What do I know?!
Reflections on a rhetorical question

The English rhetorical question what do I know?!, with emphasis on I, looks at first blush like a garden-variety wh-question in terms of its syntax, apparently featuring what as the direct object of know. What may enhance this impression is that Dutch weet ik veel?! ‘know I much/a lot’, which has the same rhetorical function, appears to have the quantifier veel as the verb’s object in situ. But the Hungarian equivalent of these rhetorical questions, mit tudom én?! ‘what.acc know.1sg.def I’, suggests clearly that mit ‘what’, despite its accusative case, is not the verb’s direct object. Regular wh-objects in Hungarian trigger indefinite agreement on the finite verb. Indefinite tudok is grammatical but the result can only be interpreted as a genuine, non-rhetorical self-addressed question; in its rhetorical use, we find definite agreement (tudom). In this short paper, I will subject this rhetorical question type to an exploratory examination aimed at showing that, rather than being an argument of the verb, the wh-expression/quantifier can serve an adverbial function in the verb’s projection.

1 What is not the object

English (1) is formally a question, but it is typically used in contexts in which it does not invite an answer: while (1) could of course be used as a self-addressed question in a monologue intérieur, when it is used in a conversational context it usually does not serve as a request for information but rather as an emphatic affirmation of the fact that the speaker has no clue about the matter brought up by his or her interlocutor.

Thus, (1), especially with a heavy pitch accent on I, is usually a rhetorical question — one that, on its face, looks just like a regular wh-question in the syntax, with what apparently serving as the wh-object of know, raised into the left periphery as in regular wh-questions.

But if what in (1) had to be mapped into the verb’s direct object position, one would be hard pressed to deal with the fact that the verb know can actually select an interrogative clause as its object in this type of rhetorical question, as in (2).(2) reproduces a number of tokens found on the internet through a search for the string “what do I know whether”. Whether interrogative complement clauses introduced by other wh-items are possible as well is something I have not systematically investigated. It may be significant that for the string “what do I know if” there are very few (if any) clear hits involving the rhetorical use of what do I know. If indeed if cannot replace whether in examples of the type in (2), this may be revealing regarding the structural position of the subordinate clause: in particular, thinking of the fact that whether does not alternate with if in subject questions (whether/*if he will win is unclear), we may want to treat the subordinate clauses in (2) as a specifier of some projection of the verb. But whether it be a complement or a specifier, the interrogative clause in (2) is definitely an argument of know. Given that know assigns at most one internal θ-role, this entails that what cannot be treated as an argument of the verb.

Apparently, the presence of what does not block the merger of an object clause, which must mean (given that know is a monotransitive verb) that what in (2) is not itself the verb’s direct object.

The Dutch equivalents of English (1), illustrated in (3), do not feature a wh-operator. They do have the syntax of a question, however — of an apparent yes/no-question this time, with its characteristic verb-initial order.

Even more so than in the case of English (1), the constructions in (3) give the impression that the verb takes the quantificational, pronominal or demonstrative element as its object, thanks to the fact that veel ‘much’, ’t ‘it’ and dat ‘that’ occur clause-internally (unlike what in (1)). But as in the case of (1), it is possible in Dutch (3a, b) to add an interrogative clause to weten ‘know’, as illustrated in (4a, b). There is also a past-tense variant of (4a) (though not of (4b)), with a declarative clause as the object: (i). To this, English has a close counterpart in the form of (ii).
(ia) wist {ik/jij/hij/zij} veel dat er iets vreselijks stond te gebeuren?!
knew I/you(sg)/he/she much that there something terrible stood to happen
(ib) wisten {wij/jullie/zij} veel dat er iets vreselijks stond te gebeuren?!
knew we/you(pl)/they much that there something terrible stood to happen
‘how could {I/you/(s)he/we/they} have known that something terrible was about to happen?’
(ii) little did {I, you, (s)he, we, they} know that something terrible was about to happen
English (ii) is interesting for the fact that downward-entailing little licenses an NPI in the embedded clause:
(iii) little did {I, you, (s)he, we, they} know that anything terrible was about to happen
The grammaticality of (iii) also confirms that the that-clause is an argument of the verb: it is impossible for little to license a polarity-sensitive element in an adverbial clause (little did we know [when {someone/*anyone} told us about it] that…). This again suggests that in (4a, b), veel ‘much’ and ’t ‘it’ are not the verb’s object.

The hypothesis that veel and ’t are being used non-argumentally in (4a, b) also holds the key to an explanation for the ungrammaticality of (4c), with demonstrative dat ‘that’: if the particular non-argumental use of veel and ’t in (4a, b) is one that dat does not support, dat can only be mapped into the object position of the verb, precluding the addition of an interrogative clause. The demonstrative dat has an established non-argumental use as a pro-predicate: Jan is slim en Marie is dat ook ‘Jan is smart and Marie is that too’. If the non-argumental use of the (wh-)pronominal or quantificational element in our rhetorical constructions were a predicative one, therefore, we would not have an account for the ill-formedness of (4c). In section 2, I will argue that the particular non-argumental use of the (wh-)pronoun or quantifier in the rhetorical constructions under discussion is an adverbial one. The fact that dat is never used as an adverbial modifier then makes (4c) fall into place. I will return to this in section 4.

That the pronoun/quantifier is not the object of the verb is suggested particularly strongly by the agreement facts of Hungarian rhetorical questions of the type in (1). Consider the examples in (5), where (5a) is a non-rhetorical self-addressed question and (5b) is the Hungarian equivalent of English (1).

Hungarian has two forms of finite transitive verbs, an indefinite (or ‘subjective’) and a definite (or ‘objective’) form. The distribution of these forms in simple transitive clauses is a function of the definiteness specification of the object. The fact that the verb is inflected for indef agreement in (5a) is entirely unsurprising: mit is the verb’s object, and it is indefinite. But the definite inflection in (5b) tells us that mit should not be treated as the object of tud ‘know’ in (5b), for otherwise def agreement would be entirely mysterious.

The mystery of the source of definiteness inflection in (5b) will be solved once we find something that could control def agreement. In (6b), where mit co-occurs with a complement clause, we have something that can plausibly be held responsible for def agreement on the verb: the finite complement clause, which we know can trigger def agreement. If the accusative case-marked indefinite wh-pronoun mit were the object of the verb and checked structural accusative Case against it, the presence of def agreement in (6b) would be entirely unexpected. Note that it is in fact possible for mit to trigger indefinite agreement on the matrix verb even in the presence of a finite complement clause, as in the wh-scope marking construction in (6a), where mit₁ is marks the scope of of mit₂, which is itself stuck in the embedded clause. But when upstairs mit₁ controls agreement in the matrix clause, we lose the rhetorical ‘I haven’t a clue’ reading — (6a) has the semantics of a self-addressed long-distance wh-question, with thematic mit₂ scoping over the matrix verb. This reading is entirely different from that of rhetorical (6b), where mit₂ scopes under the matrix verb. Concomitantly, the syntax of (6b) must be different from that of (6a). In (6a), upstairs mit₁ checks Case against tudok and controls agreement with it, as a proleptic object — a ‘herald’ of the embedded clause. But in (6b), rather than mit₁, it is the embedded clause itself that serves as the object, controlling def agreement.

For (6b), we now have a goal for the verb’s definiteness feature: the object clause. But what could be the controller of def agreement in (5b), where there seems to be nothing besides mit that could take care of this? A familiar strategy in generative syntax is to address this kind of puzzle with the aid of an abstraction: we postulate a silent definite object — a pro. This will allow us to get the morphosyntax of (5b) to come out right: tudom requires a definite object; mit cannot be that object; but pro will serve the purpose perfectly. Of course the postulation of a silent definite object will only deliver an explanatorily adequate perspective on (5b) if there is some independent way in which we can support it. Interestingly, it turns out that this theoretical postulate is not a complete phantom: it is in fact possible to have an overt azt in the complement of the verb in rhetorical questions of the type in (5b), as in (7).

The grammaticality of (7) is perhaps the clearest indication one might desire that mit in (5b) is not an argument of the verb: azt is evidently the object of tudom, so mit must be something else. In English or Dutch, it is impossible to add some pronominal or demonstrative element to rhetorical questions like (1) or (3). But for (1) and (3a, b) as well, we have been led to believe that the (wh-)pronoun or quantifier does not (always) serve as an argument of the verb. This said, now of course the question that becomes urgent is what the (wh-)pronoun or quantifier could be. I will turn to this next.

2 If what is not the object, then what?!

Though it is impossible in English (1) and Dutch (3) to add some additional nominal element, there are constructions similar to (1) and (3) in which English and Dutch can mimic Hungarian (7) closely. Consider the rhetorical questions in (8), for all three languages under investigation. The Hungarian example is once again the most transparent one, featuring both mit ‘what’ and the experiencer, engem, in the accusative.

This ‘double accusative’ pattern reminds one of constructions of the type in (9c), in which a ‘regular’ accusative object, újságot ‘newspaper’, and an accusative-marked adverbial modifier, többet ‘more’, co-occur in a single clause headed by a monotransitive verb.Csirmaz (2006:187) rejects (i); this is more a matter of lexical choice and linear order (✓János sokat olvas könyvet ‘János much.acc reads book.acc’) than compatibility of accusative modifiers and objects per se. (ii) is an attested example of the type in (9c). The accusative adjective in (9c) passes the extraction test applied in Den Dikken (to appear, Chapter 3): (iii) is grammatical, albeit somewhat marked. (Thanks to Éva Dékány for her help with these examples.) The grammaticality of (iii) indicates, in light of the analysis developed in Den Dikken (to appear), that the adverbial modifier headed by többet checks structural accusative Case against v.
(i) *János nagyot könyvet olvasott
János large.acc book.acc read
intended: ‘János reads books a lot’
(ii) ha valaki többet olvas újságot, mint amennyit tévézik, …
if someone more.acc reads newspaper.acc than s.much.acc watches.tv
(olvasas.opkm.hu/index.php?menuId=442&action=article&id=401)
(iii) kinél olvasol többet újságot?
who.to read.2sg more.acc newspaper.acc

For constructions of the type in (9b), Csirmaz (2006) argues explicitly and in admirable detail that the accusative adverbial többet has structural accusative Case. The fact that such accusative adverbials tolerate an accusative-marked direct object by their side, as in (9c), indicates that the purveyor of the structural accusative Case feature (‘little v’ in current parlance) is capable in principle of having its Case feature engage in one-to-many relations. This is theoretically very interesting — but I will set the theoretical questions aside for lack of space. What matters for our purposes here is the conclusion, emerging from (9b) and (9c), that adverbial material can check structural accusative Case.

This takes us one step closer to the answer to the question of what the (wh-)pronoun or quantifier might be doing in rhetorical questions in which it cannot be mapped into the verb’s object position. These elements could be very much like the accusative adverbial többet in (9b, c): structurally Case-marked non-arguments. For (8), this is immediately plausible: the wh-word in (8) alternates with bare nominal adverbials, as in (10):

From (10b) we learn that veel ‘much’, the same element we found in rhetorical (3a), occurs in the non-interrogative counterpart to (8b).But unfortunately, non-argumental veel does not normally combine with stative verbs such as weten ‘know’. The fact that (ib) is ungrammatical thus muddies the waters for the hypothesis that veel in rhetorical (ia) is being used adverbially.
(ia) weet ik veel wat hij uitgespookt heeft?!
know I much what he out-ghosted has
‘what do I know/heaven knows what he’s been up to’
(ib) *ik weet veel wat hij uitgespookt heeft
I know much what he out-ghosted has And in (10c) we see semmit ‘nothing.acc’, which is morphologically built on mit ‘what.acc’, being used non-argumentally, with the accusative Case we also see on többet ‘much’ in (9b, c).

Though semmit in (10c) is based on mit, it is not a question word. So the next logical step in the analysis of (5b) is to link the adverbial use of accusative-marked material directly to interrogative mit itself. We can do this with the help of examples such as those in (11), where mit ‘what’ is essentially equivalent to miért ‘why, what for’.

Transitive (11b) is particularly interesting because it shows, once again, that mit, despite its accusativity, is not the object of the verb: what serves as the object here is clearly azt a széket ‘that chair’, which controls definite agreement on piszkálod. We have here, it seems, a very close relative to the use of mit in rhetorical (5b) and (7). It serves us well in arguing for the adverbial status of mit in these Hungarian sentences.

The picture for the other two languages under examination is not quite so simple. While Dutch also features the adverbial use of the wh-word corresponding to English what seen in Hungarian (11) (wat sta je daar nou? ‘what stand you there now’), wat is peculiarly absent from the set of pronouns/quantifiers figuring in (3). I will come back to this below. And while what does show up in English (1), there is no (productive) use of what as an adverbial element in constructions of the type in (11): what are you standing there *(for)? is bad without for (but it is probably relevant that that in null-operator relatives, dropping for is possible: the reason I was standing there).

Still, its indefinite use in the form of somewhat, illustrated in (12), suggests that what does have adverbial functions. And not surprisingly, somewhat is compatible with the presence of an object clause (see (12b)), just as what is in (2).Jespersen (1961:Vol. V, p. 498) says of ‘the familiar phrase I’ll tell you what’ that ‘what has approximately the meaning of an indefinite pronoun; cf somewhat’. It is telling, however, that he treats this construction, however briefly, in a section entitled ‘Incomplete Questions’ — a characterisation that strikes me as entirely appropriate for this kind of expression: rather than functioning as an indefinite pronoun, what here seems to introduce an elliptical wh-clause (a ‘sluice’). The same is likely to be the correct analysis for you know what?, which, as Jespersen (p. 498) points out, goes back at least to Chaucer (wite ye what ‘know you what’) and Shakespeare (wot you what ‘know you what’). This latter type is found in present-day Dutch as well: weet je wat? A third ‘incomplete question’ (technically, another sluicing case) belonging to the same family of constructions is guess what. These are all structurally robustly different from the rhetorical question under discussion in the main text.
Jespersen (1961:Vol. V, p. 500) also mentions a use of what that seems potentially quite closely related to that seen in (1):
(ia) what do you say we go out to-night? (Ernest Hemingway)
(ib) what d’ye say we head for the coast? (Jack London)
(ic) what do you say we stop a few days? (Jack London)
Jespersen classifies these as ‘pseudo-questions’ of a type found in American English only. Perhaps surprisingly, Jespersen does not mention rhetorical questions of the type in (1) in his brief discussion of ‘pseudo-questions’. It may seem plausible to think that there is a connection between the use of what in (i) and that in (1), esp. if the clause following say in (i) is the object of say — but that is hard to ascertain: it may very well be that we are dealing here with two paratactically rather than hypotactically related clauses (cf what do you say? shall we go out tonight?, or what do you say to the following: we go out tonight). Though it may be possible to think of (i) as structurally related to (1), too little is known about the syntax of (i) and especially about the function of what to make it possible to say anything with confidence in this connection.

Perhaps related as well is the peculiar use of what in combination with with in expressions such as the following, ‘used to talk about the reasons for a particular situation, especially a bad or difficult situation’ (Cambridge Dictionary on-line):

There is no obvious function for the what of what with S: it seems rendundant, in light of the fact that a semantic contribution for it is hard to discern, and the examples in (14) remain grammatical with what dropped. Its optionality is compatible with what being an adverbial modifier of sorts, though it remains unclear exactly what kind of modifying role it could play here, and how that role (whatever it be) could be related to the one played by what in (1).

3 What’s more: A null operator

If the previous section makes it reasonable to think that the (wh-)pronoun/quantifier in rhetorical constructions with know can have an adverbial function, what is left to address is the fact that there is more to these rhetorical constructions than the adverbial use of a (wh-)pronoun or quantifier. In particular, what is striking about these sentences is that we systematically see the activation of a functional projection in the left periphery, resulting in fronting of the adverbially used wh-pronoun and/or the inversion of the subject and the finite verb. The examples in the right-hand column are either simply ungrammatical or, to the extent that they are well-formed, do not support a rhetorical reading equivalent to that of the examples in the left-hand column. This is what the asterisks are meant to show.

For English and Hungarian, the ungrammaticality the right-hand examples in (14) is not difficult to understand: wh-operators resist being in situ except in special registers (such as ‘echo questions’) which are irrelevant here. But the Dutch case is very interesting because of the fact that the quantifier veel itself is not being fronted — and in fact cannot be fronted: *veel weet ik?!. Nonetheless, it is definitely the case that something has fronted to the immediately preverbal position: Dutch is a Verb Second language; in all root sentences in which the verb is in surface-initial position (including yes/no questions and topic-drop sentences), it is arguably the case that the preverbal position is occupied by something — a null operator.

I would like to argue that the null operator in Dutch (14b)=(3a) is a modifier of veel, originating within the projection of veel and stranding veel under movement to sentence-initial position. Relevant in this context is the syntax of the exclamatives in (15), where the operator combining with veel is overt — and realised, interestingly in light of the English facts, as wat ‘what’.

In (15) wat veel ‘what a lot’ originates as a constituent, serving as the object of the verb weet ‘know’. It can stay together as a constituent and front as a unit, as in (15a). But in (15b), wh-fronting affects just the wh-operator wat, causing veel to be left behind clause-internally and producing a pattern similar to the one seen in (3a).

It is now plausible to hypothesise that (14b) involves splitting in the same way as (15b), and that what is raised into the left periphery of the clause in the derivation of (14b) is a null counterpart of the wh-operator wat ‘what’. This makes (14b) an instance of ‘wh-drop’, a phenomenon known independently to be possible in Dutch, esp. in rhetorical questions (which often feature the discourse particle nou) — including, most relevantly, cases in which wat has an adverbial function, as in (16b), the counterpart to Hungarian (11a).The link with ‘wh-drop’ is I think enhanced by the fact that (14b)=(3a) can be reduced in speech to just tik veel?! (where the t is the coda consonant of weet), just like (16a) can be reduced to bie nou gedaan? (where b is the coda of heb, and ie is reduction of je contracted onto heb — cf hebbie ‘have you’).

While exclamative (15b) alternates with a pied-piping derivation, as in (15a), pied-piping of veel is impossible in the case of (14b): *veel weet ik?! is bad. This is in line with the hypothesis that (14b) involves a null operator: for null operators we know that they generally cannot partake in pied-piping, as shown, for instance, by the contrast in (17).

It remains somewhat mysterious why overt wat does not occur in rhetorical questions of the type in (3), either in combination with stranded veel or on its own. Put differently (from the perspective of the analysis proposed), it is unclear why ‘wh-drop’ is apparently obligatory in (14b)=(3a). It is certainly true that rhetorical questions lend themselves particularly well to ‘wh-drop’. But I am not aware of any rhetorical wh-questions in Dutch other than (14b)=(3a) in which ‘wh-drop’ is an absolute requirement.

4 What’s up? The analysis in a nutshell

With (18) as the derivation for the exclamative with splitting in (15b), we end up with (19a) as the derivation for (14b)=(3a). The QP-constituent formed by the null operator and veel occupies an adverbial adjunction position, shown in (19b), when the object position is taken by a clausal object, as in (4a) (recall section 2).

For the variant of (3a) featuring pronominal ’t, given in (3b), I would like to tentatively suggest an analysis along similar lines, with the null operator this time stranding ’t — the coda consonant of both the pronoun het and the wh-word wat.

The treatment of ’t as a remnant of null operator movement produces a split wh-operator, with the (silent) wh-portion up in SpecCP and the restriction (-t, for [+neuter]) in the extraction site. Such splitting is certainly unusual: Dutch does not generally go about leaving ’t behind under wh-fronting, so why this does seem to happen in (3b) needs to be investigated further. But if this is indeed what is going on in (3b), we can understand why (3c) behaves differently. Demonstrative dat in (3c) evidently cannot be treated as the leftover bit of a split wh-quantifier: dat does not combine with any wh-operator; while wa+’t produces grammatical wat, *wa+dat is impossible. The fact that we cannot base-generate dat together with a null operator in a single constituent means that dat must occupy a position separate from the trace of the null operator in SpecCP. Since dat has no adverbial uses, this in turn means that, while the null operator can serve an adverbial function, dat must be mapped into the verb’s object position in (3c): see (21). And this allows us to explain the ill-formedness of (4c), as anticipated above: with dat occupying the object position, there is no room for an additional object clause.

In English and Hungarian, the operator in SpecCP is overt (what in English, mit in Hungarian), and it does not strand anything inside VP. Another difference between Dutch, on the one hand, and English and Hungarian, on the other, concerns the distribution of the adverbial adjunction strategy. In Dutch, this strategy can be exploited only when something else occupies the object position: weten ‘know’ licenses neither Unspecified Object Deletion (as English know does: oh, I know!) nor pro-drop (as does Hungarian tudom ‘know.1sg.def’). In English and Hungarian, on the other hand, the wider availability of a silent object for know/tud makes it possible (and apparently desirable) for the operator to be constistently mapped into the structure as an adverbial adjunct, with the verb’s object position being unprojected (as in the case of English UOD), occupied by object-pro (in Hungarian (5b)), or taken by an object clause (as in (2) and (6b)).

The structures in (22) and (23) complete the picture for the rhetorical wh-questions in (1)/(2) and (5b)/(6b).

Dedication

It is with great pleasure and admiration that I present this paper to Ádám Nádasdy on the occasion of his 70th birthday and retirement from the Department of English Linguistics at Eötvös Loránd University. Ádám’s generous offer to retire early if this could have helped to expedite my appointment to the Department is a deeply moving gesture revealing his support for me and his devotion to the cause of English linguistics at ELTE. Thankfully, in the end my appointment came through without Ádám’s offer being needed, which made it possible for him to serve the Department all the way until his legally mandated retirement. I wish Ádám all the very best in the years to come, spent in the company of wonderful literature (to enjoy and to translate for others to enjoy), music, people, and everything else that makes life good.

References