In situ mixed wh-coordination and the argument/adjunct distinction

One of the most important results of syntactic inquiry has been a detailed empirical and, to some extent, theoretical understanding of the argument/adjunct distinction, which underlies a wide array of superficially different phenomena. Therefore, any phenomena that appear to challenge the argument/adjunct distinction deserve scrutiny. This squib investigates an almost unremarked-upon phenomenon of just that type: apparent in situ mixed wh-coordination (ISMW: Mary ate what and when to impress Sue?!), in which argument and adjunct wh-phrases are apparently coordinated in situ. Two analyses of ISMW are compared: the Wh-Coordination Analysis, on which the conjuncts are the wh-phrases, and the VP-Coordination Analysis, on which the conjuncts are VPs whose head Vs undergo across-the-board head movement to v. The squib argues for the VP-Coordination Analysis on conceptual and empirical grounds. Conceptually, the VPbut not the Wh-Coordination Analysis is compatible with our understanding of the argument/adjunct distinction, and involves an unremarkable derivation that it would take a stipulation to rule out; hence, the VP-Coordination Analysis is preferable. Empirically, the VPbut not the Wh-Coordination Analysis makes several correct predictions: (a) that ISMW should be impossible with obligatorily transitive verbs; (b) that adverbs should be able to follow the first wh-phrase in ISMW that cannot follow it in the left periphery; and (c) that there should be apparent in situ coordination of argument wh-phrases with different θ-roles. That ISMW involves VP-coordination rather than wh-coordination indicates that it in fact does not threaten the argument/adjunct distinction, contrary to initial appearances, a theoretically welcome result.


Introduction
One of the most important results of work in syntax has been a detailed empirical and, to some extent, theoretical understanding of the argument/adjunct distinction. Thus, it is now known that arguments and adjuncts differ systematically along multiple dimensions  (Pesetsky 1991;Merchant 2019).
For many more argument/adjunct asymmetries, some seldom discussed, see Toivonen (n.d.). The argument/adjunct distinction, then, underlies a wide array of superficially very different phenomena, indicating that it runs deep in syntax, as is generally acknowledged. 2 This being so, any phenomena that appear to blur or challenge the argument/adjunct distinction deserve careful scrutiny, to determine whether our understanding of it needs revising, and what their etiology is. This squib investigates a phenomenon of just that type: apparent in situ mixed wh-coordination. It proceeds as follows. Section 2 sets the stage by discussing (left-peripheral) mixed wh-coordination, in which an argument and an adjunct wh-phrase appear to be coordinated, and reviews one existing analysis of it, on which it actually does not challenge the argument/adjunct distinction. Section 3 introduces the main phenomenon investigated here-apparent in situ mixed wh-coordination (ISMW)-and shows that it initially appears more problematic for the argument/adjunct distinction than its non-insitu counterpart.
Section 4 develops two analyses of ISMW: the Wh-Coordination Analysis (on which the conjuncts are the wh-phrases) and the VP-Coordination Analysis (on which the conjuncts are VPs whose head Vs undergo across-the-board movement to v). It is shown that the VP-but not the Wh-Coordination Analysis is compatible with our understanding of the argument/adjunct distinction, and involves an unremarkable derivation that is difficult to rule out; consequently, the former is conceptually preferable. That section then tests some empirical predictions of the two analyses, showing that, in all three cases considered, the VP-but not the Wh-Coordination Analysis makes the right prediction. Section 5, the conclusion, lays out the main finding: that a phenomenon (ISMW) that initially appears seriously problematic for the argument/adjunct distinction actually is not, a theoretically welcome result.
This would be problematic for the argument/adjunct distinction if, in such mixed whcoordinations, either the adjunct wh-phrase never occupied an adjunct position or the argument wh-phrase never occupied an argument position. For example, if what and when in (2-a) originated as the complement of V and moved to [Spec,CP], then at no derivational stage would when be adjoined to anything, contra the standard view that when is an adjunct. Fortunately, though, a number of analyses of mixed wh-coordination avoid that consequence. To mention just one, Citko & Gračanin-Yuksek (2013) argue that, in English mixed wh-coordinations, the conjuncts are actually CPs. In (2-a), what originates as the complement of V, and when originates in an adjunct position, as usual. Then, what moves to [Spec,CP] in one of the CP conjuncts, and when moves to [Spec,CP] in the other CP conjunct. The reader is referred to their paper for detailed discussion. 3 What is important here is that, despite appearances, mixed wh-coordinations like (2a-b) do not seriously threaten the argument/adjunct distinction, because they do not force the conclusion that certain "adjunct" wh-phrases never occupy adjunct positions (at any derivational stage), or the parallel conclusion about "argument" wh-phrases.

The phenomenon: Apparent in situ mixed wh-coordination
There is, however, another type of mixed wh-coordination (almost unmentioned in the literature, though see Grosu 1985: 236-237 for two possible examples, and Kasai 2016 on Japanese) that initially appears much more problematic for the argument/adjunct distinction. In this type, an argument and an adjunct wh-phrase appear to be coordinated in situ ((3)-(4)). This phenomenon will be referred to here as apparent in situ mixed wh-coordination (ISMW-apparent because it is argued below that the conjuncts are actually VPs Sentence (3-a) therefore cannot have a structure along the lines of (6): its structure must be as in (7). 5 But this raises the question of how an argument and an adjunct wh-phrase can (apparently) be coordinated in situ. The question is important, because (7) gives the impression that perhaps when is not in an adjunct position, or what is not in an argument position, contra what the argument/adjunct distinction leads us to expect.

A note on the crosslinguistic landscape
Before we analyze ISMW, it will be worth noting that ISMW also occurs in (at least) Kuria, Dholuo, Spanish, and German ((10)- (13)). This indicates that ISMW (along with the apparent problem it poses for the argument/adjunct distinction) is not an Englishspecific quirk, but a phenomenon of much broader crosslinguistic import. 6 (10) Kuria ( (iii) Mary ate what-and when-to impress Sue?! Secondly, (i) could involve CP-coordination, with multidomination of everything except what and when. But this predicts that the putatively multidominated material should be linearly contiguous (Citko & Gračanin-Yuksek 2013: 4, (6c)); in fact, some of it precedes what and when and the rest follows it. 6 Mandarin lacks ISMW, however (Zhang 2007(Zhang : 2138. Examples (10)-(11) unfortunately lack a clause-final adjunct (e.g., 'to impress Mary'), so they may involve CP-coordination with sluicing (cf. (6)

Two possible analyses: Wh-coordination and VP-coordination
We can now ask how ISMW sentences are derived-and, in particular, why they seem to allow coordination of argument and adjunct wh-phrases. Consider again the following example (= (3-a)): 7 (14) Mary ate what and when to impress Sue?! Two possible analyses of ISMW are the following: The relevant (partial) derivations follow: 8 (16) Partial derivation of (14): Wh-Coordination Analysis 7 No clause-final adjunct is included here, since the clause-finality of V rules out a CP-coordination-plussluicing derivation. 8 The VP-Coordination Analysis is briefly proposed by Zhang (2007Zhang ( : 2136 for examples like (i): (i) Grosu (1985: 232) John writes only funny letters and only to funny people.
This squib extends the VP-Coordination Analysis to ISMW, arguing for it and against the Wh-Coordination Analysis on a number of conceptual and empirical grounds. Coordinate structures are assumed here for concreteness to be ConjPs (Johannessen 1998, a.m.o.), but little to nothing will hinge on this.
(17) Partial derivation of (14): VP-Coordination Analysis All else being equal, the VP-Coordination Analysis predicts that there should be free apparent in situ argument/adjunct coordination with non-wh conjuncts. This is incorrect: (18) *Mary ate oysters and at strange times.
But all else is not equal. Grosu (1985) argues that these structures are subject to a semantic constraint: the conjuncts must "include focused operators of a common type" (231). When they do, these structures become acceptable: (19) a. Mary ate only oysters and only at strange times. b. Mary ate the most exotic oysters and at the strangest times.
The source of this constraint deserves further investigation. Now, with the Wh-and VP-Coordination Analyses spelled out, we can begin to determine which is superior. The discussion will proceed as follows. Section 4.1 argues that the VP-but not the Wh-Coordination Analysis is compatible with our understanding of the argument/adjunct distinction, and involves an unremarkable derivation that it would take something special to rule out-and therefore the former is conceptually preferable. Section 4.2 argues on several grounds that the VP-Coordination Analysis is also superior empirically.

Conceptual considerations
The VP-Coordination Analysis, but not the Wh-Coordination Analysis, is compatible with our current understanding of the argument/adjunct distinction. Consider why.
On the VP-Coordination Analysis (17), what is the complement of V in the first VP conjunct, and when is adjoined to V(P) within the second VP conjunct. That is, what and when respectively originate in an argument position and an adjunct position. This is perfectly compatible with the standard view that what and when are an argument and an adjunct Zyman: In situ mixed wh-coordination Art. 30, page 7 of 13 wh-phrase, respectively, and thus with our understanding of the argument/adjunct distinction.
Viewed against the background of the argument/adjunct distinction, the Wh-Coordination Analysis is much more troublesome. Consider the specific implementation of this analysis in (16). The analysis of what seems unproblematic. What is a conjunct within a ConjP in an argument position (complement-of-V position). Since coordinate structures generally have the same distribution as their conjuncts do when they occur independently, this is expected for an argument wh-phrase like what. 9 Now consider the analysis of when in (16). When itself is not in an adjunct position but in an argument position: the complement position of Conj. But this is unproblematic, since traditional adjuncts can be coordinated (She works skillfully and efficiently). What is far more problematic is that the coordinate structure of which when is a conjunct is also not in an adjunct position. On the analysis in (16), then, when is completely divorced from adjuncthood: neither it nor the coordinate structure of which it is a conjunct is in an adjunct position, contra the standard and well-motivated view that when is an adjunct wh-phrase. The Wh-Coordination Analysis is thus not straightforwardly compatible with our current understanding of the argument/adjunct distinction. 10 Given the robustness of the argument/adjunct distinction and its near-centrality to the theory of syntax, these considerations indicate that the VP-Coordination Analysis is conceptually preferable to the Wh-Coordination Analysis.
Furthermore, as a reviewer notes, the derivation in (17) is quite unremarkable in the current theoretical context. All else being equal, we expect it to be available: it would take something special to rule it out.
The next subsection shows that the VP-Coordination Analysis is also superior empirically.

Empirical predictions of the two analyses
We now turn to some empirical predictions of the VP-and Wh-Coordination Analyses. The investigation will proceed as follows. Section 4.2.1 tests the predictions of the two analyses about whether ISMW should permit obligatorily transitive verbs. Section 4.2.2 tests their predictions about what types of (non-wh) adverbs should be possible within the coordinate structure. Finally, Section 4.2.3 tests their predictions about whether there should be (apparent) in situ coordination of argument wh-phrases with different θ-roles. 9 Furthermore, on the ConjP analysis of coordinate structures, what is itself in an argument position: [Spec, ConjP]. 10 One might wonder whether the Wh-Coordination Analysis could be improved by positing that the ConjP is not the complement of the V eat, but adjoined to the V(P) eat:

(i) … [ VP [ V(P) eat] [ ConjP what and when]]
Unfortunately, (i) suffers from a problem that is approximately the mirror image of the problem with (16). The treatment of when is unproblematic, because it is a conjunct within a ConjP in an adjunct position, which is expected behavior for an adjunct wh-phrase. However, although the ConjP is an adjunct, its first conjunct (what) is an "argument" wh-phrase. This is unlikely to be possible, because what cannot be an adjunct on its own: (ii) Mary ate steak (*what)?! And crucially, a phrase (e.g., the morning) that can be an argument ((iii-a)) but not an adjunct ((iii-b)) cannot be a conjunct within a coordinate structure in an adjunct position ((iii-d)): (iii) a. Let's discuss the morning. b. *Let's discuss it the morning. c. Let's discuss it in the {morning/afternoon}. d. *Let's discuss it the morning and in the afternoon.
For convenience, the sample ISMW sentence in (3-a) is repeated in (20), and the structures assigned to it by the Wh-and VP-Coordination Analyses are repeated in (21) The subprediction in (23-b) is straightforward. If eat in (22) is replaced with an obligatorily transitive V like devour, the devour in the second conjunct will lack a direct object; thus, the resulting structure should be unacceptable. That the Wh-Coordination Analysis makes the prediction in (23-a) is initially less obvious. But consider (21) a. *Mary congratulated who and where to annoy Sue?! b. **Mary congratulated where and who to annoy Sue?! This provides an argument for the VP-and against the Wh-Coordination Analysis. 12

Prediction B: Adverb insertion
The Wh-and VP-Coordination Analyses also make different predictions about what types of adverbs should be able to follow the first wh-phrase in ISMW: (26) a. Wh-Coordination: If an adverb immediately follows the first wh-phrase in ISMW, it should also be able to immediately follow that wh-phrase when the latter is in the left periphery. b. VP-Coordination: It may well be possible for an adverb to immediately follow the first wh-phrase in ISMW that cannot immediately follow that wh-phrase when the latter is in the left periphery.
11 These sentences must be read in a prosodically integrated manner. Unlike (24-a), (i) is acceptable: (i) Mary devoured what-and when-to impress Sue?! But (i) is presumably a variant of (ii) (which involves CP-coordination plus sluicing) in which and when is inserted into the first conjunct as a parenthetical (cf. Lipták 2011), and therefore irrelevant to the analysis of ISMW.
(ii) Mary devoured what to impress Sue?! And when?!
Presumably the (b)-sentences in (24)-(25) are even worse than the (a)-sentences because the (a)-sentences at least contain locally well-formed strings like Mary devoured what, unlike the (b)-sentences (*Mary devoured when). 12 The VP-Coordination Analysis also predicts, correctly, that ISMW should allow double-object verbs whose objects are both optional: (i) Mary paid him how much and when to impress Sue?! Consider why. On the Wh-Coordination Analysis, the conjuncts are the wh-phrases. Therefore, if an adverb immediately follows the first wh-phrase, it must be right-adjoined to it (assuming there is no left-adjunction to the mother of Conj-the traditional "Conj"). And if that wh-phrase allows a particular right-adjunct when it is in [Spec,ConjP], then it is fully expected to also allow that right-adjunct when it is in the left periphery.
On the VP-Coordination Analysis, by contrast, the conjuncts in ISMW are VPs. Therefore, an adverb immediately following the first wh-phrase need not be right-adjoined to it: it may instead be right-adjoined to the first VP conjunct. Consequently, it is perfectly possible that the adverb will not be able to immediately follow that wh-phrase when the latter is in the left periphery: this is what is expected if the adverb is able to adjoin to VPs but not to wh-phrases.
The correct prediction, it turns out, is (26-b The same holds in the corresponding embedded questions (where, as in (27), the linearly first wh-phrase immediately follows a verb): (29) I wonder what (*greedily) (and {when/where}) Mary ate to impress John.
Likewise, publicly can immediately follow the first conjunct (with whom) in the following ISMWs: (30) a. M Mary spoke with whom publicly and when to annoy John?! b. M Mary spoke with whom publicly and where to annoy John?! But it cannot immediately follow with whom in the left periphery: (31) a. With whom (*publicly) (and {when/where}) did Mary speak to annoy John? b. I wonder with whom (*publicly) (and {when/where}) Mary spoke to annoy John.
In the examples discussed so far in this subsection, the first wh-phrase has been an argument wh-phrase, but analogous asymmetries are observed when the first wh-phrase is an adjunct wh-phrase. Thus, publicly can immediately follow the first conjunct (when or where) in the following ISMWs: (32) a. M Mary spoke when publicly and with whom to annoy John?! b. M Mary spoke where publicly and with whom to annoy John?! But it cannot immediately follow when or where in the left periphery: (33) a. {When/Where} (*publicly) (and with whom) did Mary speak to annoy John? b. I wonder {when/where} (*publicly) (and with whom) Mary spoke to annoy John.
The distribution of adverbs in ISMWs, then, provides a second argument for the VP-and against the Wh-Coordination Analysis. 14

Prediction C: Coordination of argument wh-phrases with different θ-roles
The discussion so far has focused on apparent in situ coordination of argument and adjunct wh-phrases. However, the Wh-and VP-Coordination Analyses also make different predictions about whether there should be apparent in situ coordination of argument wh-phrases with different θ-roles (cf. Whitman 2002;2004;Citko 2013): (34) a. Wh-Coordination: Argument wh-phrases with different θ-roles should not be coordinable in situ. b. VP-Coordination: Argument wh-phrases with different θ-roles should be (apparently) coordinable in situ.
On the Wh-Coordination Analysis, the entire ConjP is in an argument position of V. Therefore, the entire ConjP should receive a θ-role from V; there is no way for the individual conjuncts to receive θ-roles, let alone different ones (Zhang 2007, §2.2.1). 15 On the VP-Coordination Analysis, by contrast, the conjuncts are VPs. Therefore, if the V that undergoes across-the-board head movement is one that can assign more than one type of θ-role, then it should be able to assign different θ-roles within the two VP conjuncts.
One such V is serve (Whitman 2004):

Conclusion
This squib has examined the almost uninvestigated phenomenon of apparent in situ mixed wh-coordination (ISMW), in which an argument and an adjunct wh-phrase appear to be coordinated in situ rather than in the left periphery. ISMW differs from left-peripheral mixed wh-coordination in that the former initially appears difficult to reconcile with the argument/adjunct distinction. Two possible analyses of ISMW were considered: the Wh-Coordination Analysis, on which the conjuncts are the wh-phrases, and the VP-Coordination Analysis, on which the conjuncts are VPs whose head Vs undergo across-the-board head movement to v. The VP-but not the Wh-Coordination Analysis is compatible with our understanding of the argument/adjunct distinction, and involves an unremarkable derivation that it would take a stipulation to rule out. The VP-Coordination Analysis is also superior empirically, judging by evidence involving (a) obligatorily transitive verbs, (b) adverb insertion, and (c) apparent in situ coordination of argument wh-phrases with different θ-roles. This being so, ISMW does not in fact threaten the argument/adjunct distinction, despite appearances-a theoretically most welcome result, given how fundamental to syntax the argument/adjunct distinction appears to be.
Abbreviations a = personal a (Spanish), M = marked