1.1 Overview

This paper is largely concerned with better understanding the structure and interpretation of an extremely common but understudied why-question in English, for which I will use the descriptive term Why-VP. A naturally occurring example of Why-VP is in (1).

Why-VP can also take the form seen in (2), in which not intervenes between why and the verb phrase:

What is immediately striking about such structures as (1) and (2) is their reduced quality – there are no auxiliaries, no Subject-Aux Inversion (SAI), no tense marking, nor is there an overt subject.¹ Importantly, questions of this form only occur with why in English.

A finite why-question, a term I will adopt to draw a comparison throughout, is one that resembles (4) and is part of a larger class of wh-questions, where the choice of tense does not constrain the availability of other wh-phrases:

The interpretation of both (1) and (2) and all examples of the same type are consistently and crucially modal, best paraphrased with should or would – it will generally be odd to paraphrase (1) with (4), which conveys past tense. The departure in form and interpretation from finite why-questions suggests that the syntax of (1) and (2) should also be distinct from that which is generally assumed for (4).

Despite the frequency with which Why-VP occurs in natural speech and the striking differences which set it apart from finite wh-questions, the analytic and theoretical issues raised by the construction have not been much investigated. In this paper I show that the core properties of the structure emerge in a routine way from standard compositional mechanisms (syntactic and semantic) that are well established and well understood.

2.1 Early approaches

Early discussions on Why-VP largely focused on how it fits in to a theory of speech acts (Sadock 1974; Green 1975; Searle 1975) and by questions having to do with the interaction between illocutionary force and syntactic structure. Such discussions were initiated by Gordon & Lakoff (1971), who endeavored to link the pragmatic function of Why-VP to its syntax, arguing that a transformation they called TENSE-DELETION must be sensitive to the conversational import of an utterance. They want to relate the following two examples by way of this transformation:

They claim that (6a) is ambiguous in its conversational import between conveying an actual request for information and conveying the suggestion that one should not paint their house purple. However, (6b) is unambiguous in necessarily conveying a suggestion. If the why-question is in the form seen in (6a) and is used to make a suggestion, then TENSE-DELETION can apply and derive (6b) from (6a), whereby do you is deleted.

Gordon & Lakoff’s analysis is able to account for the interesting restriction below, observed by Johnson (1975):

If Why-VP necessarily makes a suggestion, as Gordon & Lakoff claim, then verbs like those above, which do not assign an agent theta-role to the subject position will be ruled out, as their underlying representations are unlikely to have the conversational import of a suggestion. In other words, it would be pragmatically odd to suggest a course of action of which the subject cannot be an agent.

However, Johnson observes that theta-role assignment may not be the determining factor here; verb choice, alone, does not contribute this restriction in Why-VP. The same verb will be (un)- acceptable depending on some independent factor. Consider the following:

The manner adverb in (8b) allows us to construe the addressee as having control over a particular part of the event, thus Why-VP here is completely acceptable as opposed to (8a), which seems degraded.⁵

Searle (1975) discusses the speech act status of Why-VP independently of these discussions, acknowledging also that this pseudo-agentive restriction exists, drawing a comparison between Why-VP and imperative structures, which he maintains are both similar in interpretation and in form, but that Why-VP should not itself be taken to be inherently imperative, and that it retains a literal question meaning in addition to having a suggestive or imperative force. Under Searle’s view, Why-VP structures are actually ambiguous and can be used as real information-seeking questions, contra Gordon & Lakoff, whose analysis depends on Why-VP having the conversational import of a suggestion.

An interesting point which comes to bear on the discussion above is the status of the subject in Why-VP. Johnson observes the implicit subject of Why-VP is not necessarily 2nd person pronoun, noting contexts in which something more generic, like anyone, is a better interpretation for the subject:

Further, reviewers of Johnson’s paper suggest that 3rd and 1st person interpretations are possible given the acceptability of reflexives with these features in Why-VP:

While Johnson is unsure about the judgments for (10d) and (10d), we will see in Section 4 naturally occurring data from the corpus which suggests that certain contexts do license third person human reflexives, and all seem quite natural in those contexts.⁶ This observation casts doubt on the claim that Why-VP are always used as suggestions – if they were, we might expect them to always be addressee-oriented, or have interpretation of a 2nd person pronoun. Given (10) above, this does not seem to be the case.

Johnson posits, however, that the real issue with TENSE-DELETION is that the presuppositional content of finite why-questions don’t hold in Why-VP. Consider:

In (11a), it is presupposed that you are kind to your husband, while in (11b), no such presupposition is imposed. If (11b) is to be derived from (11a) then they should have identical interpretation and should show the same presuppositional content.

Johnson (1975) ultimately proposes an amendment to Gordon & Lakoff that Why-VP structures be derived from modal why-questions, particularly from why-would or why-should structures, rather than from Why-do-you questions as in (6a). This move is warranted, according to Johnson, because of the interpretation of Why-VP questions and also because it was a solution to the problem outlined above. The modal question no longer projects the presupposition. Asking why should you be kind to your husband doesn’t presuppose that the addressee should, or has the duty to be kind to their husband. The analysis I develop below will try to incorporate these important observations about modality while avoiding appeal to the disjunctive would or should and also making a connection with other expressions of modality in wh-infinitival contexts.

2.2 Why-VP as expressing a rhetorical question

Bhatt (1998), Duffley & Enns (1996), and Francez (2017) return to the rhetorical or suggestive force of Why-VP utterances first noted by Gordon & Lakoff (1971). They converge on noting certain preferences concerning the use of Why-VP, i.e., that Why-VP favors a rhetorical interpretation; however, I will provide naturally occurring data that show this is merely a preference rather than a hard constraint on interpretation. I will ultimately argue that we will better capture the observations from this line of work by way of mechanisms that are only indirectly sensitive to syntactic and semantic structure.

2.3 Why-VP as a root phenomenon

One novel observation about the distribution of Why-VP I’d like to make at this point is that Why-VP structures are crucially and stubbornly root questions. They cannot be embedded even under predicates which otherwise routinely select wh-interrogatives:

It is worth noting also that why to is incredibly marginal in English, and in a basic context, has been deemed with much consistency as unacceptable:¹⁶

This is a point with which Shlonsky & Soare (2011) and Barrie (2007) have grappled. Both attribute this restriction to why being merged high in the structure, therefore determining a phrase of a larger and higher type than question-embedding verbs normally select. I will argue that selection plays a role in blocking the Why-VP examples in (24) as well, but the implementation will differ. We will return to this important issue in Section 5.

2.4 Interim summary

The theme that runs through all of these studies is that Why-VP is fundamentally different in both interpretation and structure from finite why-questions. For example, the why of Why-VP does not presuppose its prejacent (Johnson 1975; Francez 2017). Further, Why-VP does not seem to correspond to a past or present interpretation, but is rather modal (Johnson 1975; Bhatt 1998; Duffley & Enns 1996; Francez 2017). Obviously, finite why-questions can freely be in the past or present tense, thus the absence of tense morphology on the verb in Why-VP has an irrealis effect on the interpretation as might be expected.

Unlike finite why-questions, the why of Why-VP can only be interpreted in the matrix clause of a bi-clausal structure (Bhatt 1998; Collins 1991). This property is shared with rhetorical why-questions as Bhatt points out. But we should not take that to be the explanation for the restriction in Why-VP as the restriction remains even when Why-VP is not rhetorical.

It is also observed in Bhatt (1998) that Why-VP licenses NPIs, and counter-intuitively Why-not-VP does not. This is also attributed to the polarity reversal they undergo under his analysis of these as necessarily rhetorical. As we have seen however, this cannot be the full story given that Why-VP is not necessarily rhetorical.

Finally, Johnson (1975) makes the important observation that Why-VP hosts reflexives varying in person, suggesting that the subject isn’t always addressee-oriented in its interpretation; this is an important point for any pragmatic account, especially those which have claimed that Why-VP structures are fixed in their pragmatic import as commands or suggestions (Bhatt 1998; Francez 2017). The very possibility of a reflexive also has implications for the issue of what syntactic structure underlies all of these properties – an issue we return to in detail in Section 4.

3.1 Antecedence and given-ness

The weakest possible hypothesis that can be entertained about the licensing of ellipsis is that elided content must be given – roughly, entailed by the discourse context – in a sense that can be made precise in ways explored by Rooth (1992), Schwarzschild (1999), Heim (1997), Takahashi & Fox (2005), Merchant (2001), and others. However there is considerable evidence that a stricter requirement is imposed – one that requires that the elided material (or core subparts of the elided material) have an antecedent phrase with respect to which it is lexically and structurally parallel. Neither condition holds of Why-VP.

Why-VP was brought to our attention as a consequence of the project The Implicit Content of Sluicing, NSF #1451819 at UC Santa Cruz. A group of undergraduate and graduate student researchers are tasked with identifying a discourse-provided antecedent for a given instance of sluicing. This is generally a straightforward task, where the antecedent is immediately apparent, but root sluices in particular proved to be on average, more problematic. Why-VP was sometimes identified as a root sluice by the search mechanisms that identified candidate sluices for annotation. Upon encountering and discussing these examples, annotators unanimously agreed that it was impossible to identify an antecedent for Why-VP. To look for an antecedent for Why-VP essentially means to identify the missing subject, but since there is rarely a parallel VP to the VP of Why-VP, figuring out what would count as an antecedent proved to be nearly impossible.

Yoshida et al. (2015) largely base their argument that various Why-XP fragments are an instance of clausal (TP) ellipsis on observed connectivity effects between a provided context and the material they observe to be missing in said fragments. They propose that why is base generated in spec-CP, above a focus position (FocP), to which the non wh-remnant is ultimately raised. Everything below the focus position that hosts the non-wh remnant is elided. Their analysis primarily draws on examples of the type Why-DP. Their proposal is schematized in (26):

In (26), the material that is elided (crossed out) is recoverable based on the antecedent provided by Speaker A, permitting the utterance to simply be read as why natto? Following Pesetsky (1997), Yoshida et al. adopt a semantic view of the identity condition that warrants clausal ellipsis. They state the following:

The above is essentially an earlier variant of the semantic condition on clausal ellipsis proposed for Sluicing in Merchant (2001), which requires e-GIVENNESS between an elided clause and its antecedent such that the two mutually entail one another (modulo focus marking).

Yoshida et al. argue that in (26), natto is focus marked (F-marked) and must therefore raise out of TP which is then elided. Although the focus in their paper is mainly on cases such as (26) in which the remnant is nominal, the operations illustrated in (26) are in fact meant to apply to any syntactic constituent (below TP, presumably) that may appear with why, including AP, PP, and VP.

An illustration of this line of analysis for Why-VP is presented in (28):

Under this view, the tense and the subject in Why-VP are recoverable based on A’s utterance. However, Why-VP is not dependent on the apparent antecedent in (28A) in the same way that Why natto in (26B) is dependent on its antecedent in (26A). In (28B) the subject of shop is not obligatorily interpreted as Richard, nor is the progressive be necessarily preserved in the interpretation of Why-VP. A possible reading is paraphrased in (29):

It is possible and often preferred for the subject in Why-VP to be interpreted as impersonal with universal force, rather than as referring to a specific individual (Richard in (28)) provided in an apparent antecedent. The provided present tense and imperfective aspect in the apparent antecedent is discarded in favor of a modal interpretation akin to would in this context. Crucial aspects of the interpretation, then, seem to be entirely independent of any potential antecedent.

More to the point, in naturally occurring examples of Why-VP, there is rarely a discernible or potential antecedent at all, contrary to the constructed exchanges above. Consider the attested examples presented in (30) and (31), for example, with their discourse context:

In some abstract sense, the subject of Why-VP in (30) is related to the mention of people in the provided context. However, there is really no discernible syntactic parallelism between people not to tell their friends in the context, and the VP of Why-VP take the chance that…, nor do these seem to stand in the kind of oppositional relation to each other that would warrant a contrastive focus interpretation, as would be predicted by the Yoshida et al. account.

The presence of possessive your in the example of (30) (bound by the implicit subject) indicates also that the nominal people from the previous context does not function as an antecedent for the implicit subject of Why-VP. Consider the oddness of (32):

Similarly, in (31) the subject seems pragmatically related to their in the clause immediately before the instance of Why-VP. But there is surely no way that Why-VP can be syntactically derived from their cost performance is much better as an antecedent.

Finally, and most strikingly, Why-VP occurs freely occurs with no preceding discourse context at all. Here, repeated, are naturally occurring examples found on flyers:

In the absence of any preceding discourse at all both examples of (33) are well-formed and interpretable. This is not the case for structures generally assumed to be derived via ellipsis, like Sluicing or VP Ellipsis:

In sum, Why-VP (both in terms of well-formedness and in terms of interpretation) seem to function independently of any discourse context in a way that is entirely uncharacteristic of ellipsis.¹⁸

3.2 Against VP-fronting

To be maintained, the ellipsis analysis must also hold that the VP of Why-VP undergoes movement (VP-Fronting) in order to escape from the ellipsis site. I consider that possibility here. Consider (35):

Under the ellipsis analysis it will be understood as in (36):

In (35), as in (36) the reflexive inside VP may not be bound by the matrix subject John, even though the VP which contains it, by hypothesis will have moved through the specifier position of the clausal complement, a position in which the reflexive it contains will be commanded by the matrix subject John, An apparently, analogous restriction holds of VP Preposing in the absence of ellipsis, as shown by Huang (1993) and is illustrated in (37):

If the failure in (37) is to be attributed to movement of VP, the apparently analogous observation for (36) might suggest that VP preposing also plays a role in the derivation of Why-VP structures. Such an argument is developed, briefly and tentatively, by Yoshida et al.

But there is a much more straightforward way of understanding these facts – by assuming simply that the VP of Why-VP structures does not raise at al; the VP of (36), for instance, never moves through the specifier of VP and is therefore never in a sufficiently local relation with the matrix subject to be bound by it.

Independent evidence that there is no VP fronting in Why-VP comes from the lack of island amelioration effects, which Sluicing and certain Why-Stripping (Why-DP) cases exhibit. Yoshida et al. show that there is such an effect in genuine cases of Why-Stripping, where, for instance what would otherwise be violations of the constraint banning extraction from relative clauses are ameliorated, as seen clearly in (38):

(38B) is interpreted as Why do they want to hire someone who speaks a Balkan language? It does not mean Why does someone speak a Balkan language? However, in Why-VP, the interpretation is so restricted and the interpretation which would require movement out of the island is not available:

In the exchange above, the only possible interpretation of (39B) is Why would someone write a paper about a Balkan language? An interpretation in which the island is ameliorated is impossible: Why would they hire someone who wrote a paper about a Balkan language? (39B) supports only the reading that we would expect it to have if it involves neither ellipsis nor movement of the VP.

On this understanding, we expect the results of both (35) and (39). There is no syntactic structure that the VP raises through, so there is no intermediate scope position in (35) from which the VP could possibly reconstruct. Further, there is no island out of which the VP remnant is raising so we would not expect the meaning of the VP to include the structure outside of the island domain, i.e. the relative clause. The “correlate” in (35A) and (39A), then, is an illusion; it is by pure coincidence that the VPs match, it is not by necessity.

3.3 Weir’s amendment

A revision to the Yoshida et al. analysis was proposed in Weir (2014), who argues that FocP, which would host the VP remnant in its specifier, selects VoiceP rather than TP and that VoiceP is the target of deletion rather than TP.

Following Hacquard (2006), Weir assumes that circumstantial modality and deontic modality are generated within vP, presumably below Tense and Aspect. He notes examples in which the context licenses circumstantial interpretations which are preserved in Why-Stripping and where epistemic modality is present in the previous context, but not interpreted in Why-Stripping. I provide the deontic case in (40a), the circumstantial in (40b), and the epistemic in (40c):

Weir addresses this issue by positing that no T layer exists in Why-Stripping, which means epistemic modals that scope above T are also eliminated, therefore predicting the contrasts above. An analysis like Weir’s would have the advantage of narrowing the possibility space for the modal interpretation, restricting it to root modals, which seems to be a step in the right direction. Note that in Why-VP epistemic interpretations are impossible:

The modal in Why-VP is most saliently interpreted as would in (41), which is not epistemic. Thus, Weir’s amendment comes closer to understanding the interpretive properties of Why-VP, but inherits most of the difficulties already raised for Yoshida et al. (2015) since it is equally committed to the ellipsis component.

Additionally, the ability interpretation of can in (40a) and (40b) suggests that Why-stripping fragments permit a wider range of modal interpretations than are observed in Why-VP, as Why-VP is never interpreted as involving the modal force of can. We will see in Section 6 that we can narrow the availability of particular modal interpretations even further in adopting an analysis which treats Why-VP as having the covert modal found in infinitival clauses.

Weir’s observations, however, are very important in that they contribute a new insight – namely that Why-VP and other kinds of truncated Why-XP questions are parallel in never being interpreted as involving epistemic modal flavors. Full and finite why-questions exhibit no such restriction. Notice, though, that Weir’s insight in no way depends on the ellipsis component of the analysis developed by Yoshida et al., nor on its appeal to VP-movement. I will in fact argue below that the real mysteries in Why-VP structures center on modality and its syntactic expression, not on their clausal syntax, which is more or less routine.

4.1 Binding

We will be concerned here mainly with Principle A, which we will understand in the simplified and conventional form seen in (42):

There are at least two kinds of anaphors that are restricted by Principle A: reflexives and reciprocals. There is evidence that both may occur in Why-VP constructions. Since Principle A is evidently satisfied in such configurations, we must conclude that Why-VP structures contain an appropriate local binder – one which c-commands the object anaphors we will soon observe.

4.2 Control

Why-VP fragments can subsume structures that are in general analyzed as involving Control. That is, they may contain infinitival clauses whose silent subjects are of the type usually analyzed as controlled – bound by a c-commanding element. Consider, then, examples such as (54)–(56):

In (54), there is a clausal adjunct headed by before which does not realize a subject but is nevertheless well-formed. The subject of having is interpreted as a variable bound by whatever is in the subject position of the matrix clause. The intuitive interpretation of (55) of course is, why would x choose (for x) to add more negativity to the world? That is, the subject of the lower clause is referentially dependent on the subject of choose in exactly the way familiar from other cases of control. But if obligatory control of the sort seen in (55) and (56) is always dependent on the presence of a structural binder which commands the embedded subject (see Landau 2013 for extensive arguments in favor of this position) then there must be such a structural binder in (55)–(56).

Once again, the generalization that emerges is that there is a structural subject capable of being an appropriate binder for embedded or controlled PRO.

4.3 Passive

If we accept standard analyses, the passive construction in English is derived by A-movement, where a DP complement to V is raised to the closest commanding specifier of T position. In passive structures, no external argument is projected (at least in what is assumed to be its canonical spec-vP position), and the internal argument of the verb is therefore a legitimate goal for a T probe, and is raised to its specifier position.

Passive is possible in Why-VP fragments, which suggests two things about the structure: one, there exists a voice head (we’ll assume it’s little-v) in these constructions (responsible for the appearance of passive morphology on the main verb), and two that there is enough structure to ensure that the internal argument of the verb can be promoted to a higher “subject position.” Consider the naturally occurring example below:²³

In Why-VP the argument which raises is silent, but in all other respects the syntactic mechanisms at work here seem to be just those at work in full clauses, implying not only that we need a voice (i.e., little-v) head as part of what we have been calling informally VP, but also that we need a projection like T above Voice within Why-VP – whose morphosyntactic properties are responsible for attracting the internal argument to its specifier position, just as in infinitival full clauses.²⁴

4.4 Quantifier float

Floated all implies the presence of a structural subject, either an overt DP which has been raised or a PRO which has been raised (Baltin 1995; Bobaljik 2003). Quantifier float is an available option in Why-VP structures (most clearly in a particular sub-type, as we will see); since the subject of Why-VP is always null, we must be dealing with cases in which the host which “floats” the quantifier is null PRO.

First let’s consider floated quantifiers licensed in certain infinitival contexts:

Why-not-VP structures, also, seem to be quite natural with floated quantifiers:

However, Why-VP and Why-not-VP seem to come apart at this point. Floated quantifiers appear to be more marked in Why-VP in the absence of negation. I was unable to find any fully convincing naturally occurring examples of Why-all-VP and my own intuition is that the relevant examples are at best awkward, though it has been suggested to me that (60c) sounds slightly better than the others:

A puzzle remains, then, concerning this contrast.²⁵ But the evidence from Why-not-VP cases at lest seems clear and from them I tentatively conclude that quantifier float is in principle available in Why-VP. There is reason to believe that PRO is only licensed in the subject position (Spec, T) of nonfinite clauses (Radford 2004). To the extent that such premises hold and to the extent that the control clauses in (58) and the Why-VP examples above are well-formed, the observations again point to the existence of a structural subject position – specifier of non-finite T specifically – and evidence that that position is occupied by PRO.

4.5 Summary

What has been established so far is that Why-VP structures are surprisingly routine as far as the syntax of subjecthood is concerned. There is evidence that they contain a subject position which can host binders for reflexives, reciprocals, and PRO, and which can be the target position for A-movement processes (notably passive and raising) which standardly target the specifier of T position. We conclude from all of this that there must be a null functional head above vP which hosts the subject of Why-VP in its specifier. These conclusions suggest a structure such as that in (61):

(61)

I will remain agnostic for now about the category that hosts the null subject in Why-VP, using the label F above to refer to it but I will note that the properties must be at least close to those usually attributed to the tense head of infinitival clauses.

With this much established, the next task is to decide how why fits into the structure in (61). Traditional analyses of why and other wh-words take them to be phrasal and hold that they undergo A’-movement to the specifier of C (or one of the heads defining the C-domain in frameworks which assume a more articulated structure). However, we will see in what follows that the why of Why-VP is different in a number of ways from the why of unreduced clauses, and that the facts therefore require a different analysis of why in the contexts we are concerned with here. We must also find a way to understand that the presence and the properties of that null functional head (F above) are contingent on the appearance of why. These and related questions are addressed in Section 5.

6.1 Interpretive properties

We start with the vagueness of the interpretation of the modal in Why-VP, which, intuitively, seems to be a close semantic correlate to the overt modals should and would. It is unclear, at present, what contextual factors or what kind of semantic composition governs the evocation of the interpretation for either of these modals in Why-VP – but given the same context, some speakers report should and some, would as the best paraphrase for the following example:

Both should and would being futurate and having universal force may make finding contexts in which only one interpretation is felicitous difficult. Additionally, the semantic composition of the modal may be underspecified allowing both interpretations rather freely. These are matters to explore in future work.

It is clear, though, that context plays an important role in determining which paraphrase emerges as most appropriate. Take, for example, the initial example presented in the paper, repeated here:

This naturally occurring example above occurs with no preceding linguistic context; it was found on a poster advertising why one should take this class. Accordingly, the best paraphrase here seems to be should. However, we can manipulate the context such that would becomes a more obvious paraphrase:

While would certainly seems like the best paraphrase of the Why-VP clause above, a paraphrase with should does not seem out of the question. A complete analysis of this modal will have to provide an understanding of this vagueness. For now, we can say that some contexts seems to favor certain interpretations of the modal in Why-VP.

Evidence that there is a connection between the interpretation of the covert modal in Why-VP and the overt modals we have just been discussing comes from their interaction with NPIs in why-questions. Recall that Bhatt (1998) observes NPIs are licensed in Why-VP:³³

In finite why-questions only certain modals, namely, should and would, appear to license NPIs, which are exactly the modals which seem to figure in the interpretation of Why-VP.³⁴ Consider:

If the interaction between NPIs and modals above is a semantic one, a constraint on interpretation, as seems likely, then we should presumably conclude that the semantics of the covert modal of Why-VP structures is close to, or identical to, the semantics of would and should, given that Why-VP also licenses NPIs. Bhatt (1998) attributes the NPI-licensing possibilities in Why-VP to its rhetorical nature, which is cashed out via a semantic composition specific to Why-VP, in which the provided positive polarity is flipped to be negative. However, given the observations in (85), we can see that no such technology is needed – we can attribute the NPI-licensing properties of Why-VP to its modal character, whose interpretation we have already identified as a close approximate to should and would.

6.2 Bhatt (1999/2008): Covert modality in non-finite contexts

We turn now to the syntactic characteristics of Why-VP that should lead us to expect this modal interpretation. There are parallels between the modal interpretation of Why-VP and the covert modal identified in Bhatt (1999/2008) characteristic of embedded wh-infinitivals – exploring these will likely be a good starting point to understanding the composition of the covert modal in Why-VP. It’s also clear that Why-VP clauses should be identified with non-finite clauses more broadly – there is evidence of a structural subject that is obligatorily silent, much like the subject of non-finite clauses, PRO. The VP is obligatorily tenseless, uninflected, with an irrealis interpreation, all of which are properties associated with non-finite clauses.

Bhatt observes that non-finite wh-clauses (wh-infinitivals) are always modal, with an interpretation that varies between should, would, and could. The following examples are adapted from that discussion.

The interpretation varies depending on various factors, including choice of wh-phrases, choice of embedding predicate, and certain elements of the discourse context. Noting in particular that the modal interpretation can vary in force (though not in flavor), between having a necessity interpretation (should or would), or in having a possibility interpretation (can/could), Bhatt wants to maintain one modal operator occurs in these kinds of clauses.

The covert modal that Bhatt proposes is one that is interpreted relative to a circumstantial modal base, thus ruling out the possibility of an epistemic interpretation. The force of the modal is one that is underlyingly existential with a deontic ordering source, though it might be better described as a teleological modal in present terms. The operator postulated draws on the denotation provided in Kratzer (1991) for a possibility deontic modal, except that a restriction is added that we only consider worlds in which the matrix subject has achieved the goal set by the context. It’s the existential force of the modal that is responsible for the could interpretation, and it is the added restriction that is responsible for the should interpretation. What determines whether one part of the denotation is relevant here depends crucially on the context.

For example, in (86a) should may emerge as the best paraphrase. This is due partially to the properties of the wh-phrase who but is also due to some knowledge about Zach’s goals given in context. If we know that Zach’s goal is to become popular, then in (86a), knowing who to talk to means knowing who it is necessary to talk to in order to achieve that goal. Bhatt’s restriction that we only consider worlds in which Zach meets his goals means that we only consider worlds in which he has also talked to a specific group of people.

In (86b), depending on the context, we will either get should or could. First imagine a context in which Charlie’s goal is simply to get gas. The wh-infinitival and the goal, in this case, match – Bhatt’s restriction is trivially satisfied – we only consider worlds in which Charlie gets gas. Any and all locations in which Charlie gets gas will satisfy his goal. When the restriction is trivially satisfied, the underlying existential force will take effect and we observe a can or could interpretation. Now, consider a context such that Charlie’s goal is to get gas from some ethically commendable source – there is now additional content in the goal that is not in the wh-infinitival – the restriction is no longer trivially satisfied. His goal will not be met in getting gas from any filling station he might happen on. In that context, it seems intuitively correct, that we understand should as the best paraphrase.

In (86c), the effect of the embedding predicate decide likely has to do with invoking the futurate interpretation of would – though not much more than this is said by Bhatt. The core insight though is that the interpretation of the modal depends heavily on the context of use.

Why-VP certainly has in common with embedded wh-infinitivals both the necessity interpretations should and would.³⁶ It will be difficult to fully import Bhatt’s modal to Why-VP, however, given that much of his analysis derives from his examination of contexts in which Why-VP may not appear – the complement position of factive know. Why-VP as we have seen, can never be embedded. Furthermore, it’s not clear how would is derived in his system, a central element of the modal meaning characteristic of Why-VP. However, that embedded wh-infinitivals and Why-VP have very similar interpretive effects should not be taken to be a coincidence, especially given that Why-VP independently exhibits many properties associated with infinitival clauses. It seems natural, now, to formally characterize Why-VP as non-finite – a characteristic I will localize as a feature on the head I call F. This will have the right syntactic effects – we will expect there to be a structural subject PRO in its specifier and a bare vP as its complement. In addition, we will predict a modal interpretation that is variable in precisely the way we observe (to the exclusion of could). The analysis proposed for Why-VP contributes to a larger generalization: non-finite wh-clauses are always interpreted as modal. We should now understand Why-VP to have the following structure:

(87)

Our proposals, however, will depart from Bhatt’s on the question of where the modal properties of the construction are localized. Bhatt assumes that the modal in embedded wh-infinitivals is hosted at C_{[+wh, –fin]}, correctly predicting that wh-infinitivals will always be modal. I have argued here that Why-VP is not a CP, rather, it’s a WhyP. It does not seem right that the semantics for why and the semantics for the modal be introduced by the same head; we would be forced to claim, for instance, that the semantics of why in Why-VP is different from its semantics in other (finite) contexts. This seems wrong. Instead, I’d like to propose that the modality comes from within the clause itself, specifically at F. The why of Why-VP as a head will select this modal; thus, the cluster of properties we have been exploring fall out from this syntactic relation.

The precise composition of that modal is still an open question, but it must be related to that which was identified by Bhatt. Positing the existence of this modal in Why-VP gives us a way of understanding the intuition that underlies many of the earliest investigations of Why-VP. The oft noted rhetorical nature of Why-VP is a property of being a why-question that contains a modal akin to should and would. Note that finite why-questions with these modals are able to have the directive force associated with Why-VP:

The attitude in the modal why-questions above is essentially that which was observed by Duffley & Enns (1996) to be characteristic of Why-VP: the speaker conveys that they see no reason that the prejacent of the modal should occur. This has the effect of conveying that one should not take the course of action offered in the prejacent, given that there is no good reason to do so.

We should promptly say, though, that this is not the exclusive function of these modals in why-questions. These modals can occur in why-questions and express a genuine wondering about the reasons someone should or would take some course of action – still here, the speaker sees no reason to VP, but they ask precisely because they see no reason to VP, not because they intend to instruct their addressee – see (90) and (91) below:

The modal why-questions above need not be interpreted as making a suggestion (though they could be) – they can simply ask a question pertaining to reason and necessity; the same goes for Why-VP. We can conclude from this, then, that it is the modal, argued to be present (and covert) in Why-VP that links the rhetorical and non-rhetorical uses. The precise syntactic characterization of that modal and its compositional properties are certainly worth exploring further, as there do seem to be certain syntactic and semantic configurations that the covert modal of Why-VP cannot participate in that is available to its overt modal counterparts, issues I will explore in the following section.

6.3 Aspectual puzzles

The covert modal of Why-VP and its overt counterparts come apart in surprising and interesting ways, which will be important to identify here for future work on this construction and hopefully, for work on covert modality in general. The covert modal of Why-VP appears to be far more limited than its overt counterparts in the range of auxiliaries that it can co-occur with. In particular, aspectual auxiliaries appear to be wholly excluded in Why-VP:³⁷

However, in finite why-questions with should and would, no such restriction exists:

One reason I have remained hesitant to adopt an analysis where, what I have called F is really a variant of non-finite T is because T is generally associated with the presence of aspectual markers; given that they are rejected in Why-VP, it’s not clear that what is generally associated with a tense head should exist in Why-VP.³⁸ Furthermore, as we will discuss below, it is not clear that the semantic properties associated with T, tense, should exist in Why-VP, particularly because have does not seem to be able to contribute a past temporal perspective. Such an effect is generally possible for priority modals in the context of have.

An analysis of the covert modal in Why-VP, its compositional properties and structural representation, will need to account for this puzzling fact. It will not do to simply assert that the covert modal of Why-VP is a silent version of the operator associated with the lexical items should and would. Modals are associated in general with futurate or irrealis interpretation, so much so that it is common to assume that the future orientation is built into the denotation of modal operators. However, that does not mean that aspects of their temporal properties can’t be shifted to the past with have. This is most famously explored in Condoravdi (2002), and more recently by Rullmann & Matthewson (2018).

Condoravdi distinguishes between temporal orientation and temporal perspective, which are crucially distinct both in their syntactic and semantic representations. Temporal perspective is contributed by T, while temporal orientation is contributed either by an aspectual operator scoping below the modal, or is built into the composition of the modal itself. Semantically, the temporal perspective refers to the evaluation time of the modal with respect to the utterance time, while temporal orientation refers to the event of the modal’s prejacent with respect to the utterance time.

Condoravdi argues that depending on the flavor of the modal, auxiliary have will affect these temporal dimensions differently. For example, have in the context of an epistemic modal will shift the temporal orientation to the past (but not the perspective), while for metaphysical modals, have shifts the perspective to the past.³⁹ In the basic case, a modal utterance without have, the future orientation we observe is part of the modal’s denotation. Thus, there are two ways temporal orientation manifests under this view: either the denotation of the modal contributes it or an aspectual auxiliary (e.g. have), scoping below the modal, contributes it.

On the other hand, Rullmann & Matthewson (2018) argue that modals uniformly get their temporal orientation from an aspectual operator that scopes below the modal. In English, this non-past (future/present) aspectual operator is covert, while overt for the past aspectual operator, realized as have.

Let us assume that the modal of Why-VP is of a teleological flavor, essentially as was proposed by Bhatt for wh-infinitivals. While certain modal flavors are incompatible with a past orientation, e.g., metaphysical modals and pure circumstantial modals, Thomas (2014) argues that teleological modals are not incompatible with a past orientation. If that’s true, the modal of Why-VP then should not be semantically incompatible with have.

It may be tempting at this point to attribute the unavailability of have in Why-VP to the directive force associated with its use. It is infelicitous to suggest or request a course of action in the present take place in the past. As an anonymous reviewer points out, these aspectual restrictions are reminiscent of imperatives. However, champions of this view will need to contend with the genuine questions identified here – there should be no reason that one cannot ask a question about the past (see (94) above).⁴⁰

We might have recourse to a syntactic analysis along the lines suggested by Rullmann & Matthewson (2018) – the modal of Why-VP selects only the covert future/present aspectual operator. This would predict that Why-VP should be able to have both present and future orientation, provided we have the right lexical aspect (stative for present and eventive for future). This prediction is borne out:

An anonymous reviewer suggests a close alternative: the modal of Why-VP selects vP directly. On this view, the passive auxiliary is ruled in (presumably hosted at v), and higher auxiliaries, have and progressive be are ruled out.⁴² While it is clear that we have the theoretical tools to capture this observation, we cannot yet understand why the covert modal must be deficient in this way.

There seem to be restrictions on the interpretation of the covert modal that simply don’t hold for its overt lexical counterparts. This general finding is not a novel one – Hackl & Nissenbaum (2012) find several contexts in which the covert modal posited for infinitival relatives (e.g., many things for us to do) in which it is restricted to one particular modal force, despite being variable in force in other contexts. They consider infinitival relatives in the context of strong and weak determiners – when the head of the relative clause is weak, should and could interpretations are possible, while for strong determiners, the covert modal can only be interpreted as should. However, finite relative clauses show no such restriction – there is no ban on the overt modal could occurring in the context of a strong determiner. This is very puzzling – the covert modal can in principle express existential force, but seems to be blocked from doing so in certain contexts.

There is likely something about being syntactically non-finite and its interaction with the surrounding context that limits the interpretive possibilities. It seems to me worth exploring further the commonalities among these covert modals and where the source of these interpretive restrictions might lie – are they underspecified in their denotation in a way that overt modals are not? And if so, in precisely what way? Or does the syntactic frame in which they occur contribute these deficiencies? What would that mean for the structural decomposition of modality? And ultimately, can these covert modals be unified in their composition in some way? These are questions I leave open, but I hope to have established that they are very much worth probing further, especially as the empirical domain of covert modality expands.