Free Access

Right Node Raising

Barbara Citko, 

Barbara Citko

University of Washington, USA

Search for more papers by this author
First published: 27 February 2017
https://doi.org/10.1002/9781118358733.wbsyncom020
Citations: 1

Abstract

This chapter examines a construction known as Right Node Raising (RNR). There are many questions that RNR raises, and perhaps the most widely debated ones concern the position of the ‘right node raised’ element: is it located outside both conjuncts, inside only one of them, or perhaps inside both conjuncts simultaneously? All three possibilities have been explored in the literature. On some accounts, RNR initially involves two copies of the shared element, one in each conjunct, which undergo rightward across-the-board movement to a position above the coordination level. On other accounts, one of the copies undergoes deletion; and yet, on others, a single copy is literally shared between the two conjuncts, resulting in a multidominant structure. After introducing the existing accounts of RNR, this chapter turns to the diagnostics that have been used to weigh in on the choice between them. These include constituency, case, agreement, reconstruction effects, binding, islandhood, scope, compatibility with relational modifiers, vehicle change effects, and the availability of strict versus sloppy identity readings. Interestingly, with respect to some of these diagnostics, the shared element behaves as if it were inside both conjuncts; with respect to others, it behaves as if it were outside both of them; and with respect to still others, it behaves as if it were only in one.

1 Introduction

The focus of this chapter is the construction referred to as Right Node Raising (RNR), a simple example of which is given in (1). RNR typically (though not necessarily, as we will see in section 3) involves a coordinate structure in which a single element, such as ‘a new movie by Woody Allen’ in (1), is understood as being ‘shared’ in a sense to be made more precise below between the two conjuncts. Following the common practice in the literature on RNR, I will refer to the shared element as the pivot (in italics), and mark the conjunct internal positions in which it originated with an underscore. The two conjuncts are separated from each other and from the pivot by a sharp intonational break, which I will mark with commas (although some sources use dashes for this purpose).

(1) Terry praised ____, and Leslie criticized ____, a new movie by Woody Allen.

The term Right Node Raising is attributed to Postal (1974). The term, however, is somewhat misleading, as it implies a particular analysis, one in which the pivot undergoes rightward movement. Such an analysis is only one of the many that have been proposed, and certain aspects of it are quite problematic. Common alternatives to a movement analysis of RNR include ellipsis, in which each conjunct contains one copy of the pivot, and a multidominant analysis, in which the pivot is simultaneously present inside both conjuncts. What distinguishes movement and non-movement accounts is the position of the pivot; on movement accounts it is outside the two conjuncts, whereas on non-movement accounts it remains inside the two conjuncts throughout the derivation. This is the reason why the two types are sometimes referred to as in-situ versus ex-situ accounts (see e.g. Abels 2004), and the bulk of the discussion in this chapter is devoted to presenting the data that weigh in on the choice between them. Since the properties of RNR are presented through the prism of the three major types of accounts that RNR has received in generative literature (i.e. movement, ellipsis, and multidominance), let me thus conclude this introduction by stating my assumptions about the nature of these three syntactic mechanisms. Crucially, I take movement, ellipsis, and the kind of multidominance involved in RNR to be distinct syntactic processes., For the sake of concreteness, I assume that multidominant structures are generated by means of a Parallel Merge mechanism of Citko (2005; 2011b), which allows a single element to occupy two positions simultaneously. The very concept of multidominance, however, goes back at least to Simpson (1975), and there are many different proposals couched within many different frameworks on how to derive it.

(2) image

For ellipsis, I adopt a fairly standard PF deletion view (see Merchant 2001 and the references therein for a host of arguments in favor of this view), on which there are two copies of XP to begin with, one of which is deleted under (sometimes less than perfect) identity with the other.

(3) image

Chomsky (2004) analyzes movement as Internal Merge, which has been taken to mean that movement can be reduced to multidominance, and represented as (4a) rather than (4b).

(4) a. image b. image

While the issue of whether movement involves multidominance or the more standard Copy and Move operation is an issue that has received a fair amount of attention in the literature, it does not affect the assumption of this chapter, which is that a multidominant analysis of RNR is different from a movement analysis of RNR. Even if movement reduces to multidominance, there is a fundamental difference between the kind of multidominance involved in RNR and the kind that might result from movement. In (2), XP is shared between two elements neither of which dominates the other, whereas in (4a) one of the two mothers of XP dominates the other. Gračanin-Yuksek (2007) refers to the two kinds as horizontal versus vertical sharing, and De Vries (2012) as Internal versus External Remerge. The empirical findings concerning RNR hold irrespective of which analysis of movement we adopt, (4a) or (4b). In one case there is a copy of XP in a higher position, whereas in the other case there is an occurrence of it in this higher position.

2 Different structures, different predictions

2.1 RNR as across-the-board (ATB) movement

The idea that RNR might involve ATB rightward movement dominated early accounts of RNR (see e.g. Abbott 1976; Bresnan 1974; Hankamer 1971; Maling 1972; Postal 1974; Ross 1967), and was resurrected more recently by Postal (1998; 2004) and Sabbagh (2007). This is the derivation that the term RNR is meant to capture; we are taking the rightmost node and moving it in an ATB fashion to the right. How exactly this happens, however, will depend on how exactly ATB movement works. Even though there are many different views on the nature of ATB movement (with the major ones listed in (5a)–(5c)), what remains constant across the various accounts is the fact that the ATB moved element (the RNR pivot in the case at hand) ends up in a position above the coordination level. This is the aspect of an ATB movement analysis of RNR that is going to be most relevant for our purposes.

(5) a. Multidominance (e.g. Williams 1978; Goodall 1987; Citko 2005; 2011a; 2011b)
b. Sideward movement (e.g. Nunes 1995; 2001; 2004; Hornstein and Nunes 2002)
c. Empty operator movement (e.g. Franks 1993; 1995; Munn 1993).

The multidominant and sideward movement derivations of RNR capture the intuition that the pivot is understood as being shared between the two conjuncts by taking the pivot to originate inside the two conjuncts (either by virtue of Parallel Merge, as in (6a), or by virtue of movement from one conjunct to the other, as in (6b)).,

(6) a. image
b. image

This is not to say that in any conceivable movement derivation, the pivot moves to the high position; we could also imagine a derivation in which the pivot simply moves from the first conjunct to the next without ever leaving the conjunction phrase, as in (7). Since, strictly speaking, (7) does not represent ATB movement, I will not consider it any further.

(7) image

Setting the derivational details of ATB movement aside, let us turn to the predictions an ATB account of RNR makes. These are fairly straightforward movement predictions, and a lot of the literature on RNR is devoted to testing these predictions against the predictions made by alternative (non-movement) accounts.

The most straightforward difference between movement and non-movement accounts concerns the position of the pivot. On a movement account, the pivot ends up c-commanding elements inside the two conjuncts. On a non-movement account, it remains inside the two conjuncts. Furthermore, if the pivot moves, we should see some constraints on this movement. These might involve islands (i.e. is this movement subject to island constraints?) or constituency restrictions (i.e. does the pivot have to be a constituent?).

There are two prominent alternatives to a movement account of RNR. One involves multidominance, and the other one ellipsis. Even though in my own work (see Citko 2011a; 2011b) I have argued for a multidominant account, my goal in this chapter is somewhat more modest: to present the pros and cons of each account, without necessarily arguing in favor of any particular one.

2.2 RNR as multidominant sharing

In a multidominant approach to RNR, schematized in (8), the pivot does not move at all and is literally shared between the two conjuncts (for variants of such an account, see e.g. McCawley 1982; Goodall 1987; Blevins 1990; Muadz 1991; Wilder 1999; 2008; Abels 2004; De Vos and Vicente 2005; De Vries 2005; 2009; Fox and Pesetsky 2007; Gračanin-Yuksek 2007; Johnson 2007; Bachrach and Katzir 2009; Citko 2011a; 2011b). This means that we should not expect to see any movement or constituency restrictions on the pivot. The only restrictions we expect to see are independently motivated restrictions on multidominant structures. The pivot remains low inside the two conjuncts, so we expect it to behave as such with respect to standard c-command diagnostics. Furthermore, since there is only one pivot, we expect it to satisfy whatever requirements are imposed on it inside each conjunct.

(8) image

Given that the pivot is simultaneously in two places, a natural question to ask is in which of these two places it gets interpreted and linearized. The issue of linearization becomes even more pressing if we assume the Linear Correspondence Axiom (LCA) of Kayne (1994), on which precedence reduces to asymmetric c-command, and the same element cannot be linearized twice. One way to solve this issue has been to develop ‘special’ linearization algorithms to handle multidominant structures (see e.g. Goodall 1987; Moltmann 1992). A slightly different take is to modify the LCA so that it can linearize multidominant structures (as proposed by e.g. Wilder 1999; 2008; Fox and Pesetsky 2007; Johnson 2007; Bachrach and Katzir 2009; Gračanin-Yuksek 2013). Even though the details of these proposals differ, what they share is the idea that there is something special about multiply dominated nodes. In general terms, they capture the special status of such nodes by introducing the concept of complete dominance and arguing that only completely dominated elements are subject to the LCA. In (8), for example, neither VP1 nor VP2 completely dominates the DP pivot a new movie by Woody Allen. This means that the terminals of this DP, not being completely dominated by VP1 and VP2, are not going to be linearized till they become fully dominated. Bachrach and Katzir (2009) refer to this process as Delayed Spell-Out, which affects not only linearization but also the interpretation of pivots in RNR constructions.

2.3 RNR as ellipsis

A common alternative to a movement structure and derivation for RNR involves ellipsis (see e.g. Hartmann 2000; Bošković 2004/1996; Féry and Hartmann 2005; Ha 2006; 2008; An 2007). What ellipsis and multidominant accounts share is the lack of movement in RNR. What they disagree on is the number of copies of the pivot. In a multidominant structure, there is a single copy with multiple mothers. In an elliptical structure, there are two copies, one deleted under identity with the other:

(9) image

C-command diagnostics thus are not going to distinguish ellipsis from multidominance; on either of them, the pivot remains inside the two conjuncts. However, on an ellipsis account, we expect RNR to behave similarly to other (perhaps better understood) ellipsis types, such as VP Ellipsis (VPE). Thus, if there are any properties that are unique to ellipsis, we predict RNR to have them. The natural ones to consider are identity requirements (or the lack thereof), vehicle change effects, and directionality of ellipsis.

This concludes the overview of the major accounts of RNR. The flowchart in (10) condenses the differences between the different types of accounts into a small set of questions, which will serve as a guide in the discussion that follows.

(10) image

This flowchart focuses on prominent attested accounts of RNR; it does not exhaust all the logical possibilities. One could also imagine various combinations of these three accounts, such as the one in (11), which combines movement with ellipsis. Here, both conjuncts contain a copy of the pivot; one moves and one deletes.

(11) image
It is also quite possible that there exists no single account, and that the construction we descriptively refer to as RNR is in fact ambiguous between two (or more) structures. This is the view taken by Barros and Vicente (2011), for example, who argue in favor of such an approach (which they refer to as an eclectic approach). For them, RNR constructions are in principle ambiguous between an elliptical and a multidominant structure, and different RNR types are only compatible with one structure.

3 Testing the predictions

3.1 In situ versus ex situ

3.1.1 Possible pivots

The ability to move is a standard textbook example of a constituency diagnostic. Thus, if RNR involves movement of the pivot, we expect the pivot to be a constituent. Admittedly, the usefulness of this diagnostic is complicated by the fact that not all constituents can move, for a myriad of independent reasons. However, it is still reasonable to expect the range of possible pivots to correspond to the range of moveable constituents.

Even though many accounts of RNR focus on cases in which the pivot is a noun phrase (and the examples we have seen so far all involved noun phrases), other pivots are certainly possible (as discussed by e.g. Bresnan 1974; Postal 1974; 1998; Hudson 1976; Gazdar 1981).

(12) a. Jack may be ____, and Tony certainly is ____, [DPa werewolf].
b. Tom said he would ____, and Pete actually did ____, [VPeat a raw eggplant].
c. Terry used to be ____, and George still is ____, [AdjPvery suspicious].
d. Harry has claimed ____, but I certainly don't believe ____, [CPthat Melvin is a Communist].
(Postal 1974, 126–127)
e. I can tell you when ____, but I can't tell you why ____, [TPhe left me].
(Bresnan 1974, 618)

The next question is whether pivots have to be constituents. Postal (1974) assumes that RNR always targets constituents and uses RNR as a constituency test to argue for the raising to object analysis of ECM examples like (13). If the embedded subject does not move to the matrix object position, he reasons, the pivot remains a constituent and the ungrammaticality of (13) is unaccounted for.

(13) *I find it easy to believe ____, but Joan finds it hard to believe ____, Tom to be dishonest.
(Postal 1974, 128)

There are two other types of considerations showing that the constituency requirement on RNR is too strong. The examples in (14a)–(14d), due to Abbott (1976), show that the pivot can be larger than a single constituent (and involve multiple constituents), whereas the ones in (15a)–(15c) show that it can be smaller than a single constituent (and involve sub-constituents, so to speak).

(14) a. Smith loaned ____, and his widow later donated ____, a valuable collection of manuscripts to the library.
b. I borrowed ____, and my sisters stole ____, large sums of money from the Chase Manhattan Bank.
c. Leslie played ____, and Mary sang ____, some country‑and‑western songs at George's party.
d. Mary baked ____, and George frosted ____, 20 cakes in less than an hour.
(Abbott 1976, 639)
(15) a. Your theory over ____, and mine under ____, generates.
(Booij 1985, ex.1b, also discussed in Hartmann 2000, 57)
German
b. Philip sate Frühlings ____ und Herbstblumen.
Philip sowed spring and autumn.flowers
‘Philip sowed springtime and autumn flowers.’
(Féry and Hartmann 2005, 73)
Polish
c. Jan prze ____, a Piotr pod ______, pisał list    do prezydenta.
Jan through and Peter under wrote letter to president
‘Jan copied and Peter signed the letter to the president.’

There are also many other factors besides constituency, such as focus structure or prosodic weight, that play a role in determining what kinds of elements can be pivots in RNR constructions, as shown by the following contrast from Bresnan (1974).

(16) a. *He tried to persuade ____, but he couldn't convince ____, them.
b. He tried to persuade ____, but he couldn't convince ____, the students that he liked.
(Bresnan 1974, 615)

The next set of diagnostics that can help us decide whether the pivot remains in situ or moves out has to do with constraints on movement. If only certain types of phrases can move, we would expect only these types of phrases to be pivots in RNR. This prediction is also not borne out. The following contrast shows that CPs can move but TPs cannot (see Abels 2003). However, TPs are possible pivots in an RNR construction, which is problematic for a movement account (see Bošković 2002; Abels 2003).

(17) a. [CP That the earth is not flat]i, everyone knows ti.
b. *[TP The earth is not flat]i, everyone knows that ti.
c. Everyone knows that ____, and nobody questions if ____, [TPthe earth is not flat].

This is not the only case of pivots that are not moveable. In (18a)–(18c), as noted by Johnson (2007), we see that NPs can be pivots in RNR, but cannot undergo either leftward or rightward movement:

(18) a. I know Sally's ____, but not Mary's____, parents.
b. *Parentsi, I know Sally's ti.
c. *I met Sally's ti yesterday parentsi.
(Johnson 2007, 16)

3.1.2 Islands

The behavior of RNR with respect to islands is well documented in the literature, going back at least to Wexler and Culicover (1980), who note a difference between movement and RNR constructions with respect to a number of islands. The contrast in grammaticality between the (a) and (b) examples in (19)–(21) provides an illustration. The (a) examples, involving run-of-the-mill wh-movement, are all ungrammatical, whereas the corresponding (b) examples, involving RNR, are fully grammatical.

(19) a. *Whati did Leslie ask [wh‑island when John saw ti]?
b. Leslie asked [wh‑island when John saw ti] and Terry asked [wh‑island when Bill reviewed ti] a new movie by Woody Alleni.
(20) a. *Whati does Leslie know [complex np‑island the person who wrote ti]?
b. Leslie knows [complex np‑island the person who saw ti] and Terry knows [complex np‑island the person who reviewed ti] a new movie by Woody Alleni.

(21) a. *Whati did Leslie leave [adjunct island after seeing ti]?
b. Leslie left [adjunct island after seeing ti] and Sue arrived [adjunct island before reviewing ti] a new movie by Woody Alleni.

This is not to say that RNR is not sensitive to any island violations. The ungrammaticality of (22b) indicates that RNR is subject to the Left Branch Condition. However, its ungrammaticality could also be due to independent properties of RNR, such as the requirement that the pivot occupy (or originate from) the rightmost position inside the two conjuncts. This restriction, referred to in the literature as the Right Edge Restriction, will be the focus of section 3.1.7.

(22) a. *Whichi has Mary read ti articles on RNR?
b. *Mary read ti articles on RNR and Bill reviewed ti books on ellipsis, recenti.
The fact that in languages that do allow left branch extraction (such as the Slavic languages, as known since Ross 1967), the equivalent of (22b) is still ungrammatical (as shown in (23b) for Polish), also suggests that movement is not the culprit.
(23) Polish
a. Którei Maria przeczytała ti artykuły o RNR?
which Maria read articles about RNR
‘Which articles about RNR has Maria read?’
b. *Maria przeczytała____ artykuły o RNR a
Maria read articles about RNR and
Piotr zrecenzjonował____ artykuły o elipsie, ostatnie.
Peter reviewed articles about ellipsis recent.
‘Mary read recent articles on RNR and Peter reviewed recent books on ellipsis.’

The ungrammaticality of (24b) could be seen as evidence that RNR obeys the Subject Condition. However, this example also violates the so-called Right Roof Constraint of Ross (1967), which requires rightward movements to be clause-bound.

(24) a. *What does [sbj island that John read ti] bother Mary?
b. *[sbj island That John read ____, and Bill reviewed ____,] bothered Mary an article on Right Node Raising.
Given that RNR typically involves coordination, an obvious constraint to consider is the Coordinate Structure Constraint. The general consensus appears to be that RNR does obey the Coordinate Structure Constraint (see e.g. Wexler and Culicover 1980; McCawley 1982; Postal 1998; Johnson 2007; Sabbagh 2007).

There are two parts to CSC; one prevents movement of an entire conjunct, and the other prevents movement of any element contained within one conjunct. The following examples, due to Postal (1998), show that both are operative in RNR.

(25) a. *Tom is writing an article on Aristotle and ____, and Elaine has just published a monograph on Mesmer and ____, Freud.
b. * Tom may have bought sketches of Gail and photos of ____, and Bob saw ____, Louise.
(Postal 1998, 122)

The fact that ungrammatical examples become grammatical when movement takes place in an ATB fashion from both conjuncts simultaneously further supports the conclusion that RNR obeys the CSC:

(26) a. * Terry watched ____, Leslie ignored the preview, and Bill reviewed ____, a new movie by Woody Allen.
b. Terry watched ____, Leslie ignored ____, and Bill reviewed ____, a new movie by Woody Allen.
If RNR is rightward movement, we might also wonder whether it obeys constraints on rightward movement, if such exist. One discussed by Barros and Vicente (2010) involves the so-called Right Roof Constraint of Ross (1967), which requires rightward movement to be clause-bound, as shown by the contrast between (27a) and (27b). The further contrast between (27b) and (27c) shows that RNR does not have to be similarly clause-bound.
(27) a. Alice saw ti, yesterday, [the new headmaster]i.
b. *Alice said [that Beatrix loves ti] yesterday, [the new headmaster]i.
c. Alice claimed [that Beatrix loves ti], and Claire proved [that Diana hates ti], [the new headmaster]i.
(Barros and Vicente 2010, 9)

A more general issue I would like to conclude this section with concerns cross-linguistic variation. If there are languages in which RNR does obey island constraints, it may well be the case that RNR in such languages does involve movement. This is the argument Sabbagh (2008) makes for Tagalog, showing that in this language RNR is subject to the so-called subject extraction restriction, one of the hallmarks of Austronesian syntax.

Furthermore, if RNR involves rightward ATB movement, we would expect to find the same cross-linguistic variation with respect to RNR as we find with respect to ATB movement. To illustrate with one example, in a multiple wh-fronting language like Polish, multiple ATB movement is possible as long as both wh-phrases are extracted from both conjuncts simultaneously:

(28) Coj komui Tomasz obiecał ti tj a Ewa kupiła ti tj?
who.acc who.dat Tom promised and Eve bought
‘What did Tomasz promise to whom and Ewa bought for whom?’

Multiple RNR is also possible:

(29) Tomasz obiecał ____ ____, a Ewa kupiła ____ ____, nowy
Tomasz promised and Ewa bought new.acc
computer Jankowi.
computer.acc Jan.dat
‘Tomasz promised and Ewa bought new computer for Jan.’
This seems to support the movement account; however, we have seen similar examples in English (see the discussion in endnote 13), whose grammaticality casts doubt on the correlation between multiple fronting and the availability of multiple pivots in RNR.

Another set of potential arguments bearing on the choice between in-situ and ex-situ accounts comes from the interaction of RNR with VP ellipsis. McCawley (1982) points out that in VP ellipsis contexts, the VP ellipsis site includes the pivot. This is only possible if the pivot does not move.

(30) a. Tom talked ____, and is sure that everyone else talked ____, about politics, but of course you and I didn't [VP Ø].
b. VP = talk about politics
c. VP ≠ talk
(adapted from McCawley 1982, 100)
A related argument comes from Abels’ (2004) discussion of the interaction between VP ellipsis and RNR in cases like (31). If the pivot moves, it is not clear why its movement cannot feed VP ellipsis in this case.
(31) *Jane talked about ____, and/but Frank didn't ____, the achievements of the syntax students.
(Abels 2004, 52)

Perhaps the most forceful recent case for a movement account of RNR comes from Postal's (1998) book. Postal treats RNR as a type of extraction (L-extraction in his terms) and argues that RNR involves the same type of extraction as ATB movement. Even though Postal frames his discussion of RNR as a response to McCawley's (1982) multidominant account of RNR, his arguments also apply to other types of multidominant accounts. Postal points out that with respect to quite a number of constraints, RNR behaves similarly to other constructions involving movement. Among the constraints he considers are: the Coordinate Structure Constraint (which we discussed in this chapter), the Indirect Object Constraint (a constraint that prohibits A-bar movement of the indirect object in a double object construction, illustrated in (32a)), the Genitive Constraint (a constraint prohibiting movement of postnominal of genitives, illustrated in (32b)), the Reflexive Constraint (a constraint preventing so-called inherent reflexives from being extracted, illustrated in (32c)), and a constraint against extraction from exceptive phrases (illustrated in (32d)).

(32) a. *I first offered ____ apples, and then sold ____ peaches, the immigrant from Paraguay.
b. *Glen was looking for nieces of ____, but only found cousins of ____, Ted and Alice's.
c. *Lois may have perjured/exerted ____, and should have perjured/exerted ____, herself.
d. *The nurse wanted to watch everything except ____, and did watch everything except ____, the most difficult and lengthy operation of the day.
(Postal 1998: 123–129)
(33a)–(33d) provide parallel ATB movement examples.
(33) a. *Which immigranti did you offer ti apples and then sell ti peaches?
b. *Of Ted and Alice'si, Glen was looking for nieces ti but only found cousins ti.
c. *Herselfi, Lois may have perjured ti and exerted ti.
d. *Which operationi did the nurse want to watch everything except ti and participate in everything except ti?
The behavior of RNR pivots with allege-class verbs and parasitic gap licensing, Postal (1998) argues, also points toward a movement analysis. The fact that (34a) is grammatical (and parallels (34b), in which the object of allege undergoes movement, rather than (34c), in which it does not) suggests movement.
(34) a. They may have alleged to be pimps ____, and probably did allege to be pimps ____, all of the Parisians who the CIA hired in Nice.
(Postal 1998, 131)
b. Who may they have alleged t to be pimps and probably did allege t to be pimps?
c. *They may have alleged them to be pimps.

So does the fact that RNR can license parasitic gaps, a property associated with overt movement:

(35) Greg decided to buy ____ after reading about pg, and Gail agreed to lease ____ before test driving pg, that new model electric car which actually doesn't work.
(adapted from Postal 1998, 133)

However, in this case RNR appears to be fed by Heavy NP Shift within each conjunct, so the grammaticality of this example is also compatible with the Heavy NP Shift (rather than RNR of the pivot) licensing the parasitic gap inside each conjunct.

3.1.3 Stranding

The next set of arguments that bears on the ex-situ versus in-situ status of the pivot comes from various types of stranding phenomena. In this section, we will look at two types of stranding: preposition stranding and complementizer stranding. One of the classic arguments against a movement analysis comes from McCloskey's (1986) observation that RNR can strand prepositions in languages in which P-stranding is generally disallowed. His evidence comes from Irish, which disallows P-stranding in movement, including rightward movement, but allows it in RNR. This is shown by the contrast between the grammatical RNR in (36a) and the ungrammatical Heavy NP Shift in (36b).

(36) a. Brian Mag Uidhir … ag glacadh le agus ag cabhrú le plandáil
a dtailte féin.
Brian Magquire take.prog with and help.prog with planting
their lands refl
‘Brian Maguire … accepting, and helping with, the planting of their own lands.’
b. *Bhi me ag eisteacht le inne clar mor fada ar an raidio
faoin toghachan.
was I listen.prog with yesterday program great long on radio
about‑ the election
‘I was listening yesterday to a great long program on the radio about the election.’
(McCloskey 1986, 184–185)
English also provides a relevant, albeit slightly different, contrast. Even though English is generally a language that allows P-stranding, it disallows it in cases of rightward movement, as noted by Ross (1967) and illustrated in (37a). The fact that English RNR allows P-stranding (discussed by Bošković 2004/1996) thus suggests that it is not (rightward) movement.
(37) a. *John talked about ti yesterday [the man you met in Paris]i.
b. John talked about ____, and Mary ignored ____, the man you met in Paris.
c. Mary ignored ____, and John talked about ____, the man you met in Paris.
(Bošković 2004/1996, 19)

Complementizers also cannot be stranded by movement but can be ‘stranded’ by RNR, which suggests that the pivot remains in situ (see also the discussion of examples (12e) and (17c)):

(38) a. *[TP Anything would happen]i, nobody thought that ti.
(Abels 2003, 10)
b. Nobody expected that ____, but everyone wondered if ____ [TPJohn would run for president].

3.1.4 Binding

The next set of data that bears on the ex-situ versus in-situ debate comes from binding effects. If the pivot undergoes ATB rightward movement, it should c-command the material inside the two conjuncts. And, conversely, if it does not move, it should remain c-commanded by higher elements inside the two conjuncts. Levine (1985) notes that with respect to binding effects, the pivot behaves as if it were in situ. The examples in (39a)–(39c) provide an illustration.

(39) a. Terryi liked ____, and Lesliej disliked ____, a picture of herselfi/j on the wall.
b. *Terryi liked ____, and Lesliej disliked ____, a picture of herj on the wall.
c. *Hei liked ____, but Lesliej disliked ____, a picture of Terryi on the wall.

Sabbagh (2007), however, points out that this argument is weakened given the possibility of reconstruction. If reconstruction is responsible for the binding effects in wh-questions in (40a)–(40b), it is not clear why the same mechanism (syntactic reconstruction) could not apply to RNR.

(40) a. Which picture of herselfi on the wall did Lesliei dislike?
b. *Which picture of Terryi on the wall did hei dislike?
However, as noted by Phillips (1996, 53), reconstruction is not going to explain the NPI-licensing data below. The ungrammaticality of (41a) shows that NPIs generally cannot be licensed under reconstruction. Parallel RNR examples, pointed out by Kayne (1994), are grammatical, which argues against a movement and reconstruction analysis.
(41) a. *Which picture of anyone did Leslie not like?
b. Mary bought ____, but John didn't buy ____, any books about linguistics.
(Kayne 1994, 146, crediting Paul Portman)
Variable binding also shows that the pivot (containing a bound pronoun) can be interpreted inside the two conjuncts:
(42) Everyonei suspected ____, but nobodyi really believed ____, that hei was being investigated by the FBI.
(Phillips 1996, 53)

3.1.5 Scope

So far, most of the arguments we have seen point toward the in-situ analysis of RNR. However, quantifier scope seems to point in the opposite direction. Sabbagh (2007) points out that the quantified pivot can have scope above the conjunction level, as shown in (43a). This cannot be the result of covert ATB movement, given the unavailability of this interpretation in (43b).

(43) a. Some nurse gave a flu shot to ____, and administered a blood test for ____, every patient who was admitted last night. someone > every; every > someone
b. Some nurse gave a flu shot to every patient, and administered a blood test for every patient. someone > every; *every > someone
(Sabbagh 2007, 365)
Interestingly, such wide scope is also possible when the quantified pivot is embedded inside an island, which suggests that movement cannot be responsible for the wide scope of the pivot in this case.
(44) John knows [someone who speaks ____], and Bill knows [someone who wants to learn ____], every Germanic language. someone > every; every > someone
(Sabbagh 2007, 367)

Bachrach and Katzir (2007; 2009), however, show that there is a way to understand wide scope in a non-movement account. To do so, they employ the concept of Delayed Spell-Out alluded to in section 2.2, and argue that the multiply dominated element is not transferred to the interfaces till it is fully dominated. This captures the seemingly paradoxical behavior of the pivot; it does not move, but it can nevertheless scope high.

3.1.6 Relational modifiers

This section discusses the interpretation of RNR examples with modifiers like similar, same, or different. What is interesting about these modifiers is that, on one reading, they require semantic plurality. Consider the interpretation of different or the same in (45). On one reading, the so-called external reading (which will be mostly irrelevant for our purposes), it means that John and Bill whistled tunes that were different from some other contextually specified set of tunes. On the other reading, the so-called internal reading, it means that the tunes John whistled were different from the tunes Bill whistled. This is the reading we will be concerned with here.

(45) John and Bill whistled different/the same tunes.

With a semantically singular subject, the internal reading disappears. (46) can only mean that the tunes John whistled are different from (or the same as) some other contextually defined tunes.

(46) John whistled different/the same tunes.

What is relevant for our purposes is Jackendoff's (1977) well-known observation that such relational modifiers with internal readings are allowed in RNR constructions:

(47) a. John avoided ____, and Bill ignored ____, similar issues/the same man/men with the same birthday.
b. John whistled ____ and Mary hummed ____ the same tune/at equal volumes/together.
(Jackendoff 1977, 192)

However, Jackendoff is quite explicit about not analyzing such examples as RNR “since the presumed sources would either have the wrong reading or be ungrammatical” (Jackendoff 1977, 192). This, however, presumes that (48a)–(48b) are the only possible sources for the RNR examples in (47a)–(47b), which we have seen is not the case.

(48) a. John avoided similar issues/the same man/men with the same birthday and Bill ignored similar issues/the same man/men with the same birthday.
b. John whistled the same tune/at equal volumes/together and Mary hummed the same tune/at equal volumes/together.
(Jackendoff 1977, 192)
This is consistent with Abels’ (2004) observation that internal readings are also impossible in VP ellipsis, as shown in (49a)–(49b).
(49) a. John sang two quite different songs and Mary did, too.
b. John sang (together) (at {equal/similar/different} volumes) and Mary did, too.
(Abels 2004, 55)

3.1.7 Right Edge Restriction

One of the most unique properties of RNR is the so-called Right Edge Restriction. There are many other formulations (and names) for this restriction: Right Edge Generalization (Abels 2004), Right Edge Effect (Johnson 2007), and Right Edge Condition (Wilder 1999), to name a few. I give Sabbagh's (2007) formulation in (50), as it remains nicely agnostic about the source of RNR.,

(50) In the configuration:
[[A….X…] Conj [B…X…]]
X must be rightmost within A and B before either (i) X can be deleted from A; (ii) X can be rightward ATB‑moved; or (iii) X can be multiply dominated by A and B.
(Sabbagh 2007, 356)
The contrast in (51a)–(51b) can thus be attributed to a violation of this restriction. (51a) is grammatical because the pivot corresponds to the rightmost gap inside each conjunct, whereas (51b) is not, due to the presence of the PP ‘to Terry’ in the first conjunct.
(51) a. Leslie wrote ____, and Terry reviewed ____, a new manuscript.
b. *Leslie sent ____ to Terry, and he reviewed ___, a new manuscript.

However, the Right Edge Restriction can be obscured by the presence of rightward movement inside the conjunct (such as Heavy NP Shift or extraposition) feeding RNR. This is presumably what happens in (52):

(52) Leslie has already sent ti to Paris ____, and Terry will soon send ti to London ____, the best designs from the spring collection.
Even though the Right Edge Constraint does not bear directly on the ex-situ versus in-situ approach to RNR, any descriptively adequate approach to RNR has to capture it. One question to ask, then, is whether the ex-situ (movement) approaches or in-situ (ellipsis or multidominant) approaches are better suited to handle it. It is not generally the case that movement needs to target peripheral elements; leftward ATB movement is not limited to leftmost elements, thus there is no reason to expect that rightmost ATB movement (which is what RNR reduces to on a movement account) should be subject to such a restriction. It is also not generally the case that ellipsis has to target peripheral elements. Consider gapping, for example. If anything, it becomes ungrammatical if the gapped verb is right-peripheral:
(53) a. Terry eats beans and Leslie eats rice.
b. *Terry eats beans and Leslie eats.
However, on many existing multidominant approaches to RNR (e.g. Wilder 1999; Fox and Pesetsky 2007; Johnson 2007; Gračanin-Yuksek 2013), the right peripherality of RNR falls out from linearization considerations. In most general terms, on these approaches, if the pivot is not the rightmost element in its conjunct, the structure cannot be linearized. Let me illustrate with Wilder's approach. The ungrammatical example (51b), repeated as (54a), has the structure in (54b).
(54) a. *Leslie sent ____ to Terry, and he reviewed ____, a new manuscript.
b. image

Within the first conjunct, the DP pivot ‘a new manuscript’ is going to c-command (and thus precede, given the LCA) the PP ‘to Terry’. However, since the first conjunct (TP1) c-commands the second conjunct (TP2), this PP should also end up preceding the pivot, since anything contained inside the first conjunct will end up preceding anything contained in the second conjunct. This leads to a violation of antisymmetry, with the PP ‘to Terry’ both preceding and following the DP ‘a new manuscript’. Consequently, the structure cannot be linearized. The problem does not arise if there is no PP inside the first conjunct; the pivot is not going to c-command anything inside the first conjunct.

To sum up briefly, we have seen in this section the data that bear on the choice between ex-situ versus in-situ approaches to RNR. We have seen that the behavior of RNR with respect to islands, stranding, and binding point toward the in-situ approach, whereas scope and the behavior of RNR with respect to some constraints on movement seem to point in the opposite direction. In section 3.2, we turn to the issue of what kind of in-situ approach to RNR captures best the properties of this construction. Here the choices are between ellipsis and multidominance, the major difference between them lying in the number of the copies of the pivot.

3.2 One copy versus two copies

3.2.1 Morphological (mis)matches

It is well-known that ellipsis allows certain inflectional mismatches (see e.g. Sag 1976; Lasnik 1995; Potsdam 1997). Bošković (2004/1996) points out that the same inflectional mismatches that are possible in VPE are possible in RNR, as shown in (55a)–(55b). And, conversely, the mismatches that are impossible in VPE are also impossible in RNR, as shown in (56a)–(56b).

(55) a. John has slept in her house, and now Peter will sleep in her house.
(VPE: slept/sleep)
b. John has slept in her house, and Peter definitely will, sleep in her house.
(RNR: slept/sleep)

(56) a. *John won't enter the championship, but Jane is entering the championship.
(VPE: enter/entering)
b. *John is entering the championship, but Jane won't, enter the championship.
(RNR: entering/enter)
(Bošković 2004/1996, 15)
Crucially, ATB movement does not allow analogous mismatches:
(57) a. *[Sleeping in her office]i, (Peter was ti and) John will ti.
b. *[Slept in her house]i, (John has ti and) Peter will ti.
c. *[Questioning our motives]i, (John may be ti and) Peter hasn't ti.
(Bošković 2004/1996, 16)

These kinds of mismatches favor an ellipsis analysis in which there are two copies of the pivot and one gets deleted under (less than perfect) identity with the other. Interestingly, other types of morphological mismatches favor an analysis with one copy of the pivot. Taken together with the evidence we saw in section 2 pointing toward the low position of the pivot, they favor an analysis in which the pivot is multiply dominated by material inside the two conjuncts. The relevant examples involve what Grosz (2015) dubs cumulative agreement and Yatabe (2003) summative agreement, which refers to plural agreement with two singular DPs, also discussed by Postal (1998). (58) provides an illustration.

(58) Sue's proud that Billsg ____, and Mary's glad that Johnsg ____, havepl/?*hassg traveled to Cameroon.
(Grosz 2015, 6)

For Grosz (2015), cumulative agreement of this sort is due to a single head (T in this case) being shared between the two conjuncts. The T head in such a sharing configuration can undergo Multiple Agree (i.e. Agree between a single Probe and two Goals, proposed by Hiraiwa 2001) with two singular DP subjects simultaneously, which is what gives rise to cumulative agreement.

(59) image

Interestingly, sometimes instead of the cumulative agreement we have just seen, the pivot agrees with the closer (i.e. the second) conjunct. This is, for example, what Kluck (2009) shows for Dutch. Kluck (2009) examines agreement with verbal pivots in Dutch and shows that even though matching (i.e. the situation in which the verb matches both conjuncts) is the norm (expected on a multidominant account), non-matching is sometimes possible. In such cases, there is a preference for the pivot to agree with the second (closer) conjunct. In (60a)–(60b), I provide examples from Polish to illustrate this proximity effect. It can be seen with both verbal and nominal pivots (for a discussion of the latter, see also Citko 2011a; 2011b). In (60a) the subject of the second conjunct determines agreement on the verbal pivot, and in (60b) the verb inside the second conjunct determines case on the nominal pivot.

(60) a. Maria myśli, że ona ____, a Jan myśli, że on ____,
Maria thinks that she and Jan thinks that he
był/*byłanajlepszym składniowcemnanaszymwydziale.
was.m/*f best syntactician on our department
‘Maria thinks that she, and Jan thinks that he, was the best syntactician in our department.’
b. Jan polubił ____, a wszyscy inni unikali
Jan likedacc and all others avoidedgen
nowego kolegi/*nowego kolegȩ.
new.gen/*acc colleague.gen/*acc
‘Jan liked, but everyone else avoided, the new colleague.’
Sabbagh (2007) points out a similar proximity effect with respect to NPI-licensing:
(61) a. %John immensely enjoyed ____, but few other people liked ____, any of the talks onRNR.
b. *Few people liked ____, but John immensely enjoyed ____, any of the talks on RNR.
(Sabbagh 2007, 363, fn. 11)
And An (2007) observes a similar proximity effect with respect to the distribution of honorific agreement in Korean (and Japanese). He notes that in RNR, honorific agreement is possible if the subject of the second conjunct is socially superior to the speaker but not when the subject of the first conjunct is:
(62) Korean
a. Tomo‑nun bap‑ul, kuliko kyoswunim‑un ppang‑ul,
Toma‑top rice‑acc and professor‑top bread‑acc
Nina‑ekeycwu‑si‑ess‑ta
Nina‑dat give‑hon‑past‑dec
‘Tomo (gave) rice (to Nina) and Professor gave bread to Nina.’
b. *kyoswunim‑un ppang‑ul, kuliko Tomo‑nun bap‑ul,
professor‑top bread‑acc and Tomo‑top rice‑acc
Nina‑ekeycwu‑si‑ess‑ta
Nina‑dat give‑hon‑past‑dec
(An 2007, 119–120)

3.2.2 Strict versus sloppy identity

The next set of data that bears on the choice between a structure with a single copy of the pivot and a structure with multiple copies thereof comes from the distribution of strict versus sloppy identity readings. Ha (2006; 2008) points out the following parallelism between ellipsis and RNR: in both of them, pronouns are similarly three-way ambiguous. The ambiguity is between a strict reading, a sloppy reading, and what Ha refers to as a third-party reading.

(63) John likes his father, and Bill does like his father, too. VPE
a. John likes John's father and Bill likes John's father, too. strict reading
b. John likes John's father and Bill likes Bill's father, too. sloppy reading
c. John likes Chris’ father and Bill likes Chris’ father, too. third‑partyreading
(Ha 2008, 11)

(64) John likes ____, but Bill hates ____, his father. RNR
a. John likes Bill's father, but Bill hates Bill's father. strict reading
b. John likes John's father, but Bill hates Bill's father. sloppy reading
c. John likes Chris’ father, but Bill hates Chris’ father. third‑party reading
(Ha 2008, 11)

3.2.3 Vehicle change effects

Ha also points out that RNR exhibits vehicle change effects, a property that is typically associated with ellipsis rather than movement or multidominance. We see this with respect to binding and negative polarity item licensing. (65a) is an example of VP ellipsis, in which the R-expression John is ‘replaced’ by a pronoun to avoid a Principle C violation. (65b) shows that the same is possible in RNR.

(65) a. Sue said Bill wrote a mean joke about Johni on the blackboard, and hei told us that Mary did write a mean joke abouthimi, too. VPE
b. Hei hopes Susan WON'T firehimi, but the secretary knows that she WILL, fire Johni at the end of this year. RNR
(Ha 2008, 10)

The examples in (66a)–(66b), also due to Ha (2008, 81), show analogous vehicle change effects with respect to NPI-licensing.

(66) a. Mary didn't read any of the articles, but John did.
b. John read ____, but he hasn't understood ____, any of my books.

Considerations of this sort lead Ha to argue in favor of an ellipsis account. However, we have seen a number of problems for such an account, such as the lack of equivalence between the (a) and (b) examples in (67)–(68).

(67) a. I borrowed ____, and my sisters stole ____, a total of $3000 from the bank.
b. I borrowed a total of $3000 from the bank, and my sisters stole a total of $3000 from the bank.

(68) a. John gave Mary ____, and Joan presented to Fred ____, books that looked
remarkably similar.
b. John gave Mary books that looked remarkably similar, and Joan presented to Fred books that looked remarkably similar.
(Abbott 1976, 642)

Another potential issue for ellipsis accounts of RNR concerns the directionality of ellipsis. It is not clear why, given the existence of backward deletion (which is what RNR amounts to), forward deletion is not possible. If anything, in head-initial languages, other attested cases of ellipsis (gapping, sluicing, VP ellipsis, and pseudogapping) typically operate in a forward (not backward) fashion.

(69) a. Bill praised a new movie by Woody Allen and Pat criticized a new movie by Woody Allen.
b. *Bill praised a new movie by Woody Allen and Pat criticized a new movie
by Woody Allen.

And, lastly, if RNR is ellipsis, we might expect it to be limited to coordination. This is not the case, as shown by the following examples from Hudson (1976):

(70) a. Of the people questioned, those who liked ____ outnumbered by two to one those who disliked ____ the way in which the devaluation of the pound had been handled.
b. I'd have said he was sitting on the edge of ____ rather than in the middle of ____ thepuddle.
c. It's interesting to compare the people who like ____ with the people who dislike ____ the power of the big unions.
(Hudson 1976, 550)

3.2.4 Conflicting requirements

In this chapter, we have seen both data that point toward a single multiply dominated copy of the pivot in an RNR construction and data that point toward more than one copy. This apparent contradiction leads Barros and Vicente (2011) to propose what they dub an eclectic theory of RNR, in which there is no single structure and derivation for RNR. They argue that both ellipsis and multidominance are possible sources for RNR. More specifically, they take morphological mismatches and vehicle change effects to be only compatible with an ellipsis account, and cumulative agreement and internal readings of relational modifiers to be only compatible with a multidominant account. What Barros and Vicente (2011) further point out is that when the properties associated with ellipsis (vehicle change or morphological mismatches) are combined with the properties associated with multidominance (cumulative agreement and relational modifiers) in a single RNR construction, the result is ungrammatical. The conflicts they examine are represented schematically in (71).

(71)

Cumulative Agreement Relational Modifiers Vehicle Change Morphological Mismatches
Cumulative agreement √ * *
Relational modifiers √ * *
Vehicle change * √
Morphological mismatches * √

The examples in (72) are modeled upon Barros and Vicente's (2011) examples. The paradigm in (72a)–(72b) shows the interaction of relational modifiers and morphological mismatches; (72a) shows that the internal reading of relational modifiers is possible in RNR, and (72b) shows that cu5mulative agreement is likewise possible. We have seen similar examples in sections 3.1.6 and 3.2.1, respectively. However, when the two are put together in (72c), the internal reading of different disappears.

(72) a. Terry wants to ____ but Leslie refuses to ____, work on different languages.
b. Terry has ____, and Leslie will ____, work on a new language.
c. Terry has ____, and Leslie will ____, work on different languages.

Larson (2012), however, shows such conflicts do not always result in ungrammaticality, which suggests that there must be a way to account for morphological mismatches in a way that does not necessarily rely on ellipsis. The examples in (73a)–(73b) confirm this intuition: they combine wide-scope interpretation of the pivot (which is indicative of a movement or a multidominant structure) with a morphological mismatch (which is indicative of ellipsis). Interestingly, the result is grammatical, contra the prediction made by the eclectic account. In both cases, inverse scope is possible.

(73) a. Some woman must ____, and some man ought to be ____, working with every student. (Larson 2012, 149)
b. Every professor must have ____, and every research assistant should ____, work on a famous RNR puzzle.

4 Conclusion

To conclude, we have seen how the three major accounts of Right Node Raising handle major properties of this construction. The accounts we saw differed along two parameters: the position of the pivot (in situ versus ex situ) and the number of the copies of the pivot (a single copy versus multiple copies with deletion). On some accounts, RNR initially involves two copies of the shared element, one in each conjunct, which undergo rightward ATB movement above the coordination level. On others, one of them undergoes deletion. And yet on others, a single copy is literally shared between the two conjuncts, resulting in a multidominant structure. We also examined the diagnostics that are used to weigh in on the choice between movement, multidominance, and ellipsis accounts. These included constituency, case and agreement, reconstruction effects, binding, islandhood, scope, relational modifiers, vehicle change effects, and the availability of strict versus sloppy identity readings. The list in (74) summarizes the behavior of RNR with respect to these diagnostics.

(74) a. RNR does not obey (all) island constraints.
b. RNR can target non‑constituents.
c. RNR allows P‑stranding and C‑stranding in languages that (otherwise) disallow such stranding.
d. The pivot is treated as if it were in situ with respect to binding and negative polarity item licensing.
e. The pivot can have scope over both conjuncts.
f. RNR element has to be right‑peripheral (but RNR can be fed by Heavy NP Shift inside the conjuncts).
g. The pivot allows certain types of morphological mismatches (i.e. cumulative agreement and proximity).
h. RNR allows vehicle change effects.

Acknowledgments

I would like to thank the editors and the anonymous reviewers for useful feedback; I alone remain responsible for any errors and omissions. Since the chapter was originally submitted in 2013, it does not include more recent research on RNR.

SEE ALSO: Across‐the‐Board Phenomena; Conjoined Wh‐Questions; Extraposition; Preposition Stranding; VP‐Ellipsis

    Notes

  1. 1 Given the syntactic nature of the contributions in this volume, I will not discuss the prosodic properties of Right Node Raising (for discussion, data, and references, see Hartmann 2000; Selkirk 2002; Féry and Hartmann 2005; An 2007).
  2. 2 The sharing alluded to here does not imply that (1) is equivalent to (i).
    (i) Terry praised a new movie by Woody Allen and Leslie criticized a new movie by Woody Allen.
    The lack of equivalence between the two was made explicit by Geach (1970), who noted the lack of equivalence between (i) and (ii):
    (ii) All the girls admired ____, but most of the boys detested ____, one of the saxophonists.
    (iii) All the girls admired one of the saxophonists, but most of the boys detested one of the saxophonists.
    (Geach 1970, 8)
  3. 3 Postal (1974), however, is not the first one to describe the construction. He himself credits Hankamer (1971) and Maling (1972), who use the term node raising with previous discussions of Right Node Raising (see also Ross 1967; Dougherty 1970).
  4. 4 This is fairly standard but not completely uncontroversial, as movement is sometimes reduced to multidominance, and some cases of ellipsis have been reduced to movement (see e.g. Johnson's (2004; 2009) analysis of gapping).
  5. 5 I will limit my attention here to what might be broadly characterized as transformational generative treatments of RNR. This choice reflects my own theoretical background and expertise, and does not imply the lack of literature on RNR in other frameworks. For representative treatments of RNR in other frameworks, I refer the interested reader to Cann et al. (2005) (Dynamic Syntax), Yatabe (2003), Chaves (2007) (HPSG), Postal (1998) (Arc Pair Grammar), Chen-Main (2006) (Tree Adjoining Grammar), and Steedman (1985) (Categorical Grammar). See also Geach (1970) for an early categorical grammar treatment. I thank an anonymous reviewer for bringing Geach's work to my attention.
  6. 6 This does not mean that Parallel Merge can combine more than two objects at a time, as is it sometimes understood. (2) is a result of a two-step binary Merge; first XP merges with Z, and next XP merges with Y.
  7. 7 It is not my goal here to defend the existence of multidominance in the grammar. For relevant recent discussion, I refer the interested reader to Citko (2011a; 2011b), Gračanin-Yuksek (2007; 2013), Wilder (2008), Johnson (2012), Van Riemsdijk (2000; 2006a; 2006b), De Vries (2012), and the references therein.
  8. 8 I refer the reader to Larson and Hornstein (2012) for a discussion of the differences between (4a) and (4b), which they refer to as the Occurrence Theory of Movement (OTM) and the Copy Theory of Movement (CTM), respectively. They show that while in many cases OTM and CTM appear to be notational variants, there are situations in which the two make distinct empirical predictions. One such case involves Lebeaux-style late adjunction effects, illustrated in (i)–(ii) below. The CTM allows the relative clause to be adjoined to one copy without (necessarily) being adjoined to the other, whereas the OTC does not allow it.
    (i) Which picture that Billi hated did hei say that Mary took?
    (ii) Which habit that Billi hated did hei say that Mary finally kicked?
    (Larson and Hornstein 2012, 14)
    Their argument shows that a multidominant approach might not be the right approach to movement; it does not bear on the question of whether multidominance is the right approach to RNR. It does nevertheless show that movement and multidominance are two distinct mechanisms, and therefore is consistent with the assumption that a movement and a multidominant approach to RNR are not notational variants of each other. Sauerland (2007) makes a similar point about ellipsis versus multidominance, also showing that the two are distinct processes with different empirical predictions.
  9. 9 On the empty operator approach to ATB movement, the wh-phrase (or the pivot) moves from one conjunct, and the empty operator moves within the other conjunct. Such an account would presumably capture the sharing of the pivot by means of a mechanism responsible for ‘identifying’ these two movement chains, akin to the mechanism employed in parasitic gap formation or tough-movement constructions. Zhang's (2010) approach also involves wh-movement from one conjunct and a pro-element in the other conjunct.
  10. 10 We might wonder about the status of (6a) or (6b) with respect to the Coordinate Structure Constraint (CSC). Strictly speaking, neither (6a) nor (6b) involves movement from a single conjunct, which is what the CSC cares about. In (6a), movement happens from both conjuncts simultaneously. In (6b), movement happens across two ‘conjuncts’ before they even become conjuncts (i.e. before they are combined to form a coordinate structure).
  11. 11 The relevant definitions that accomplish this goal are:
    (i) a. X fully dominates α if X dominates α and X does not share α.
    b. α is shared by X and Y if (i) neither of X and Y dominates the other, and (ii) both X and Y dominate α.
    c. X c‑commands Y only if X does not fully dominate Y.
    d. d(A) = the set of terminals fully dominated by A.
    (Wilder 1999, 590–591)
  12. 12 Chaves (2011) makes a similar proposal, couched in HPSG terms.
  13. 13 The ungrammaticality of (i)–(ii) has also been attributed to the fact that the pivot has to be a constituent.
    (i) *John offered ____, and Harry gave ____, Sally a Cadillac.
    (ii) *John told ____, and Harry showed ____, Seymour that Sally was a virgin.
    (Gazdar 1981, 80, crediting Hankamer 1971)
    The argument is not as straightforward as it might first appear, given the existence of many structures for double object constructions in which ‘Sally’ and ‘a Cadillac’ (or ‘Seymour’ and ‘that Sally was a virgin’) are a constituent, such as a small clause structure of Hoekstra (1988), a low applicative structure of Pylkkänen (2008), or a VP shell structure of Larson (1988), to name a few. It is also not the case that such examples are generally ungrammatical. As pointed out by an anonymous reviewer, (iii) improves quite a bit:
    (iii) Peter (only) offered ____, but Harry actually gave ____, her/Sue a Cadillac.
  14. 14 The Polish example in (15c) involves the same verbal root with different perfective prefixes. These prefixes (which are prepositional in origin), in addition to contributing perfectivity by making an imperfective root perfective, modify the core meaning of the verb.
    (i) po‑pisać after+write = write a little
    pod‑pisać under+write = sign
    prze‑pisać across+write = copy
    do‑pisać to+write = add by writing
  15. 15 Bresnan, however, interprets this to mean that constituency is a necessary but not a sufficient factor.
  16. 16 To illustrate, the Right Roof Constraint accounts for the following contrast:
    (i) [That [a review ti] came out yesterday [of this article]i] is catastrophic.
    (ii) *[That [a review ti] came out yesterday] is catastrophic [of this article]i.
    (adapted from Ross 167, 305)
  17. 17 Postal (1998) points out a number of other similarities between movement and RNR constructions (involving constraints on the extraction of indirect objects, post-nominal of genitives, and licensing of parasitic gaps, among others), arguing in favor of a movement account.
  18. 18 This restriction states that only subjects (and oblique arguments) may undergo movement (see Rackowski and Richards (2005) and Aldridge (2008), among others, for data and analysis). What is interesting for our purposes is Sabbagh's (2008) observation that, in Tagalog, RNR constructions’ non-subjects cannot be pivots:
    (i) * N‑agsara si Juan , at n‑agbukas si Pedro , ng pintuan.
    agr.asp‑close s Juan and agr.asp‑open s Pedro ns door
    ‘Juan closed, and Pedro opened, a door.’
    (Sabbagh 2008, 506)
    Larson (2011), however, points out that once factors like specificity and focus structure are taken into consideration, the parallelism between elements that can be A-bar moved and function as pivots disappears.
  19. 19 The argument weakens, however, on the copy theory of movement, on which even if the pivot moves, there is a copy of it inside the elided VP.
  20. 20 The list is not exhaustive; I refer the reader to Postal (1998) for others.
  21. 21 Interestingly, though, Postal (1998) notes that in English the prepositions that resist pied-piping are also impossible in RNR:
    (i) * At what idea could Jane have scoffed?
    (ii) *Jane could have scoffed ____, and should have scoffed ____, at that idea.
    (Postal 1998, 128)
  22. 22 This is not the conclusion Sabbagh (2007) draws. For him, wide scope is indicative of a movement analysis.
  23. 23 As noted by Carlson (1987) and Moltmann (1992), among others, the internal reading can also be licensed by coordination of other categories besides DPs:
    (i) Different people discovered America and invented bifocals.
    (Carlson 1987, ex. 18a)
    And (ii) below shows that the internal reading requires semantic (not necessarily syntactic) plurality:
    (ii) The couple likes different types of movies – and that might lead to problems, eventually.
    Thank you to a reviewer for pointing this out to me.
  24. 24 The modifier ‘different’ behaves similarly:
    (i) John whistled ____, and Mary sang ____, different tunes.
    Hartmann (2000), however, notes variation in the availability of internal readings of ‘different’, noting that not all English speakers accept it and that German generally disallows it.
  25. 25 Wilder's formulation, given in (i), allows the pivot to be non-final in the final conjunct.
    (i) If α surfaces in the final conjunct (RNR), gap(s) corresponding to α must be at the right edge of their non‑final conjuncts.
    (Wilder 1999, 587)
  26. 26 The discussion in this section relies on the not uncontroversial assumption that the Right Edge Restriction is a syntactic (rather than a phonological) constraint (see Hartmann 2000 for the latter view).
  27. 27 I am borrowing Gračanin-Yuksek's (2007) convention and enclosing offending elements in exploding starbursts.
  28. 28 Interestingly, the pivot receives plural agreement but nevertheless cannot contain a collective predicate. Grosz (2015) refers to it as an anti-collectivity effect.
    (i) *Sue's proud that Bill ____, and Mary's glad that John ____, have finally tBill/John met.
    (Grosz 2015, 4)
  29. 29 The subscripts on the verbs indicate what cases these verbs require; the verb ‘lubić/like’ takes accusative objects, whereas the verb ‘unikać/avoid’ takes genitive ones.

      30. The full text of this article hosted at iucr.org is unavailable due to technical difficulties.