2. Properties of swiping in LFF

In this section, we describe the external distribution of swiping in LFF in general terms, before turning specifically to the wh-phrases and prepositions that can (and cannot) appear in the construction.

2.1 Distribution

Merchant (Reference Merchant2001) establishes the generalization that a language will allow P(reposition)-stranding under sluicing only if that language allows P-stranding under wh-movement in non-elliptical clauses. English is a prime example of such a language: since it permits P-stranding generally (3a), it permits prepositions to be stranded and deleted in sluicing (3b), in addition to optional pied-piping (3c).

1. (3) Jack borrowed money from someone, but…

   1. a. I don't know who_i he borrowed it from t _i.

   2. b. I don't know who_ihe borrowed it from t _i.

   3. c. I don't know [from who(m)]_ihe borrowed it t _i.

In his seminal work on sluicing, Ross (Reference Ross1969) observed that English also allows a subtype of sluicing in which the preposition is retained, but unlike in (3c) follows rather than precedes its wh-complement. Thus, the continuation in (4a) has the elliptical variant in (4b).

1. (4) Jack borrowed some money, but…

   1. a. I don't know who he borrowed it from.

   2. b. I don't know who from.

According to Merchant (Reference Merchant, Zwart and Abraham2002), the construction is also found in Danish and “some varieties of Norwegian” (see his article for examples; also Hasegawa Reference Hasegawa2007).

Conversely, a language like German, which does not allow P-stranding under overt wh-movement (5a), permits neither preposition omission under sluicing nor swiping (5b).

P-stranding under regular, non-elliptical wh-movement is thus a necessary precondition for preposition omission under sluicing as well as swiping. It is not a sufficient condition for swiping if Merchant's and Hasegawa's claims about the infelicity of swiping in Swedish, Icelandic, and (dialects of) Norwegian are correct: for reasons that remain unclear, these languages do permit P-stranding in non-elliptical contexts but do not permit swiping under sluicing (see also footnote 2).

European French patterns with German in allowing neither P-stranding nor swiping (Dagnac Reference Dagnac, van Craenenbroek and Temmerman2019). However, some dialects of Canadian French are known to permit P-stranding productively (see King and Roberge Reference King and Roberge1990, Roberge and Rosen Reference Roberge and Rosen1999, among others), as illustrated in (6) for LFF.Footnote ⁴

Apparently as a result of this, LFF also permits swiping in both embedded and matrix contexts, as shown in (7).

The non-inverted order (pour qui) is equally if not more natural.Footnote ⁵

Following Rosen (Reference Rosen1976), Merchant (Reference Merchant, Zwart and Abraham2002) and Hartman and Ai (Reference Hartman, Ai, Grohmann and Panagiotidis2009) claim that swiping is most felicitous when the sluiced PP is a sprouted adjunct in the sense of Chung et al. (Reference Chung, Ladusaw and McCloskey1995), that is to say, when it has no overt correlate in the antecedent. Hartman and Ai provide the contrast in (8).

1. (8) a. She has a date tonight, but she won't tell me who with.

   b. *She has a date with some guy, but she won't tell me who with.

They note that this judgment is not very robust and provide a number of counterexamples. In the judgment of the second author of this article, non-sprouted instances of swiping as in (8b) are fairly natural. The judgment extends to swiping in LFF in both embedded (9a) and matrix contexts (9b), where the non-sprouted variant is at most slightly degraded.

The presence of a PP correlate in the antecedent thus does not appear to significantly affect the felicity of LFF swiping, and we will consequently abstract away from this issue in what follows. We also remain agnostic as to whether (in-)sensitivity to the presence of a correlate constitutes a difference between LFF and English, given the inconsistency of judgments reported in the literature.

As in English, P–wh inversion as witnessed in swiping is entirely illicit in non-elliptical contexts like (10), and with non-wh fragments, as in (11) (Merchant's Reference Merchant, Zwart and Abraham2002Sluicing Condition on swiping).Footnote ⁶

In multiple sluicing, only the linearly first remnant can invert with its preposition, as in (12); swiping is illicit for any subsequent remnants, as shown in (13).

While Merchant (Reference Merchant, Zwart and Abraham2002: 314, fn. 13) questions the possibility of swiping in English multiple sluicing, Richards (Reference Richards1997: 164) reports judgments analogous to those above.

The wh-phrase and the preposition can be separated by certain kinds of non-sentential adverbs, as in (14a), as well as by an unelided main clause, as in (14b), but not at all by an island boundary, as shown in (15).

Note that, as indicated above, island-sensitivity obtains whether or not the swiped wh-remnant is sprouted.Footnote ⁸

Swiping is not clause-bounded, however: (16) permits both a short and a long construal, depending on whether the ellipsis is resolved against the entire preceding matrix clause or only against the embedded clause.

What we can glean from the facts presented in this subsection is that swiping exists in LFF, that it does not categorically require sprouting, that it occurs only in (initial) sluicing remnants, and that the wh-phrase and the preposition can be separated as long as locality of movement is respected. We turn next to the question of which wh-phrases and prepositions can appear in LFF swiping.

2.2 Wh-phrases in LFF swiping

Swiping is fairly limited with regard to the wh-phrases that can appear in the construction. For English, Merchant (Reference Merchant, Zwart and Abraham2002) claims that only morphologically simplex wh-elements are felicitous in swiping; by contrast, Hartman and Ai (Reference Hartman, Ai, Grohmann and Panagiotidis2009) and other authors report that certain complex wh-elements are felicitous as well. The picture in LFF is similarly diffuse: while by and large, swiping in LFF prefers simplex wh-phrases, not all of them can appear in swiping; furthermore, LFF swiping does tolerate some complex wh-phrases.

The examples from LFF in (17) are in line with Merchant's (Reference Merchant, Zwart and Abraham2002) claim (based on English data) that complex wh-phrases are typically infelicitous in swiping:

All of the above are fully felicitous in their non-elliptical form, as in (18), showing that their deviance is not due to illicit P-stranding.

The same is true when the wh-phrase and the preposition appear in the uninverted order, as in (19).

The cases in (17) contrast with instances of swiping with simplex wh-phrases, as in (20).

As before, the judgments track the status of non-elliptical variants:

With the exception of lequel (and possibly combien), the wh-phrases in (20) are monomorphemic.Footnote ¹⁰ However, it would be premature to conclude based on this contrast that only simplex, head-like wh-phrases can undergo swiping. Wh-phrases of the form combien de N ‘how much/many N’ and quel(le) N ‘which N’, like those in (22), are acceptable in some cases.

The paradigm in (23) summarizes the observed restriction. Beyond the simplex wh-phrase qui, only a minimal [D_wh N ] remnant is potentially licit in swiping; any additional complexity renders the configuration unacceptable.

Thus, while the claim that “swiping is perfectly well-formed with simple wh-phrases [but] systematically excluded with complex ones” (van Craenenbroeck Reference van Craenenbroeck, Merchant and Simpson2012: 57) is too strong at least for LFF, we see that LFF swiping is restricted with regard to the wh-phrases that can occur in the construction, as also reported for English (Culicover Reference Culicover1999, Culicover and Jackendoff Reference Culicover and Jackendoff2005).

2.3 Prepositions in LFF swiping

Next, we consider the question of which prepositions can appear in LFF swiping. We will not attempt, within the confines of this article, to comprehensively test all possible prepositions; rather, our discussion will aim to develop a first sketch of the general picture, to be refined in future research.

As was observed for English by Ross (Reference Ross1969, 266), strandability of a preposition P is a precondition for P to appear in LFF swiping. For instance, the complex preposition jusqu’à cannot be stranded under regular wh-movement (24a), and is likewise excluded from swiping (24b).

Turning now to strandable prepositions, functional de ‘of/from,’ semi-functional pour ‘for’ and par ‘by,’ and locative sous ‘under’ are permissible in swiping:

The lexical prepositions contre ‘against’ and entre ‘between’ are likewise licit in LFF swiping:Footnote ¹¹

Unlike de (25a), the other purely functional preposition à ‘to’ is excluded from swiping, as are comitative avec ‘with,’ locative (de)dans ‘in,’ dessus ‘on (top of)’ and dessous ‘under,’Footnote ¹² as well as temporal après ‘after’ and avant ‘before’.Footnote ¹³

As indicated in the translations, English swiping appears to be somewhat more liberal than LFF swiping with regard to the range of permissible prepositions in at least some cases (permitting who with and who to).Footnote ¹⁴

Prepositional complexes such as à côté de and en faveur de are likewise excluded:

Importantly, LFF permits stranding of all illicit prepositions above under wh-movement in non-elliptical constructions; their infelicity when following the associated wh-phrase in linear order is specific to the swiping construction. The examples in (29) are representative.

The empirical picture emerging here is rather murky and in need of further clarification, which we hope future work can provide. The prepositions permissible in LFF swiping do not seem to form a natural class syntactically, phonologically, or semantically. In addition, the set of permissible prepositions is not coextensive with the corresponding set in English swiping.

3. Theoretical implications

In this section, we discuss some rather significant implications of LFF swiping for the theory of swiping and sluicing. After briefly sketching existing approaches to swiping, we suggest an alternative analysis according to which swiping reduces to P-stranding, as originally suggested by Ross (Reference Ross1969). While this approach will not immediately derive all properties and constraints observed above, we show that it avoids certain serious problems faced by alternative analyses.

Existing analyses of swiping fall into two major categories. Building on a suggestion in van Riemsdijk (Reference van Riemsdijk1978), Lobeck (Reference Lobeck1995: 61f.) and Merchant (Reference Merchant, Zwart and Abraham2002) propose that swiping is the result of P–wh inversion internal to the remnant PP; on Merchant's analysis, this is achieved by wh-to-P head movement, whereas Lobeck assumes that the wh-phrase moves to the PP's edge.Footnote ¹⁵ Abstracting away from this difference in implementation, the analysis is schematically illustrated in (31) for the example in (2), repeated in (30) for convenience.

As an alternative to this internal-inversion approach, Richards (Reference Richards2001), van Craenenbroeck (Reference van Craenenbroeck2004), Hartman and Ai (Reference Hartman, Ai, Grohmann and Panagiotidis2009), and Radford and Iwasaki (Reference Radford and Iwasaki2015) develop variants of what we refer to as an external-inversion analysis. On this approach, the PP containing the wh-phrase is raised to some left-peripheral position (labeled XP in (32)),Footnote ¹⁶ and subsequently the wh-phrase is subextracted to an even higher position:

1. (32) External inversion

   [_CPqui _i[_XP [_PP pour t _i ]_k … [_TP … t _k…] ]]

What both types of approach have in common is the assumption that sluicing is derived by TP-deletion, and consequently that all and any remnants must be evacuated from TP.Footnote ¹⁷ This assumption turns out to be problematic, since the relevant configuration cannot be created outside of elliptical contexts. However the inversion of P and wh-phrase is assumed to come about, it can never occur in wh-in situ configurations or when no deletion takes place (recall the analogous examples from LFF in (10)).

1. (33) a. *He was talking WHAT about?!

   b. *What about was he talking?

To account for this fact, Merchant's (Reference Merchant, Zwart and Abraham2002) head-movement operation, which adjoins a minimal wh-complement to its selecting P-head, is stipulated to apply only under sluicing; but no principled reason for this restriction is given. It remains unclear why this operation can apply only in initial remnants in multiple sluicing (recall (12) vs. (13)). In fact, it is not even clear on this approach why P–wh inversion could not take place in a language like German (a non-P-stranding language, which consequently lacks swiping; recall (5) above), since the postulated head-movement operation, assumed to apply at PF, is formally entirely distinct from P-stranding under phrasal ${\bar{\rm A}}$-movement. Furthermore, an analysis of swiping in terms of head movement falsely rules out any swiping with non-minimal wh-phrases (as in (23b)), as proponents of external-inversion analyses have pointed out (e.g., Hartman and Ai Reference Hartman, Ai, Grohmann and Panagiotidis2009).

While permitting complex wh-phrases in swiping, external-inversion approaches as illustrated in (32) above do not fare much better with regard to the other points mentioned. To rule out cases like (33b) (and, by extension, their LFF counterparts in (10) above), these analyses likewise restrict the application of swiping to sluicing contexts by mere stipulation. Similarly, the asymmetry between initial and non-initial remnants in multiple sluicing again remains unaccounted for.Footnote ¹⁸ Nonetheless, external-inversion approaches might appear to have an edge over Merchant's (Reference Merchant, Zwart and Abraham2002) approach by establishing a more direct link between inversion in swiping and genuine P-stranding (as explicitly claimed, e.g., in Hartman and Ai Reference Hartman, Ai, Grohmann and Panagiotidis2009): the possibility of wh-extraction from PP within the left periphery is claimed to be an instance of P-stranding, which must be independently licensed in the language for swiping to occur. But as observed by Merchant (Reference Merchant, Zwart and Abraham2002: 300) in response to Richards (Reference Richards2001), the postulated inversion operation has very little in common with bona fide P-stranding, stranding the preposition in a derived, left-peripheral position.Footnote ¹⁹ In fact, as Merchant points out, this kind of derivation violates a robust and general ban against P-stranding in left-peripheral positions (first discussed in Postal Reference Postal1972).Footnote ²⁰

1. (34) *Who_i do you think [_CP [to t _i]_k (that) John talked t _k]?

But illicit peripheral P-stranding as in (34) is essentially what all external-inversion analyses of swiping explicitly countenance (recall (32)). Consequently, P-stranding in the derivation of swiping can only be exceptional P-stranding on these approaches, licensed – for unspecified reasons – under sluicing but not otherwise. But then, there is no longer any direct, principled link between swiping and bona fide P-stranding, which means in turn that external-inversion approaches fail, just like internal-inversion analyses, at explaining why English and LFF permit swiping whereas German does not. In short, a major generalization about swiping – its dependence on P-stranding, recognized clearly by Ross (Reference Ross1969) – ultimately remains unaccounted for on all existing approaches.Footnote ²¹

We would like to suggest an alternative approach that does not suffer from this defect, by rejecting the assumption that sluicing is deletion of a syntactic constituent (IP/TP). Instead, we propose that deletion simply targets all given and prosodically demoted material in the clause (the clausal background, excluding any F-marked material including wh-phrases; see Reich Reference Reich, Schwabe and Winkler2007):

1. (35) … je ne sais pas qui_iqu'il a acheté un cadeau [pour t _i]_f

The preposition is spared not due to some sort of exceptional, ellipsis-induced evacuation movement, but simply by virtue of being part of a surface-discontinuous focal constituent; inversion of preposition and wh-phrase is effected by P-stranding wh-movement of qui alone. This analysis of swiping is not new: it spells out Ross's (Reference Ross1969) original suggestion that “it is possible to delete everything in [a question] but the question word and a stranded preposition” (p. 265).Footnote ²² This view of deletion aligns it with deaccentuation, an alternative means of prosodic givenness-marking, which can likewise affect non-constituents (see Tancredi Reference Tancredi1992, Chomsky and Lasnik Reference Chomsky, Lasnik, Jacobs, Stechow, Sternefeld and Vennemann1993).Footnote ²³ The insistence on single-constituent deletion has been argued to be problematic by Bruening (Reference Bruening2015) and Ott and Struckmeier (Reference Ott and Struckmeier2018) and is rejected in earlier works such as Hankamer (Reference Hankamer1979) and Morgan (Reference Morgan, Kachru, Lees, Malkiel, Pietrangeli and Saporta1973).

By eschewing exceptional operations, our alternative view of clausal ellipsis in swiping as the purely prosodic deletion of recoverable material straightforwardly explains why swiping can occur only where P-stranding is independently possible, and why the swiping pattern can only arise in elliptical contexts: as shown in (35), the order wh $\prec $P arises simply as a result of ordinary P-stranding under wh-movement (potentially with intervening material: (14)).

By the same token, this analysis correctly predicts the availability of long-distance construals in swiping (20) as well as the island-sensitivity of the construction (15).Footnote ²⁴ Multiple sluicing with swiping in the first remnant is the result of combining P-stranding with a second in situ remnant, as shown below for (12):

1. (36) … je ne sais pas qui_iqu'il a gagné [contre t _i]_f [dans quel match]_f

This reduction of swiping to P-stranding makes sense of the fact that initial, but not non-initial remnants in multiple sluicing permit the swiping pattern (recall (13)). The latter always necessarily remain in situ, LFF not being a multiple-wh-fronting language. As far as we can see, none of the existing, inversion-based analyses make accurate predictions about swiping in multiple sluicing.

In the remainder of this section, we present two empirical arguments in favor of our alternative approach, based on LFF data. The first argument concerns the immobile wh-phrase quoi ‘what’, which fails to participate in swiping; we claim that only our approach provides a principled explanation for this behavior. The second argument shows that where P-stranding and pied-piping of P are not in free variation but correlate with meaning differences, it can be detected that swiping involves genuine P-stranding rather than initial pied-piping of P, contrary to what is assumed in all previous accounts.

Like European French, LFF has wh-phrases that robustly resist fronting. One example is the French counterpart of English what, which has a strong (tonic) form quoi and a weak (clitic) variant que.Footnote ²⁵ The former only appears in situ, whereas que must surface ex situ:

As observed by Dagnac (Reference Dagnac, van Craenenbroek and Temmerman2019), quoi can be sluiced; in fact, only quoi, but not its weak ex-situ counterpart que, can appear in a sluiced question:

The sluiceability of immobile quoi is unproblematic for a theory of sluicing permitting deletion of a non-constituent string, as suggested here: on such an approach, quoi in (38) is simply an in-situ remnant of deletion. Approaches that insist on single-constituent deletion must resort to either exceptional evacuation movement, or else a process of ellipsis-conditioned allomorphy.Footnote ²⁶ Either type of approach will struggle to capture the facts discussed immediately below, however.

There is one important exception to the general immobility of quoi:Footnote ²⁷ when it is the complement of a preposition, this quoi-containing PP can be fronted, as long as the preposition is pied-piped.

As before, only strong quoi but not clitic que can appear in this context, including under sluicing, as shown in (40). Outside of sluicing, quoi can never invert with its preposition, whether in situ or ex situ, as shown in (41).

Given that quoi can be fronted as part of a PP (39), Merchant's (Reference Merchant, Zwart and Abraham2002) analysis predicts that it should optionally invert with its selecting preposition just in case sluicing applies. The prediction is not borne out, however:Footnote ²⁸

By contrast, on our analysis, an illicit swiping construction as in (42a) is derived as shown in (43a), making it underlyingly identical to non-elliptical (43b), which is equally infelicitous.

1. (43) a.  *quoi_ia-t-il parlé [de t _i]_f

   b.  *Quoi a-t-il parlé de?

Recall that external-inversion analyses establish no direct link between illicit P-stranding in non-elliptical contexts as in (43b) and the infelicity of swiping, as in (42), since these approaches uniformly resort to exceptional P-stranding subsequent to pied-piping of P. In other words, on such an approach there is no reason that the infelicity of P-stranding in (43b) should block the application of the sluicing-specific inversion operation illustrated in (32) above, since PP-contained quoi can move to the left periphery, as in (39b) (and can appear in sluicing, as in (40)). We thus submit that the categorical absence of quoi-sluices in LFF points to the conclusion that pied-piping-plus-inversion analyses of swiping are flawed: swiping reduces to P-stranding and deletion of recoverable material separating the fronted wh-phrase from its selecting preposition.Footnote ²⁹

Our second argument militates even more directly against both internal- and external-inversion analyses of LFF swiping. This argument capitalizes on meaning differences between otherwise identical questions with and without P-stranding. The following paradigm illustrates a case in point.Footnote ³⁰

The in situ question in (44a) is most naturally interpreted with de qui construed as the complement of photos. The same “of who” reading obtains in (44b), indicating that the wh-phrase is subextracted from the complement PP to the exclusion of its selecting preposition. Extraction of the entire PP (44c), however, yields a different salient interpretation, which instead construes the extracted PP as a modifier in the underlying structure, asking about the source of the picture rather than its content (“from who”).Footnote ³¹

These cases allow for a direct comparison of the predictions of inversion analyses (both internal and external) of swiping on the one hand and our P-stranding analysis on the other, since the former approaches assume that the source of swiping is PP-extraction as in (44c), whereas we derive it directly from P-stranding as in (44b). To see this, consider the swiped variant of the above questions, in an analogous context:

According to all TP-deletion-based analyses of swiping, B's response must involve fronting of the PP de qui and subsequent inversion, either internal to the PP or by means of subextraction of qui. The latter analysis is illustrated below.

1. (46) a. [_XP [_PP de qui ]_k … [_TP…t _k… ]] →

   b. [_CPqui_i [_XP [_PPde t _i ]_k … [_TP…t _k… ]]]

As we saw with (44c), fronting of the entire PP yields the modifier reading (source of the picture, “from who”); consequently, the swiped question in (45) should have the same reading, given that it derives from an analogous source (46a). Crucially, however, this is not the case: its interpretation matches that of (44b), not that of (44c).Footnote ³² While inversion analyses thus make the wrong prediction about the meaning of B's response in (45), our approach derives the swiping order directly from the meaning-identical question with P-stranding in (44b):

1. (47) Qui_iaimerais-tu avoir une des photos [de t _i]_f?

   ‘Who would you like to have one of the pictures of?’

We thus submit that the interpretation of swiping constructions as in (45) strongly suggests that LFF swiping – and presumably swiping in general – involves bona fide P-stranding and prosodic deletion of the informational background, not exceptional P-stranding fed by PP-fronting and TP-deletion.

An analogous argument based on English data is mentioned in passing by Merchant (Reference Merchant, Zwart and Abraham2002: 314, fn. 13). He observes that the combination of for and what can have an idiomatic reading (roughly meaning ‘why, for what reason’) if and only if for is stranded:

1. (48) a. What did he do that for? (≈Why did he do that?)

   b. #For what did he do that?

Merchant points out that a corresponding swiping construction has the idiomatic reading (as already noted by Ross Reference Ross1969: 265), suggesting that it does not derive from the same source as (48b).

1. (49) He did it, but I don't know what for. (≈ …why he did it)

As with the above LFF case, it is unclear how any approach assuming movement of the entire PP to the left periphery could account for this fact, given that such an approach necessarily postulates an underlying structure for the swipe in (49) that is isomorphic to that of (48b) in relevant respects. By contrast, the P-stranding-cum-deletion analysis advocated here does not face this problem, since it analyzes (49) as shown in (50), correctly predicting interpretive equivalence with (48a).

1. (50) what_idid he do it [for t _i]_F

The above observations strongly suggest that inversion analyses are untenable for English and LFF swiping alike, and that swiping should instead be analyzed as ordinary wh-movement and P-stranding in syntax and subsequent prosodic (non-constituent) deletion at PF.

What is left open by this approach (and any other, as far as we can tell) is an account of the contrasts and subtleties described in sections 2.2 and 2.3, where only a subset of the possible combinations of (strandable) prepositions and (mobile) wh-phrases was found to be permissible in swiping, while others are perceived as less natural. We leave it to future research to address this gap in the current understanding of swiping. Given that the observed restrictions on permissible prepositions and wh-phrases have no obvious characterization in syntactic or semantic terms, it seems likely to us that the wh $\prec $P surface sequence remaining after deletion must satisfy prosodic constraints and conform to phonotactic preferences. Future work should seek to unravel these factors which, jointly with syntactic constraints pertaining to the mobility of the wh-phrases and strandability of the prepositions involved, determine the range of felicitous swiping configurations. Hopefully these investigations will also shed light on the vexing question – left unanswered by all approaches, including ours – why certain languages that permit P-stranding under non-elliptical wh-movement, such as Quebec French and Icelandic, nevertheless do not appear to tolerate swiping.