2.1 Licensing conditions

In an influential paper, Johnson (Reference Johnson, Baltin and Collins2001) observes that vP-preposing is subject to a number of very specific licensing conditions that are not familiar from A $^{\prime }$ -movement of other elements, and that these conditions are the same as those found for vP-ellipsis.Footnote ^[5] Most strikingly, the ‘trace’ must be governed by an auxiliary; lexical verbs alone license neither vP-preposing nor vP-ellipsis, and do-support is always triggered when no other auxiliary is present.Footnote ^[6] The examples in (8) are thus well-formed:

All of the auxiliary elements that are in bold in (8) license vP-ellipsis in the same configurations:

When there is no auxiliary element to act as a licensor, however, both vP-ellipsis and vP-preposing fail.

The (a) examples show this for auxiliary-less cases, and the (b)–(d) examples show that both vP-ellipsis and vP-preposing fail when the target vPs occur in a number of other auxiliary-less non-finite embedding contexts, such as go+vP constructions, feel like complements and perception verb complements.

The correlation between failure to license vP-ellipsis and vP-preposing also holds when we look at other kinds of passive embedding contexts, such as get-passives and the more dialectally restricted need-passives (see Edelstein Reference Edelstein, Zanuttini and Horn2014), as illustrated by (12) and (13). Although get and need are auxiliary-like, they license neither vP-ellipsis nor vP-preposing.

Under the hypothesis that vP-preposing is simply A $^{\prime }$ -movement, these restrictions on vP-preposing are mysterious: no such conditions exist for topicalisation of arguments or wh-movement, for instance.Footnote ^[7] This mystery is further compounded by the fact that these restrictions are not shared by processes of vP-movement in all languages. To see this, consider vP-fronting in German. In this language, there is no requirement for an auxiliary to license the gap position created by fronting of the vP, as shown by the examples in (14).

Note furthermore that it is not the case that it is a requirement of predicate movement for the trace to be licensed. This can be demonstrated for English for wh-movement of non-verbal predicates, as in the examples in (15).

What this tells us is that there is no general requirement that vP-gaps or predicate gaps need to be licensed by an auxiliary. Rather, it indicates that the conditions observed with English vP-preposing might be of a different nature.

Another restriction shared by vP-preposing and vP-ellipsis, but unfamiliar from other movement processes, is that they both resist ‘stranding’ of adverbs left-adjacent to the empty position (Sag Reference Sag1978, Abels Reference Abels2003), as illustrated in (16). One possible analysis of these facts is to postulate that both vP-ellipsis and vP-preposing obligatorily target a constituent that contains these adverbs, and that as a result they may not be stranded. The problem with this analysis is that high adverbs like probably resist both being stranded at the left edge of the empty position and being part of a fronted constituent, (17); these facts indicate that the ban on stranding has nothing to do with the size of the constituent targeted.

Once again, no such restriction is found with German vP-fronting, as shown by (18).

How, then, should these similarities between vP-ellipsis and vP-preposing be accounted for? According to Johnson (Reference Johnson, Baltin and Collins2001), the similarities fall out naturally if we assume that ellipsis is licensed by preposing, with an unpronounced vP constituent moved to the clausal left periphery. However, Aelbrecht & Haegeman (Reference Aelbrecht and Haegeman2012) demonstrate that such an account runs into various problems. Most significantly, vP-ellipsis is licensed in a number of configurations in which vP-preposing is not licensed: for instance, vP-preposing is generally a root phenomenon, only occurring in embedded positions that generally allow for embedded root phenomena, whereas there is no such restriction on vP-ellipsis. This is demonstrated in (19) for factive environments, which resist embedded root phenomena (from Aelbrecht & Haegeman Reference Aelbrecht and Haegeman2012); the same is true of relative clauses.

It seems that the environments that allow vP-preposing are a subset of those that allow vP-ellipsis. vP-preposing is thus not capable of accounting for vP-ellipsis in any general way.Footnote ^[8] Even if these problems are put to one side, however, the restrictions on vP-preposing (and consequently vP-ellipsis) identified above remain wholly mysterious; that is, no such restrictions on stranding of adjacent material that does not form a constituent with the moved element, or licensing of traces/copies by dominating non-lexical heads, are seen with other kinds of movement. On the other hand, many of these restrictions can in fact be understood as restrictions on vP-ellipsis in the context of an explicit theory of ellipsis licensing, such as Merchant’s (Reference Merchant2001) E-feature-based approach to licensing (which would derive the auxiliary licensing condition from subcategorisation restrictions on the distribution of E) or Thoms’ (Reference Thoms2010) movement-based account (which would derive the role of auxiliaries straightforwardly from the fact that it is only auxiliaries that move into the IP-domain, and the adverb restriction from the fact that adverbs do not move to their surface position in English). This suggests that it is more sensible to reverse the picture: instead of vP-ellipsis being licensed by vP-preposing, we claim that vP-preposing is licensed by vP-ellipsis.

2.2 A detour on types of predicate ellipsis

We should be more specific about the ellipsis process that we take to be involved here, as stating that vP-preposing involves vP-ellipsis overlooks the fact that ‘vP-ellipsis’, or predicate ellipsis to use a more general term, is not a unitary phenomenon; rather, it turns out that there are at least two different subtypes of predicate ellipsis in English, and one in particular is relevant to our analysis of vP-preposing.

The background to this is the claim in Merchant (Reference Merchant2013) that we can explain the different properties of ellipsis constructions in terms of the size of the antecedent for ellipsis. Empirically, Merchant is concerned with the fact that ellipsis processes seem to differ in the extent to which they allow for mismatches in form between the antecedent clause and the ellipsis clause, in particular with respect to grammatical voice: while predicate ellipsis allows for a mismatch with respect to voice (Kehler Reference Kehler2002), the same seems not to be possible with clausal ellipsis constructions like sluicing.

This presents a problem for stating the ellipsis identity relation solely in terms of semantic representations, since passive–active pairs have the same meaning. However, Merchant argues that it can be explained if the ellipsis identity relation is a syntactic condition which imposes identity of (at least) the featural content of syntactic heads in the ellipsis site. With such a condition in hand, Merchant claims that the difference between predicate ellipsis and clausal ellipsis is that grammatical voice is encoded by a verbal head, Voice, above vP in the thematic domain: this head will always be included in the TP antecedent to clausal ellipsis, enforcing identity of the content of Voice, whereas with predicate ellipsis the antecedent may be the vP layer below, thereby allowing for mismatches.

Thus, interaction of the identity condition and the size of the antecedent for ellipsis determines grammatical properties of the ellipsis processes in question.

This becomes relevant to our account of vP-preposing when we take into account the fact that not all instances of predicate ellipsis are the same. In an in-depth study of predicate ellipsis in English, Sailor (Reference Sailor2014) argues that we can distinguish at least two kinds of predicate ellipsis on the basis of the syntactic size of the antecedent: low predicate ellipsis, which targets just the lowest vP layer containing the lexical verb and the verbalising head v, and high predicate ellipsis, which targets higher functional structure including Voice.

In line with Merchant’s findings, low predicate ellipsis is characterised by the possibility of mismatches, and largely by the extent to which parallelism holds between the elliptical clause and the antecedent. Low predicate ellipsis seems to largely be reserved for configurations where the ellipsis clause is embedded in a subordinate clause, or it stands in some kind of subordinating discourse configuration with respect to the antecedent (Kehler Reference Kehler2002). This is the case with all of the voice mismatch cases from Kehler and Merchant. Importantly, they stand in quite striking contrast to examples where the antecedent clause and ellipsis clause are simply coordinated with and and linked together by an additive focus particle like too or also; in such configurations, voice mismatches are completely impossible.

Sailor suggests that this effect is due to the size of the ellipsis site, and so he characterises plain coordination of antecedent and ellipsis clause (with neither clause embedded) as prototypical cases of high predicate ellipsis. Sailor does not speculate as to why this configuration would force high predicate ellipsis, but we would like to suggest that it may be down to the use of the additive particles too/also. The semantic contribution of these particles is to mark the subject as the focus of the clause and the rest of the constituent that they dominate as the background; given that these particles are adjoined above the ellipsis site, they therefore extend the backgrounded domain beyond the ellipsis site, right up to their adjunction position. Since backgrounded constituents require an identical antecedent (Tancredi Reference Tancredi1992), the effect is to extend the identity-dependent domain of the elliptical clause, hence the higher antecedent.

Sailor shows that a number of other grammatical properties that can be understood in terms of the height of ellipsis show asymmetries between these two kinds of antecedent–ellipsis configuration. One that is of relevance in what follows is the availability of strict readings with reflexive antecedents, where the ellipsis site and antecedent seem to mismatch with respect to the form of the pronoun (reflexive in antecedent and regular pronoun in ellipsis site). Sailor observes that such strict readings are available with low predicate ellipsis configurations but not with high predicate ellipsis ones, these only getting the sloppy reading. This contrast can be explained, Sailor argues, if we adopt Ahn’s (Reference Ahn2015) proposal that reflexives move to Spec,VoiceP in English: with low predicate ellipsis, the antecedent will exclude the reflexive, meaning that there is no need to have a reflexive in the ellipsis site and so a strict reading may be derived by using a simple pronoun in the ellipsis site, whereas this option is not possible with high predicate ellipsis.

The high predicate ellipsis option seems to be a strong preference for symmetric coordinations with additive focus particles, although it is not clear whether it can be ruled out entirely in such cases; that is, it might be possible to access a low predicate ellipsis analysis with cases like (26) in the right circumstances, with the apparent absence of the strict reading following from cumulative markedness effects (strict readings are marked even in cases like (25)). However, Sailor argues that the high predicate ellipsis option is forced with a number of polarity-reversing elliptical constructions, such as tag questions and emphatic retorts.

Significantly, these constructions share at least three properties in common with vP-preposing. First, these constructions feature some kind of operator that encodes an anaphoric dependency between the elliptical constituent and the antecedent. In the case of the retorts, it is the polarity particle that does this, and with tags this work is arguably done by whatever head or operator it is that encodes the paratactic relation between the two clauses; in the case of vP-preposing, on our analysis this is done by the operator that links the subjects of the two constituents. Second, the ellipsis process involved is obligatory, unlike with most other instances of predicate ellipsis.Footnote ^[9] This might be understood in terms of the fact that the operators involved directly encode some degree of dependency between antecedent and ellipsis clause in the syntax, making reduction a matter of syntactic economy, whereas with other kinds of predicate ellipsis these relations are mediated by discourse and so are not subject to such hard conditions.

Finally, all of these constructions require that the subject of the elided clause covaries with the subject of the antecedent clause, a requirement that can be understood if the ellipsis site for high predicate ellipsis must contain a variable bound by the subject (in Spec,VoiceP or lower). The same requirement is enforced with vP-preposing, in that the initial clause subject cannot covary with some other nominal; thus, the preposed vP in (37) cannot be interpreted as first person plural, even though it would be compatible in principle with binding of PRO by the higher subject we.

This is part of a broader tendency with high predicate ellipsis which Sailor calls ‘neophobia’, where the elliptical clause shows resistance to introducing additional information not contributed by the antecedent clause. This means that there is a high degree of matching of the set of auxiliaries present, with the introduction of additional modals or progressive morphology being impossible, although it does seem to be marginally possible to have the perfect in the elliptical clause with the simple past in the antecedent. We can see that strict morphological identity is not enforced, as different forms in (40) show (for British English).

We return to this issue in more detail below with respect to vP-preposing. For now, we take these commonalities between Sailor’s high predicate ellipsis constructions and vP-preposing to motivate treating vP-preposing as another high predicate ellipsis construction, and so we do so in what follows. This will be significant as at a number of points we will show that while the relation between the initial verbal constituent and the gap position in vP-fronting constructions is looser than what we would expect with movement, it is not quite so loose as predicate ellipsis, at least when this is understood as a unified phenomenon. However, the right looseness is captured if we treat vP-preposing as involving high predicate ellipsis.

A note on terminology: for convenience we resume talking in terms of vP-preposing and vP-ellipsis, even though on this analysis high predicate ellipsis actually involves a larger constituent than just the vP.

2.3 Morphological mismatches

If the lower and higher vP are two copies in a chain, as in the standard analysis, then they are expected to show identity effects: no vP should be able to appear in Spec,CP that is not able to appear in the position taken by the lower copy. Yet, as observed by many authors (e.g. Emonds Reference Emonds1976: 115; Andrews Reference Andrews1994: 218; Oku Reference Oku, Dubach Green and Motapanyane1996, Reference Oku1998; Urushibara Reference Urushibara, Ukaji, Nakao, Kajita and Chiba1997; Bresnan Reference Bresnan2001: 18; Dorfman, Leu & Levon Reference Dorfman, Leu, Levon, van Koppen, Sio and de Vos2002: 5; Kayne Reference Kayne2005: 99; Breul Reference Breul2014), the preposed vP need not be able to appear in this lower position, as shown by (41)–(42). See in particular Breul (Reference Breul2014) for dozens of naturally-occurring examples of this type.Footnote ^[10]

(41) shows that the preposed vP can be a bare infinitive when the base form would normally bear participial morphology; in other words, vP-preposing seems to bleed affixation here. Conversely, (42) shows that the verb in the preposed vP can be in an inflected form where it would normally be in an uninflected form in the base position (since tense morphology is absorbed by do-support), yielding an apparent tense-doubling effect. Kayne (Reference Kayne2005) suggests that (41) may be related to the Infinitivus Pro Participio (IPP) effect found in other Germanic languages, where the participle surfaces in the less specified form as an infinitive. However, this does not generalise to (42), where the verb is more specified than the expected form.

vP-preposing is also compatible with British do for the relevant speakers, as shown by (43).

Even more strikingly, whole auxiliaries can be doubled with little or no degradation, as in (44). In (44c)–(44d), one of the two non-finite auxiliaries must be present, but not both.Footnote ^[11]

The examples in (43) and (44) show that the mismatches between the preposed vP and what is possible in its supposed base position can in some cases be very great indeed. This is not at all expected on a vP-movement account.

On the basis of these data, one might propose that there are no restrictions on the morphological form of the vP when it is moved to the initial position, with all morphological affixation processes being suspended when the vP is moved. However, it is not the case that the mismatches are completely unrestricted, as (45)–(46) show that it is not possible for there to be a mismatch between the fronted verb and the verb that would normally occur in the base position with respect to progressive and passive morphology.

Similarly, not all doubling options are possible either. (47) shows that progressive being cannot be doubled in the same way that the passive auxiliary been can, even though it can be fronted productively (Harwood Reference Harwood2015: 550).

Moreover, as pointed out to us by an anonymous reviewer, doubling is not possible with be when its form in the initial vP is different from what it would be in the base position.

Thus, the challenge to the vP-movement approach in accounting for bleeding and doubling is to restrict them so that it does not overgenerate these bad cases. It is far from clear how one would formulate the vP-movement rule to obtain this result.

By contrast, the distribution of doubling and bleeding effects in vP-preposing is expected on our ellipsis approach. On this account, mismatches arise when the vP that is base-generated in the initial position is different in morphological form from the one in the ‘base position’, and the limitations on what kinds of mismatches we can get are the limitations on what kinds of morphological mismatches are allowed between elided vPs and their antecedents, in particular in cases of high vP-ellipsis. We recall from Section 2.2 that this is allowed in limited circumstances with high vP-ellipsis; in fact, it turns out that the possible mismatches line up quite closely with those that are allowed with what we get with vP-preposing.

First, let us consider bleeding, where the initial vP shows up in a less specified form than the one in the base position, and so there is ellipsis of a vP which carries verbal morphology that is not present in the antecedent. This was demonstrated by (41) above, and the schematic for what would be found underlying is given in (49). As (50) shows, high vP-ellipsis is possible in the same kind of configuration, where the verb in the antecedent clause is bare but the one in the putative ellipsis site ought to bear past participle morphology (see also (39) above for a similar mismatch). The mismatch here is slightly marginal for some speakers, and a comparable effect is found with vP-preposing cases like (41).

As for the failure of examples (45)–(46), these are expected to be ungrammatical on the ellipsis analysis because the underlying structures would be as in (51)–(52), where the ellipsis site hosts progressive and passive morphology respectively. Mismatches with respect to progressive morphology are impossible with high vP-ellipsis, as shown by (53), and we have already seen in Section 2.2 that mismatches with respect to voice are not possible with this kind of ellipsis either.

The ungrammaticality of the bleeding configurations in (45)–(46) thus follows from the fact that the initial vP is not able to provide a suitable antecedent for ellipsis of the vP in the ‘base position’.

Now consider doubling, where the initial vP shows up in a more specified form than expected. For the simplest case in (42), this is expected since it involves ellipsis of a bare verb form under identity with a finite antecedent, as shown by (55). It is shown in (56) that this is possible with high vP-ellipsis.

Doubling of the passive auxiliary as in (44), meanwhile, follows from the fact that it can optionally be included in the ellipsis site in cases of high vP-ellipsis, as (57) illustrates; for us, there is a strong preference for the (a) option, but (b) is also possible. With vP-preposing, then, doubling then arises when the smaller ellipsis site is chosen, with a vP containing the relevant auxiliary base-generated in the initial position, as shown by (58).

The ellipsis account also predicts that doubling is ungrammatical with being because it is generally the case that vP-ellipsis is degraded with being adjacent to the ellipsis site (at least for many speakers), (59). Thus, (47) has the analysis in (60), which is ungrammatical for the same reason that (59) is (see Thoms Reference Thoms2010, Rouveret Reference Rouveret2012, Sailor Reference Sailor2014 for suggestions regarding how to derive this restriction).

Finally, the proposal also allows us to account for the degradation of doubling where the non-finite forms of be do not match, whether or not the lower form is elided, since mismatches of this kind seem to be degraded with high predicate ellipsis.

It seems that the introduction of some additional semantic information with the modal will leads to too much of a mismatch between the antecedent clause and the ellipsis clause. Whatever causes this ‘neophobia’ with high vP-ellipsis can therefore be blamed for the degradation of similar mismatches with vP-preposing on our analysis.

In sum, the evidence from morphological mismatches casts serious doubt on the standard analysis of vP-preposing, while at the same time supporting the claim that there is a parallelism between preposing and ellipsis. Importantly, these morphological mismatches between the initial vP and the form found in the base position are not seen in German vP-preposing: participles do not surface as infinitives under fronting in cases such as (63), even though participles sometimes show up as infinitives under adjacency with other participles (the IPP effect).

In effect, German vP-movement shows the expected identity effects between the higher and lower positions in the proposed movement chain. Given that the preceding discussion has given us good reason to believe that German vP-fronting is indeed a movement rule, this should be unsurprising, and it only serves to underline the fact that the English facts are uncharacteristic of movement and so demand a non-movement analysis.

2.4 Evidence against predicate movement

If vP-preposing involved no movement, we would expect it to be immune to locality constraints, as shown by (64) for classical resumption (McCloskey Reference McCloskey, Everaert and van Riemsdijk2006). As (65)–(66) illustrate, however, this is not the case.

Both Complex NP islands (65) and do so (66), which blocks A $^{\prime }$ -extraction, disallow extraction here. These examples suggest that movement is involved in the derivation of vP-preposing.Footnote ^[12] The question then is: movement of what? On the standard analysis, it is predicate movement, whereas on our analysis, it is movement of an operator. There are a number of properties that ought to distinguish operator movement and predicate movement, chief among these being reconstruction, since null operators should not show reconstruction effects at all, and locality, since null operators are known to show different boundedness properties from some other kinds of movement. We address each of these in turn and show that once we control for the role of ellipsis in vP-preposing, both of these diagnostics indicate that the movement involved is operator movement rather than predicate movement.

First, we consider reconstruction, and to do this we need to describe the reconstruction profile of predicate movement, taking wh-movement of predicates in English (discussed in Section 2.1) as a standard. It is well known that wh-moved predicates reconstruct obligatorily with respect to binding Conditions A and C, as shown by (67) and (68); note that in this respect they differ from wh-movement of arguments, which can create new Condition A binding configurations, (69), and sometimes bleed Condition C, (70).

Huang (Reference Huang1993) argued that the obligatory reconstruction effect is due to the fact that the predicates contain traces of the subject they apply to, and so the presence of this trace in the fronted constituent ensures that the binding conditions apply as if the argument were inside the predicate, with no need for reconstruction of the predicate to its base position. However, Heycock (Reference Heycock1995) points out that the reconstruction effect in (68b) remains unaccounted for on this analysis, and she also points out that predicates seem to show Condition C reconstruction effects even when a predicate-internal trace is not wholly plausible, as in (71a).

Heycock proposes that the asymmetry between wh-moved arguments and predicates may be due to semantic factors like referentiality, such that non-referential phrases like predicates obligatorily reconstruct for semantic reasons.Footnote ^[13]

A final property of wh-moved predicates which will be important here is that they do share some reconstruction properties with wh-moved referential arguments: they show antireconstruction effects . In particular, they show the effect whereby adjuncts seem not to reconstruct with the phrases they attach to, also known as the Freidin-Lebeaux effect (Freidin Reference Freidin and Lust1986, Lebeaux Reference Lebeaux1988). This is demonstrated for arguments in (72) and predicates in (73), where the adjunct is a relative clause. Note that in (73) the relative clause adjuncts are added onto arguments that are deeply embedded in the AP, indicating that late adjunction can target an embedded constituent (contra Sauerland Reference Sauerland1998: ch.2).

The classic analysis of this phenomenon in Lebeaux (Reference Lebeaux1988) is that adjuncts can be inserted countercyclically, with the antireconstruction effect being derived by adding the adjunct to the wh-phrase in Spec,CP; the lack of a copy of the adjunct in the lower position thus means that it cannot trigger Condition C effects even if the lower copy of the host whP is interpreted. We adopt Lebeaux’s analysis here, assuming that it can also account for examples like (70), although the exact details of how late adjunction is derived are not important. What is important, however, is that the late adjunction effect is fundamentally about the timing of adjunction, with adjunction to a higher copy in the chain resulting in the antireconstruction effect.

With this as background, we can now consider the case of reconstruction effects in vP-preposing. Huang (Reference Huang1993) provides the following examples to demonstrate that vP-preposing seems to show the same obligatory reconstruction profile as wh-fronting of predicates, in that it shows obligatory reconstruction for Conditions A and C.

Although this represents a serious problem for an analysis where there is only movement of an operator and no further structure in the ‘base position’, it can be handled by the matching analysis that we have proposed here, since the base position is occupied by another vP that is elided under identity with the initial vP. First, consider the Condition A data in (74). In Section 2.2, we argued that vP-preposing involves high vP-ellipsis, and we saw there with example (76), repeated here, that this kind of ellipsis does not allow for mismatches between ellipsis site and antecedent with respect to reflexives; that is, if there is a reflexive in the antecedent, one must also be interpreted in the same position in the antecedent. If this is implemented with vP-preposing, then (74) follows straightforwardly, since the reflexive in the ellipsis site will always need a local binder.

A similar explanation would carry over for the Condition C data in (75) if we assume that the strictness of the ellipsis identity condition ensures that the ellipsis site contains the same r-expression that is contained in the antecedent.

One potential problem here is that it ought to be possible to avoid a Condition C violation in (78) if the r-expression were changed to a pronoun by vehicle change (Fiengo & May Reference Fiengo and May1994). Vehicle change is what is said to be involved in ensuring that the underlying structure for (79) is (a), where the ellipsis contains a pronoun, rather than (b), where the r-expression is copied in and a Condition C violation ought to be present. This poses no problem for an approach to ellipsis where the ellipsis identity relation is purely a semantic one (e.g. Merchant Reference Merchant2001), but it poses a non-trivial problem for theories that hold that the identity relation holds over syntactic structures (e.g. Chung, Ladusaw & McCloskey Reference Chung, Ladusaw and McCloskey1995, Merchant Reference Merchant2013).

If vehicle change were possible with vP-fronting, with the lower vP containing only pronouns, then we might expect Condition C obviation in examples like (75). This would get the wrong result, and so we must address this possibility; otherwise, the Condition C data in (78) would present a problem for our analysis.

The question that we need to ask is whether vehicle change is always possible in unambiguous cases of high vP-ellipsis. Unfortunately, this is quite difficult to test, since there is no possibility of creating the correct test configurations with retorts or tag questions due to the fixity of the subject and the unembeddability of these constructions. Cases of symmetric coordination seem to allow for vehicle change, in that examples like (80) allow the subject to corefer with an r-expression in the ellipsis site. This is problematic if examples like (80) must have a high vP-ellipsis analysis and symmetric coordinations like (81) do not allow the indicated reading.

However, both of these premises are somewhat hazy. Grodzinsky & Reinhart (Reference Grodzinsky and Reinhart1993) provide examples that are quite similar to (81), where there is strong parallelism, repetition of the r-expressions involved, and stress on the pronominal subject, and they note that many speakers allow for the coindexation indicated; they characterise these as instances of accidental coreference, and they submit that they are not regulated by binding theory proper. It is also not clear that a low vP-ellipsis analysis can be ruled out with examples like (80) entirely. This is particularly important because Drummond & Shimoyama (Reference Drummond, Junko, Iyer and Kusmer2014) have shown that vehicle change is not equally possible in all ellipsis configurations, and in fact it seems to be ruled out in configurations that are quite similar to high vP-ellipsis. Their crucial examples involve comparative stripping where the ellipsis remnant is the pronominal subject of the standard as in (82): in such configurations, this pronoun cannot corefer to expressions in the antecedent that would potentially occur in the ellipsis site. The contrast to (83) indicates that it is the size of the ellipsis site that is crucial.

If comparative stripping involves ellipsis of the whole clause except the subject, then it is very similar to high vP-ellipsis, where a large chunk of the clause is taken as the antecedent for ellipsis, and so this might be taken as an indication that there is some problem with vehicle change with big ellipsis of this kind more generally. At the very least, this suggests that it is wrong to assume that vehicle change is always available with ellipsis.Footnote ^[14] We take this as grounds to dismiss the challenge from vehicle change to our ellipsis analysis, although we admit that this is a frailty of our account which requires further attention.

Now, let us consider the antireconstruction effects associated with late merger of adjuncts, as it is with these that the predictions of the different accounts of vP-preposing differ. The standard predicate movement account predicts that it should be possible to late-adjoin a relative clause to an argument inside the preposed predicate, as (73) showed that this is possible with wh-moved predicates. However, this prediction is not borne out, as (84), from Landau (Reference Landau2007: 156), indicates that antireconstruction does not seem to be available with vP-preposing. In (85), further examples are provided that demonstrate the same point: the pronoun needs to be interpreted as disjoint from the r-expression contained in the relative clause that modifies the argument.

Although the lack of antireconstruction is unexpected on the predicate movement account, it is precisely what is predicted by the matching analysis we are proposing here. To see this, consider the schematic in (86), which represents (84) on a matching analysis.

Here, the ‘trace’ position contains a complete vP which is elided under identity with the initial vP, and since the initial vP contains the relative clause, the ellipsis identity condition dictates that the lower one must too.Footnote ^[15] The vP in the initial position is not moved to its surface position, so there is no question of adjoining the relative late to ensure that it is not part of the antecedent, and thus late adjunction is unable to evade the Condition C violation that obtains by putting the r-expression in the relative in the c-command domain of the matrix subject. Thus, the matching analysis correctly predicts there to be no effect of late adjunction, because there is no movement of a vP that can be adjoined to late.

Now, we consider evidence from locality, which is in fact intrinsically connected to the argument from reconstruction just stated. The empirical observation that we begin with (from Chomsky Reference Chomsky1986: 20) is that vP-preposing does not seem to be sensitive to wh-islands, as in (87), which shows only the mild degradation associated with extraction of DPs.

This is unexpected on the predicate movement account, since it is known that predicates and other non-referential expressions are sensitive to weak islands (Cinque Reference Cinque1990b), a fact that Cresti (Reference Cresti1995) attributes to weak islands being impervious to reconstruction. This is demonstrated in (88) for non-referential arguments and AP predicates (from Heycock Reference Heycock1995: 562–563), and (89) demonstrates that the same holds for DP predicates.

By contrast, it has been independently observed that many constructions analysed as operator movement can escape weak islands (Lasnik & Stowell Reference Lasnik and Stowell1991), so these data cause no problems for the matching analysis.

An additional, similar argument in favour of the operator-movement analysis comes from British do, which vP-preposing can escape, as seen above in (43), (43a) repeated here. Building on Haddican (Reference Haddican2007) and Baltin (Reference Baltin2012), Thoms (Reference Thoms and Surányito appear) notes that British do can be characterised as a selective island, disallowing wh-extraction but allowing comparative extraction and QR over negation.Footnote ^[16]

Thoms (Reference Thoms and Surányito appear) argues that the relevant generalisation regarding extraction from British do is that, like (other) weak islands, it bans reconstruction of extracted material. This accounts for the difference between (93a) and (93b), where the problem with (93a) is not the inability of QR to escape British do (as proposed by Haddican Reference Haddican2007), but rather the inability to reconstruct the subject, since subject reconstruction is a prerequisite for inverse scope (Hornstein Reference Hornstein1995); moreover, it also accords with the restrictions on wh-movement, since wh-movement seems to reconstruct obligatorily (Chomsky Reference Chomsky1995). The matching analysis then predicts that vP-preposing should be able to escape British do, since the operator does not need to reconstruct; this prediction is borne out, as we have already seen. Under the standard analysis, on the other hand, vP-preposing would also be the only form of overt movement to escape British do, and hence paradigmatically highly anomalous.

2.5 Reconstruction into vPs

One final empirical issue that we should address is reconstruction of the subject into a preposed vP. This is one empirical domain in which vP-preposing and wh-movement of predicates pattern together entirely, so it does not tease apart the matching and movement analyses, but nevertheless it is important for us to show that the facts follow from the matching analysis.

The received wisdom on reconstruction into moved predicates is that it is impossible: see Barss (Reference Barss1986), Sauerland & Elbourne (Reference Sauerland and Elbourne2002), van de Koot (Reference van de Koot2004) and much subsequent literature. Sauerland & Elbourne (Reference Sauerland and Elbourne2002) dub this Barss’ generalisation, and their example (94) demonstrates this for scope reconstruction with wh-moved predicates; (95) (from Lasnik & Saito Reference Lasnik and Saito1992) does so for reconstruction for idiom interpretation.

Now consider vP-preposing. The examples in (96) (adapted from Bobaljik & Wurmbrand Reference Bobaljik and Wurmbrand2012) and (97) show that the subject fails to reconstruct with this construction as well.Footnote ^[17]

The same pattern of judgments holds irrespective of which kind of vP is fronted (demonstrated above for progressive and passive vPs), so aspectual and categorial factors are not to blame here. These facts indicate that wh-movement of predicates and vP-preposing both fall under Barss’ generalisation: reconstruction into the moved predicate is in both cases impossible.

However, in Thoms & Walkden (Reference Thoms and Walkden2015) and in Section 4.1 below, we demonstrate that reconstruction into a moved constituent is not impossible, since it is possible to reconstruct into extraposed relatives (Heycock Reference Heycock2012, contra Hulsey & Sauerland Reference Hulsey and Sauerland2006), so Barss’ generalisation (or whatever is supposed to derive it; see, e.g., Sauerland & Elbourne Reference Sauerland and Elbourne2002) cannot be relied upon to account for the lack of reconstruction into moved predicates and vPs alike. Thoms & Walkden (Reference Thoms and Walkden2015) defend an approach to the lack of reconstruction into wh-moved predicates (from Lasnik & Saito Reference Lasnik and Saito1992) according to which there was no A-chain connecting the surface subject in Spec,TP and the subject position within the predicate, but rather the surface subject was base-generated in Spec,TP and the predicate contained a PRO subject; the lack of reconstruction thus followed from the fact that there is no reconstruction in control dependencies.

The same analysis extends to vP-preposing, although it requires some elaboration. Recall that, on the matching analysis, the internal subject position in the preposed vP is also occupied by PRO, as schematised for (96b) in (98).Footnote ^[18]

Regarding (98), the lack of inverse scope in the initial vP follows again from the fact that the subject is PRO. With the vP in the ‘base position’, one might expect that the subject could reconstruct into the elided vP to derive an inverse scope interpretation there. However this option would be blocked by Scope Parallelism , the condition that ensures that an ellipsis site and its antecedent are identical with respect to scope relations of quantificational elements (Fiengo & May Reference Fiengo and May1994, Fox Reference Fox2000). Reconstructing the subject into the vP to scope below the object in the elided would result in a violation of Scope Parallelism since there is no corresponding quantificational subject that would reconstruct in the antecedent vP in the initial position (it is PRO).Footnote ^[19]

We can see, then, that the lack of reconstruction into vPs is compatible with the matching analysis, while the movement account needs to say something extra to account for this; this something extra could be some stipulation that rules out reconstruction into moved vPs specifically. However, we will see in the next section that such a stipulation will not work.

2.6 Summary

To summarise, we have seen that the matching analysis we proposed for English vP-preposing accounts for a large number of curious properties of the construction: the ellipsis licensing condition on the gap position, the distribution of morphological mismatches between the expected and possible forms of the initial vP, the (apparent) reconstruction properties of the dependency between the initial vP and the gap, and the impossibility of reconstruction into the fronted vPs. All of these properties are problematic for the standard movement account, so much so that they ought to lead us to the conclusion that it must be rejected in favour of an alternative, either the matching analysis or something like it. However the details work out, what is important is that this seems to show us that true vP-movement is not possible in English, at least as an analysis of vP-preposing.

The crucial question at this point is why vP-movement should be ruled out in this instance. This question becomes especially pressing in light of our argument that German vP-preposing does involve movement (since it contrasts with English vP-preposing in not displaying idiosyncratic licensing restrictions or morphological mismatches). We will return to this question in Section 4, once we have examined the other major vP-fronting construction in English: participle preposing.

3.1 A subject extraposition analysis

The two key word order properties of participle preposing that any analysis of the construction needs to capture are the non-canonical initial position of the vP and the non-canonical rightward position (to the right of the auxiliaries) of the subject. These two issues are also at the heart of all analyses of locative inversion (LI), another inversion construction which has been examined in much more detail in the literature.

Locative inversion and participle preposing have a number of properties in common (Haegeman Reference Haegeman2008), including the distinctive information structural properties identified for LI by Birner (Reference Birner1992) and Levin & Rappaport Hovav (Reference Levin and Rappaport Hovav1995),Footnote ^[20] so it is useful to consider the arguments brought to bear on the former to arrive at an analysis of the latter.

First, let us consider the position of the fronted XP. With LI, the initial PP is located in the subject position by Levin & Rappaport Hovav (Reference Levin and Rappaport Hovav1995), Culicover & Levine (Reference Culicover and Levine2001), and Doggett (Reference Doggett2004); Bresnan (Reference Bresnan1994) adopts a similar stance in an LFG framework, according to which the PP is the subject but it occupies a topic position in c-structure, roughly equivalent to a transformational analysis where the subject undergoes topicalisation at S-structure. Two relevant sources of evidence are provided for the PP being in a subject position: (i) further extraction of the PP is subject to the that-trace effect, (101) (Bresnan Reference Bresnan1994: 97), and (ii) the PP can be raised to a higher subject position like regular DP-subjects, (102) (data from Postal Reference Postal and Postal2004: 18).

Examples (103) and (104) show that the same properties hold of the fronted verbal projection in participle preposing: it can be topicalised into a higher clause but is subject to the that-trace effect, and it can undergo raising when embedded under seem.

This indicates that the preposed participle occupies Spec,TP or some such neighbouring position at at least one point in the derivation. Now, consider the position of the subject of the vP. With LI, it was previously proposed by Coopmans (Reference Coopmans1989), Hoekstra & Mulder (Reference Hoekstra and Mulder1990), and Levin (Reference Levin1986) that the postverbal subject is an in-situ subject; this was motivated by the fact that LI is most productive with unaccusative and passive verbs, with which an in-situ analysis for the subject would correctly locate it in a postverbal position (assuming the vP-internal subject hypothesis). However, Levin and Rappaport Hovav (Reference Levin and Rappaport Hovav1995: Chapter 6) argue convincingly against this proposal, showing that LI occurs productively with a number of unergative verbs, such as walk (as in (100)) or stand (as in (102)). Building on Birner (Reference Birner1992), Levin and Rappaport Hovav show that the tendency for the use of unaccusative and passive verbs in LI, as well as the apparent exceptions with unergatives, follows from the discourse properties of the construction.Footnote ^[21] Importantly, since unergative subjects would not be base-generated postverbally on any account, it follows that the postverbal subject in LI sentences like (100) and (102) must be in some rightward-branching position. One argument for the rightward position of the subject DP that is put forth in the literature (Bresnan Reference Bresnan1994, Doggett Reference Doggett2004) is that extraction of or subextraction from the postverbal subject is impossible, (105a), a fact that would follow from the extraposition analysis given that extraposed XPs are generally islands to extraction, (105b).Footnote ^[22]

Another, perhaps stronger, argument comes from the relative position of the subject and other elements that are normally vP-internal or higher. For instance, Doggett (Reference Doggett2004: 39) shows that the subject always occurs to the right of the vP-internal particle in LI, as shown by (106), indicating that it is at least as high as vP-adjoined; if the subject were in-situ inside the vP, this would be unexpected. Similarly, Levin & Rappaport Hovav (Reference Levin and Rappaport Hovav1995: 266) show that the subject may also follow IP-level modifier PPs, as in the attested example in (107). Intervening material is generally quite restricted due to the discourse conditions on LI, which dictate that the pre-subject material introduces a minimal amount of information in addition to what is presented by the locative in the initial position.

We conclude that subject extraposition is possible in at least some cases of LI, although the launching and landing sites of this extraposition are not easily determined.

Now, let us consider participle preposing. Like LI, this might be analysed either as involving rightward extraposition of the subject or as involving leftward movement to some specifier position outside the vP-shell followed by remnant movement of the vP-shell. An implementation of the latter analysis is proposed by Samko (Reference Samko and Santana-LaBarge2014), who argues that the subject moves from vP to the specifier of a PredP projection, which is projected above vP. Note that this position must also be above the progressive layer of the IP-domain, since progressive auxiliaries can be fronted in participle preposing as well.

It is not clear to us whether there is any evidence for locating such a PredP projection so high in the IP-domain, or indeed for the existence of such a projection in addition to vP at all, so this analysis has a somewhat stipulative character, and leaves the commonalities of LI and participle preposing somewhat mysterious. By contrast, the extraposition analysis is lent great plausibility by the preceding analysis of LI, and providing a unified analysis captures the fact that the discourse properties of the two constructions seem to be identical.Footnote ^[23] We can also replicate the two arguments for subject extraposition for participle preposing. First, extraction of or subextraction from the subject is impossible, even though further extraction of the vP itself is possible ((103) above).

Second, the subject appears to the right of right-adjoined IP-level adjuncts, (110); as before, though, this is restricted by the discourse conditions of the construction, and in general the preference is for the adjuncts to be fronted along with the participle.Footnote ^[24]

Thus, the subject extraposition analysis of participle preposing seems to be well motivated.Footnote ^[25]

We assume this subject extraposition analysis in what follows, although we will not get into the details of determining the precise timing of subject extraposition in the derivation; all that matters is that the subject is extraposed to a position to the right of the base position of the vP.

In the following subsections, we discuss key properties of participle preposing that are relevant for the comparison with vP-preposing. The discourse restrictions on participle preposing prevent us from testing some of the properties that we tested with vP-preposing, but nevertheless we argue that some of participle preposing’s distinctive properties indicate that it must be treated as a true vP-movement operation, rather than as involving a matching analysis.

3.2 Reconstruction in participle preposing

Perhaps the most important distinctive property of participle preposing (hitherto unnoticed as far as we can tell) is that the subject may reconstruct into the initial vP. This is demonstrated by (112), where the subject a different bottle of wine can scope below the universal quantifier every table, receiving a distributive interpretation. (The subject may take widest scope if different is omitted.)

Further examples demonstrating the possibility of subject reconstruction into the fronted predicate are given below. In (113), a further example is provided of subject reconstruction (this time interacting scopally with a direct object) on the assumption that inverse scope requires subject reconstruction (see references in Section 2.5). In (114), we see that bare plural subjects of fronted verbal and adjectival predicates may receive existential interpretations, an interpretation that is argued by Diesing (Reference Diesing1992) and Fox (Reference Fox1995) to require reconstruction of the subject back into the predicate. This last example is important since it allows us to dismiss an analysis of the effect in the other cases where the wide-scoping universal quantifiers scope over the subjects by QRing out of the initial vP into a position c-commanding the subject: no such trick could be performed to get the existential reading of the subject in (114), since it is derived by reconstruction.

These contrast strikingly to their regular vP-preposing counterparts. Recall from Section 2.5 that vP-preposing always disallows reconstruction of the subject back into the fronted constituent. This is shown by (115)–(117), which differ minimally from (112)–(114) and which all disallow the reconstructed readings.Footnote ^[26]

This difference is predicted on the present account. We recall from Section 2.5 that the lack of reconstruction with vP-preposing is predicted on the matching analysis, since there is no trace of the overt scope-taking subject in the initial vP. By contrast, we argue that participle preposing may be derived by true vP-movement, with a trace of the subject in the moved verbal constituent, so reconstruction into that constituent ought to be possible unless some other factors interfere.

The fact that reconstruction into the fronted vP is possible with participle preposing also has implications for the analysis we provided for vP-preposing above. We recall that vP-preposing generally disallows reconstruction of the subject back into the moved predicate, a generalisation established by Barss (Reference Barss1986) and called ‘Barss’ generalisation’ by Sauerland & Elbourne (Reference Sauerland and Elbourne2002). As we noted above, movement-based analyses of vP-preposing can appeal to this generalisation and whatever derives it in accounting for the lack of reconstruction back into the initial vP in vP-preposing, and so they may usefully counter that our alternative account in terms of ellipsis parallelism would fare no better. However, Barss’ generalisation is clearly falsified by the data in this section, and so the movement analyses of vP-preposing are left without an account of the reconstruction facts. The availability of reconstruction is arguably the most reliable source of evidence for the presence of a movement dependency that we can draw upon in studies of this kind, so the contrast between vP-preposing and participle preposing presents a very strong argument for our proposal that the two constructions have very different derivations.

3.3 Restrictions on participle preposing

Another property that distinguishes the two constructions is that vP-preposing is possible in a number of vP-ellipsis-licensing configurations that disallow participle preposing. Thus, in Section 2.1, we saw that vP-preposing is compatible with bare infinitives as well as perfect, simple past, passive, and progressive participles. By contrast, only passive and progressive participles participate in participle preposing.

This tells us that the matching analysis is unavailable for participle preposing, since if it were available, all of these would be grammatical: the initial vPs are all vP types that can normally be elided. There are two reasons why matching might not be available for participle preposing, both relating to the rightward position of the subject. One possibility is that the rightward subject is adjoined quite low, within the vP that would be targeted by vP-ellipsis in a matching analysis, and so a matching analysis would derive subjectless clauses rather than the structure that we see with the subject finally. Alternatively, it could be that subject extraposition causes problems for the operator movement step; for instance, if it is the case that the operator is adjoined to the subject of the lower vP, extraction of that operator would be rendered impossible by subject extraposition, given that extraposed constituents in English are typically islands for subextraction. We will not settle on a specific analysis here, since it is not clear how one would decide upon one analysis or another, but it seems sufficient to us that there are a few reasons why matching would not work for participle preposing.

Of course, the data in (118) are not straightforwardly predicted by a movement account of participle preposing, since it is not clear why only progressive and passive participles may front. We return to this question in Section 4.2, where we argue that it is a reflex of a very general (possibly universal) condition on movement, one that only reveals itself once we reanalyse vP-preposing in terms of base-generation.

3.4 Summary

In this section, we have shown not only that evidence for the matching analysis is lacking for participle preposing, but also that the matching analysis is unavailable for participle preposing. We have proposed an analysis in which the subject is extraposed rightward from its vP-internal position, followed by leftward remnant movement of the vP. The restricted nature of participle preposing has prevented us from testing a number of properties that might distinguish it from true vP-preposing, such as island effects and morphological mismatches, but nevertheless we have seen that reconstruction effects, arguably the most reliable diagnostic for movement dependencies that we have, force us to draw a distinction between the two constructions. Why these two different types of vP-fronting turn out to require such different derivations, when they are examined in detail, will be the subject of the next section.

4.1 The Excluded Middle Constraint

The standard analysis of vP-preposing is reproduced in (119a), together with our analysis of participle preposing in (119b). Here we adopt the copy theory of movement (Chomsky Reference Chomsky1995) and use struckthrough copies to represent traces of movement.

Both (119a) and (119b) involve extraction of the subject from the vP followed by leftward remnant movement of the vP, and so both involve remnant movement , that is, movement of a constituent containing a trace. Numerous authors have argued for imposing restrictions on remnant movement, with some banning it in certain movement rule combinations (Müller Reference Müller1998, Abels Reference Abels2008) and others ruling it out entirely (Saito Reference Saito1985, Reference Saito2002; Lasnik & Saito Reference Lasnik and Saito1992), so it is possible that the restriction in (119a) is attributable to some such constraint. However, if remnant movement is the problem, why does it not also rule out (119b)?

We propose that the crucial difference between (119a) and (119b) is where the subject chain is linearised. Looking again at the flat representations in (119), we can see that the copy of the subject that is linearised in vP-preposing is an intermediate one in the linear sequence, since there are unpronounced copies in the initial vP, which is to the left, and the base copy of that vP, which is to the right. In Thoms & Walkden (Reference Thoms and Walkden2015), we propose the constraint in (120).

We call (120) the Excluded Middle Constraint (EMC), and our proposal is that it is a general constraint on the linearisation of all dependencies formed by Remerge, rather than a remnant movement-specific constraint. The EMC clearly gets the right result with standard spec-to-spec cyclic movement chains, as these typically involve spelling out the highest and leftmost copy of the moved element, and in those cases where some PF-constraint rules out the pronunciation of the highest occurrence, the result is that we spell out the lowest and rightmost copy of the moved element, rather than some intermediate copy (Bošković Reference Bošković2002, Bošković & Nunes Reference Bošković, Nunes, Corver and Nunes2007).Footnote ^[27] The EMC also accounts for the fact that in right node raising (RNR) constructions, where a given XP seems to be shared by a number of selecting heads at once, it is the rightmost occurrence of the shared XP that is linearised and never an intermediate one.Footnote ^[28] In Thoms & Walkden (Reference Thoms and Walkden2015), we show that the effects of (120) follow from a simple linearisation algorithm for multidominance trees modelled on the proposal in de Vries (Reference de Vries2009), and we argue that it requires no more stipulation than alternative accounts that derive copy pronunciation in terms of feature strength and copy deletion.

Returning to the case of remnant movement, (120) has the effect of ruling out all instances of remnant movement where both the container XP and the containee YP move leftwards, as in (121), since all such derivations will require the linearisation of the containee in an intermediate position like in (119a) (assuming that cyclicity is respected). On the other hand, remnant movement derivations like (122) and (123), where the container and containee move in different directions, can be linearised by (120), and so if this constraint is what is responsible for remnant movement restrictions, in general, we should expect that these should be possible.

In Thoms & Walkden (Reference Thoms and Walkden2015), we provide ample evidence to demonstrate that this correctly describes the availability of true remnant movement derivations. In many cases, the key evidence comes from reconstruction: no construction that seems to instantiate a well-formed instance of (121) allows reconstruction of the XP back into moved YP, as seen above with vP-preposing. By contrast, cases of remnant movement corresponding to the configurations in (122)–(123) fare differently, as we saw earlier that (122), which corresponds to our analysis for participle preposing in (119b), does allow reconstruction of the containee into the container. However, in Thoms & Walkden (Reference Thoms and Walkden2015), we show that there are also cases of (123) that also allow reconstruction of the containee XP into the container YP, such as with extraposed relative clauses which require reconstruction of the head NP. Example (124), from Heycock (Reference Heycock2012), demonstrates that such relative clause extrapositions are grammatical (contra Hulsey & Sauerland Reference Hulsey and Sauerland2006), with reconstruction for anaphor binding and idiom interpretation being possible.

These show that independent conditions restricting reconstruction into moved constituents cannot be maintained, thereby strengthening the case for the non-movement analysis of apparent cases of (121). We discuss various other sources of evidence for using (120) to restrict remnant movement and show that it is superior to some alternative accounts and compatible with others that impose distinct but complementary restrictions.

Let us talk through vP-movement in more detail to clarify how the EMC accounts for the facts. The EMC rules out a movement analysis for vP-preposing as in (119a) because there is a copy of the subject in the base position of the initial vP, and in the base copy of that vP, and so linearising the subject in Spec,TP requires spelling out an intermediate copy of the subject. Since this is ruled out, the matching analysis is required to generate an initial vP. With participle preposing, there is no such problem in spelling out the subject chain, since the subject is in an extraposed position, so it is rightward of both the initial copy of the vP and the base copy; this means that it is the rightmost copy of the subject that is to be spelled out, with no violation of the EMC.

On this analysis, there is no specific restriction on fronting verbal constituents per se, nor is there a specific condition against remnant movement; rather, vP-movement restrictions are a product of the EMC and the base positions of subjects. This means that a vP may also be moved leftward if there has been no subject extraction at all. We hypothesise that German vP-fronting is such a case. We recall from Sections 2.1 and 2.3 that German vP-fronting does not display the licensing restrictions characteristic of English vP-preposing, nor does it permit morphological mismatches; thus, it seems to display the properties of true A $^{\prime }$ -movement. Following Fanselow (Reference Fanselow, Alexiadou, Anagnostopoulou, Barbiers and Gärtner2002), we propose to derive this difference between English and German from an independent property of German syntax. A-scrambling is possible in German, and since A-scrambling involves base-generation of an argument in a higher position (see Neeleman Reference Neeleman1994 for a powerful defence of this view), this makes it possible to derive true  vP-fronting in German without remnant movement: we simply base-generate the subject outside the vP and move that vP to the initial position. Such a derivation does not involve remnant movement, so this is regular A $^{\prime }$ -movement of the vP and thus it is not necessary for vP-ellipsis to be licensed in the trace position.Footnote ^[29] This analysis of German vP-fronting is defended in more detail in Thoms & Walkden (Reference Thoms and Walkden2015).

4.2 The N-only constraint

Having accounted for the fact that only participle preposing can be derived by true vP-movement and provided an ellipsis-based analysis of the restrictions observed with vP-preposing, what remains to be addressed is the fact that participle preposing is still itself quite restricted, applying only to passive and progressive vPs. Here, we propose an account of this restriction in terms of a quite general constraint on which categories can be targeted for movement, one that only reveals itself now that we have reanalysed vP-preposing as not involving true movement: only constituents that occur in nominal-licensing positions may be moved.

4.3 Explaining the N-only constraint

Finally, we turn to the matter of how to explain the N-only constraint. In many ways, the constraint is exactly the kind of thing we might have expected to find in the early days of government and binding theory, according to which movement involved inserting a pronoun into the base position which is then construed as a bound variable: in that case, the constraint would fall out of the fact that pronouns are nominals, and so movement would only be possible when the pronoun to be inserted would satisfy the selectional restrictions of the dominating head. Our explanation of the constraint is similar in spirit but employs the modern equivalent of classical trace theory, namely a version of Fox’s (Reference Fox1999, Reference Fox2002) trace conversion rule (see also Sauerland Reference Sauerland1998, Reference Sauerland2004; Rullmann & Beck Reference Rullmann, Sigrid, Katz, Kim and Winhart1998). Fox proposes an account of the interpretation of traces working with Chomsky’s (Reference Chomsky1995) copy theory of movement, where lower copies in chains are converted into restricted variables that are bound by the higher copy of the moved phrase. This is done by inserting a variable, removing the determiner of the lower DP, and replacing it with a definite determiner.

Lambda abstraction over the variable thus ensures that the chain receives the appropriate bound variable interpretation.

As Fox (Reference Fox2002: 67) points out, trace conversion so defined is restricted to application in environments where Ds can be inserted, leading to the expectation that it is only compatible with nominal traces, since D is only selectionally compatible with nominals.Footnote ^[42] Fox notes that this leads us to expect that VP-traces cannot be interpreted by trace conversion, and he suggests this might account for their curious reconstruction behaviour. Relatedly, Takahashi (Reference Takahashi2010) observes that the categorial requirements of trace conversion may explain the fact that CP-fronting is only possible where DP-traces could be accommodated, as evidenced by the DP-requirement above. We therefore propose that this is what explains the N-only constraint, assuming that it is universal: trace conversion is required for the interpretation of movement chains, and the only gap positions that are compatible with trace conversion are those that are compatible with nominals in terms of selection and licensing. It may be possible in principle to move non-nominal categories, but the chains created will not be interpretable at the LF-interface. Although this account has the virtue of simplicity, numerous explanatory questions remain: for instance, why is trace conversion the only way to interpret movement chains? Why this rule in particular? How do we deal with apparent non-nominal traces in other languages? Interesting as they are, these questions are beyond the scope of the current paper.