Conjunction saves multiple sluicing: How *(and) why?

Our goal in this paper is to analyze coordinated wh-sluicing in English and compare its properties to the properties of multiple sluicing, coordinated wh-questions and coordinated clefts. We show that none of these constructions has the properties of coordinated sluices, and conclude that coordinated sluices are not derivationally related to them. We propose, instead, that coordinated sluices underlyingly involve coordination of two interrogative CPs with a single fronted wh-phrase and an elided TP in each CP. In addition, we propose that the elided TP of the second conjunct in a coordinated sluice contains a pronominal element (an E-type pronoun) coindexed with the trace of the fronted wh-phrase in the first conjunct. We show that this analysis derives the properties of coordinated sluices and explains why they differ from multiple sluices, coordinated wh-questions and coordinated clefts.


Introduction
In this paper, we investigate the effects of coordination on the status of multiple sluicing constructions in English. We build on the observation that coordination sometimes improves the status of constructions that are ungrammatical without coordination. This is, for example, what happens in multiple wh-questions: the ungrammatical sequence of two fronted wh-phrases in (1a) becomes grammatical when the two wh-phrases are coordinated, as shown in (1b) (Browne 1972;Kazenin 2002;Whitman 2002;Gračanin-Yuksek 2007;Gribanova 2009;Scott 2012;Citko and Gračanin-Yuksek 2013;2016, among many others): (1) a. *When where did you see John? b. When and where did you see John?
Similarly, the ungrammatical sequence of a complementizer followed by a wh-phrase in (2a) becomes grammatical when a conjunction separates them, as shown in (2b) (Giannakidou and Merchant 1998;Citko and Gračanin-Yuksek 2017).
(2) a. *Let me know if when you see John.
b. Let me know if and when you see John.
b. Someone saw something, but I can't remember who or what.
We examine possible sources for coordinated sluices, and show that coordinated sluices like (3b) above cannot be derived from multiple sluices in (3a). We also show that they cannot be derived from coordinated wh-questions or coordinated clefts, illustrated in (4a-b), respectively.
(4) a. Coordinated Wh-Questions *Someone saw something, but I can't remember who or what saw.
b. Coordinated Clefts Someone saw something, but I can't remember who it was or what it was.
We propose instead that coordinated sluices involve coordination of two CPs, with a single wh-phrase sluiced in each, as shown in (5) We proceed as follows. In Section 2, we investigate coordinated wh-questions as a possible source for coordinated sluices and show that deriving coordinated sluices from coordinated wh-questions makes an incorrect prediction for English that the two should be subject to the same restrictions. In Sections 3 and 4, we argue against deriving coordinated sluicing from multiple sluicing and coordinated clefts, respectively. In Section 5, we present our proposal, and in Section 6, we address the question of why coordinated wh-questions and coordinated sluices cannot share the same underlying structure. Section 7 is the conclusion. 1 To the best of our knowledge, Bolinger (1978: 145) was the first to note the improvement that coordination brings about (even though he was not explicit about it). His insight, based on the contrasts in (i-ii), was that the ungrammatical multiple sluicing examples improve if "the words are separated": inserting a conjunction between the wh-phrases is one of the ways they can be separated.
(i) Bolinger (1978: 145) a. I know somebody gave somebody something. *But who who what? (*But who, who, what?) b. I know somebody gave somebody something (something to somebody). ?But who, what, to who?
(ii) Bolinger (1978: 145) I know they're taking something someplace. *But what where why? But what, where and why? Hoyt and Teodorescu (2012: 86) attribute the observation that "the unacceptability of multiple remnants in English can be mitigated by the addition of a conjunction between the two remnants" to Richards (1997). We were not able to locate the original discussion.
(iii) Hoyt and Teodorescu (2012: 86) a. John gave someone something, and I want to know who and what. b. John gave something to someone, but I don't know what or to whom. 2 Here and throughout the paper we boldface wh-phrases in example sentences even if the original source didn't.
2 Coordinated wh-questions as possible source for coordinated sluicing

Background on English CWHs
Coordinated Wh-Questions (henceforth CWHs) are questions in which two wh-phrases, not necessarily of the same category, appear conjoined in a left-peripheral position, as shown in (6a-b). The grammaticality of (6a) is not surprising given that it involves coordination of two Adverbial Phrases (so they could be part of a larger Adverbial Phrase), but in (6b) coordination would have to target a DP and an AdvP, in violation of the Law of the Coordination of Likes (Williams 1981). This has led to alternative analyses for CWHs, on which coordination involves larger (clausal) constituents, or which regard the fact that coordinated constituents are wh-phrases as sufficient for the purposes of evaluating "likeness" (see Gračanin-Yuksek 2017 for an overview of existing approaches to CWHs). English CWHs have been shown to be subject to a number of restrictions (discussed by Browne 1972;Kazenin 2002;Whitman 2002;Gračanin-Yuksek 2007;Gribanova 2009;Scott 2012;Citko 2013;Citko and Gračanin-Yuksek 2013;2016, among many others). In order to capture these restrictions, Gračanin-Yuksek (2007) proposes that English CWHs have the structure in (7b), which she dubs a bi-clausal non-bulk sharing structure. 3 It is bi-clausal because coordination is at the CP level, it is sharing because the two CPs share lexical material due to the multidominant nature of the structure (in this particular case, the two CPs share everything except for the wh-phrases). And it is non-bulk sharing because multiple individual nodes are being shared (rather than a larger chunk of structure being shared in bulk).
(7) a. What and when did you eat? This structure captures a number of restrictions on English CWHs. The first one concerns the ban on coordinating two wh-arguments, illustrated in (8a). This example is ungrammatical for the same reason (8b) is ungrammatical. The verb put requires two VP internal arguments, but since according to the structure in (7b) it has only one per conjunct, its selectional requirements are not satisfied in either conjunct.
(8) a. *What and where did you put? b. *What did you put and where did you put?
The second restriction concerns CWHs with obligatorily transitive verbs like buy. Such CWHs are ungrammatical if one of the wh-phrases is an adjunct and the other one a direct object, as shown in (9a), whose ungrammaticality also reduces to that of (9b).
(9) Obligatorily transitive V; wh-DO & wh-adjunct a. *What and when did you buy? b. *What did you buy and when did you buy?
In this respect (9a) contrasts with (10a); instead of the obligatorily transitive verb buy, (10a) has the optionally transitive verb eat, which can be interpreted intransitively in the conjunct that is missing the direct object.
(10) Optionally transitive V; wh-DO & wh-adjunct a. What and when did you eat? b. What did you eat and when did you eat?
The third restriction concerns CWHs in which one of the coordinated wh-phrases is a subject, such as the one in (11a). This example is ungrammatical because in English all finite clauses require an overt subject, due to the EPP requirement (Chomsky 1981). In (11a), however, since the subject is one of the wh-phrases, it is only part of the first conjunct, but not the second one, as shown in (11b).
(11) a. *Who and when sang? b. *Who sang and when sang?
Finally, the structure in (7b) captures the interpretation of CWHs; they are interpreted as two coordinated CPs with a single wh-pronoun in each CP, as we would expect given this structure. This is most obvious in CWHs in which one of the coordinated wh-phrases is a direct object, and the other one is an adjunct, such as the one in (12a). This example can have the so-called at-all reading, paraphrased in (12b), on which the wh-object is not part of the interpretation of the question introduced by the wh-adjunct. Since the wh-object is not part of the conjunct introduced by the wh-adjunct, the verb ate in the where-conjunct is interpreted intransitively, giving rise to the at-all reading. The at-all reading contrasts with the so-called it reading, paraphrased in (12c), on which the direct object is present in both conjuncts for interpretive purposes. 4 4 A reviewer asks what other readings CWHs might have. Gračanin-Yuksek (2007) and Citko and Gračanin-Yuksek (2013) report that English CWHs only have the at-all reading, as predicted by (7), which they argue is the only possible structure for CWHs in English. This structure has been challenged by experimental evidence reported in a series of presentations by Kush, Larson and Lewis (e.g., Kush 2012 andLarson 2013 With this background on CWHs (and the structure in (7b)) in mind, we turn to the question of whether coordinated sluices can be derived from coordinated wh-questions. This is the focus of the next section.

Against coordinated wh-questions as the source of coordinated sluicing
If coordinated sluices were derived from coordinated wh-questions, the only difference between the structure of CWHs, given in (13b), and the structure of coordinated sluices, given in (14b), would be that in coordinated sluicing, TPs in the two conjuncts undergo ellipsis (indicated in (14b) by shadowing and dotted lines This finding might be taken as evidence that in (ia) a copy/trace of what is present in both conjuncts (otherwise the verb in the first conjunct would be missing an obligatory argument). Kush, Larson, and Lewis do not take this stance; they argue for an analysis in which only the second wh-phrase is syntactically related to the rest of the structure. If a copy of what is indeed present in both conjuncts in a subset of CWHs, then such CWHs would be predicted to have the it reading. However, the above mentioned experiments also show that any CWH with an obligatorily transitive verb (even those in which what introduces the second conjunct, as in (ia)) is less acceptable than any CWH with an optionally transitive verb, suggesting that a structure which might give rise to the it reading is not readily available. Whitman (2002;  Given that all these experiments tested the relative acceptability of CWHs with obligatorily transitive verbs rather than the availability of the at-all versus the it readings, we do not take the contrast in (ia-b) to be a counterexample to our generalization that CWHs in English only have the at-all reading, at least for some speakers. 5 We do not address crosslinguistic variation here. In principle, languages that allow different structures for CWHs might also allow different structures for coordinated sluices. Interestingly, this is not what seems to be happening: Citko and Gračanin-Yuksek (to appear) show that Croatian and Polish coordinated sluices also involve bi-clausal structures, in spite of allowing mono-clausal structures for CWHs. In the previous section, we reviewed the restrictions on English CWHs and showed how these restrictions follow from the structure in (13b). Assimilating coordinated sluices to CWHs predicts that coordinated sluices should be subject to the same restrictions. This is not what we find: coordinated sluices are not subject to the same restrictions as coordinated wh-questions. First, coordination of obligatory arguments is impossible in CWHs, but possible in coordinated sluicing, as shown in (15a-b). These differences between CWHs and coordinated sluices lead us to conclude that coordinated sluices are not derived from CWHs. We have not said anything yet about why this should be the case; we will discuss this question in Section 6. First though, we turn to multiple sluicing as a possible source for coordinated sluicing.
3 Multiple sluicing as possible source for coordinated sluicing

Background on multiple sluicing
Multiple sluicing in English has been analyzed in various ways (see Vicente 2018 for an overview). On one analysis (Nishigauchi 1998;Lasnik 2007;2014), the first wh-phrase in a multiple sluicing configuration moves leftwards to [Spec CP], while the second one is extraposed/moved rightwards to an adjoined position, as shown schematically in (20b).
b. …but I don't know who i [t i was talking t j (yesterday)] to who j .
If (20b) is the correct analysis of multiple sluicing, could coordinated sluicing be derived in an analogous way, and, if so, what would such a derivation look like? One possible structure is given in (21b); the first wh-phrase moves to [Spec CP], and the second whphrase moves in a sidewards manner (Nunes 2001; and merges with the conjunction head. The result, however, violates the Law of the Coordination of Likes (Williams 1981): it involves coordination of a CP with a PP. The two conjuncts are not both whphrases nor are they similar in any other respects, so it is hard to imagine how a violation of the Law of the Coordination of Likes might be avoided in this structure.
(21) a. Someone was talking (yesterday) to someone, but I don't know who or to whom. (22) a. Park and Kang (2007: 396) ?Mary showed something to someone, but I don't know exactly what to whom.

b. …but I don't know exactly [ CP what i [ C' to whom j [ TP Mary showed t i t j ]]]
What is unusual about this derivation for a language like English is that both wh-phrases undergo overt wh-fronting: multiple overt wh-fronting in English is exceptionally allowed in multiple sluicing because the deletion of the TP removes the violation induced by the pronunciation of the highest copy of the second wh-phrase. For Abels and Dayal, after deletion removes lower copies of both wh-phrases from the structure, the remaining copies count as simultaneously highest and lowest. Thus, it is possible to characterize the examples in (22a-b) as involving the pronunciation of the highest copy of what and the lowest copy of to whom. 6

Against multiple sluicing as the source of coordinated sluicing
If coordinated sluices were derived from multiple sluices and if multiple sluices have the structure in (22b), coordinated sluices would presumably involve sideward movement of wh-phrases to form a Coordination Phrase in [Spec CP] (Zhang 2007;, followed by the deletion of the TP in which the wh-phrases originated, as shown in (23b).
(23) a. ?Mary showed something to someone, but I don't know exactly what or to whom. However, there are sufficient differences between multiple sluicing and coordinated sluicing to argue against deriving coordinated sluicing from multiple sluicing. We turn to these differences next. The first one involves the so-called clause-mate condition. Multiple sluicing crosslinguistically is well-known to be subject to this condition, which requires all wh-remnants in a multiple sluicing configuration to originate in the same finite clause (see Takahashi 1994;Nishigauchi 1998;Merchant 2001;Marušič and Žaucer 2013;Lasnik 2014;Citko to appear, among others). This is shown in (24a-b b. Kotek and Barros (2018: 799) Every boy likes some girl, but I don't know which boy which girl.
Coordination of simplex wh-DPs, by contrast, is perfectly fine in coordinated sluicing, as (28) shows.
Another difference concerns interpretation; the interpretation of coordinated sluicing is different from the interpretation of multiple sluicing. Abels and Dayal (2017a: 24) note that multiple sluicing can in principle either have a single-pair (SP) or a pair-list (PL) reading, and that the readings of the sluice are disambiguated by the readings of the antecedent. 9 (29) Abels and Dayal (2017a: 24) a. Single-Pair Reading Some student has published on some topic, but I couldn't tell you which student on which topic.

b. Pair-List Reading
Every student has published on some topic, but I couldn't tell you which student on which topic.
Coordinated sluices, however, only allow single-pair readings, as noted by Abels and Dayal (2017a) (also by Gribanova 2009 for Russian, who builds on Grebenyova's 2004;2006 work on multiple sluicing). This is indicated by the infelicity of the sluice in (30b), whose antecedent forces a pair-list interpretation.
(30) a. Single-Pair Reading Some student has published on some topic, but I forgot which student and on which topic.
b. Pair-List Reading #Every student has published on some topic, but I forgot which student and on which topic.
The last difference between coordinated sluices and multiple sluices involves swiping (Sluiced Wh-phrase Inversion with Prepositions In Northern Germanic) (we thank Anikó Lipták for bringing this fact to our attention). The paradigm in (31a-e), due to Richards (2001), shows that in a multiple sluicing configuration, only the first wh-PP remnant can undergo swiping. This is shown by the grammaticality of (31a). All the other logically possible swiping options are out: in (31b) and (31c)  These differences between multiple sluicing and coordinated sluicing lead us to conclude that coordinated sluicing is not derived from multiple sluicing.

Coordinated clefts as possible source for coordinated sluices
On some accounts, sluicing can be derived from non-isomorphic pre-sluice sources (i.e., short sources, predicative sources, cleft sources) (Erteschik-Shir 1973;Merchant 2001;Marušič and Žaucer 2013;Barros, Elliott and Thoms 2014;Vicente 2018, among others). In more concrete terms, this means that the source of the sluice in (33) is the "short" string in (33b) or the cleft string in (33c), rather than the isomorphic string in (33a) containing an island. Since neither (33b) nor (33c) involves an island, the availability of such short sources is one way to derive the lack of island effects in sluicing. 11 (33) Merchant (2001); Barros, Elliott and Thoms (2014) They hired someone who speaks a Balkan language -guess which! a. …which they hired someone who speaks! b. …which she speaks! c. …which it was! Abels and Dayal (2017a: 25) suggest that coordinated sluices, like the one in (34a), "are derived from a different pre-sluice", indicated by their cleft source and/or predicative source paraphrases in (34b).
(34) Adapted from Abels and Dayal (2017a: Fn. 16) a. Some linguist spoke to some philosopher but Bill doesn't know which linguist and which philosopher. 11 The same logic has been used to explain exceptions to the so-called P-stranding generalization. However, as pointed out to us by a reviewer, there are also island repair phenomena where the short sluice analysis is ruled out. These are cases like (i), in which the sluice contains a parasitic gap (PG), discussed in Yoshida et al. (2015), who argue that (i) cannot have a cleft or copular structure as the source because such a source would not contain the "real gap" (RG) that would license the parasitic gap in the remnant. The same point is made in Vicente (2018).
(i) Yoshida et al. (2015Yoshida et al. ( : 1453 The editor told me which book I must review __RG soon after receiving __PG, but I don't remember exactly how soon after receiving __PG.
b. Some linguist spoke to some philosopher but Bill doesn't know which linguist and which philosopher it was/they were.
In this section, we focus on cleft sources and the question of whether coordinated sluices in general can be derived from coordinated clefts along the lines schematized in (35).
(35) I know that someone saw something but I don't know who it was or what it was.
The same considerations that rule out a cleft source for singular sluices rule out a cleft source for coordinated sluices. We show this by applying to coordinated sluices a subset of the ten diagnostics that Merchant (2001) used to argue against deriving singular sluices from clefts. Van Craenenbroeck (2010) showed some of Merchant's original diagnostics to be inconclusive due to the fact that they do not distinguish between the cleft source and the non-elliptical wh-question source for the sluice. 12 Thus, we focus primarily on the diagnostics that do make this distinction. The first diagnostic involves compatibility with adjuncts. Merchant shows that adjuncts can appear in sluices but not in clefts, as shown in (36a). The fact that adjuncts are possible in coordinated sluicing, as shown in (36b), shows that the coordinated sluice in (36b) cannot be derived from the coordinated clefts in (36c).
(36) a. Merchant (2001: 121) He fixed the car but I don't know how/why/when (*it was). b. He fixed something somehow, but I don't know what or how. c. He fixed something somehow but I don't know what it was or how (*it was).
The second diagnostic concerns implicit arguments. Example (37a) shows that implicit arguments are possible as correlates of wh-phrases in sluicing constructions, but not as pivots of clefts. Implicit arguments are also possible as correlates of wh-phrases in coordi-(i) Merchant (2001: 122) Someone dented my car last night-a. I wish I knew who! b. I wish I knew who the hell it was! c. *I wish I knew who the hell! Not surprisingly, coordinated sluices pattern with regular sluicing in that they also disallow aggressively non-D-linked wh-phrases, as shown in (iia), which suggests that they cannot be derived from the coordinated cleft in (iib).
(ii) a. *Someone saw something, but I can't remember who the hell or what the hell. b. Someone saw something, but I can't remember who (the hell) it was or what (the hell) it was.
However, as brought to our attention by one of the reviewers, the distribution of aggressively non-D-linked wh-phrases is not a valid diagnostic for establishing that a sluice is derived from a non-cleft source. If it were, we would expect (iii), the non-elliptical non-cleft source for the ungrammatical sluice in (ic), to also be ungrammatical. We thank one of the reviewers for bringing Van Craenenbroeck's work to our attention and for pointing out the example in (iii).
(iii) Van Craenenbroeck (2010: 1721 I wish I knew who the hell dented my car! nated sluices, as shown in (37b), which shows that coordinated sluices cannot be derived from coordinated clefts in (37c). 13 (37) a. Merchant (2001: 121) They served the guests but I don't know what (*it was). b. They served someone but I don't know whom or what. c. They served someone but I don't know who it was or what (*it was).
The next diagnostic involves what Merchant refers to as 'mention-some' modification. 14 He shows that wh-pivots in pseudoclefts are impossible with modifiers like 'for example'. However, such modifiers can modify sluiced wh-phrases. This is shown in (38).
(38) Merchant (2001: 122) A: You should talk to somebody in the legal department for help with that. B1: Could you tell me who (*it is), for example? B2: Who (*is it), for example?
Coordinated sluices also allow 'mention-some' modification, as shown in (39) a. He fixed the car, but I don't know how it was that he fixed the car. b. They served the guests, but I don't know what it was they served the guests. 14 Two related diagnostics that Merchant discusses are 'else' modification and 'mention-all' modification. 'Else' modification is possible in sluicing but not in clefts (as shown in (ia)), whereas 'mention-all' modification has the opposite distribution, as shown in (ib).
(i) Merchant (2001: 122) a. Harry was there, but I don't know who else (*it was). b. A bunch of students were protesting, and the FBI is trying to find out who all *(it was).
Van Craenenbroeck (2010) shows that the 'mention-all' diagnostic is not conclusive, given that it does not distinguish between the wh-question and the cleft source for the sluice; both are grammatical (as shown in (iia-b)), which leaves the ungrammaticality of the corresponding sluice mysterious.
(ii) Van Craenenbroeck (2010: 1722) a. A bunch of students were protesting, and the FBI is trying to find out who all it was. b. A bunch of students were protesting, and the FBI is trying to find out who all was protesting. 15 Merchant also uses swiping to argue against a cleft source for sluicing, based on the contrast in (ia-b).
However, as Van Craenenbroeck (2010) points out, since swiping is also impossible in non-elliptical whquestions (as shown in (ii), modeled after Van Craenenbroeck's (2010) example (41c): *God knows who for she bought a robe.), this diagnostic by itself does not distinguish between clefts and non-elliptical whquestions as potential sources for sluicing. Given the differences between coordinated sluicing and coordinated clefts discussed in this section, it is reasonable to conclude that the two constructions are not derivationally related.
In the next section, we present an analysis that derives the properties of coordinated sluicing and explains why these properties differ from the properties of constructions that we have entertained as possible sources of coordinated sluicing: coordinated whquestions, multiple sluicing and coordinated clefts.

Analysis
The analysis of coordinated sluicing must account for the fact that coordinated sluicing is not subject to the clause-mate condition (unlike multiple sluicing), that there are no restrictions on the category of wh-phrases in coordinated sluicing (unlike in multiple sluicing), that coordination of arguments is possible in coordinated sluicing (unlike in CWHs), and that coordinated sluicing does not have the at-all reading (unlike CWHs).
Our starting point is the observation that in examples involving coordinated sluicing, such as the one in (41a), modeled after Merchant's (1999: 484) examples, the interpretation of the second sluice is (41b), and not (41c): The speaker is not wondering whether Abby called some suspect, the speaker is wondering whether Abby called the suspect mentioned in the first conjunct (whoever he/she might be).
(41) a. I know Abby called some suspect. I wonder which suspect and when. b. I know Abby called some suspect. I wonder which suspect Abby called and when she called him. c. #I know Abby called some suspect. I wonder which suspect Abby called and when she called some suspect.
This interpretation is reminiscent of the interpretation, given in (42b), of single sluicing examples like (42a) below, discussed first by Merchant (1999). The bi-clausal structure we propose for coordinated sluices explains why coordinated sluicing does not have the properties found in multiple sluicing, coordinated wh-questions, or coordinated clefts. While coordinated sluices involve coordination of single whquestions with ellipsis in each conjunct, multiple (non-coordinated) sluices, coordinated wh-questions, and coordinated clefts have different underlying structures, given (in a somewhat simplified manner) in (45a-c), respectively.
(45) a. Multiple Sluicing Someone saw something, but I can't remember who i or what j <t i saw t j >.

b. Coordinated Wh-Questions
Someone saw something, but I can't remember who i <t i saw> or what j <saw t j >.
c. Coordinated Clefts Someone saw something, but I can't remember who i <it was t j > or what i <it was t j >.
The structure we propose straightforwardly derives two properties of coordinated sluicing: the absence of the clause-mate condition and the absence of the restriction on the category of wh-phrases that can be coordinated. The clause-mate condition -the requirement that multiple moving wh-phrases originate in the same clause -arises only when the multiple instances of wh-movement interact so that their paths either nest or cross. Abels and Dayal (2017a;b), for example, account for the clause-mate condition in multiple sluicing by assuming that covert wh-movement is clause-bound and subject to superiority (for these authors the second wh-phrase in a multiple sluicing configuration undergoes covert wh-movement). The fact that the restrictions that derive the clause-mate condition in multiple sluicing hold of covert movement implies that there is another, overt instance of wh-movement and the fact that one instance of wh-movement is overt and the other one is covert indicates that the two are, in some relevant sense, evaluated in the same domain, i.e., that they interact. On our analysis of coordinated sluicing, schematized in (46b), the two instances of wh-movement happen in separate clauses, unlike in multiple sluicing, schematized in (46a).
Since in (46b) there is no interaction between the two instances of wh-movement (neither is non-first in a single domain), it is expected that the clause-mate condition does not constrain coordinated sluicing, as shown by the grammaticality of the ( (47) and (48).
(47) a. Lasnik (2014: 12) A certain boy said that Fred talked to a certain girl. *I wish I could remember which boy to what girl.

b. A certain boy said that Fred talked to a certain girl. I wish I could remember which boy and to what girl.
(48) Adapted from Abels and Dayal (2017a: 25) a. *Some linguist was upset because Harry spoke to some philosopher but Bill doesn't know which linguist to which philosopher.
b. Some linguist was upset because Harry spoke to some philosopher but Bill doesn't know which linguist and to which philosopher.
Our analysis also derives the fact that there are no restrictions on the category of whphrases that can be coordinated in multiple sluicing. Recall that English multiple sluicing is degraded if both wh-remnants are simplex wh-DPs and improves if the second wh-phrase is a PP (Bolinger 1978;Richards 1997;Lasnik 2014;Cortés Rodríguez 2019). This is shown by the contrast in (49) We have argued that coordinated sluicing is not derivationally related to multiple sluicing. Since on our analysis of coordinated sluicing, both wh-phrases undergo leftward wh-movement, there is no extraposition, and the absence of the preference for PPs is to be expected. Richards (2010) accounts for the ban on multiple wh-DP remnants in multiple sluicing in a different way. As stated in Section 3.1, Richards assumes that multiple sluicing involves the structure in (52), where both wh-phrases undergo movement to [Spec CP].
Richards considers the ban on two wh-DP remnants in English sluicing to be a special case of a more general restriction on linearization that "rejects trees in which two nodes that are both of type α are to be linearized in the same Spell-Out domain" (Richards 2010: 5). He refers to this restriction, given in (53), as Distinctness.
In a multiple sluice, two wh-DP remnants cause the derivation to crash: fronting both whphrases to their left-peripheral positions creates the linearization statement <DP, DP>, which violates Distinctness.
Since on our analysis of coordinated sluicing the two wh-phrases are in separate clauses, they are never required to be linearized in the same Spell-Out domain. Thus, even when both wh-remnants are DPs, the illicit linearization statement is never created and Distinctness is obeyed.
The bi-clausal nature of our analysis does not in and of itself derive the availability of wh-argument coordination in coordinated sluicing, or the absence of the at-all reading. However, both of these properties follow from the presence of an E-type pronoun in the second conjunct (they), co-indexed with the trace of the wh-phrase in the first conjunct, as shown in (44b), repeated here as (54).
(54) Someone saw something, but I can't remember who i <t i saw something> or what j <they i saw t j >.
The presence of the pronoun in the second conjunct makes wh-argument coordination possible in coordinated sluicing by ensuring that the selectional properties of the verb are satisfied in both conjuncts; in the first conjunct, by the wh-phrase and the indefinite pronoun, and in the second conjunct, by the wh-phrase and the E-type pronoun (co-indexed with the trace of the wh-phrase in the first conjunct). Since the selectional properties of the verb are satisfied in both conjuncts, regardless of the type of wh-phrases or the type of the verb, coordination of wh-arguments is possible. The presence of the pronoun in the second sluice also derives the absence of the at-all reading in coordinated sluicing. The at-all reading arises in coordination of wh-questions in which a wh-object that introduces one conjunct is not interpreted in the conjunct introduced by a wh-adjunct, as in the CWH in (55a). Since in coordinated sluicing the whobject is interpreted in the second conjunct (via the E-type pronoun), the at-all reading is not available, as (56a-b) shows.

(55)
At-all reading a. What and where did John eat? b. What did John eat and where did John eat at all? Finally, since coordinated sluicing involves coordination of two single questions, it captures the observation that coordinated sluices have only single pair readings and contrast in this respect with multiple (non-coordinated) sluices. This is shown by the contrast in (57), repeated from (29), where coordinated sluicing is unacceptable when the antecedent clause contains a distributive quantifier every and acceptable with the non-distributive some.
(57) a. Some student has published on some topic, but I forgot which student and on which topic. b. #Every student has published on some topic, but I forgot which student and on which topic.
Our analysis makes the prediction, brought to our attention by Joe Emonds, that each of the two conjuncts should be able to undergo ellipsis independently of the other. We think this prediction is correct as long as no independent restrictions are violated. These independent restrictions include MaxElide, constraints against cataphoric dependencies, conditions on the interpretation of indefinites and/or the licensing of E-type pronouns. In the remainder of this section, we present examples that show that this prediction is borne out. We start with examples in which ellipsis, instead of targeting TPs in the two conjuncts (as in (58a)), targets a TP in one conjunct and a VP in the other conjunct, as in (58b-c), or a VP in both conjuncts, as in (58d We attribute the ungrammaticality of (58b-d) to violations of MaxElide, which, in general terms, requires deletion of the largest possible constituent (see Fiengo and May 1994;Lasnik 2001;Takahashi and Fox 2005;Merchant 2008; Kimura 2013, among others). 19 Our conclusion that (58b-d) are excluded because they violate MaxElide is corroborated by the contrast in (59a-b). The ungrammatical (59a) parallels (58b) in that the first of the coordinated sluices involves VP ellipsis, in violation of MaxElide. In the grammatical (59b), however, the sluice involving VP ellipsis contrasts with the antecedent clause, which results in the auxiliary being focused. The focused auxiliary is not part of the maximal string that can be deleted, so (59b) does not violate MaxElide even though, like the ungrammatical (59a), it involves VP ellipsis. 20 (59) a. *I know Mary saw some movies but I don't know which ones she did or when. b. I know which movies Mary DIDN'T see, but I don't know which ones she did or when.
We turn next to examples in which ellipsis targets a TP only in one conjunct. The (a) examples in (60) and (61)  (60) a. Some students sang but I don't know which students i [t i sang] or when j [they sang t j ]. b. Some students sang but I don't know which students i t i sang or when j [they sang t j ]. c. Some students sang but I don't know which students i [t i sang] or when j they sang t j . d. Some students sang but I don't know which students i t i sang or when j they sang t j . 19 Here we follow the literature mentioned in the main text and refer to MaxElide as a grammatical constraint. However, as a reviewer points out, the preference to elide more rather than less has received a non-uniform treatment in the literature and should therefore be taken as more of a descriptive term (the "MaxElide effect") than as a bona fide constraint (see, for example, Griffiths 2019, where this effect is attributed to a parallelism condition on ellipsis recoverability rather than to the requirement that ellipsis should apply to as much material as possible). 20 We are grateful to an anonymous reviewer for suggesting adding focused material into the offending sluice. 21 Both (60a-d) and (61a-d) involve an antecedent TP with an optionally transitive verb sing; they differ in that the second conjunct in (60a-d) is introduced by the adjunct when, whereas in (61a-d), it is introduced by the optional argument what. This difference does not affect the generalization that the two conjuncts can undergo ellipsis independently of one another.
(61) a. Some students sang but I don't know which students i [t i sang] or what j [they sang t j ]. b. Some students sang but I don't know which students i t i sang or what j [they sang t j ]. c. Some students sang but I don't know which students i [t i sang] or what j they sang t j . d. Some students sang but I don't know which students i t i sang or what j they sang t j .
The examples in (60) and (61) show that the two instances of TP ellipsis in multiple sluicing are in principle independent, as one can take place without the other. However, things get more complicated in (62), which shows that not all logically possible TP deletions are grammatical. What distinguishes the examples in (62) from the ones in (60) and (61) above is the fact that the wh-pronoun what, which introduces the second sluiced conjunct, has an overt correlate something in the antecedent clause. The ungrammaticality of (62b) might be taken to indicate that in this case the second conjunct cannot be deleted. However, the fact that the non-elliptical (62d) is equally ungrammatical shows that the ungrammaticality of (62b) has nothing to do with ellipsis.
(62) a. Some students sang something but I don't know which students i [t i sang something] or what j [they sang t j ]. b. *Some students sang something but I don't know which students i sang something or what j [they sang t j ]. c. Some students sang something but I don't know which students i [t i sang something] or what j they sang. d. *Some students sang something but I don't know which students i sang something or what j they sang.
The problem with the examples in (62b) and (62d) appears to stem from the presence of two instances of the indefinite pronoun something: one in the antecedent TP Some students sang something, and the other one in the first conjunct which students sang something. Either instance of the indefinite something could in principle be the correlate of the wh-pronoun what in the second conjunct. However, the two indefinites necessarily introduce two different discourse referents (Heim 1982). Thus, what may be causing the degradation of (62b) and (62d) is the fact that both TPs (the TP Some students sang something and the first conjunct which students sang something) contain the correlate something, but the two correlates are non-identical since the referents of the two indefinites are different. There is another question related to the grammatical coordinated sluicing example in (62a), where the TP in the first conjunct is deleted (together with the TP in the second conjunct). The question is whether the non-elliptical source of the first conjunct contains the indefinite something to being with. The interpretation of singular sluices such as the one in (63a) below suggests that it does not. The most natural interpretation of (63a) is not the one in (63b), in which the sluice contains the indefinite present in the antecedent TP, but the one in (63c), in which the sluice contains a pronoun coindexed with the indefinite in the antecedent.
(63) a. Some students sang something but I don't know which students. b. Some students sang something but I don't know which students sang something.
c. Some students sang something but I don't know which students sang it/the thing that they sang. Kim (2010) arrives at a similar conclusion based on the examples in (64) and (65a). She first notes that in (64) the two instances of someone refer to two different individuals, and discusses the source of the sluice in (65a) in light of this observation.
(64) Kim (2010: 164-165) Someone committed a crime on Monday and someone committed a crime on Tuesday.
(65) Kim (2010: 164-165) a. John asked where someone had committed a crime but he does not know when. b. John asked where someone had committed a crime but he does not know when someone committed a crime. c. John asked where someone had committed a crime but he does not know when he committed a crime.
Kim states that if the source of the sluice in (65a) were (65b), we would also expect (65a) to be a statement about two different individuals, one committing a crime at a certain place and the other committing a crime at a certain time. This is not what (65a) means; it asks about the time and place of the crime committed by one and the same person. This leads Kim to propose (65c) as a source for (65a). If this reasoning is on the right track, the source for the multiple sluice in (66a) does not contain an indefinite in the first conjunct (as in (62a), repeated as (66b)), but rather a pronominal element, as shown in (66c).
(66) a. Some students sang something but I don't know which students or what. b. Some students sang something but I don't know which students i [t i sang something] or what j [they sang t j ]. c. Some students sang something but I don't know which students i [sang it j ] or what j [they sang t j ].
However, even when the indefinite is replaced with a pronoun that can be construed as an E-type pronoun, as in (66b), the sentence is degraded if this pronoun is pronounced. This is what examples (67a-b) show.
(67) a. *Some students sang something but I don't know which students sang it or what. b. *Some students sang something but I don't know which students sang it or what they sang.
We attribute the ungrammaticality of these examples to a cataphora constraint, which in this case prevents the pronoun it from being coindexed with the wh-pronoun what following it (Kazenin 2002, for example). We take it to be the same constraint that excludes example (68a), which contrasts with (68b), where the pronoun stands in an anaphoric relationship with the preceding wh-phrase.
(68) a. *I wonder who sang it i and which aria i amazed the audience. b. I wonder which aria i amazed the audience and who sang it i .
We thus conclude that in coordinated sluicing ellipsis may apply to each conjunct without applying to the other conjunct(s), as long as no independent constraints are violated. We hypothesize that (69b) is excluded by economy. Our reasoning relies on an independently motivated assumption that operations apply only when they have an effect on pronunciation and/or interpretation (see Fox 2000). 22 In the case at hand, the idea is that ellipsis can only apply if it has an effect on pronunciation. We adopt Merchant's (2001: 60) proposal that ellipsis is triggered by an E(llipsis) feature, which instructs "the PF system to skip its complement for purposes of parsing and production." In (69b), the E feature is located on C, and since both TP 1 and TP 2 are its complements, both have to be deleted. However, since deleting one TP (say TP 1 ) will already have deleted the string John sang, deleting TP 2 will have no further effect on pronunciation; such a vacuous application of ellipsis is banned by economy considerations. 22 We build on Fox (2000), who argues that covert semantic operations that have no effect on the interpretation of an utterance are banned from applying. In Fox's words: "scope shifting operations (e.g., Quantifier Raising (QR) and Quantifier Lowering (QL)) are allowed to apply only when they are necessary to achieve a designated semantic interpretation" (Fox 2000: 3). Fox proposes the principle of Scope Economy in (i) to formalize this ban: (i) Scope Economy (Fox 2000: 3) Scope-shifting operations (SSOs) cannot be semantically vacuous.
In our analysis of coordinated sluicing, given in (70b), ellipsis also applies to two syntactic objects: TP 1 and TP 2 , just as it does in (69b). However, each TP dominates a separate string, so both ellipsis operations have an effect on pronunciation. Thus, no ellipsis operation applies vacuously. We next turn to a related question: why can't CWHs be derived from the structure we posited for coordinated sluicing? What prevents the CWH in (71a) from having the structure in (71b) or (71c)? We first consider (71b) as a possible source of the CWH in (71a). In this structure, the verb sing in the second conjunct does not contain a direct object; this forces an intransitive interpretation of the verb and yields the at-all reading of the CWH, as desired. However, if the string in (71a) were derived from (71b), it would mean that the sluicing remnant in the first conjunct -the wh-phrase what -has no overt correlate in the second conjunct, i.e., that its correlate is an implicit argument. Implicit arguments always take the lowest scope in their clause, as indicated, for example, by the fact that (72b) does not give rise to the same ambiguity as (72a) and that (73) is not ambiguous (Fodor and Fodor 1980). These examples show that an implicit argument (interpreted as an existential) does not interact with other quantifiers in the sentence; instead, it is always outscoped by them.
(72) Fodor and Fodor (1980: 759) a. Everyone ate something. 'Everyone is such that there is something that he ate.' 'There is something such that everyone ate it.' b. Everyone ate.
'Everyone is such that there is something that he ate.' #'There is something such that everyone ate it.' (73) Fodor and Fodor (1980: 760) Everyone was kissed.
'Everyone is such that there is someone who kissed them.' #'There is someone such that that person kissed everyone.' We conclude that the analysis in (71c) is impossible for CWHs because on such an analysis, the non-elided E-type pronoun in the second conjunct is always co-indexed with a copy of a wh-phrase that has undergone ellipsis. If this is illicit, we have an explanation for why CWHs cannot be derived in the same way as coordinated sluices.

Conclusion
We started with a puzzle concerning the ameliorating effect of coordination in multiple sluicing: why do the ungrammatical cases of multiple sluicing in English improve when the wh-phrases are coordinated?
We presented several arguments against the claim that coordinated sluicing is derived from coordinated wh-questions (CWHs). First, coordinated sluices allow wh-coordination to contain obligatory argument(s) of the verb, while CWHs do not. Moreover, coordinated sluices disallow at-all readings, while CWHs allow such readings. We also showed that coordinated sluicing cannot be derived from multiple sluicing: while multiple sluicing is, coordinated sluicing is not subject to the clause-mate condition. Also, unlike multiple sluicing, coordinated sluicing allows coordination of simplex DP wh-phrases. Finally, we showed, based on contrasts involving compatibility with wh-adjuncts, compatibility with implicit arguments, "mention-some" modification, and case considerations that coordinated sluicing is not derived from a coordinated cleft construction. Instead, we proposed a bi-clausal structure for coordinated sluices, which involves coordination of two CPs, with a single wh-phrase in the specifier of each CP and TP ellipsis in each CP. We justified the existence of this structure alongside the structure of CWHs by showing that the two constructions cannot share the same underlying source (either the one we posited for coordinated sluicing or the one of CWHs) because if they did, the derivation of one construction would involve a violation of a principle that is independently operative in the grammar.