for the true answer to “Is Sam here?”. Indeed, it appears that B has tacitly accepted that liability, even if B does not address the question “Is Sam here?” immediately. Once B’s question “Who is Sam?” is resolved by A, B returns to resolve A’s question “Is Sam here?” in canonical fashion, here by committing to ¬p and pushing the conversation towards a projected set containing the proposition that Sam is not here. Certainly, if B were to ask “Who is Sam?” and then overtly reject their liability for the true answer to “Is Sam here?”, their behaviour would be interpreted as uncooperative, with a sense of having ‘strung A along’:

B also cannot use meta-why after asking another question of their own. The following exchange is infelicitous to the point that it is unclear what B’s why is challenging:

The infelicity in (99) occurs despite the felicity (if uncooperativity) of B’s second utterance in (98), which is akin to “Why are you asking me…”. This shows that the distribution of meta-why is tightly tied to the point in the discourse at which the addressee’s liability needs to be accepted or rejected—apparently, immediately following the utterance of the question.

Meta-why works differently from constituent questions like “Who is Sam?” because rather than introducing a new question onto the table, it is a method of overtly failing to accept liability for giving the true answer to the speaker’s question. It is not, however, an overt rejection either, but rather a method of reserving or delaying acceptance.

We claim that this effect is derived as follows. Meta-why marks removal of B’s acceptance of liability from the projected set as projected by B by expressing (though not asserting) a new proposition that B believes there’s a reason for the original question that they don’t understand. It cannot remove that liability from A’s discourse commitments—that would be far too powerful a move and is not directly reflected in the information that meta-why conveys.

Remember that the form of the utterance containing meta-why contains the original question and is therefore interrogative in form (even in the case where ellipsis has applied and there is no non-why-remnant). However, its effect on the discourse is modified such that neither the addition of p nor ¬p to the common ground (via the projected set) constitute canonical responses. What, then, is the canonical response to a meta-why-utterance?

Let us first lay out what we claim a meta-why-utterance adds to the table (with a small ‘t’). The projected set will, like an assertion, contain a new proposition. Unlike an assertion, however, the proposition conveyed by meta-why is not directly at issue (in the sense of Simons et al. (2010) and cannot be directly challenged, so goes directly into the new projected set and is not added to the Table (with a big ‘T’). The original question remains unanswered, but is also not part of the projected set resulting from a meta-why utterance. A meta-why utterance does not add {p, ¬p} to the table again. This is the real ‘freezing’ effect of meta-why, that it stops the content it modifies from being at issue (a little like embedding) but does not add its own at issue content (unlike embedding). What meta-why utterances add to the Table, then, is the sentence form meta-why plus the material it modifies, without any at-issue propositional content (represented below by the empty set). Given that meta-why utterances do, however, add not-at-issue content that must be added to the common ground, and a sentence form whose utterance also needs to be recorded, it still constitutes a discourse move that needs to be dealt with. The resolution of the original question, and the presence of a true answer to it in some projected set, must therefore wait for a further update.

This means that there the canonical response to a meta-why-utterance is for A to accept—and they can only do so tacitly—that B believes there’s some information that they are missing. We refer to this proposition expressed by B via meta-why as q and visualise the use of meta-why in response to a polar question “Is Sam here?” in Table 3.

Table 3

K₃: B responds “Why?” relative to the input context K₂.

The projected set in K₃ is coherent and achievable, in that it is possible and internally consistent. Moreover, the meta-why utterance will be removed from the Table after its utterance as it is resolved—there is no propositional content related to it that is on the Table and not in a projected set. The problem is that as A’s acceptance of the projected set in Table 3 must be tacit and so the conversation will stall—there is no prescribed next overt discourse move by either A or B, yet A’s unresolved question is still on the Table, their commitment that B should take liability for answering their question is still publicly live (and has not been overtly rejected), and A cannot answer their own question. A therefore has to persuade B to take liability for the answering A’s question, which A can do by providing a reason as to why they asked the question. This results in the discourse situation in Table 4, where the set projected by A includes both the acceptance of their new assertion and the acceptance by B of liability for the question that is still on the Table. Note that the projected set does not contain either p or ¬p, because only one discourse move on the Table can be dealt with at any one time (see Farkas & Bruce (2010) for more on the mechanics of this). B will have to accept A’s new assertion (their reason for asking the question, here r), before either can return to the question of {p,¬p}.

Table 4

K₄: A asserts “I have a meeting with them” relative to the input context K₃.

In Table 4, A has tacitly agreed to go with the projected set from Table 3 so B’s discourse commitments are empty. A is still the only discourse participant who is committed to B’s liability for the truth of p or ¬p and additionally they are committed to r and to being liable for its truth. The projected set is biased towards future common grounds containing r and A’s liability for r, as this is the canonical result of making an assertion. It is also biased towards B accepting liability for the truth of some answer to {p,¬p}, as the aim of conversation is to try to make all of public commitments shared ones, and A has not yet indicated that they are happy to drop B’s liability for the truth of some answer to {p,¬p}, nor have A and B negotiated an agreement to disagree on it.

If B accepts the projected set in Table 4, they should then explicitly proffer liability for some true answer to {p,¬p} by asserting p or ¬p as appropriate. Once this is done, A’s original question will itself be removed from the Table as either p or ¬p will move into the common ground via the projected set proffered by B.

To summarise this section, we have claimed that meta-why is a discourse level modifier that takes a dynamic question speech act object—an utterance event with an interrogative form that projects multiple possible discourse continuations that the addressee is urged to choose between—and modifies what those available discourse continuations are. It does this without adding any new asserted or questioned content to the Table, but (a) by proposing to add a proposition directly to the new projected set that there is some reason for the question that B does not know and (b) by removing B’s liability for answering the question from the projected set. This results in no next discourse move being defined, forcing the original questioner to find a way to persuade B to accept liability for the original question, i.e. to clarify why they asked the question.

Before moving on, we would highlight that this proposal groups meta-why together with other elements that modify specific types of utterance by altering how they may be responded to or expressing elements of the relationship between the speaker, the addressee and what each wants the other to know or do. This group of elements includes the Romanian particle oare briefly discussed in Farkas & Bruce (2010) and Canadian English confirmational eh (Wiltschko et al. 2018). Taking the latter as an example, we see how the addition of eh to a command modifies its interpretation and the range of felicitous responses to it:

In naturalistic example (101), the command Call me later is modified by eh such that it additionally requires the addressee to confirm overtly that they agree to comply. As Wiltschko et al. (2018: 582) note, this extra restriction on the discourse “is evidenced by the addressee’s response, yeah, which would be an infelicitous response to the command.” Note that OK, a typically felicitous response to a command, is also still possible. What is not felicitous is a failure to respond verbally by the addressee, which is typically a perfectly appropriate response to commands in general.

Another example of using particles to modify speaker commitment involves the use of response particles like yes and no to express whether the speaker believes a previous speech act to be a valid one in the discourse context, as discussed by Wiltschko (2018).²⁹ Using a very similar approach to speech act syntax to the one used in this paper, Wiltschko proposes that responses in discourses like (103) have the structure in (104):

To translate Wiltschko’s terms into those used here, GroundP encodes the commitment of the speaker towards the propositional content and takes the typed clause (in Wiltschko’s terms, CP, in ours, ForceP) as its complement. Abstracting away from details such as the identity of the IA/Ground head, and Wiltschko’s inclusion of an an abstract argument GroundS (referring to the speaker’s attitudes) in SpecGroundP, the structure proposed for yes as a validator of a speech act is very similar to that which we propose here for meta-why, which encodes the user’s recognition that there is a non-shared reason for uttering an information-seeking question (and therefore implicitly questions its validity). It really does seem, therefore, that not only is the speech act syntax enterprise on the right lines, but that there is a whole class of discourse elements to which meta-why belongs and which is yet to be fully explored formally.

5.3.2 More on the focus properties of meta-why

One last step is to account for the effect of meta-why’s focus-association properties, as laid out in Section 3, and address how bare and non-bare meta-whys compare.

Let us assume that, with meta-why in SpecSAP, the focused non-wh-remnant in meta-why-stripping moves to SpecIAP. This is far above the typical focus positions in Cartographic approaches such as Rizzi (1997; 2002) inter alia and much of the material in this space is not compatible with a focal interpretation, e.g. discourse particles (cf. Zimmermann 2011; Ott & Struckmeier 2018 i.a.). Indeed, as we have been claiming, these projections are concerned with interlocutor relationships and attitudes towards propositions rather than propositional material itself. However, the meaning of meta-why-strips and typical why-strips differ in exactly that respect. A meta-why-strip asks for the reason why the question contains the content of the non-wh-remnant (i.e. why that specifically is important to the speaker in contrast to all other relevant things that they could ask about), whereas a typical why-strip contrasts the content of the non-wh-remnant with some other possible content of the same type independent of any speaker motivation or interest. Let us consider the following near-minimal pair, where (105) is taken from Gavin Lyall’s novel The Crocus List:

In (105), the strip is understood as meaning something like What’s your interest in knowing about his stance on Berlin? rather than Why did Barling take a stance on Berlin as opposed to some other issue? Although the overt non-wh-remnant only consists of Berlin in (105), the contrast that is set up here is between the question asked about Berlin and the set of all other questions that the speaker might have asked (indeed, utterances that they might have uttered). In (106), the contrast that emerges is between Berlin and the set of other issues that Barling might take a stance on. In other words, meta-why-strips always query some focused element relative to all the utterances that a speaker might make, rather than relative to other possible propositions. This motivates the proposal that the non-wh-remnant in meta-why-strips moves to a projection concerning interlocutor attitudes (the structure of the discourse) rather than one concerning the information structure of the proposition.

This approach to focus finds some support from Zimmermann (2008: 348–349), who argues that analyses of contrastive focus should take into account “the speaker’s assumptions about what the hearer considers to be likely or unlikely [and] contain [information] on the background assumptions of speaker and hearer” on the basis of the variable realisation of contrastive focus in a range of languages. Zimmermann argues that focus marking, whether by intonational or morphological means, appears to be mediated by the speaker’s judgment on whether the hearer will expect the focus content based on the common ground that they share (Zimmermann 2008: 354–355). What meta-why-strips appear to do is mark that the utterer of the meta-why-strip, the original addressee, does not share the common ground required to, in their opinion, adequately answer the original question, but knows that some information exists that the original speaker has not yet shared (given that they asked a question with this content in the first place).

By proposing that the non-wh-remnant in meta-why-strips moves into the discourse layer, we can also account for the fact that typical why-strips are contingent on ellipsis, i.e. never permit full spell out of the ellipsis site, while meta-why-strips do permit full reproduction of the original ForceP.³¹ The clause or utterance itself cannot be in focus in a typical why-strip because both the focus position and the why remnant are lower than the discourse-structural positions. In other words, they are themselves part of the propositional structure. In contrast, meta-why and the non-wh-remnant in a meta-why-strip are above ForceP so ForceP can be taken as the largest non-why-remnant. This predicts that material that is base-generated in IAP or SAP, e.g. the highest types of speech act adverb, should not be able to be part of the non-why-remnant in a meta-why-strip, and this appears to be borne out.

The focus association property of meta-why also helps us (tentatively) explain the paradoxical data from Fenno-Swedish as outlined in Section 5.2.1. Recall that the equivalent of bare meta-why, hur så (lit. ‘how so?’) did not participate in meta-why-stripping, and varför (‘why’) could not be used in its bare form, but the meta-why-strip Varför Sally? (‘Why Sally?’) is permitted. Both bare varför and meta-varför-stripping are available in the Swedish spoken in Sweden, whereas Finnish, the majority language of the communities in which Fenno-Swedish speakers tend to live, has a meta-why equivalent miten niin that also literally means ‘how so?’ and does not permit meta-why-stripping.³² We assume that contact with Finnish has influenced Fenno-Swedish’s bare meta-why, but that meta-varför-stripping persists as long as the non-wh-remnant is overt, presumably because there is some structural difference between hur så and meta-varför, both of which are present in and reinforced in the input from both languages that make up the language repertoire of Fenno-Swedish speakers. We make no further speculations about the nature of language contact in Fenno-Swedish, or exactly how it would result in the ‘dual approach’ to metacommunicative acts that we see in Fenno-Swedish.

However, this leads us to query whether bare meta-why in languages like English and German is subject to the same ellipsis mechanism as meta-why-stripping, or whether it is plausibly a different, non-sentential, element as per Ginzburg (2012). We claim that bare meta-why still results from this stripping-like ellipsis mechanism for three reasons. Firstly, we still need to account for its restriction to ‘freezing’ interrogatively-typed information-seeking questions. Secondly, recall that the meaning of meta-why does not change from bare to stripping contexts, unlike other speech act modifiers and particles. For example, Canadian eh can appear as a standalone utterance, but it loses its confirmational meaning and is only interpretable as a request to repeat the preceding utterance (Wiltschko & Heim 2016: 309). The same holds for huh and right, while other discourse particles like those in German cannot stand alone at all (Ott & Struckmeier 2018: 396). Thirdly, recall that in Yoshida et al.’s (2015) account of typical why-stripping, the focus-associated movement was driven by ellipsis and not vice versa. We therefore claim that bare meta-why is also a product of meta-why-stripping, simply without the focus-associated remnant. Its high position is further confirmed by the absence of pair-list answers to bare meta-why, just like in meta-why-strips (cf. (56) in Section 3).