Prototypical cases of split extraposition are nominal arguments with a relative clause or a postnominal clause/PP. We illustrate this again in the examples in (80): the primeless examples show the structures of the noun phrases in the non-split pattern, while the primed examples show the split extraposition pattern.

For completeness’ sake, we add the examples in (81) to show that split extraposition is possible not only with prepositional phrases but also with post and circumpositional phrases; cf. Veld (1993: §4.3).

Until the mid-1990s it was generally assumed that the split patterns in (80) and (81) are derived by movement, because this immediately takes into account the facts illustrated in (82), that the postverbal phrase obeys selection restrictions imposed by the presumed selecting head (hoop selects an op-PP), and that the pre and postverbal PP are in complementary distribution: cf. Corver (1991).

However, the nature of the movement is not entirely clear. One generally accepted derivation involves an extraposition transformation (sometimes called PP-over-V in the case of PPs), which optionally moves the postnominal clause/PP to some postverbal position, as illustrated by structure (83a). Another view, originally proposed in Vergnaud (1974) and which became quite popular after the publication of Kayne (1994), is the so-called raising (or promotion) analysis. According to this analysis, the noun phrase is generated to the right of the verb and then moved to the left of the verb, optionally stranding its post-nominal part; this is indicated by structure (83b), in which NP* stands for a slightly larger nominal projection than the moved NP-projection.

Despite the popularity of both proposals, there are many theoretical and empirical problems with them; we will outline some of the most important issues below and refer the reader to Koster (1973/1995/2000), Kaan (1992), De Vries (2002: §7), Boef (2013: §3) and the references cited there for more detailed discussions.

A rather problematic aspect of the extraposition analysis in (83a) is that it presupposes that relative clauses and postnominal PPs can be extracted from noun phrases, while there is actually no independent evidence to support this claim. For example, while virtually any clausal constituent can be moved into the clause-initial position, topicalization of relative clauses and postnominal clauses/PPs is excluded, as shown by the primed examples in (84). The number sign in (84c') indicates that this example is acceptable if the met-PP is interpreted as a comitative adverbial phrase; this reading is irrelevant here.

The unacceptability of the primed examples follows from the hypothesis that noun phrases are islands for movement (cf. Section 11.3.1.1, sub VB), but this hypothesis would make the extraposition analysis in (83a) highly implausible anyway. Of course, there are also arguments in favor of the extraposition analysis, but they do not seem very strong. For example, it has been argued that noun phrases such as het debuut van Hella Haasse do allow topicalization of their postnominal PP. However, such topicalization is only possible if the PP is headed by van or over, and Section N15.2.1, sub VC, has shown that such PPs can be analyzed as restrictive adverbial phrases.

A more convincing argument for the analysis in (83a) might be that scrambling of the object across a clause adverbial has a degrading effect on extraposition; this might follow from the so-called freezing effect, according to which moved phrases are islands for extraction; cf. Koster (1978: §2.6.4.4), Corver (2006b/2017) and Ruys (2008). However, note that Guéron (1980) has argued on the basis of English that extraposition is possible only from noun phrases that are part of the focus (new information) of the clause, whereas scrambled nominal arguments are typically presuppositional.

Another argument against the freezing approach and in favor of Guéron’s proposal is that De Vries (2002:244) claims that split extraposition is possible in the case of topicalized phrases. However, it is not so clear whether examples such as (87) actually involve extraposition or whether we are dealing with a form of right dislocation; the percentage signs in these examples indicate that, according to some speakers, an intonation break is preferred, which would support the right-dislocation analysis. Unfortunately, the VP-topicalization test from Section 12.1, sub IV, cannot be used to help us in this case, because the clause-initial position is already occupied by the topicalized noun phrase itself; we leave this issue to future research.

Guéron’s claim may also be consistent with the fact that extraposition from noun phrases with definite articles is difficult and perhaps even impossible in English; cf. Baltin (2006). Note, however, that replacing the demonstrative diethat with the definite article dethe in Dutch examples such as (86a) does not have the same far-reaching effect on acceptability judgments as it does in English, as is clear from the full acceptability of the examples in (80); cf. also Koster (2000). Whatever the reason for this striking difference between English and Dutch, the main conclusion for the moment is that it is not a priori clear that an appeal to the syntactic notion of freezing is necessary to explain the contrast in acceptability between the two examples in (86). This conclusion seems to be supported by the acceptability judgments on the examples in (88), which show that split extraposition generally becomes more difficult when more material intervenes between the extraposed phrase and its intended associate; cf. Corver (1991:134).

Let us now turn to the raising analysis in (83b). A possible problem for this analysis is related to the fact that extraposition is possible not only from direct objects, but also from other nominal arguments, although the result sometimes seems to be marked in the case of an indirect object; cf. Section N16.3.2.3.2, sub IIA. Example (89) provides cases of extraposed relative clauses: the split nominal argument is again italicized.

The examples in (89) involve indefinite nominal arguments, but the examples in (90) show that split extraposition is also possible with definite nominal arguments (although the result again seems to be marked in the case of an indirect object), provided that they are part of the focus (new information) of the clause and thus follow clause adverbials such as waarschijnlijkprobably (if present); placing de man further to the left leads to a degraded result.

Split extraposition with PPs is given in (91). The case with an indirect object in (91b) is again somewhat marked, but the case with a subject in (91c) seems impeccable. Note that the acceptability of the (b) and (c)-examples in (90) and (91) refutes the claim in De Haan’s (1974:176-7) that split extraposition is excluded with (definite) indirect objects and subjects; the marked status of split extraposition with the indirect object in the (b)-examples may be due to the intervention effect noted in (88).

The problem that the acceptability of the examples in (89) to (91) poses for the raising analysis is that it presupposes that relative clauses can appear postverbally only if the noun phrases they modify are base-generated in a position following the surface position of the clause-final verbs. While this is plausible for objects, it is quite unlikely for subjects: the assumption that the subject in (89c) is base-generated to the right of the surface position of the main verb is incompatible with the standard assumption presented in Section 9.2 that the clause-final verb is within VP and thus follows the base position of the external argument (subject) of the main verb (in vP). The raising analysis therefore requires a revision of the standard analysis of Dutch clauses, which should not be done lightly; cf. also Koster (2000:8). Note in passing that the so-called scattered deletion approach proposed in Wilder (1995) and Sheehan (2010), which we have not discussed here, has the same flaw (which Sheehan actually presents as a virtue on the basis of English); cf. De Vries (2002: §7) for a more extensive review of this approach.

A problem for both proposals in (83) is that extraposition is also possible from a noun phrase that does not function as a clausal constituent itself, but is embedded in a clausal constituent. This is illustrated in (92) for cases where the noun phrases function as complements of a prepositional object.

The problem for the extraposition analysis in (83a) is that we have to assume that the extraposed phrase is not only extracted from a noun phrase, but also from the containing PP: cf. ... [PP P [NP ... N ti ]] ... V XPi, where XP is a (relative) clause or a PP. The fact that examples such as *Wiei wacht je [PP op ti]? Who are you waiting for? are unacceptable shows that Dutch PPs behave as islands for movement, and this makes the extraposition analysis quite implausible, because the extraposed phrase is not only extracted from a noun phrase, but also from a PP. The problem with the raising approach is of a different nature: the putative leftward movement involves the non-constituent op die man (cf. [PP op [NP die man [rel-clause die ...]]]), and we would therefore not expect that movement of this PP could strand the postnominal phrase. However, this objection only applies to theories that assume that the PP is base-generated as a unit; if we assume that PP-complements are created in the course of the derivation, as in Kayne (2004), this problem need not arise.

It is also generally assumed that extraposition from noun phrases embedded in a postnominal PP is possible, although there seem to be several restrictions on this possibility that are not yet well understood. Example (93a) has two alternating versions with extraposition. The first version is given in (93b) and simply involves extraposition of a postnominal PP from a direct object. The second version, given in (93c), is the one that is relevant here: it involves extraposition of a relative clause from a noun phrase embedded in a postnominal modifier; that the extraposed relative clause must be construed with plaatjes and not with boek is clear from the fact that the relative pronoun die only agrees with the former (cf. de plaatjes die/*dat ik zag the pictures I saw versus Het boek dat/*die ik las the book I read).

Example (93c) may be marked in comparison to (93b), but it seems perfectly acceptable; the contrast may be computational in nature, in the sense that speakers simply tend to link extraposed relative clauses with the (structurally) closest antecedent. In (93c), this is the nominal projection een boek met plaatjes, and not the more deeply embedded phrase plaatjes. For one reason or another, this effect seems to be stronger when the extraposed phrase is of the same category as the postnominal modifier. This is illustrated in (94) for PPs.

Example (94c) is reasonably acceptable, but there are cases with a similar structure that are considered unacceptable by at least some speakers; cf. Haeseryn et al. (1997:1381ff) for a number of cases that they claim resist split extraposition of the kind under discussion, and Johnson (1991: §3.3.4) for similar data from English. For instance, examples such as (95c) are given as unacceptable, although some of our informants tend to accept them; we have indicated this with a percentage sign.

Although it is unclear to us what determines whether extraposition of a more deeply embedded PP leads to a generally accepted result or not, we suspect that the restrictions are not of a syntactic nature, but that considerations of processing, semantic coherence, prosody, etc. are involved; since we are not aware of any in-depth investigations of this, we have to leave this to future research. If our tentative conclusion that all the (c)-examples in (93) to (95) are syntactically well-formed turns out to be well-founded, it would lead to problems of the kind that were already pointed out for the examples in (92). At the moment, we are not aware of any existing proposal that can be used to solve the same problem for the raising analysis; note that the scattered deletion approach, which we rejected earlier, would be able to handle this problem; cf. De Vries (2002: §7) for this.

Finally, we note that the split extraposition pattern is also possible when the noun phrase is the complement of a locational/temporal adverbial PP; this is illustrated with a relative clause in (96). The acceptability of the primed examples is again a serious problem for the movement analyses in (83), since such adverbial phrases are often regarded as absolute islands for movement. Moreover, the raising approach is problematic because it requires the adjunct PPs to be base-generated postverbally and moved into their preverbal surface position, whereas there are good reasons to assume the opposite: that the adverbial phrase is base-generated in preverbal position can be supported by the fact that this is the unmarked position for non-prepositional adverbial phrases like morgentomorrow and gisterenyesterday; cf. Section 12.3, sub IV. Note in passing that this problem also applies to the scattered deletion approach mentioned above.

All in al, we can conclude from the data in this subsection that the split extraposition pattern cannot be accounted for by the two movement analyses in (83); these proposals can only be maintained if we allow the proposed movements to violate otherwise well-motivated island constraints on movement. The raising (as well as the scattered deletion) approach also requires that we adopt the rather unorthodox claim that the external argument (≈ subject) of the verb has a base position that is structurally lower than (or, in linear terms, to the right of) the surface position of the clause-final verb.

Subsection I has discussed split extraposition for nominal phrases. Although this is the prototypical case, it has long been known that split extraposition also occurs with other categories; cf. Koster (1974). We illustrate this in (97a) for complementive adjectival phrases with a PP-complement. Note that such cases cannot easily be used to argue against a movement analysis of extraposition, because the PP-complements can also be moved leftward, as shown in the primed examples by topicalization.

Things are different when the extraposed phrase is part of a modifier of the adjective. This is illustrated in (98) with the discontinuous degree phrase zo ... dat hij overal pastso .. that it fits everywhere. Although Section A25.1.3.1, sub II, has shown that the finite degree phrase is part of the AP (they can be extraposed together), it is usually in extraposed position; placing the clause before the copular verb zijn leads to a quite marked result. Nevertheless, the fact illustrated by (98b) that the degree clause cannot be topicalized in isolation strongly suggests that it cannot be extracted from the AP; cf. Rijkhoek (1998).

The unacceptability of (98b) thus again suggests that the split extraposition pattern in (98a) is not island-sensitive. This is further supported by the examples in (99), which show that the AP can easily be embedded more deeply: in (99a) the split AP is part of a direct object, and in (99b) it is part of a PP-complement.

That extraposition of degree clauses is not island-sensitive is also shown by the fact that they can be associated with modified manner adverbs such as hardloud in (100), despite the fact that such adverbial phrases are often considered absolute islands for movement.

We find essentially the same with dan/als-phrases accompanying comparatives; cf. Chapter A26. The examples in (101) first show that despite the fact that the dan/als-phrases cannot be topicalized, the split extraposition pattern is possible (and perhaps even preferred). This in turn suggests that split extraposition is not island-sensitive.

More support comes from the fact that the comparative can easily be more deeply embedded: in (102a) the split AP is part of a direct object, and in (102b) it is part of a PP-complement.

That extraposition of dan/als-phrases is not sensitive to islands is also shown by the fact that they can be associated with modified manner adverbs such as snellerfaster in (103), despite the fact that such adverbial phrases are often considered absolute islands for movement.

For completeness’ sake, note that split extraposition is not possible in the case of attributively used adjectives. This is illustrated by the examples in (104); while the PP-complement of the adjective verliefd can be extraposed if the AP is used as a complementive, it cannot if it is used as an attributive modifier.

Subsections I and II have shown that split extraposition is not sensitive to islands for extraction, suggesting that we are not dealing with movement, which then raises the question of what extraposition is. One possibility is that we are dealing with right dislocation. However, this does not seem to be the correct solution either, since Section 12.1, sub IV, has shown that right-dislocated phrases tend to be stranded under VP-topicalization, while postverbal phrases in split extraposition constructions tend to be pied-piped, as illustrated in (105) for extraposed postnominal phrases. Recall from Kaan’s generalization in (78) that the preverbal part of split topicalization patterns must also be pied-piped; this is expected, since Subsection I has shown that scrambling blocks split extraposition.

Note that the examples in (106) with stranding are only acceptable with the typical intonation contour of an afterthought, i.e. with an intonation break and an additional accent in the phrase following that break. This suggests that while right-dislocated phrases are external to the preposed verbal projection, the extraposed phrases in (105) are internal to it.

The examples in (107) show essentially the same as the examples in (105), but now we are dealing with cases where the split noun phrase is embedded in a PP-complement. In accordance with Kaan’s generalization, VP-topicalization requires pied piping of both parts of the phrase split by extraposition; the primed examples are only possible with an afterthought reading. As with nominal objects, the whole PP can be stranded under VP-topicalization: cf. Gewacht heeft Jan niet op die man die hem wil helpen.

The examples in (108) illustrate the same again, but now for split APs. The degraded status of (108a'') is especially instructive, since dol meaning “fond (of)” obligatorily takes an op-PP as its complement, and we have seen in Subsection IC that such obligatory PPs can only be right-dislocated if a pronominal PP is present in preverbal position.

Note in passing that the complementive and the manner adverb in the singly-primed examples must be pied-piped under VP-topicalization; this is expected, since it also holds for cases of VP-topicalization with a simple adjective: cf. *Gebleven ben ik mijn hele leven onzeker (intended: “I have remained insecure all my life”) and *Spelen zal de band niet hard (intended: “The band will not play loudly”).

Example (110) concludes by giving examples with a comparative dan/als-phrase: in accordance with Kaan’s generalization, the phrase split by extraposition must again be pied-piped under VP-topicalization.

The examples above show that extraposed phrases in the split extraposition construction differ from right-dislocated clauses in that they are internal to the preposed verbal projection. Consequently, we need a different (non-movement) account for the split extraposition pattern.

Koster (1995/2000) proposes to analyze split extraposition as a form of juxtaposition. The initial motivation for this was that we find the split pattern also in coordinate structures; a rightward movement analysis of an example such as (111a) would violate the coordinate structure constraint, which is held to be universally valid. De Vries (2002) further claimed that split coordination resembles split extraposition in that the postverbal part can be pied-piped under VP-topicalization, and we do indeed detect a contrast between the pied piping case in (111b) and the stranding case in (111b'), which is severely degraded (even when the second part of the conjunction is preceded by an intonation break). The percentage sign in (111b) is used to indicate that while De Vries gives this example as perfectly acceptable, we find the result marked, and certainly less comfortable than the VP-topicalization cases in Subsection III.

That the split pattern cannot be derived by rightward movement also becomes clear when considering subjects: while the non-split pattern in (112a) triggers plural agreement on the finite verb, the split pattern in (112b) does not; Koster (2000) notes that this would be unexpected if (112b) were derived from (112a) by movement.

Another fact not expected under the rightward movement approach is that while the non-split pattern is subject to the coordinate structure constraint, which prohibits extraction of (or from) a single coordinands, the split pattern is not subject to this constraint. This is illustrated by the contrast between the two (b)-examples in (113).

Koster proposes that the split patterns differ from the non-split patterns in that they do not involve coordination of equals, as in (114a), but rather have the form in (114b), with the equal of the second coordinand (here a noun phrase) embedded in a larger verb phrase (a VP, TP, or CP).

The forms of coordination in (114b) raise many questions, especially the fact that the two conjuncts are not parallel in categorial status, syntactic function, and meaning; they thus violate the co-occurrence restrictions on coordinands in a coordinate structure discussed in Section C38.3. We will not go into this here, because De Vries (1999/2002/2011) has proposed an alternative, according to which we are dealing with the coordination of two verbal projections plus the deletion of identical material. According to this proposal, the three examples in (114b) have the representations in (115), which do satisfy the co-occurrence restrictions mentioned above.

Note in passing that structures similar to (115c) are analyzed as right-dislocation constructions in Section C37.3, sub VII; the second coordinand actually has all the features of a fragment clause (cf. Section 5.1.5). For the sake of argument, we will ignore this here, while noting that we cannot apply the VP-topicalization test to this case, so there is no a priori reason to reject a right-dislocation/fragment-clause analysis, which in principle could also be considered for the other cases of the split pattern; future research should take this alternative analysis into account.

Koster proposes that split extraposition is a specific instantiation of parallel construal; this term refers to a larger set of structures in which two (or more) elements are juxtaposed and in which the second phrase specifies the first. For concreteness, we follow De Vries’ implementation, which analyzes the split extraposition pattern as specifying asyndetic coordination plus ellipsis; cf. also Bianchi (1999:264ff). The primed examples in (116) illustrate this analysis of split extraposition for a direct object; the element &: marks a phonetically empty conjunctive coordinator with a specifying meaning.

Since the examples in (117) show that ellipsis can affect subparts of phrases and words, it is not surprising that split extraposition can likewise affect subparts of phrases, leaving remnants like the relative clause and postnominal modifier in (116).

Following this line of reasoning, we can expect that the extraposed phrase may originate in quite deeply embedded positions. This is illustrated in (118a) for a split extraposition construction with a noun phrase that functions as the complement of a prepositional object, and in (118b) for a noun phrase that is part of a postnominal modifier.

Another advantage of De Vries’ analysis is that it can account for the fact, shown in (119), that the extraposed PP obeys the selection restrictions imposed by its associate noun phrase de hoop, for the simple reason that the two form a unit in the second conjunct. It is not immediately clear how Koster’s proposal could account for this.

Furthermore, De Vries’ analysis immediately derives the fact that the extraposed part of the split phrase cannot be stranded under VP-topicalization. The representations in primed examples in (116) and (118) show that stranding can only be derived by moving the first conjunct (here: VP) of the coordinate structure, but this would violate the coordinate structure constraint. Since this constraint also prohibits subextraction from one of the coordinands, we may have a principled account for Kaan’s generalization that it is impossible to pied-piped the postverbal part while stranding the preverbal part (thus making an appeal to Guéron’s semantic restriction on split extraposition superfluous). Finally, we can also derive Ross’ (1967) right roof constraint on extraposition, illustrated in (120), according to which the postverbal part cannot be “moved” out of its own minimal finite clause (here the subject clause). The reason is that coordination always involves clause-internal elements; example (120b) is excluded because the hypothesized reduced phrase [VP [NP de vrouw die hier net was] kent] should be coordinated with the VP of the subject clause.

A possible drawback of De Vries’ proposal is that it requires forward deletion (deletion in the second conjunct) of material in the right periphery of the second conjunct, while this type of conjunction reduction can normally only be applied backward, as is clear from the contrast between (121a&b); cf. Section C39.1.

Of course, there are also cases that can be analyzed as forward deletion. De Vries (2011) refers to examples such as (122), which are known as gapping; cf. Section C39.2 for a detailed discussion.

However, it seems that such gapping constructions have quite different properties. Vanden Wyngaerd (2011), for example, points out that unifying the deletion operation assumed in De Vries’ analysis of split extraposition with the deletion operation that derives the gapping construction in (122b) overgenerates: the remnants in the gapping constructions are usually clausal constituents and not parts of clausal constituents; cf. the clausemate restriction on gapping in Section C39.2, sub IIB. Unifying the two deletion operations thus wrongly predicts the gapping constructions in (123) to be acceptable.

We will leave this issue to future research and refer the reader to the discussion between De Vries and Vanden Wyngaerd (2011) for more details.

This subsection has shown that there are several problems with analyzing split extraposition as a result of rightward movement. We have therefore concluded our discussion with a fairly recent proposal to consider split extraposition as a form of juxtaposition (with or without deletion). Since this approach seems relatively successful in deriving the basic facts, it is not surprising that attempts have been made to derive a wider range of data from the same mechanism: non-split extraposition (Koster 1995/1999), appositional constructions (Heringa 2012), contrastive left dislocation (Ott 2014), backgrounding right dislocation (De Vries & Ott 2012 and Ott & De Vries 2015), etc. We will return to the cases of left and right dislocation in Sections C37.2 and C37.3. However, there are still some serious issues that need further investigation; cf. Vanden Wyngaerd (2011).