The two types of RD constructions, illustrated again in the (a)-examples in (102), are similar in several ways to the two types of LD constructions discussed in Section 37.2, illustrated in the (b)-examples.

First, all four types of dislocated phrases in (102) seem to be clause-external, since they all have a clause-internal pronominal associate, namely the pronouns hemhim and diethat. The fact that the thematic role of the verb is assigned to the pronoun suggests that the dislocated phrases are not licensed within the clause but in some other way. That the dislocated phrases are not part of the clause is also supported by the fact that (like parentheticals) they are separated from the intonation contour of the clause, which, at least in the case of afterthoughts and contrastive left-dislocated phrases, goes hand in hand with a distinct intonation break. Further reasons for assuming that dislocated phrases are clause-external are (i) that left-dislocated phrases precede the main-clause initial position and (ii) that right-dislocated phrases cannot be assigned a sentence accent, which is again indicated by a grave accent in the (a)-examples.

Second, all four types of dislocated phrases provide more specific information than their clause-internal associate: in (102) they all provide information about the function of the person referred to by the pronoun.

Third, LD and RD both come in two types: one in which the dislocated phrase is typically contrastively accented, and another in which the dislocated phrase is usually pronounced with a flat intonation contour. Note, however, that contrastive LD and afterthought RD differ in that the former invites a set of alternative propositions, while the latter simply provides discourse-new information. This can be seen in the examples in (103): while (103a) is fully compatible with a contrastive maar-phrase, the use of a contrastive maar-phrase leads to a somewhat marked result in (103b); the latter example becomes acceptable if the weak referential pronoun ʼm is replaced by a contrastively stressed pronoun hem, but then the contrastive maar-phrase is licensed by the pronoun, not by the afterthought.

Finally, example (104a) shows that right-dislocated phrases can provide information that helps the speaker identify the intended reference of the clause-internal nominal correlate, but example (104b) shows that the right-dislocated phrase can also be an epithet. When the nominal correlate is predicative, as in (104c), the right-dislocated phrase provides a more precise qualification.

The fact that left-dislocated phrases do not seem to have such a modifying function is the main reason why we will use different notions for the relationship between left and right-dislocated phrases and their clause-internal associate: resumption versus correlation. Left-dislocated phrases provide information that is simply taken up again by their clause-internal associate, whereas right-dislocated phrases provide more specific information about their clause-internal associate.

Right-dislocated and extraposed phrases both follow the clause-final verbs (if present), so it can sometimes be difficult to distinguish between the two cases. This subsection will review a number of possible cases, and will provide a number of tests that can be used to distinguish between the two cases.

Backgrounding RD and hanging-topic LD (cf. Section 37.2) are similar in that they produce structures that are pronounced with a flat intonation contour, but the examples in (126) show that backgrounding RD is more flexible with respect to categorial status; while hanging topics are typically nominal in nature, backgrounded phrases can be nominal, clausal, adjectival, or adpositional. The examples in (126) also show that the clause-internal correlate of the backgrounded phrase can serve different syntactic functions: it can be an argument, as in the (a)-examples, a complementive (126b), or an adverbial phrase (126c). The correlate is usually a phonetically light element, like the pronouns ʼmhim and hetit or the R-word erthere, although phonetically heavier demonstrative forms like die/datthat and daarthere are occasionally found.

The judgments on the examples in (126) do not seem to change when we give a contrastive accent to the right-dislocated phrase; this shows that afterthoughts have the same properties as background phrases, although their correlate can more easily be heavy or phrasal. This is illustrated in (127).

Right-dislocated phrases add to the information expressed by their correlates: (128a) assumes that the hearer does not know that Mr. Jansen is the manager, and (128b) suggests that the hearer may confuse the intended referent with someone who is not the manager. The right-dislocated phrase and its correlate can be interchanged but then it is assumed that the hearer does not know the name of the manager or may confuse the intended referent with someone who is not named Jansen.

The examples in (129) show that afterthoughts of the type in (128a) can occur in German either as an accusative or as a nominative noun phrase. The two cases have a slightly different meanings, which Ott & De Vries (2015: §6) tries to clarify with the paraphrases given here as translations. In (129a) the referent of the correlate is contextually given and the afterthought simply gives more specific information about this referent. In (129b) the correlate may introduce a new referent into the discourse, and the afterthought is then used to identify this referent as the speaker’s neighbor. Because of their meaning, these cases are called specificational and predicative afterthoughts, respectively, and given different analyses in Ott & De Vries (2015).

The same difference in meaning is found in Dutch, but since Dutch has no morphological case, we will largely ignore predicative afterthoughts in this section and refer the reader to Ott & De Vries’ (2015) for more discussion. For completeness, note that Van Riemsdijk (1997) and Van Riemsdijk & Zwart (1997:fn.5) observe a similar optionality in case agreement in German LD constructions, although in such constructions a mismatch in case assignment does not seem to trigger a similar predicative reading.

That right-dislocated phrases must provide more specific information than their clause-internal correlates accounts for the acceptability contrast between the examples in (130): referential pronouns and definite noun phrases both presuppose that the hearer is able to identify the intended referent, but definite noun phrases are more informative thanks to their descriptive content, and consequently the pronoun cannot be the right-dislocated phrase. Note that we have marked (130b) with a dollar sign because the construction is certainly not ungrammatical, as is clear from the fact that it becomes felicitous if the right-dislocated pronoun is accompanied by specific extra-linguistic information, such as a pointing gesture.

The acceptability contrast between the examples in (131) is also expected because the two coordinated phrases provide more precise information than their clause-internal pronominal correlates. We have again used a dollar sign because these examples may become acceptable when the right-dislocated phrase is supplemented with more material: for example, replacing these pronouns with ons/jullie tweeënthe two of us/you makes these examples felicitous by explicitly excluding other invitees from the set of contextually determined persons who could have been invited.

Backgrounded noun phrases are generally definite noun phrases, since they express discourse-familiar information; cf. the contrast between the two (a)-examples. Afterthoughts, on the other hand, can be indefinite if they are more informative than their clause-internal correlates: this implies that the correlate must also be indefinite.

The (a)-examples in (133) show that RD is like LD in that it cannot be applied to non-referential expressions; while definite noun phrases can easily be right-dislocated, quantified noun phrases cannot. The (b)-examples illustrate the same by showing that non-referential parts of idiomatic expressions resist right dislocation.

De Vries (2009) claims that nominal right-dislocated phrases do not show connectivity effects for binding, and at first glance this seems to be confirmed by the unacceptability of example (134a): RD of reflexive/reciprocal pronouns leads to unacceptability because the pronoun is not bound by a local antecedent. However, it is not clear whether the unacceptability of (134a) is due to the lack of connectivity; it may also be due to the fact that, under the given coindexation, the referential pronoun zethem is incorrectly bound within its local domain, the clause. This means that in order to investigate whether connectivity effects occur, we need to consider more complex examples such as (134b), where the anaphor is embedded in a right-dislocated phrase. The acceptability status of such examples is somewhat unclear: De Vries (2009) gives similar examples as unacceptable, but we find them acceptable and certainly much better than examples such as (134a).

That connectivity effects occur with reflexive pronouns is also clear from the fact that they can occur as an afterthought in examples such as (135a), taken from Ott & De Vries (2015); example (135b) provides a similar case with a reciprocal.

Example (136) shows that connectivity effects can also be demonstrated using bound variable readings: De Vries (2009) claims that this reading in not available, but similar examples are given as perfectly acceptable in Ott & De Vries (2015); we agree with the latter.

Example (137a) suggests that connectivity effects also occur with referential expressions: the proper noun Peter cannot be interpreted as coreferential with the subject pronoun hijhe. However, De Vries (2009) rightly points out that this is not conclusive, because coreferentiality is also blocked when the pronoun and the proper name occur in two consecutive clauses; linear order may be the crucial factor here.

If the correlate of a right-dislocated phrase is a non-obligatory clausal constituent, it can be omitted. Several cases were already given in Subsection II, but we illustrate this again here in (138a) for a temporal adverbial phrase. Some right-dislocated phrases never have a correlate, simply because there is no proform available; this is true for modal adverbs such as misschienmaybe in (138b).

Although the previous subsection has shown that afterthoughts and backgrounded phrases, as well as their clause-internal correlates, behave similarly in several respects, there are also differences; cf. Ott & De Vries (2015) and the references cited there. We have already mentioned that afterthoughts provide new information, whereas backgrounded phrases express information that is already known to the hearer. This is easily demonstrated by the question-answer pair in (139): RD of the noun phrase Marie is possible in the answer, but only if it is pronounced with a flat intonation.

Related to this difference in information load is that afterthoughts can be preceded by epistemic modal adverbials such as waarschijnlijkprobably when their correlates are indefinite; this possibility does not arise with backgrounded phrases, since their correlates usually refer to entities known to both the speaker and the hearer. For the same reason, afterthoughts, but not backgrounded phrases, can be combined with a modality marker such as wellichtmaybe or a hesitation marker such as toch.

Because afterthoughts add more specific information to the assertion in the preceding clause, the hearer can negate the added information independently of the clause, as can be seen in the (a)-examples. For the same reason, the afterthought can be provided by the hearer. Note that examples like (141b) show that afterthoughts can, at least sometimes, be independent of the clause containing their correlate; Subsection VI will provide more evidence for this conclusion.

This subsection has shown that there are a number of differences in the use of afterthoughts and backgrounded phrases that can be traced back to the role they play in the information structure of the discourse; afterthoughts provide discourse-new information, while backgrounded phrases provide discourse-old information.

Section 37.2, sub V, has shown that hanging-topic LD is not island-sensitive, because the resumptive pronoun hemhim remains in situ. Contrastive LD is different in this respect, as the resumptive demonstrative pronoun die must be moved into main-clause initial position; the acceptability contrast in (142) can be attributed to the fact that the demonstrative die is extracted from an interrogative clause (i.e. a wh-island).

Because RD is like hanging-topic LD in that it does not involve wh-movement of the clause-internal correlate of the right-dislocated phrase, we also expect it not to be island-sensitive regardless of its type (i.e. backgrounding or afterthought RD). The examples in (143) show that this is indeed borne out for wh and adjunct islands.

The situation is less clear for non-clausal islands, an issue to which we will return in Subsection VII. The examples in (144) show that RD seems to be like LD in that it is not sensitive to the islandhood of PPs: the preposition can be used, but is not required. However, we have added a percentage sign to (144b), because both Zwart (2011:78) and Ott & De Vries (2015:40ff) have claimed that the preposition must be realized; for them RD may be sensitive to the islandhood of PPs. Note that an anonymous reviewer of De Ott & De Vries (2015) also pointed out that not all Dutch speakers require a preposition to be present.

The primeless examples in (145) show that there is a sharp acceptability contrast between hanging-topic LD and RD when the correlate is embedded in a coordinate structure; this may suggest that RD is sensitive to the islandhood of these structures. Note, however, that the primed examples show that a similar contrast is found when the correlate is simply embedded in e.g. a direct object, suggesting that we cannot attribute the ungrammaticality to the presence of the coordinate structure.

Finally, we have to point out that De Vries (2002) suggests on the basis of example (146a) that attributively used APs can be used as afterthoughts, which would of course be another example of island insensitivity. However, Veld (1993:132ff) has pointed out that this is only apparent: we are dealing with reduced noun phrases, as is also clear from the fact that the indefinite article eena appears obligatorily when the correlate is singular, as in (146b). Like (145b), this seems to show that RD is sensitive to certain non-clausal islands.

This subsection has shown that there is some ambiguity in the evidence for/against island effects in RD constructions. There is evidence showing that RD is not sensitive to prototypical island configurations (wh and adjunct islands), while other evidence suggests that RD may be sensitive to some restrictions on movement. It is clear, therefore, that this is a topic for further study.

Right-dislocated phrases usually follow extraposed phrases such as the obligatory prepositional object naar meer informatie in (147): placing the right-dislocated modal adverb waarschijnlijk in a position between the clause-final verb and the extraposed PP leads to a degraded result. For completeness’ sake, we contrast example (147b) with example (147b'), in which the PP is right-dislocated.

Right-dislocated phrases also obey certain order restrictions: cf. Ott & De Vries (2015). Before illustrating this, the example in (148) shows that there can be more than one backgrounded phrase or afterthought (although some speakers may have difficulty with this). Note in passing that the first afterthought in (148b) is marked not only by a contrastive accent, but also by the fact that it is preceded by althans; we will use this as a diagnostic in (149).

Backgrounded phrases and afterthoughts can also co-occur, but then the former must precede the latter; together with our finding in Subsection IV that afterthoughts can occasionally occur as separate utterances, this shows that backgrounded phrases are more closely related to the preceding clause than are afterthoughts.

That backgrounded phrases are more closely related to the preceding clause is also suggested by the fact, illustrated in (150), that they must be adjacent to the minimal clause containing their correlate, whereas afterthoughts follow the main clause containing their correlate; we return to these instances in Subsection VII.

Although the examples in (148) have shown that there can be more than one backgrounded phrase or afterthought, this does not always lead to an acceptable result. The examples in (151b&c) show that modal and temporal adverbials can easily be right-dislocated, but for unknown reasons the (d)-examples show that it is difficult to have them together in a right-dislocated position.

While LD has received considerable attention in the theoretical literature, this is much less the case for RD. Because of the similarities between LD and RD constructions discussed in Subsection III, Ott & De Vries (2015) argue that it would be preferable for the two types of dislocation to receive similar analyses. Since RD does not involve movement of the correlate of the right-dislocated phrase, this may be a reason to reject the various movement approaches to contrastive LD discussed in Section 37.2, sub IX. This would leave us with Ott’s (2014) hypothesis that contrastive LD constructions consist of two juxtaposed clauses, the first of which is reduced under identity with the second clause. Ott & De Vries (2015) therefore provides a similar analysis for RD; it is argued that the derivation of RD differs from that of LD only in that the reduction applies not to the first but to the second clause of the juxtaposition. They analyze the right-dislocated phrase in (152a) in the same way as the fragment answer in (152b), i.e. the juxtaposition analysis of RD appeals to an independently motivated deletion operation.

Although it is not very clear what would trigger topicalization in the derivation of RD in (152a'), Ott & De Vries (2015) argues that examples of the type in (153) provide independent evidence for topicalization. Zwart (2011:79) has noted that the two examples in (153a&b) differ in their relative scope of the indefinite subject twee mensentwo persons and the epistemic modal vermoedelijk. The most prominent reading of (153a) is that the modal is in the scope of the numeral; two people have seen something that is presumably a wolf. The most prominent reading of the RD construction in (153b) is that the modal takes scope over the whole proposition, including the numerals; it is presumably the case that two people have seen a wolf. This would follow immediately from the proposed analysis of RD because the wide scope reading of the modal is also the most prominent one for the topicalization construction in (153c).

A possible problem for the assumption of topicalization in the second conjunct is the fact that RD is not sensitive to e.g. wh and adjunct islands; cf. Subsection V. However, Section 37.2, sub IX, has shown that this also holds for fragment clauses and LD constructions, which may mean that we are again dealing with the more general fact, discussed in Section V5.1.5, sub IB, that the ellipsis operation found in fragment clauses has the side effect that it also removes island violations; cf. Merchant (2001/2006) and the references cited there. Ott & De Vries (2015) develops an alternative to the island insensitivity of RD, according to which the proposed juxtaposition involves the minimal clause of the associate of the dislocated phrase; we illustrate this here only for RD. The minimal clause restriction requires that we analyze example (154a) as in (154b); the asterisk is used here to indicate that structure (154b') is ungrammatical (i.e. excluded by the minimal clause restriction).

A possible drawback of the minimal clause restriction is that we must allow for the coordination of an embedded clause and a main clause; cf. the structure in (154b), in which the first conjunct is a non-main clause while the second conjunct is a main clause. Although this kind of unbalanced coordination is not normally possible, the minimal clause restriction is empirically supported by the fact, illustrated in the primeless examples in (155), that backgrounded phrases must be adjacent to their minimal clauses: structure (155a') shows that this follows directly if we coordinate minimal clauses, while structure (155b') shows that the alternative of coordinating main clauses would incorrectly predict that (155b) is acceptable.

The acceptability of (155a) thus illustrates that the island insensitivity of RD follows directly from the minimal clause restriction, since wh-extraction from subject clauses is normally impossible; cf. *Peteri was [dat ti weg was] vervelend. Essentially the same is illustrated in (156); the structure of the utterance in (156a) must be as in (156b), since wh-extraction from a wh-island, as in the alternative structure in (156b'), usually leads to a highly degraded result; cf. *Elsj weet ik [wati tj ti al gegeten heeft].

In short, the island insensitivity of RD is now explained by the fact that the minimal clause restriction simply prevents island violations from occurring by blocking the structures in (155b') and (156b'). The minimal clause restriction further seems superior to Merchant’s analysis, according to which ellipsis removes island violations, because it accounts for the fact that RD is usually sensitive to certain non-clausal islands, such as the coordinate structure in (157a); topicalization in the minimal clause in the second conjunct in (157b) violates the coordinate structure constraint.

However, there are also problems. A first possible problem is that LD does not seem to be sensitive to the coordinate structure constraint; cf. Section 37.2, sub V. Indeed, this is quite damaging for the above-mentioned claim of Ott & De Vries that LD and RD should be analyzed in essentially the same way. And perhaps the problem is even more general, since Ott & De Vries (2015:fn.50) point to a similar possible problem with afterthoughts, which we have not discussed here.

A second problem is that afterthoughts can be detached from the clause containing their correlate, i.e. example (155b) becomes acceptable, at least for some speakers, when the right-dislocated phrase is given a contrastive accent; cf. Subsection VI. This may be related to the fact that afterthoughts can be used as independent expressions, but if afterthoughts are indeed independent of the preceding main clauses, the minimal clause restriction cannot, by definition, apply to them.

A third possible problem for the analysis based on the minimal clause restriction is that, at least for some speakers, RD of a prepositional object does not require the preposition to be present: since preposition stranding is not possible in the case of topicalization, as can be seen from *Mijn vaderi wacht ik op ti, example (158a) would be predicted to be unacceptable without the preposition.

This problem is not new, as Merchant (2001: §3, fn.6) already found that Dutch speakers show a great deal of variation with respect to preposition stranding in ellipsis constructions in general. The fact that we find this variation also in the case of RD can be used to support of a unifying ellipsis analysis of fragment clauses, LD and RD; cf. Section V5.1.5, sub IB, and Section 37.2, sub IX, for further relevant discussion.