Abstract and Keywords
This chapter provides a comprehensive analysis of Italian right dislocation, which is claimed to (i) occur clause externally, (ii) not require clitic doubling, (iii) involve movement. An accurate understanding of right dislocation is necessary for the analysis of discourse-givennes as well as focalization, as discussed in detail in Chapter 5. The analysis of right dislocation in this chapter challenges widely held but incorrect assumptions concerning the necessity of clitic doubling and the position of right dislocation, both of which can lead to incorrect conclusions about the position of focus when focus is followed by right dislocated constituents. The chapter also includes a detailed comparison between right dislocation and clitic left dislocation as well as a detailed discussion of alternative analyses of right dislocation to date.
Besides being marginalized in situ, Italian discourse-given constituents are routinely displaced to the right periphery of the clause (Antinucci and Cinque 1977; Calabrese 1988; Vallduví 1992, 1994; Cecchetto 1999; Villalba 2000; Cardinaletti 2001, 2002; Frascarelli 2004; Samek-Lodovici 2006; De Cat 2007; Frascarelli and Hinterhölzl 2007; López 2009). Since right-dislocated and marginalized phrases both follow focus, an accurate understanding of the representation and properties of right dislocation—or ‘RD’ for short—is essential for determining the dislocated or marginalized status of post-focus phrases, which in turn is a crucial prerequisite for an accurate analysis of the distribution of focus and the structure of Italian clauses. For example, the analysis of left-peripheral focus in Rizzi (1997, 2004) implies that the TP following focus is marginalized, since it occurs in situ. In the next chapter, I will instead claim that the TP following left-peripheral foci is right-dislocated, forcing a different analysis of left-peripheral focus itself.
This chapter examines the syntactic properties of RD in Italian, which, as we will see at the end of the chapter, are not necessarily shared by constructions that go under the same name in other languages (for a review of the possible semantic function of RD, see Villalba and Mayol 2013). I will show that in Italian declaratives right-dislocated phrases occur clause-externally, i.e. outside the TP where they have been generated. I will also claim that although RD may involve clitic doubling, RD without clitic doubling is possible too, as also claimed in Frascarelli and Hinterhölzl (2007). This result directly affects the analysis of Italian focalization, because it implies that sentences like (1) where a focused phrase precedes one or more higher-generated phrases cannot be proposed as evidence that the focused phrase has moved leftwards. The same word order obtains if focalization occurs in situ and the higher generated constituents have right-dislocated to the right of focus without clitic doubling. In fact, the absence of focus movement when RD is (p.76) controlled for, discussed in Chapter 3, confirms that this is the correct analysis of these sentences.
This chapter also aims at a better understanding of RD for its own sake and is therefore likely to also interest scholars working on dislocation phenomena. Specifically, I will claim that there are two distinct RD operations, one involving clitic doubling, dubbed ‘RD+’, and another lacking it, not even involving null clitics, which I will call ‘RD–’. But for clitic doubling, the two operations appear similar in all other respects, including their interpretation. The presence/absence of clitic doubling, however, affects the binding and wh‑extraction properties of the dislocated constituent. These differences are accounted for by assuming that phrases undergoing RD+ are generated as the specifier of a big DP headed by the clitic (Cecchetto 1999), while phrases undergoing RD– involve no such DP and are dislocated directly from their base generated position. The full representation will be introduced in Section 4.2.2.
As for the position of RD and its base-generated or movement-based nature, I will show that Italian RD moves the dislocated item to the specifier of an RP projection above the extended projection of the verb (mostly TP, but also CP when the right-dislocated phrase is an entire interrogative clause). Its right-peripheral position, in turn, follows from the remnant movement of the TP-node containing the original clause to an even higher projection, thus following in this respect the remnant movement analysis attributed to Kayne in Cecchetto (1999) and reproposed in Samek-Lodovici (2006). Unlike these studies, however, I will not assume that RD shares the position of clitic left-dislocated (CLLD) phrases. Much of the chapter will indeed examine several properties that distinguish RD from CLLD.
The table in (2) illustrates the different analyses proposed by different authors across Romance languages classified relative to the TP-internal vs. TP-external position and the moved vs. base-generated nature attributed to RD. The analysis proposed here belongs in the top left box, in contrast with other scholars arguing for base-generation, or a clause-internal position, or both.
The obligatoriness or optionality of overt clitic doubling is also subject to debate, with Villalba (2000), Cecchetto (1999), and Cardinaletti (2002) assuming obligatory clitic doubling and maintaining that apparent cases of RD lacking it should be reinterpreted as involving marginalization (Cardinaletti 2002; Cecchetto 1999: 65). These and other differences will be addressed as the discussion unfolds. (p.77)
The chapter is organized as follows. Section 4.2 examines the availability of right dislocation without clitic doubling and provides the structural representations for RD– and RD+. Section 4.3 examines the position of right-dislocated phrases, claiming that they are located above TP on the basis of evidence from clitic doubling, word order, NPI‑licensing, binding, extraction, and agreement. Section 4.4 discusses the evidence for a movement-based analysis, investigating how RD– and RD+ diverge from CLLD on the tests for movement and base-generation developed in Cinque (1990) in support of a base-generated analysis of CLLD. Section 4.5 compares the proposed analysis with the alternative accounts in the table in (2), highlighting which properties of Italian RD they cannot explain and also how any relevant property of RD captured by these analyses is addressed by the analysis proposed here. Finally, Section 4.6 examines how Italian RD differs from Catalan and French RD with respect to NPI-licensing, reconstruction, and island sensitivity, showing that RD is crosslinguistically non-uniform.
In Italian, RD is an extremely productive process that optionally applies to one or more discourse-given constituents independently of their syntactic category and grammatical function, as shown by the examples in (3) where all right-dislocated constituents are marked by the subscript ‘R’ and follow a presentationally focused clause. I am here exclusively interested in instances of right dislocation occurring in declarative clauses; right dislocation in yes/no questions will not be considered because as Crocco (2013) showed it has very different properties and appears to constitute a distinct process serving a separate discourse function.
Like marginalized phrases, right-dislocated constituents in Italian declaratives always follow the word carrying main stress, never carry main stress themselves, and are introduced by an intonation break and optional short pause here represented as a comma. Unlike marginalized constituents, right-dislocated constituents can be doubled by a clitic that agrees in person, number, gender, and case with the dislocated constituent whenever a clitic expressing these traits exists (for example, gender specification is expressed in third person object clitics, but not in first and second person ones).
(p.79) When clitic doubling involves a silent pro, as proposed by Cardinaletti (2001, 2002) for right-dislocated subjects, or it is absent, two properties permit us to distinguish RD from marginalization. First, right-dislocated constituents may occur in any order (Antinucci and Cinque 1977; Vallduví 1992), whereas in-situ marginalized phrases obligatorily follow the base-generated order. See example (4) involving simultaneous RD of a subject (doubled by pro), an indirect object, and a second prepositional argument. The six possible orders are all grammatical and contrast sharply with the ungrammatical examples involving freely ordered marginalized phrases discussed in Section 2.3.
(p.80) Second, postverbal negative phrases and negative polarity items can be marginalized but not right-dislocated. See the contrast between (5)(a), involving a marginalized object, and (5)(b), where the object is right-dislocated, as demonstrated by the presence of clitic doubling. The same holds when clitic doubling is absent, as is the case in (6) where the negative object follows a clitic-doubled right-dislocated indirect object (further examples will be discussed in the next section).
We thus have three properties that help us assess the status of unstressed discourse-given phrases following focus that are potentially ambiguous between a marginalization and a right dislocation analysis. If they are clitic-doubled, or if they do not follow their base-generated order, they are right-dislocated. If they are negative phrases or NPIs and the sentence is grammatical, they are marginalized. The only truly ambiguous cases involve non-clitic doubled phrases which occur ordered along their base-generated order and are neither negative phrases nor NPIs.
4.2.1 Right dislocation without clitic doubling
Before proceeding with the structural representation of RD, it is worth examining in detail the optional vs. obligatory status of clitic doubling. Though possible whenever a suitable clitic exists, I claim that clitic doubling of right-dislocated constituents is not mandatory, thus following Benincà (1988) and Frascarelli and Hinterhölzl (2007) (p.81) but contra Cardinaletti (2002), who claims that clitic doubling is obligatory, as well as Cecchetto (1999: 65) and Cruschina (2010), who maintain that any case of RD not involving clitic doubling is only apparent, actually involving marginalization.
There are two distinct sets of cases involving RD without clitic doubling. The first instance occurs when a suitable clitic is available and yet clitic doubling is optionally absent. This is illustrated by the following examples where each sentence ends with two right-dislocated constituents, but only the first one is clitic-doubled.
The two sentences in each example are identical except for the order of the right-dislocated phrases and the corresponding omitted clitic. Their interpretation is also identical, as expected if the final two phrases share the same right-dislocated status (see also the examples in Benincà 1988: 147). Their right-dislocated status is also confirmed by the negative phrase/NPI diagnostics. Consider for example the final object of (7)(a) and the final indirect object in (7)(b). If they were marginalized, they should be replaceable by a negative phrase, but this is not possible, see (10)(a–b). Sentences (11)(a–b) act as control, showing that the same negative phrases are fine when the entire clause is presentationally focused and right dislocation is not a factor. It follows that the non clitic-doubled phrases in the above examples are right-dislocated and cannot be analysed as marginalized, as proposed in Cruschina (2010). It also follows that overt clitic doubling with right-dislocated phrases is optional, as originally observed by Benincà (1988). (p.82)
The second set of cases occurs when RD affects a constituent for which a suitable clitic is unavailable. Temporal, instrumental, or benefactive adjuncts, for example, lack a corresponding clitic and consequently are never clitic-doubled, yet they can be right-dislocated. This is shown by examples (12)–(14), where the adjuncts at issue follow a clitic-doubled, right-dislocated phrase and therefore must themselves be right-dislocated. Their right-dislocated status is confirmed by the ungrammaticality of (15)(a), where the negative counterpart of the adjunct yields an ungrammatical sentence, as expected if the adjunct is right-dislocated. Sentence (15)(b) acts as control, showing that the same negative adjunct is fine when the entire clause is focused and right dislocation is not a factor.5
(p.83) Cardinaletti’s claim for obligatory clitic doubling is based on constructions involving direct object DPs. A first piece of data, from Cardinaletti (2002: 33) but here slightly adapted to express the associated intonation and focusing, is shown in (16). The post-sentential object requires clitic doubling. A second example follows in (17) (Cardinaletti 2002: 33, p.c.) and concerns Central and Southern varieties of Italian where right-dislocated accusative objects can be preceded by the case marking preposition a ‘to’. As (17)(b) shows, the same preposition becomes ungrammatical when a clitic is absent, showing that its presence is necessary in these RD constructions.
When considered in the contexts of the above discussion, where overt clitic doubling was shown to be either optional or impossible, maintaining that clitic doubling is obligatory appears too strong a claim. Rather, factors unrelated to RD, and possibly associated with case-assignment, appear to be forcing the presence of a clitic in (16) and (17). Indeed, when we consider simple sentences involving object DPs, such as the ditransitive clauses in the following examples, we see that object DPs may be right-dislocated without clitic doubling. (p.84)
On the basis of the above discussion, we may conclude that RD does not require overt clitic doubling. What we do not yet know is whether clitic doubling can be structurally absent, i.e. there is no clitic, or whether clitic doubling remains present but involves a null clitic.
22.214.171.124 No null object clitics
A first piece of evidence comes from the observation that Italian object clitics force obligatory past participle agreement, as shown in (21). If null object clitics were possible, they should trigger agreement too. Instead, agreement is obligatorily absent: past participles are obligatorily inflected with default third person singular masculine morphology. This is shown by (22) where the object le tue zie is ambiguous between a marginalized and a right dislocation analysis. Crucially, even when right-dislocated it does not trigger agreement.
A second piece of evidence comes from Cardinaletti’s (2002) observation that if null clitics were possible in Italian they should be able to also occur elsewhere. For example, they should be able to replace the obligatory clitic required by left-peripheral object topics. Yet, minimal pairs like (23) show that this is not the case.
(p.85) The third argument follows from the analysis of right-dislocated quantified DPs like (24). At first, these sentences may appear to involve right dislocation of a quantifier. Yet this impression is incorrect because unambiguous cases of quantifier extraction are ungrammatical; see (25)(b) and (25)(c), which are derived from (25)(a) via wh-extraction and right dislocation of the quantifier.
Sentence (24) therefore involves right dislocation of the entire object DP with no overt clitic doubling, as shown in (26) (the partitive clitic ne ‘of them’ refers to the quantified NP, not the entire DP).
Crucially, however, the dislocated DP disallows for any overt clitic doubling whenever the partitive clitic ne is present. This is shown in (27)(a), which is ungrammatical under any permutation of the involved clitics, and (27)(b), showing that the same holds even when the indirect object is absent. The conditions that disallow the simultaneous occurrence of the overt object clitic and the partitive clitic ne would arguably also apply to a null object clitic, showing that no null object clitic can be present in sentence (24).
A final argument for the existence of clitic-less right dislocation emerges when examining the binding properties of RD. Consider the anaphoric object se stesso (p.86) (himself) in the three sentences in (28). Its right-dislocated status is certain across all three: in (a), it follows a clitic-doubled right-dislocated indirect object and therefore it too is right-dislocated. In (b) and (c), it is overtly clitic doubled, with (c) also overtly doubling the indirect object.
Sentences (b) and (c) are ungrammatical due to the presence of the overt object clitic li (them). The object clitic corefers with the dislocated object anaphor and therefore also with the binding subject, causing a condition B violation.
If sentence (a) involved clitic doubling too, albeit by a null object clitic, its structure would be identical to that of sentence (c), and therefore it too would violate condition B. But (a) is significantly more acceptable than (b) and (c) and therefore it cannot involve a null clitic.
That the ungrammaticality of (28)(b–c) follows only from the presence of the object clitic, rather than some other unrelated factor, is confirmed by the sentences in (29), where the anaphoric object in the sentences in (28) is replaced by the object il proprio lavoro ‘his own work’, which though anaphoric with the subject, it is not co-indexed with the clitic (since the latter doubles the entire dislocated phrase), thus removing the binding relation between the subject and the object-clitic that violated condition B in the sentences in (28). As expected, binding is now successful even for the clitic-doubled sentences in (b) and (c).
(p.87) In conclusion, the evidence examined in this section shows that Italian does not allow for null clitics, in accord with Cardinaletti (2002). It also shows that RD allows for clitic doubling but does not require it, against the assumption that clitic doubling is necessary, whenever a suitable clitic exists as maintained in Cardinaletti (2002), Cecchetto (1999), Villalba (2000), López (2009), and Cruschina (2010).
Furthermore, since these results have been shown to hold for right-dislocated constituents while controlling for marginalized status, it is not possible to maintain that right-dislocated phrases lacking clitic doubling should be re‑analysed as marginalized phrases as proposed in Cruschina (2010). Conversely, the absence of clitic doubling is not a sufficient cue to assume marginalized status.
It follows that any post-focus discourse-given phrase that is neither clitic-doubled nor an NPI or negative constituent is ambiguous between a marginalized or right-dislocated analysis until other diagnostics determine its status. Therefore, any analysis in the literature about focus and givenness in Italian where constituents of this kind have been considered as occurring in situ solely because they lacked clitic doubling should be re-examined, as the structural implications associated with a marginalized and right-dislocated status are radically different. Marginalized phrases occur in situ, but right-dislocated phrases are extracted and moved to a clause-external position, as discussed in the following sections.
4.2.2 The representation of right dislocation
So far we have established that RD is a highly productive operation that optionally applies to discourse-given constituents across distinct syntactic categories, dislocating them to the right of focus, leaving them freely ordered, and occurring both with and without clitic doubling (provided a suitable clitic exists). What we still need to establish is the structural position of right-dislocated phrases and whether their dislocation involves movement or base-generation. In this respect the literature is divided. On one side, Cardinaletti (2001, 2002) and Frascarelli (2004) for Italian and De Cat (2007) for French maintain that right-dislocated constituents are base-generated outside their clause, even though the details of their analyses vary significantly. On the other, Cecchetto (1999) for Italian and Villalba (1998, 2000) and López (2003, 2009) for Catalan, though again with significant differences, maintain that right-dislocated constituents are clause-internal, located lower than T but above vP and reaching this position through movement.
The analysis that emerges from the array of evidence discussed in the rest of this chapter diverges from both claims. As for the position of RD, which will play a crucial role in the analysis of left-peripheral focus in Chapter 5, I will show that (p.88) Italian right-dislocated phrases are located above TP, in accord with Cardinaletti (2001, 2002), Frascarelli (2004), and De Cat (2007). As discussed in Section 4.3, this analysis accounts for the properties of right-dislocated phrases relative to NPI-licensing, binding, wh-extraction, word order, and agreement.
With respect to the movement/base-generation debate, I will show that right dislocation involves movement, in accord with Cecchetto (1999), Villalba (1998), and López (2003, 2009). Movement will be shown to be necessary whenever clitic doubling is absent, since only a movement-based analysis accounts for the reconstruction properties displayed by right-dislocated phrases in these cases. The clitic-doubled cases are less clear-cut, but as discussed in detail in Section 4.4, the available evidence still favours a movement analysis, in contrast with Frascarelli and Hinterhölzl (2007) where movement is restricted to RD– alone, while RD+ is assumed to involve base-generation.
A final choice concerns the leftward vs. rightward nature of the movement involved by right dislocation. It is a difficult choice because the two alternative representations emerging from it are consistent with all the properties of right dislocation examined in this chapter and they also account for most of the focalization patterns discussed in Chapter 5. Which representation is best is thus an independent issue; the claims and arguments presented in this chapter about right dislocation and in this book about the position of focus and its interaction with right dislocation remain valid under both representations.
There are however two patterns, one concerning rightward focus extraction and the other right-dislocated phrases sandwiched between fronted foci and right-dislocated TPs, where the representation of right dislocation based on leftward movement turns out to be empirically superior to a rightward movement and is therefore the one being chosen here. The comparison of both representations relative to these cases requires first considering the interaction of right dislocation and left-peripheral foci in Chapter 5, and is therefore provided outside this chapter in Appendix B.
The following two subsections present the representations for RD– and RD+ adopted throughout this book. Both are based on leftward movement, thus being consistent with an antisymmetric perspective, and both place right-dislocated phrases above TP. They differ only in the absence vs. presence of clitic doubling.
126.96.36.199 The structure of RD–
When clitic doubling is absent, I maintain that right-dislocated phrases move to the specifier of a projection RP located above the extended projection of the verb (Grimshaw 2000). In most cases, this extended projection coincides with TP, but it may extend to CP when wh-phrases are extracted from right-dislocated clauses, see for example the data in Section 4.4.4 as well as Section 5.4.5.
(p.89) The operation just described is always followed by remnant movement of the remnant TP (or CP where necessary) to the specifier of a higher projection XP located above RP, which is responsible for stranding right-dislocated phrases in clause-final position.6
Consider for example sentence (30) involving a right-dislocated object. As (31) shows, first the object is extracted from its base-generated position and moved to specRP, then the entire TP-remnant moves to the specifier of XP, thus eventually preceding the right-dislocated object. The final structure is provided in (32).
The analysis instantiates the double topicalization analysis attributed to Kayne’s 1995 Harvard lectures by Cecchetto (1999) and further developed in Villalba (2000), Samek-Lodovici (2006), and Frascarelli and Hinterhölzl (2007). Unlike Kayne’s (p.90) original proposal and Samek-Lodovici (2006), however, I do not maintain that RD follows from CLLD, as this would imply that right-dislocated and CLLD phrases share the same syntactic position, a prediction that is refuted by the several syntactic and pragmatic properties distinguishing CLLD and RD examined in Section 4.4 (further differences are mentioned in Section 5.3 and in Brunetti 2009).
188.8.131.52 The structure of RD+
When a clitic is present, I assume with Cecchetto (1999), Torrego (1995), and Uriagereka (1995), that the dislocating constituent is generated as the specifier of a complex DP headed by the clitic (see also Belletti 1999, where the dislocating phrase constitutes the complement of the clitic rather than its specifier, a difference that has no bearings for the analysis presented here). Support for this complex DP analysis is also provided in Kayne and Pollock (2008), who point out how it better accounts for the agreement between the clitic and the dislocated phrase.
The presence of the clitic DP determines some interesting asymmetries between RD+ and RD– which will be examined in Section 4.4, but otherwise the analysis of RD+ resembles the analysis for RD– just described. First, the dislocated phrase moves to specRP, then the remnant TP moves to specXP. As for the final position of the clitic, I follow Cecchetto (1999) in assuming that the DP moves at least as high as the specifier of the aspectual projection hosting past participles, accounting for clitic–past-participle agreement. The clitic head then moves to T, or whichever head hosts the agreement and tense features of the clause. The corresponding derivation is given in (34), yielding the final structure in (35).
184.108.40.206 Structural properties shared across RD– and RD+
Note how in both RD– and RD+ the right-dislocated phrase neither c-commands nor is c‑commanded by any of the items in the original TP. Furthermore, right-dislocated phrases necessarily follow focus, because focus cannot be discourse-given and therefore is necessarily part of the remnant TP eventually preceding all right-dislocated phrases. Since focus attracts main stress, this also derives the post-stress location of RD.
I assume that the speaker’s conceptual/intentional interface selects which phrases are targeted by RD among the available discourse given phrases, consistently with the optional nature of RD. The pool of discourse-given phrases themselves is determined from the discourse context as discussed in Schwarzschild (1999).
4.3 Right dislocation is located above TP
There is robust evidence for identifying the position of right-dislocated constituents as external to TP. This includes the properties of RD relative to clitic doubling, word order, wh‑extraction, right roof condition, NPI-licensing, and reconstruction. They are examined in turn below.
4.3.1 Clitic doubling
Dative constructions aside,7 Italian strongly disallows clitic doubling within the clause, which in turn supports a clause-external position for clitic-doubled right-dislocated phrases.
(p.92) For example, Calabrese (1988: 557) observes that when main stress is rightmost in the clause, thus ensuring that all involved constituents lie within the clause, the internal arguments of a verb cannot be clitic-doubled. Some of Calabrese’s original examples follow, here slightly adapted to show focalization and stress marking. For completeness, a few additional examples involving different clitics and lacking a final adverbial are listed in (37).
The absence of clitic doubling is independent from the informational status of the doubled phrase: clitic doubling is absent whether the doubled phrase is focused or discourse-given. For example, as pointed out in Vallduví (1992) and Zubizarreta (1994a) with respect to Catalan and Spanish, if clitic doubling of unfocused phrases were possible, the arguments preceding a narrowly focused subject in clause-final position should allow for clitic doubling, as they do in Greek (Anagnostopoulou 1999). In Italian, however, clitic doubling remains impossible in these cases as well, see (38). These sentences become grammatical when the clitic is absent, showing that it is the presence of clitic doubling that makes them ungrammatical. (Note that object clitics trigger past participle agreement. Omitting or changing the agreement suffix does not affect the ungrammatical status of the following (b) sentences.)
(p.93) Some languages extend clause-internal clitic doubling to contrastively focused items, see the dialogue in (40) from Limeño Spanish in Sanchez (2005). But for some marginal exceptions in colloquial registers,8 this is not possible in Italian, where focused constituents disallow for clitic doubling. See the corresponding dialogue in (41), with clitic doubling impossible in sentence (b).
The unavailability of clause-internal clitic doubling in Italian is also confirmed by the empirical study in Kuchenbrandt, Kupisch, and Rinke (2005) which compares the properties of weak and strong pronouns for objects and indirect objects across Italian, French, Portuguese, Romanian, and Spanish. After testing native speakers with structures similar to Calabrese’s ones, they conclude that Italian resists clitic doubling, in contrast with Portuguese, Romanian, and Spanish where clitic doubling is more freely allowed. The same conclusion is also reached in Gerlach (1998), who shows that standard Italian, unlike Romanian, Spanish, French, and specific Italian dialects, resists clause-internal clitic doubling irrespective of the pronominal/non‑pronominal, and specific/non-specific nature of the argument being doubled. Similar observations are also mentioned in Fontana (1993), Gerlach (2002), Cardinaletti (2002), and Belloro (2007).
(p.94) Overall, the robust evidence against clause-internal clitic doubling forces a clause-external analysis of Italian right dislocation whenever clitic doubling is present. As pointed out by Cardinaletti (2002), a clause-external analysis also offers a straightforward explanation for the differences between marginalization and right dislocation relative to clitic doubling. Marginalized constituents cannot be clitic-doubled because they occur clause-internally, whereas right-dislocated phrases can be clitic-doubled because they are clause-external. In contrast, a clause-internal analysis of right dislocation would have to stipulate that right-dislocated phrases are exceptional in allowing for clause-internal clitic doubling while at the same time being unable to explain why clitic doubling does not extend to marginalized constituents.
4.3.2 Relative order of marginalized and right-dislocated phrases
The distribution of right-dislocated phrases relative to marginalized ones provides further support for the clause-external position of right-dislocated items.
Under the proposed analysis of RD, marginalized constituents are predicted to obligatorily precede right-dislocated ones because they occur in situ and hence within the remnant TP that eventually precedes all right-dislocated phrases. The prediction is borne out. Sentences (42)(a) and (42)(b), containing the same marginalized negative subject and clitic-doubled right-dislocated object, differ with respect to their linear order. Crucially, only sentence (a) with the marginalized subject preceding the dislocated object is grammatical.9
(p.95) The examples in (43) follow the same logic, testing the order of a marginalized NPI object and a right-dislocated indirect object. Once again, the marginalized object must precede the dislocated indirect object.
If right-dislocated constituents occur clause-externally, the above ordering constraint follows straightforwardly. If, on the other hand, right-dislocated phrases were located clause-internally, the order between marginalized and right-dislocated phrases would closely depend on the structural details of the analysis being assumed. Current clause-internal analyses of right dislocation, such as Cecchetto (1999) and Kayne (1994), make incorrect predictions. In Cecchetto (1999), right-dislocated constituents raise to the specifier of a topic projection located between TP and VP. Since they precede VP, right-dislocated items should precede any marginalized phrase destressed in situ within VP, thus incorrectly predicting (a) to be ungrammatical and (b) grammatical in both examples (42) and (43).
A similar problem affects Kayne (1994: 81), where right-dislocated constituents are stranded in situ. Since marginalized constituents, too, occur in situ, marginalized and right-dislocated phrases should be ordered according to their base-generated order. Consequently, right-dislocated constituents would precede marginalized ones whenever generated above them. Yet this is not the case. As (44) shows, a sentence including a marginalized indirect object and right-dislocated subject and object must order the marginalized phrase first, as in (44)(a), and cannot follow the base-generated order in (44)(b). (p.96)
Cecchetto’s and Kayne’s analyses were proposed before Cardinaletti (2001, 2002) and therefore could not take advantage of the clear distinction between marginalization and right dislocation that Cardinaletti’s research made available, nor the tests for marginalization and right-dislocated status developed in this book. Nevertheless, their analyses show that clause-internal analyses of RD, whether involving leftward movement or placement in situ, are inconsistent with the observed relative order of marginalized and right-dislocated phrases, thus supporting a clause-external analysis of right dislocation.
4.3.3 Failure in licensing n-words and NPIs
Independent support for the clause-external nature of RD comes from the distribution of n‑words and NPIs. As discussed in Appendix A, Italian n-words located lower than T must be licensed by a c‑commanding licenser at the surface and the same holds for NPIs (Zanuttini 1991; Longobardi 1991; Acquaviva 1999; Penka 2011). This condition makes them an ideal testing tool for the position of right dislocation (Samek-Lodovici 2006; Villalba 2000). If RD is TP‑external, we expect right-dislocated n‑words and NPIs to be ungrammatical, since they are located above TP and hence outside the licensing domain of a preceding neg‑marker in T. For example, the right-dislocated NPI alcunché and the n-word nessuno in (45) would not be licensed because they are not c‑commanded by the neg-marker non in the corresponding structure (46). If, on the other hand, RD is TP-internal, i.e. lower than T as in Cecchetto (1999), we expect the same n-words and NPIs to be grammatical because they would still be c-commanded by the neg-marker in T. Note that the fact that right dislocation allows for reconstruction, as discussed later in this chapter, is irrelevant because negative licensing under c-command must hold in the surface structure, not the reconstructed one (see Appendix A). (p.97)
As the following examples show, the data support the TP-external analysis of RD (see also Calabrese 1992: 93). In each example, sentence (a) lacks right dislocation and the NPI/n-word is licensed by the preceding neg-marker non ‘not’ here assumed to be cliticized to T as in Belletti (1990). Sentence (b) right dislocates the NPI/n-word either on their own, as in (47)(b), or as part of a constituent containing them, as in (48)(b) and (49)(b). In all (b) sentences, the occurrence of right dislocation is confirmed by the presence of clitic doubling and the intonation break and optional pause preceding the right-dislocated items here represented by a comma. All (b) sentences are ungrammatical, as predicted.
Crucially, any conceivable alternative explanation for the ungrammatical status of the (b) sentences does not stand up to scrutiny. Bocci (2013: 220.127.116.11), for example, appropriately observes that Italian positive indefinite quantifiers like qualcosa ‘something’ cannot right dislocate, and suggests that the same property (p.98) blocking right dislocation of qualcosa might be responsible for the failed right dislocation of n-words and NPIs in the Italian sentences in the previous examples. If this were the case, however, n-words and NPIs would be expected to resist right dislocation even in languages where right dislocation occurs demonstrably lower than T, thus excluding negative licensing from being a factor. One such language is Catalan, but as Villalba (2000) and Feldhausen (2008) show, Catalan right-dislocated n‑words and NPIs are actually grammatical, refuting Bocci’s hypothesis. See for example (50), where the right-dislocated NPI res and the n‑word ningú are successfully licensed by the initial neg-marker no. This shows that n-words and NPIs are compatible with right dislocation, leaving the high position of right dislocation and the consequent negative licensing failure as the only possible explanation for the ungrammaticality of the Italian sentences in the previous examples.
Bocci (2013: 18.104.22.168) also suggests that negative items in the Italian sentences might fail locality conditions on neg-concord that could apply even if they were right-dislocated lower than T. The following Catalan examples, however, refute this hypothesis as well, since they show that when right dislocation is demonstrably low, the locality conditions on negative licensing are satisfied. Once again, this shows that the problem with the Italian sentences in the previous examples is the high position of right dislocation, and not right dislocation itself.
It is worth adding that the right-dislocated status of the constituent containing res and ningú in the Catalan examples is not in doubt, since it is clitic-doubled by the clitic ho. The corresponding structure is provided in (51), based on Villalba’s (2000) analysis of right-dislocated items as located in the specifier of a TP-internal topic projection located between TP and vP. Since (50) is a copular sentence, vP was replaced with the DP projected by the clitic ho necessary for clitic doubling. Villalba’s topic projection was renamed ‘RP’ to ease comparison with the analysis proposed here. Additional Catalan examples of this kind are provided in section 4.6.1.
(p.99) Still playing devil’s advocate, we could also wonder whether the ungrammaticality of the Italian sentence (47)(b) might follow from the clitic doubling of the referenceless items alcunché and nessuno. This explanation, however, cannot account for the ungrammaticality of (48)(b) and (49)(b) where the NPI/n‑word is contained within a larger constituent that does allow for clitic doubling. Furthermore the same sentences become grammatical again once a c-commanding licenser is inserted within the right-dislocated constituent itself as in (52) and (53). This shows that licensing under c-command is the only condition being violated in (48)(b) and (49)(b), since any other conceivable cause for their ungrammaticality would be expected to also apply in (52) and (53), but it doesn’t.10
Playing devil’s advocate even further, we may finally wonder whether it is the intervening focalization that adversely affects NPI and n-word licensing in examples (47)–(49), and ceases to do so in (52) and (53) where focalized items no longer intervene between licenser and licensee. But as the following examples show, this is not the case either. A neg‑marker in the main clause successfully licenses an NPI or n-word in an embedded clause whether they are part of a larger focused constituent as in (54), contrastively focused as in (55)–(56), or following an intervening focus as in (57)–(58). (p.100)
In conclusion, the licensing failure suffered by Italian right-dislocated NPIs and n-words follows from their being positioned above TP. This is particularly evident when comparing Italian right-dislocated n-words and NPIs against their Catalan (p.101) counterparts or even their marginalized counterparts in Italian. In all these cases the n-words and NPIs at issue are all discourse-given and always following a focused constituent, but when marginalized in Italian or right-dislocated in Catalan they occur lower than T and can be licensed by a licenser in T, whereas when right-dislocated in Italian they occur above TP, fail licensing, and are therefore ungrammatical.
The binding relations of RD structures show that right-dislocated phrases reconstruct into their base-generated position. The presence of reconstruction, in turn, prevents binding from providing a test for the final position of right-dislocated constituents. As Cecchetto (1999) showed, their position may still be revealed by the asymmetric behaviour displayed by reconstructed arguments and adjuncts relative to condition C. Closely following Samek-Lodovici (2006), this section exploits these asymmetries to argue for the clause-external position of right-dislocated constituents.
As discussed by Freidin (1986), Lebeaux (1988, 1990), and Chomsky (1995), the absence of theta assignment makes it possible to insert adjuncts at later stages of the derivation, which, in turn, allows them to avoid reconstruction. Consider for example the sentences in (59) and (60) from Safir (1999), each followed by their respective structure, with silent copies shown in angled brackets.
In (59), John and he cannot co-refer, whereas they can do so in (60). As Labeaux (1988) and Chomsky (1995) explain, in (59)(a) the phrase that Mary had offended John is a selected argument of claim and therefore part of the larger phrase which claim that Mary has offended John merged as the complement of repeat. As (59)(b) shows, the raising wh-phrase leaves a copy behind containing John, which causes a condition C violation when he and John are co-referential. In (60)(a), instead, the phrase that offended John, is just an adjunct modifying claim but not selected by it. It may therefore be added when which claim has already been extracted from its initial position. As a result, he does not c-command any copy of John—see (60)(b)—thus allowing for a co-referential reading.
A similar structural asymmetry is found in Italian RD structures. The following sentences, from Samek-Lodovici (2006), test for condition C violations incurred by the initial null subject pro when binding a human referent within the right-dislocated object whose head is shown in bold. The bound referent is located either in a relative-clause adjunct of the dislocated object or in its sentential complement. For each pair (p.102) of sentences, sentence (a) shows the relative-clause case, involving late insertion of the relative-clause adjunct, whereas sentence (b) shows the sentential complement case, involving reconstruction. The associated tree-structure under the proposed RD analysis is provided in (63), with binder and bindee underlined.
My own judgements, in (61) and (62), only allow for a bound reading in the adjunct case. The asymmetry was also tested with 18 native speakers, nine linguists tested via email, and nine non-linguists tested through informal one-to-one interviews. As the (p.103) following table shows, the relative-clause sentences were considered more acceptable than the corresponding complement sentences in 94% of the sentence pairs assessed across both groups.11 The complete results are available in Samek-Lodovici (2006). Cecchetto (1999), and López (2009: 253) report different judgements for comparable constructions, both discussed in detail in Section 22.214.171.124.3 and 4.6.1.
The asymmetry supports the proposed clause-external analysis of right dislocated phrases. If RD were clause-internal, no asymmetry should arise because the initial (p.104) pro subject would c‑command the dislocated DP independently from the complements or adjunct nature of the phrase containing it. The asymmetry is instead predicted under the proposed clause external analysis. In the (a) sentences, the relative-clause is added when the object has already been dislocated into its clause-external position. Therefore, it remains outside the c‑command domain of the initial pro subject, thus satisfying condition C. See (65) which provides the structure for (61)(a) with the silent copy of the the right-dislocated object provided in angled brackets.
As a further control that the above results are not spurious, note that when the object is not right-dislocated, whether because marginalized in situ or because focused or part of focus, condition C is violated in the adjunct case as well, as expected. This is shown in (67) for the focused case and (68) for marginalization. This latter sentence requires a negative object to ensure that RD is absent.
In summary, the distribution of condition C violations across complements and adjunct of dislocated and non-dislocated objects confirms the clause-external position of right dislocated constituents.
RD has been described as subject to the right roof constraint (Ross 1967), which prevents right-dislocated phrases from occurring beyond the boundaries of their own clause (Kayne 1994; Villalba 1998; Cecchetto 1999). While this appears true for Italian tensed clauses, about which more in Section 4.4.5, right dislocation from non-finite complements is not upward-bound in Italian and right roof violations are also found in French tensed and untensed clauses (De Cat 2007: 515). Crucially, the presence of right roof violations is inconsistent with a clause-internal analysis of RD.
Examples of right roof violations by right-dislocated phrases are provided in examples (69) and (70), where the focused subject of the higher clause intervenes between a lower infinitival complement and its right-dislocated object.12,13 (p.106)
The presence of right roof violations in the above examples is certain. To begin with, the right-dislocated status of the objects is not in doubt, given the presence of clitic doubling and the typical intonation pattern associated with RD. Furthermore, the focused subject cannot be analysed as part of the lower clause, since the non-finite complement clause lacks a case-assigner for it, and, in (70), even a theta-role, since the lower PRO subject is controlled by the higher clitic object ci ‘us’. We may also further check the position of the focused subject by replacing it with its negative counterpart, as in (71), and note that as expected it fails to be licensed by a (p.107) neg‑marker placed in the subordinate clause, thus confirming that it is part of the main clause, arguably positioned in specVP.
The right roof violations in (69) and (70) are incompatible with clause-internal analyses of RD à la Cecchetto (1999) and Villalba (2000) where right-dislocated constituents raise to an intermediate topic projection between TP and VP. These analyses predict that the right‑dislocated object in (69) and (70) would remain within the lower clause and adjacent to the precedent verb; they cannot account for the position of the focused subject. To see this, consider the associated derivation. Assume the focused subject lies in specVP within the main clause as shown in (72)(a), this being the lowest possible position it may take. As (72)(b) shows, right-dislocating the object within the complement clause would leave the focused subject before the verb of the subordinate clause, thus not matching the order in (69) and (70). Even raising the entire complement to the left of the focused subject, as in (72)(c), would not produce the attested linear order because the object never leaves the subordinate clause. (To avoid excessive cluttering, the derivation in (72) does not represent auxiliaries and clitic doubling and the same holds for the other derivations later in this section.)
A more complex analysis, still based on a clause-internal analysis of RD, could raise the object OR all the way up to the clause-internal topic projection of the higher clause, as in (73)(b). The entire subordinate clause would then have to raise to an even higher topic projection in the main clause as in (73)(c) to ensure that the subordinate verb precedes the object OR, but even these operations do not yield the order of (69) and (70).
To get the right order, the focused subject would finally have to raise to an intermediate position between the infinitival complement [TP PRO V-Fin tk]i and the dislocated object OR. As we saw in Chapter 3, however, this kind of short-range raising is unavailable to focused constituents: postverbal subjects focus in situ and cannot raise past any immediately higher discourse-given phrase. It follows that even this analysis cannot be maintained.
(p.108) Any other conceivable derivation based on a clause-internal analysis of right dislocation faces the same problems. On one hand, the object in the infinitival clause must raise to the higher clause, or else it will be impossible to separate it from its clause. On the other hand, the assumed clause-internal position for RD necessarily places the raised object before the focused subject, thus still requiring the short-range focus raising shown to be impossible in Chapter 3.14
In contrast, a suitable derivation is readily available under the clause-external analysis of RD proposed in this book. As (74) shows, first the object dislocates to the higher clause. Then the infinitival complement shifts to the left of the focused subject; this being the independently attested raising of lower-generated discourse-given constituents above a higher focus examined at length in Chapter 3. Finally, the main clause raises as a remnant again in accord with the analysis of RD, yielding the correct word order.
As we saw in Chapter 2, agreement loss in Anconetan Italian provides an additional test distinguishing subjects higher than T from postverbal subjects lower than T (Cardinaletti 2001: 131). The relevant data are repeated here: (75) shows that agreement is necessary with specTP subjects, while (76)(a) and (76)(b) show that agreement loss is possible with focused and marginalized subjects in specVP (for further discussion, see Section 2.3.3).
If right-dislocated phrases were situated lower than T, we would expect right-dislocated subjects following focus to allow for agreement loss, analogously to the post-focus marginalized subjects in (76)(b). But as (77) shows, agreement loss is impossible with right-dislocated subjects, confirming their clause-external position. The subject of (77) follows a clitic-doubled object to ensure its right-dislocated status.
López (2009: 267) wonders whether agreement loss might be a quirky property of focused subjects, rather than being sensitive to their position. The data in (76)(b), however, shows that agreement loss is also possible with unfocused marginalized subjects, thus confirming that agreement loss is sensitive to the position of subjects, not their discourse status. We may therefore conclude that the distribution of agreement loss confirms the clause-external position of right-dislocated phrases.
4.3.7 Some apparent exceptions
Bocci (2013) proposes some interesting cases where apparently right-dislocated items precede a focus. The most typical and frequent examples involve sentences with multiple clauses, like (78). In these cases, a right-dislocated item can occur at the right-edge of its clause while still preceding the focus of another following or containing clause, as is the case here where the right-dislocated la medaglia ‘the (p.110) medal’ precedes the focused MARIA. The right-dislocated object here occurs within a clausal CLLD-topic, and hence inevitably precedes the in-situ focus of the main clause. This class of data only shows that right-dislocated items do not follow foci in absolute terms, but only foci within the same clause. These data are consistent with the analysis provided here, since the right-dislocated item can still be analysed as located higher than TP with respect to the clause containing it, here the bracketed left-peripheral CLLD-topic.
As Bocci (2013) points out, apparently more problematic cases of pre-focal right dislocation may occur within a root clause, although less naturally so; see (79) where the clitic-doubled Giovanni precedes the focused MARINA and is grammatical under the appropriate intonation. There are at least two reasons to question the right-dislocated status of this object, though. The first concerns its interpretation. Giovanni is interpreted contrastively, clarifying that it is Giovanni, not other men, who must be introduced to Marina. That this is indeed the case is shown in (80), which under the appropriate intonation allows for the explicit negation of potential alternatives to Giovanni. This is significant, because as shown by Benincá and Poletto (2004), Brunetti (2009), and Samek-Lodovici (2009), and as further discussed in Section 126.96.36.199.4, genuine right-dislocated items are never contrastive. Indeed, when Giovanni occurs as a genuine post-focus right-dislocated object, as in (81), it can no longer be contrasted against potential alternatives, see (82).
A second piece of evidence against the right-dislocated status of the prefocal object comes from the observation that it can be easily turned into the object of a typical afterthought expression such as ‘I mean’, suggesting it might actually constitute an (p.111) afterthought. Bocci (2013) provides evidence against an afterthought analysis for the multiclausal sentences examined in (78), but not for the root-clause prefocal items like Giovanni in (79), thus leaving an afterthought analysis a distinct possibility.
More research is needed to understand the discourse function and syntactic representation of these constructs. Yet, they clearly are not right-dislocated phrases, and hence not an exception to the TP-external position of right-dislocated phrases established in the previous sections.
4.4 Right dislocation is movement-based
Having determined that RD is clause-external, we need to examine whether right-dislocated phrases are moved or base-generated. In this section, I consider the main tests proposed in support of base-generation in Cinque’s (1990) study of CLLD, as well as other properties, and show that they support a movement analysis of RD–.
The evidence for a movement analysis of the clitic-doubled variant RD+, discussed toward the end of the section, is more limited. Nevertheless, I will argue that a movement analysis remains the most convincing account available for RD+ too. I will also briefly consider the claim in López (2009) that CLLD itself is movement-based, which, if correct, makes a base-generatation analysis of both RD– and RD+ even harder to contemplate.
A particularly informative test in Cinque (1990) distinguishes movement from base-generation through the distribution of the Italian pronominal clitic ne ‘of them’.
In Italian, the NP complement of a quantified DP of the form ‘[Q NP]’ can be pronominalized by ne, as illustrated in structure (84) (Belletti and Rizzi 1981; Rizzi 1982; Cinque 1990: 69). When the quantified phrase occurs in object position and the NP is unexpressed, ne-cliticization is obligatory (more precisely, ne constitutes the pronominal realization of the NP). This is shown in (85)–(87) for a transitive, passive, and unaccusative clause respectively, where (a) shows the full object DP and (b) the obligatory ne-cliticization when the NP is omitted. Ne-cliticization, passivization, and unaccusativity all trigger agreement on the verbal past-participle.
(p.112) Crucially, the obligatoriness of ne-cliticization is preserved under movement (Cinque 1990; Rizzi 1981). For example, wh-extracted direct objects still require ne‑cliticization when the quantified NP is absent: compare (88)(a) showing ne-cliticization against the ungrammatical (88)(b) lacking it. The same holds for wh-extracted subjects generated in object position in passive and unaccusative constructions, see (89) and (90). In all examples, stress falls clause-finally.
(p.113) Cinque exploits this property to test for the presence of movement in CLLD. If CLLD quantifiers involved extraction from an object DP with an omitted NP, the DP would obligatorily trigger ne‑cliticization. But this is not the case, as (91) shows. Therefore, Cinque argues, the CLLD phrases in (91) should be analysed as base-generated DPs with the structure ‘[Q PRO]’. Cinque’s analysis is also supported by the obligatory presence of the DP‑related object clitic le ‘them’ for the CLLD phrase in (91)(a), showing that clitic doubling is unproblematic and in fact required. (In the passive and unaccusative cases (91)(b) and (91)(c) the CLLD DP can be analysed as doubled by a resumptive pro in specTP).15
By the same logic, we expect object quantifiers targeted by RD– to require ne-cliticization if movement-based, and lacking it if base-generated as ‘[Q PRO]’. As the following examples show, ne-cliticization is always necessary, confirming the movement nature of RD–. Compare the grammatical sentences in (a) against their ungrammatical counterparts in (b) lacking ne-cliticization (clitics in bold). In all examples, the dislocated quantified phrase follows a clitic-doubled, right-dislocated indirect object or locative, thus ensuring its right-dislocated status and excluding a marginalization analysis.
In conclusion, the ne-cliticization test strongly supports a movement analysis for the clitic-less right-dislocation variant RD–.16
Another property that Cinque (1990) associates with base-generation is the obligatory presence of clitic doubling for left-dislocated object DPs. This is shown in (95)(a), where the base-generated CLLD phrase il vino ‘the wine’ must be doubled by the object clitic lo. Compare it with the corresponding clitic-less RD case in (95)(b). The judgement is particularly clear when the sentence is assessed as a reply to the context sentence. As usual, the presence of a right-dislocated indirect object ensures that the object is not marginalized in situ.
Cinque maintains that the obligatory object clitic is a direct consequence of the base-generated status of CLLD. If the CLLD object had been extracted from object position, it would have created an operator-variable chain, allowing for a variable in object position which would make overt clitic doubling unnecessary. Clitic doubling is instead obligatory precisely because CLLD does not involve movement, hence excluding the presence of a variable in object position, while all other potential empty categories are also excluded for independent reasons (pro is not identified, PRO is governed, an anaphoric NP-trace is unlicensed because unbound in its governing category).
By the same logic, movement must be present when the object clitic can be omitted, since a variable is the only possible item available in object position in this case. Therefore, in the cliticless sentence (95)(b) the right-dislocated object must have been extracted, consistently with a movement analysis of RD.
Reconstruction effects are present with both CLLD and RD structures. López (2009) makes a strong case against interpreting reconstruction effects in any other way than as movement-based, strongly suggesting that both CLLD and RD involve movement.
(p.116) Even if Cinque (1990) were correct in maintaining that reconstruction effects per se cannot be considered sufficient evidence for movement, the reconstruction cases examined in Section 188.8.131.52 support a movement analysis of RD. In particular, the reconstruction properties of the anaphoric object in (96)(a) cannot be explained through the presence of a silent object clitic, because if a silent clitic were present an overt clitic ought to be possible too, but it is not, as (96)(b) shows (see Section 184.108.40.206 for discussion).
If no clitic is present, however, the attested reconstruction effects can only be explained through a movement analysis involving a silent copy of the right-dislocated anaphoric object in object position, and letting this copy be bound by the subject.
As base-generated constituents in a specifier position, CLLD phrases unsurprisingly constitute an island to wh-extraction, as this would require movement out of an unselected specifier. As (98) and (97) show, wh‑extraction is ungrammatical whether the left-dislocated clause is non-finite or finite respectively. The sentences in (a) show a simple CLLD construction, and the ungrammatical (b) sentences the corresponding wh-extraction case.
If RD, like CLLD, were base-generated in the specifier of the unselected RP, it too would disallow wh‑extraction. Instead, wh-extraction from right-dislocated clauses is possible provided clitic doubling is absent, i.e. under RD–. Extraction is more readily available with non-finite clauses, see (99) and (100), but marginally grammatical instances are also possible with finite clauses, see (101). Since the final clause is right- (p.117) dislocated, the typical raising intonation of interrogative clauses is not available here. The most natural intonation is close to that of a declarative.
The availability of wh-extraction is unsurprising under the movement analysis of RD– proposed in Section 4.2.2. The wh-phrase is extracted before the complement clause is dislocated. This is shown in the derivation in (102) for sentence (100). Wh-extraction occurs at step 2, followed by the cliticless dislocation of the complement clause at step 3 (auxiliaries and prepositions have been omitted). Note that the final remnant movement must include the wh-operator in specCP (or the relevant projection under a splitCP analysis), and consequently the RP projection hosting right dislocation must here occur above CP (for additional similar cases, see Section 5.4.5). The final structure is in (103). (The symbol ‘|’ closes all preceding square brackets.)
The availability of wh-extraction thus provides independent evidence for the movement analysis of RD–. (The lack of wh-extraction under the clitic-doubled RD+ is addressed in Section 220.127.116.11.)
The movement-based nature of right dislocation is also confirmed by its unavailability with tensed complements of factive verbs, in direct contrast with the base-generated CLLD which remains available in these contexts as well.
As Truswell (2007a, 2007b, 2009) showed, movement of wh-operators out of the complement of factives is possible when the complement is non-finite, as in (104), but not when the complement has finite tense, as in (105) (examples adapted from Portolan 2005: 48. For English, see Erteschik-Shir 1973). Truswell maintains that movement is sensitive to the event articulation of the sentence. When the complement has non-finite tense, it forms a single macro event with the main clause, which, in turn, enables movement. When the complement has finite tense, each clause corresponds to an event of its own and movement is blocked.17 (p.119)
Truswell’s analysis provides a diagnostics for movement. We expect base-generated CLLD constructions involving factives to be grammatical independently from the tense or untensed status of the lower complement, whereas movement-based RD– should display the same tense-related alternation observed above with respect to wh-movement. Both predictions are borne out.
Starting with RD–, examples (106) and (107) show that it is possible to right dislocate an argument or a locative adjunct out of non-finite complements, whereas (108) and (109) show that the same is not possible with their tensed counterparts. In all examples, the dislocated phrase lies outside the lower clause, since it follows the focused subject of the matrix clause (this subject cannot be situated in the lower clause, where it would not get case and would violate condition C, see section 4.3.5).
The same alternation is absent in CLLD constructions. As (110)–(111) show, both the untensed and, crucially, the tensed CLLD counterparts of the RD– cases are grammatical. The assumed context sentences are the same as those provided for the above RD– cases.
RD– thus patterns with wh-extraction in showing sensitivity to tenseness when extracted from the complements of factives, CLLD shows no similar sensitivity. In the context of Truswell’s analysis, the insensitivity of CLLD to tense finiteness must follow from its base-generated status. By the same logic, however, clitic-less right dislocation must involve movement.18
For completeness, this section examines two additional tests from Cinque (1990) that do not provide conclusive tests when applied to CLLD and RD.
18.104.22.168 Successive cyclicity
The first test exploits the observation that movement-less chains lack the intermediate traces made available by successive cyclic movement in chains built by genuine movement operations such as wh-extraction. As Cinque points out, this predicts an asymmetry between CLLD and wh-movement when they apply to adjuncts lacking corresponding clitics. Since no clitic is present, the presence (p.122) of intermediate traces becomes essential to establish the required chain of antecedent government relations necessary for licensing the original trace. Consequently, wh-extraction of these adjuncts is possible, whereas they disallow CLLD. For example, he notices the contrast between the grammatical sentences in (112), where the wh‑phrase can be interpreted as modifying the lower clause, and those in (113), where the initial CLLD phrases are ungrammatical when interpreted as modfying the lower clause.
As (114) shows, however, CLLD constructions involving non-finite subordinate clauses can easily be interpreted as modifying the lower clause, undermining a base-generated analysis of CLLD.
If CLLD is indeed base-generated, as argued in Cinque (1990), the data in (114) show that a lower-clause interpretation of CLLD adjuncts without movement is possible, under the appropriate circumstances. It follows, that factors other than the absence of intermediate traces must cause the asymmetry between tensed and untensed clauses in (113) and (114). This undermines the above test as a tool for distinguishing base-generated chains from genuine movement, making its application to RD irrelevant.
Furthermore, as we saw in the previous section, RD resists extraction from tensed clauses independently from the argument or adjunct nature of the extracted item. Consequently, the unavailability of a lower-clause interpretation for the right-dislocated adjuncts in tensed clauses, shown in (115), cannot be attributed to the absence (p.123) of intermediate traces and taken as evidence for a base-generated analysis of RD, because the same pattern is displayed by right-dislocated arguments lacking intermediate traces, as shown in (108) in the previous section.
As in the CLLD cases, the factor determining the licensing of a lower clause interpretation is the finite or non-finite status of the lower clauses. As discussed in the previous section, tensed clauses block extraction, whereas untensed ones allow for it, as also shown by the examples in (116).
More research is needed for a proper understanding of the causes of the above CLLD and RD extraction asymmetries, with Truswell’s (2009) analysis of how tense affects event structure and interferes with movement extraction providing a promising research platform. Until the factors affecting extraction are properly identified, the successive cyclicity extraction test is uninformative, as it leads to opposite conclusions depending on whether extraction occurs from a tensed or untensed clause.
22.214.171.124 Parasitic gaps
Cinque’s (1990) diagnostics for movement also include parasitic gaps. CLLD does not license parasitic gaps, whereas constructions involving wh-extraction or focus movement do.
With respect to this test, RD patterns with CLLD, independently from the presence of clitic doubling. See the examples in (117), showing parasitic gaps licensed by wh- and focus-fronting in (117)(a) and (b), but absent under CLLD and RD in (117)(c) and (d).
As argued in detail in Villalba (2000: 253), however, the asymmetry between CLLD and RD on one side and wh- and focus-extraction on the other might just reflect the non-quantificational nature of CLLD and RD, which contrasts with the quantificational nature of wh- and focus-operators (see also Rizzi 1997). If the licensing of parasitic gaps requires a quantificational operator, lack of licensing under CLLD (p.124) and RD is expected and therefore uninformative with respect to their movement or base-generated nature. (But see also López 2009: 225 where parasitic gaps failures extend to wh- and focus-extraction constructions, suggesting that other factors might be relevant too.)
The parasitic gap test is also undermined by the following data, showing that CLLD can license parasitic gaps when narrow focus is present even if it is not binding the parasitic gap; see (118)(a). The same appears possible with RD+, although not with RD–, see (118)(b)–(c).
While interestingly puzzling and calling for further research, these data show that the parasitic gap test is inconclusive.
4.4.7 Clitic-doubled RD+ is movement-based too
Some of the evidence for movement discussed so far with respect to RD– cannot be applied to the clitic-doubled variant of right dislocation RD+. This includes evidence (p.125) based on data lacking clitic doubling, such as the discussion of dislocated adjuncts in the previous sections and the availability of right dislocation without clitic doubling examined in Section 4.4.2. Nevertheless, the evidence for a movement analysis of RD+ remains significant.
As we saw in Section 4.4.1, the availability of ne-cliticization allows us to assess the underlying structure of dislocated object quantifiers. When the quantifier is base generated as part of a more complex DP of the form ‘[Q PRO]’, as in Cinque’s analysis of CLLD quantifiers, clitic doubling requires DP-related clitics and excludes ne-cliticization, because ne only doubles NPs, not DPs. When the quantifier is instead extracted via movement from an object of the form ‘[Q NP]’ where the NP is not lexically expressed, the stranded NP is obligatorily realized as the pronominal clitic ne, triggering ne-cliticization (see Section 4.4.1 for details). The test was then used to support the base-generated analysis of CLLD quantifiers, where ne-cliticization is obligatorily absent, and the movement analysis of quantifiers dislocated via RD–, where ne-cliticization is obligatory.
A similar contrast is found between CLLD and RD+. As (119) shows, clitic-doubled right dislocation of the sole quantifier is not possible. Yet, under a base-generation analysis of RD+ this sentence should be grammatical on a par with the CLLD sentence in (120). The dislocated quantifier would be base-generated as part of the DP ‘[Q PRO]’ and doubled by the DP pro-form le (them). The ungrammaticality of (119), on the other hand, follows immediately under a movement analysis, since the obligatory pronominalization of the stranded NP‑complement via ne-cliticization is absent.
Quantifier dislocation by RD+ remains ungrammatical even when ne-cliticization is present, as shown in (121). Under a movement analysis this is expected, as the presence of clitic doubling—inevitable in RD+—prevents the successful extraction of the clitic ne, as explained below. (p.126)
In accord with the analysis of RD+ proposed in Section 126.96.36.199, the object DP ‘[Q ne]’ must be generated in the specifier of a larger DP headed by the object clitic le responsible for clitic doubling. Ne-cliticization is then inevitably blocked, as it would require extraction of ne from the unselected specifier QP, which according to Cinque (1990) constitutes an island to extraction. The initial structure of the clitic DP is provided in (122).19
As shown in Section 4.4.1, ne-cliticization is instead possible under RD–, where the absence of clitic doubling allows for the object ‘[Q ne]’ to be generated in object position, which in turn allows for a successful extraction of ne in parallel with any other instance of ne-cliticization. An example of successful ne-cliticization under RD– is provided in (123), with the corresponding derivation and final structure supplied in (124) and (125).
We thus observe three distinct patterns with respect to ne-cliticization in sentences involving quantifier dislocation by CLLD or RD. Ne-cliticization is obligatory under the movement-based RD–, whose structure allows for ne-extraction. It is impossible under CLLD, where movement is not involved and the CLLD-ed DP binds a DP-related object clitic. It is also impossible with RD+, because its structure disallows ne extraction. However, the unavailability of right dislocation without ne-cliticization shows that ne-cliticization is still required, supporting a movement analysis of RD+.20
A second argument supporting a movement analysis of RD+ comes from the reconstruction asymmetries of Section 4.3.4. They show that the complements of right-dislocated nouns necessarily reconstruct. This, in turn, causes condition C violations in sentences involving right-dislocated complements like (126)(a), whereas similar sentences involving right-dislocated adjuncts are grammatical because adjuncts do not need to reconstruct, see (126)(b).
(p.128) If RD involved base-generation, and reconstruction involved a non-movement chain between the right-dislocated item and the object clitic, we would expect the dislocated object to reconstruct in its entirety, independently from the adjunct/argument nature of its modifiers, thus neutralizing the above asymmetry. Movement thus appears a necessary prerequisite for the observed alternation.
188.8.131.52 Dislocation from tensed and untensed complements
The dislocation asymmetry examined in Section 4.4.5 with respect to RD– seamlessly carries over to RD+ as well. As the following examples show, dislocation to a higher clause is possible from the untensed complements, but not from the tensed complements, of factives.
As already discussed in Section 4.4.5, the same alternation is present with wh-extraction but not with the corresponding CLLD constructions, suggesting that tenseness differentially blocks movement, and consequently that RD+, too, is movement-based (see Section 4.4.5 for further discussion).21
Unlike RD–, which allows for wh-extraction from untensed complements (Section 4.4.4), RD+ always disallows it. This is shown in the following two examples, where (a) and (b) respectively illustrate the cases with untensed and tensed completements.
The absence of wh-extraction follows from the representation of RD+. As explained in Section 4.2.2, the dislocated complement is generated in the specifier of the DP headed by the doubling clitic; for example, the initial representation of (129)(a) prior to wh-extraction would be as in (131). This specifier is unselected, and therefore it constitutes an island to extraction (Cinque 1990), preventing wh-extraction. Nor can wh-extraction occur after the right dislocation of the complement, since this too would involve extraction from the unselected specifier hosting the dislocated clause.
It follows that the unavailability of wh-extraction under RD+ does not constitute evidence for a base-generated analysis, since it is also expected under a movement (p.130) analysis under the representation proposed in Section 4.2.2, which correctly accounts for the asymmetry between RD– and RD+.22
The asymmetry in wh-extraction also provides further evidence for the claim that the representation of RD– lacks a doubling clitic, null clitics included (Section 184.108.40.206). If a null clitic were present, it would block wh-extraction in the same way as overt clitics do for RD+.
220.127.116.11 Evidence from López (2009) and Villalba (2000)
The previous sections argue against a base-generated analysis of RD+ by contrasting its properties against those of CLLD, which is claimed to be base-generated in Cinque (1990). Villalba (2000: 234–64) and López (2009: 213–38), however, argue forcefully that CLLD involves movement too, providing a detailed reassessment of the evidence supporting base-generation. To begin with, both authors argue that CLLD shows some of the properties typically associated with wh-movement. Villalba shows that CLLD mimics wh-movement in requiring across the board extraction. He also shows that even weak crossover effects become possible when considering long-distance CLLD. López shows that CLLD is sensitive to both weak and strong islands, against Cinque’s claims. Both authors also show that subjacency is respected by CLLD. López’s evidence involves sentences with complex NPs rather than Cinque’s embedded CPs (López 2000: 226). His Catalan examples extend to Italian too; see the examples in (132).
Villalba and López also examine Cinque’s claim that CLLD involves movement‑free binding chains responsible for its binding and reconstruction properties. López argues that only an analysis based on copy theory, and hence movement, may properly explain the binding and reconstruction properties of CLLD and RD and their contrast with genuinely base-generated constructions such as hanging topics, where reconstruction is absent. The same contrast is unaccounted for under Cinque’s (p.131) analysis of CLLD. López also examines and either refutes or reinterprets in movement terms most of the remaining arguments for a base-generated CLLD provided in Cinque (1990), Iatridou (1995), Anagnostopoulou (1997), Frascarelli (2004), and Suñer (2006).
If Villalba and López are correct, a movement analysis of right dislocation becomes inevitable, as even those properties that are shared with CLLD could not be interpreted as support for a base-generated analysis.
The following table summarizes the evidence discussed so far in support of a movement analysis of right dislocation.
For the clitic-less variant RD–, the evidence for movement is robust. A base-generated analysis would have to explain why RD– diverges from base-generated CLLD on the first four properties. The remaining two properties, where RD– and CLLD converge, support movement.
The evidence concerning the clitic-doubled RD+ variant is more restricted but still compelling. The differences between RD+ and CLLD with respect to ne‑cliticization and right dislocation from tensed complements support a movement analysis, and so does the presence of reconstruction and parasitic gaps. The absence of wh-extraction follows from the representation of RD+, where the complex DP headed by the doubling clitic blocks extraction; the convergence with CLLD in this respect is thus accidental and cannot be interpreted as support for base-generation.
The evidence for a moved analysis of right dislocation has been supplied under the least favourable conceivable scenario, namely one where CLLD is base-generated as argued in Cinque (1990), as this analysis enables a parallel base-generated analysis of right dislocation. The above results, however, also show that despite the significant (p.132) progress in Cinque (1990), Villalba (2000), and López (2009), the status of Italian CLLD deserves further investigation, since properties 1–4 support a base-generated analysis but properties 5–7 favour a movement one. If CLLD turns out to be movement-based, as argued in López (2009), the movement-based nature of Italian right dislocation will be further strengthened, while at the same time raising the interesting issue—though one not relevant to this book—of how to account for the differences between right dislocation and CLLD in table (133).
4.5 Alternative analyses of right dislocation and related issues
This section compares the analysis of RD proposed in this chapter with alternative analyses in the RD literature. It examines the evidence favouring the analysis proposed in Section 4.2.2 over specific alternatives. It also examines those empirical observations from the RD literature that at first appear problematic for the analysis proposed here, showing that they are actually either invalid or need to be reinterpreted.
The first section discusses clause internal analyses of RD. After briefly discussing Kayne (1994), I examine at length Cecchetto (1999) and his critique of remnant movement analyses as well as similar points raised by the comparable analyses in Villalba (2000) and López (2009). The discussion also extends to the issues raised by Villalba (2000) and López (2009) about the binding properties of RD and the interaction of RD and CLLD.
The clause-external analyses of RD are discussed in Section 4.5.2, including the binding and relativized minimality properties highlighted in Frascarelli (2004) and her IP-inversion analysis, where RD is derived from base-generated CLLD.
4.5.1 Clause-internal analyses
One of the first clause-internal analyses was proposed in Kayne (1994), where right-dislocated phrases are assumed to be generated in situ and raised to the position of CLLD at LF. When considered in the context of the knowledge gathered so far, this proposal treats right-dislocated phrases as marginalized phrases destressed in situ, and therefore it is unable to predict the observation that marginalized phrases must precede right-dislocated ones. The availability of clitic doubling becomes problematic too, as it is normally excluded within the clause for the reasons examined in Section 4.3.1. As Frascarelli (2000, 2002) points out, Kayne’s analysis also incorrectly predicts right-dislocated phrases to be ordered in accord with theta-assignment, whereas in reality they are freely ordered. Further critical remarks are provided in Villalba (2000) and De Cat (2007).
A different and highly seminal analysis was proposed in Cecchetto (1999), where right-dislocated phrases raise to the specifier of an intermediate topic projection located between VP and a higher FocusP located between VP and TP; see (134) and the related structure (135) (slightly adapted from Cecchetto 1999: 58). (p.133)
Similar analyses—maintaining that right-dislocated phrases move to an intermediate position between TP and VP—have been proposed by Belletti (2004), Villalba (1998, 2000), and López (2009). Belletti, working on Italian, follows Cecchetto (1999) but allows non-dislocated phrases to raise to an intermediate topic projection above focus, so that they can precede both focus and right-dislocated constituents. Villalba, working on Catalan, raises right-dislocated phrases to a topic projection immediately above vP. López, also working on Catalan, has RD trigger A‑movement to a higher specifier of vP (the lower specifier being taken by thematic subjects).
18.104.22.168 Problematic aspects of clause-internal analyses
The main problems affecting all clause internal analyses of RD, as far Italian is concerned, have already been discussed in Section 4.3 and need not be considered here. They included the observation that overt clitic doubling within a clause is impossible in Italian (Section 4.3.1), whereas clause-internal analyses must assume that RD is an exception to this generalization, and the observation that right-dislocated phrases always follow marginalized constituents (Section 4.3.2), whereas clause-internal analyses incorrectly predict the opposite order because the dislocated items dominate VP and hence also any item marginalized in situ within VP.
A problematic aspect worth further discussion concerns the licensing of negative phrases and NPIs. In clause-internal analyses, right-dislocated constituents are c‑commanded by T and hence remain within the licensing domain of a sentential neg-marker in T (or right above T, depending on where neg-markers are assumed to be located). Consequently, right-dislocated negative phrases and NPIs should be licensed, but as we saw in Section 4.3.3 this is not the case.
The clause external analysis proposed in this book correctly predicts this licensing failure, because right-dislocated phrases are not c-commanded by the licensing (p.134) neg-marker. In contrast, there does not seem to be any plausible way of modifying a clause-internal analysis so as to derive the same observation.
Consider for example the licensing of the negative adverb mai ‘ever’, which, as (136) shows, requires obligatory licensing by a c-commanding neg-marker when occurring in post-auxiliary position.
As expected, mai can be marginalized after a focused past participle with no effects on its licensing, see (137)(a). It cannot, however, be right-dislocated, since in this case it is no longer c‑commanded by the licensing neg-marker, see (137)(b). Since clitic doubling is unavailable for adverbs, the dislocated adverb follows a clitic-doubled dislocated object, which ensures its right-dislocated status.
Clause-internal analyses incorrectly predict (137)(b) to be grammatical, since the adverb remains c-commanded by the higher neg-marker. Indeed, the availability of this licensing relation in Catalan is offered as evidence for a clause-internal analysis of Catalan right dislocation in Villalba (2000). By the same logic, however, a similar analysis is not valid for Italian, where licensing fails.
There is also no obvious way to rescue clause-internal analyses by identifying other factors as responsible for the licensing failure. For example, we could consider whether licensing fails to extend to items in specifier positions, including the dislocated negative adverb’s one in (137)(b). But according to Cinque (1999: 44), the adverb is in a specifier position also in (137)(a), where it is successfully licensed.
Similarly, stipulating a ban on adverbial right dislocation, however worded, would not account for the observed NPI‑licensing failure with non-adverbs. It would also run counter to fact: for example, the adverb sempre ‘always’, which according to Cinque (1999: 9) is the positive counterpart of mai, can be right-dislocated and, like mai above, can occur after a dislocated object, see (138).
22.214.171.124.2 Interaction with clause-wide focus
Another problematic aspect of clause-internal analyses concerns the interaction of focus and right dislocation. In Cecchetto (1999: 58), the focus projection posited immediately above the topic projection responsible for right dislocation is used to represent sentences where a single focused constituent precedes right dislocation, as in example (139) by Cecchetto.
The same analysis, however, cannot account for sentences where focus encompasses the entire TP while still allowing for right dislocation of a specific constituent. Consider, for example, the derivation of (140a), where focus is TP-wide, except for the object Gianni which is discourse-given and right-dislocated.
Clause-internal analyses à la Cecchetto would have to use the base-generated structure in (141), with its Focus and Topic projections, and dislocate the object Gianni to the specifier of TopicP. The problem is that nothing forces the constituents following Gianni to raise to the left of the dislocated object, as they are part of the focused TP but they are not independent foci.
(p.136) Sentence (140a) is instead straightforwardly derived under the proposed clause-external analysis. As shown in (142), first Gianni is raised to specRP, then the entire TP is raised to specXP, thus preceding the right-dislocated object as required. As usual, stress falls on the rightmost item of the focused constituent, i.e. Roma. The final structure is in (143).
126.96.36.199.3 Reconstructions effects
Cecchetto (1999) argues against the presence of adjunct/argument binding asymmetries of the kind discussed in Section 4.3.4, claiming that binding into adjunct modifiers of dislocated complements is ungrammatical. This is an important observation, which if correct would provide support for a clause-internal representation of RD, since even late-inserted adjuncts would have to be c-commanded by a subject in specTP.
Cecchetto’s observations, however, are limited to sentence (144), here repeated in its original format. A more appropriate test would consider more sentences, and, crucially, compare the adjunct cases against their argument counterparts, such as the one provided in (145). This is essential because the complexity of these sentences makes them rather unnatural and it is hard to distinguish grammaticality from pragmatic felicity when assessing them in isolation. Only by examining minimal pairs can we factor out pragmatic felicity and assess the presence of a grammaticality contrast.
Personally, I find (144) acceptable, and its argument counterpart in (145) strongly ungrammatical. As reported in Samek-Lodovici (2006), the same contrast was observed by all of the 18 informants consulted on similar minimal pairs, including informants who like Cecchetto did not find the adjunct case acceptable (the individual assessments are listed in Appendix A of Samek-Lodovici 2006). This sharp grammaticality contrast is expected under a clause-external analysis, as explained in Section 4.3.4, whereas it is incorrectly predicted absent by clause-internal accounts of RD.
It is also worth mentioning that two important factors might have interfered with the assessment of (144) reported in Cecchetto (1999). First, as it stands the sentence is pragmatically implausible because the singular definite object l’annuncio ‘the announcement’ appears to suggest that politicians make just one press announcement during their careers. Using a plural object and a plural bound referent makes the sentence more natural and acceptable, see (146).
Second, the sentence was reported without indicating the position of main stress. This can easily lead to assessing it under intonational contours that do not correspond to the desired right dislocation structure.
188.8.131.52 Cecchetto’s arguments against clause-external analyses
Cecchetto (1999) offers an insightful discussion of five issues considered problematic for clause-external analyses of RD, including Kayne’s double topicalization analysis which is structurally identical to the analysis proposed here but for the claim that RD and CLLD share the same specifier position.23 While Cecchetto’s issues are indeed problematic for some (p.138) clause-external analyses, they actually do not affect the remnant movement analysis proposed here. The first issue, concerning the reconstruction properties of dislocated adjuncts, was already examined in Sections 4.3.4 and 184.108.40.206.3, where I showed how these properties actually support the proposed analysis of RD. Two other issues, concerning ECP effects and Aux-to-Comp constructions, are irrelevant because, as Cecchetto acknowledges, they are not problematic for analyses exploiting Kayne’s remnant movement structure (Cecchetto 1999: 51,53). The two remaining issues concern right-roof effects and proper binding, each discussed in detail below.
220.127.116.11.1 Right-roof effects
Cecchetto correctly observes that Kayne’s remnant movement analysis predicts long-distance RD to be possible. He is, however, incorrect in maintaining that the prediction is not borne out. As the several examples in Sections 4.3.5, 4.4.5, and 18.104.22.168 show, dislocation to a higher clause from a non-finite complement is amply possible.
Furthermore, the ungrammatical sentence proposed as evidence against long-distance RD extraction in Cecchetto (1999: 52), repeated in (147), is inadequate for two reasons. First, right dislocation occurs from a tensed complement, which as we saw in Sections 4.4.5 and 22.214.171.124, blocks extraction for reasons unrelated to RD, since extraction from untensed complements remains possible. Second, (147) actually involves right dislocation from a CLLD phrase, namely the initial clause che gliela presti.24
Consequently, right dislocation here takes place from a tensed CLLD clause. Since neither right dislocation to higher clauses from tensed complements (Section 4.4.5) nor extraction from CLLD (Section 4.4.4) is possible, the ungrammaticality of (147) is expected, and therefore uninformative with respect to the properties of long distance RD.
126.96.36.199.2 Proper binding
Cecchetto (1999) also argues that a remnant movement analysis à la Kayne fails proper binding when assessed relative to its surface structure since right-dislocated phrases no longer c-command their traces. The same criticism applies to the analysis proposed in this book. See sentence (148) and its structure in (149) where Gianni does not c-command its trace. (Remember that the clitic lo responsible for clitic doubling is not a resumptive pronoun; the trace ‘ti’ is still part of the structure.) (p.139)
The same c-command relations, however, are also inevitably present under clause-internal analyses à la Cecchetto and therefore cannot be appealed to as an argument distinguishing them from clause-external alternatives. Consider for example sentence (150), which is very natural when uttered under an appropriate intonation contour invoving prosodic downstep after each comma. It involves three right-dislocated phrases: the sentential complement di riuscire a trovarlo, doubled by the clitic ne in the matrix; the temporal adjunct quando torneremo, which must be right-dislocated since it follows the sentential complement; and the final DP un idraulico, extracted from the sentential complement, which is clitic doubled by the clitic lo. Under Cecchetto’s analysis these three phrases must have moved to the specifier of three corresponding topic projections above the adjectival phrase for SICURI (which eventually must raise to the higher FocusP). As the corresponding structure in (151) shows, the traces of un idraulico and the CP di riuscire a trovarlo are inevitably not c-commanded by their antecedents.
Indeed, the order of the dislocated constituents in the example guarantees a failure of proper binding under any conceivable anti-symmetric representation, independently (p.140) of the clause-external or clause-internal position assigned to right dislocation. Proper binding of phrasal traces does not distinguish between these competing representations, both of which must assume that proper binding is satisfied at the time of extraction.
188.8.131.52 Other potential issues from Villalba (2000) and López (2009)
Like Cecchetto, Villalba and López discuss a series of concerns which at first might appear to also apply to the clause-external analysis proposed here. This section discusses those not already examined or implicitly addressed in the previous sections.
184.108.40.206.1 Condition C.
López (2009: 90) maintains that Spanish right-dislocated objects are in an A-position and bind lower constituents. His analysis accounts for the alternation in (152) where an object containing a referential expression bound by a preceding indirect object violates condition C when in situ, as in (152)(a), but not when right-dislocated, as in (152)(b). Under López’s analysis, the right-dislocated phrase is no longer c-commanded by the indirect object and its A-status blocks reconstruction into its base-generated position.
At first, the corresponding Italian data appear to confirm López’s analysis. The pattern in (153) matches that in (152). Violation of condition C even in the right-dislocated case, however, becomes sharply evident as soon as we ensure that the bound referent is an argument of the dislocated noun, as in (154). (Stress on the pronoun LEI naturally induces a contrastively focused interpretation, hence the ‘F’ marking. The listed judgements hold even when imposing a presentational focus interpretation on the TP containing LEI.)
(p.141) The difference between (153)(b) and (154) concerns the adjunct/argument asymmetry already examined at length in Section 4.3.4. Sentence (153)(b) is marginally acceptable because the phrase di Anna Tusquets can be analysed as a possessive late-inserted adjunct not subject to reconstruction. The same is not possible in (154), where the argument of the dislocated noun necessarily reconstructs. A careful testing of condition C thus confirms the presence of reconstruction, consistently with the analysis of RD advocated in this chapter.
220.127.116.11.2 Relativized minimality effects.
In his analysis of RD in terms of A-movement, López (2009: 91) also claims that a right-dislocated indirect object blocks subject raising in raising verb constructions. The corresponding Italian sentence in (155) is marginal.
If this analysis were correct, RD should also block subject raising when affecting the indirect object of the matrix clause. This is not the case; the corresponding sentence (156) is clearly grammatical.
Furthermore, sentence (155) is ungrammatical even when right dislocation is absent, provided a dative clitic remains present in the lower clause.
It follows that the ungrammaticality of (155) is determined by the as yet still poorly understood constraints governing the distribution of dative clitics. The same constraints are bound to affect right dislocation whenever it involves dative clitic doubling in complement clauses, irrespective of its clause-external or clause-internal nature. Consequently, sentence (155) cannot be considered as evidence for a clause-internal analysis of RD.
18.104.22.168.3 Pronominal binding by quantified phrases.
Villalba and López also examine various cases of pronominal binding. Let me consider them in turn.
Preverbal subjects binding into RD—According to Villalba (2000: 191), Catalan preverbal quantified subjects can bind pronouns in RD phrases but not in CLLD (p.142) ones. This is shown by the following examples provided under Villalba’s original format. The subject ningú can bind the null subject pro in the object starting with totes when the object is right-dislocated, as in (158)(a), but not when the same object occurs as a CLLD phrase in (158)(b).
According to Villalba, if RD is located above TP, both sentences will be ungrammatical because ningù would not c-command the null subject in both. This conclusion, however, does not consider reconstruction. Since Italian RD reconstructs, the grammaticality of (158)(a) is uninformative because the subject will c-command and bind the reconstructed object independently of whether RD is TP-internal or TP-external.25
The corresponding Italian data differ from the Catalan ones and are consistent with the analysis of RD provided so far. Quantified subjects can bind pronouns in both RD and CLLD phrases. Since RD and CLLD both allow for reconstruction, this is expected.
Postverbal subjects binding into RD—López (2009: 92) claims that in Spanish postverbal subjects in spec-vP can bind into a following in-situ object but not into a right-dislocated one. According to López, this shows that RD involves A-movement, since A'-movement would enable reconstruction and make binding into right-dislocated objects grammatical.
The corresponding Italian data differ sharply. Postverbal subjects can bind into right-dislocated objects, see the following examples. This would be unexpected if RD involved the same unreconstructable A‑movement claimed for Spanish, whereas it (p.143) follows straightforwardly if right-dislocated objects are A'-moved and able to reconstruct into their base position.
The reconstruction of right-dislocated phrases is also visible under anaphoric binding, with postverbal subjects successfully binding right-dislocated anaphoric objects.26
(p.144) Right-dislocated phrases binding postverbal subjects—López (2009: 92) also considers data from Catalan where right-dislocated indirect objects bind a preceding postverbal subject stranded within VP, as in the corresponding Italian sentence in (163), which is grammatical both with or without clitic doubling.
When the indirect object is clitic-doubled, as in López’s original examples, binding is expected independently from the position and A vs. A'-status of the dislocated indirect object, since the corresponding clitic in T c-commands the postverbal subject inside VP.
The clitic-less case, too, follows from the reconstruction properties of RD and therefore does not support right dislocation to a spec-vP A-position. As the following data show, quantified objects and indirect objects involving ogni (every) may precede and bind postverbal subjects even if the latter are neither right-dislocated nor focused. The dislocated quantified indirect object in (163) may thus reconstruct into this position and bind the postverbal subject from there.
As we know from Chapter 2, the same position is not available to non-quantified phrases. As the following examples show, it is also unavailable to negatively quantified constituents. This, too, is not surprising, as the higher scope attained by universal quantifiers relative to negative ones is well documented; see the survey in Szabolcsi (2001), Beghelli (1993), and Beghelli and Stowell (1997).
Right-dislocated phrases binding into CLLD and vice versa—López’s (2009: 91) evidence for a clause-internal analysis also includes Catalan data in Villalba (2000) showing CLLD phrases binding into right-dislocated phrases, but not vice versa.
(p.145) This asymmetry is actually problematic for López's analysis, which maintains that both CLLD and RD phrases A‑move to spec-vP with CLLD phrases moving further to specFinP via A'‑movement (López 2009: 113). Since right-dislocated phrases are freely ordered, nothing prevents the intermediate spec-vP position of a CLLD phrase from occurring immediately lower than the final spec-vP position of a right-dislocated phrase. Since right-dislocated phrases have A‑status in López’s analysis, they should be able to bind CLLD phrases when they reconstruct into spec-vP, incorrectly predicting RD binding into CLLD to be grammatical. A similar problem applies to Villalba’s analysis, where CLLD moves through the same clause-internal topic position targeted by RD (Villaba 2000: 275).
Italian diverges from Catalan in this respect, as it allows for quantifier binding from CLLD into RD and vice versa, as shown in (167)(a)–(b).
This is exactly what is expected under the remnant movement analysis advocated here, independently of whether CLLD is movement-based or base-generated. Under a movement analysis of CLLD, binding occurs under reconstruction, since the quantified indirect object can bind the remaining argument when both occur in situ, see (168).
If CLLD is base-generated, pronominal binding remains possible provided quantifier binding from A'-positions is allowed. Binding may then occur because right-dislocated phrases c-command base-generated CLLD-phrases before the final remnant movement, as illustrated by step 2 of derivation (169) for sentence (167)(b).
(p.146) In summary, the structural inferences that Villalba and López draw from their Spanish and Catalan data do not carry over to Italian nor undermine the analysis of RD being proposed. Unlike Spanish, Italian postverbal subjects can bind right-dislocated objects, as expected if the latter reconstruct into their original position. Unlike Catalan, Italian preverbal subjects and right-dislocated quantified phrases can bind pronouns in CLLD phrases, again consistently with the availability of reconstruction for CLLD phrases. Finally, as in Catalan, Italian right-dislocated indirect objects involving universal quantifiers like ogni ‘every/each’ may bind postverbal subjects, consistently with their reconstruction into the position above postverbal subjects that is available to them.
22.214.171.124.4 Interactions between CLLD and RD
In their argument for a clause-internal analysis of RD, Villalba (2000) and López (2009) also consider the relation between CLLD and RD. Some of the proposed arguments are not valid, while others apply to Catalan but not to Italian, as explained below.
Extraction of RD from tensed CLLD clauses—Villalba (2000: 211) claims that the ungrammaticality of the following Catalan sentence is unexpected under a remnant movement analysis of RD like the one proposed in this book The sentence right dislocates the object el llibre ‘the book’ from the CLLD complement que se l’havia comprat ‘that she had bought it’. It is, therefore, fully analogous to Cecchetto’s right-roof case discussed in 126.96.36.199.1 and as explained there its ungrammaticality follows from the unavailability of extraction from CLLD islands and the additional unavailability of right dislocation from tensed complements.
Villalba argues that remnant movement incorrectly makes the sentence grammatical through the derivation in (171). But the proposed derivation violates conditions on extraction and is therefore ungrammatical. At step 2, it right-dislocates the object el llivre across both the subordinate and the matrix clause to the specifier of the higher projection Top1, thus violating the ban against right dislocation from tensed complements discussed in Sections 4.4.5 and 188.8.131.52. Furthermore, at step 4 it extracts the complement que se l’havia comprat from the specifier hosting the remnant TP m’ho va dirs, thus involving extraction from an unselected specifier, i.e. across an island.27 The final structure is shown in (172). (p.147)
Extraction of RD from CLLD prepositional phrases—Building on Villalba, López (2009: 91) considers additional data involving non-clausal CLLD phrases where tense is not an issue and claims that Catalan disallows right dislocation from CLLD phrases.
In this respect, Italian diverges from Catalan. RD is clearly possible provided the right-dislocated constituent is not an argument of the noun in the CLLD phrase; see (173) and (174), which provide the Italian counterparts to the corresponding construction in López (2009: 91, ex.3.13). When argument-hood is involved, as in (175), grammaticality is slightly marginal, but it significantly improves if the CLLD constituent is followed by a short pause.
The availability of RD extraction from CLLD phrases is consistent with the analysis of RD proposed in this book provided we assume a movement analysis of CLLD, since a base-generated CLLD phrase would constitute an island to extraction. The RD phrase can then be extracted before CLLD, and hence before the creation of the CLLD island. A derivation consistent with Chomsky’s (1995: 248) root condition is (p.148) provided in (176). First, the RD phrase evacuates the larger PP targeted by CLLD (step 2), then CLLD occurs (step 3), then the RD phrase is raised (step 4), and finally the entire clause is remnant-moved completing the right dislocation operation.28 See (177) for the final structure.
Extraction of CLLD from RD prepositional phrases—López (2009: 91) also considers the opposite operation: the extraction of CLLD phrases from right-dislocated constituents. These constructions are grammatical in Catalan and the same is true for Italian, see the data in the following examples. Once again, when the extracted CLLD phrase is an argument of the dislocated noun, grammaticality is marginal but it improves if the CLLD phrase is followed by a brief pause.
The grammaticality of these sentences follows straightforwardly under the proposed analysis of RD. If CLLD is base-generated, the CLLD phrase simply binds the relevant empty category within the right-dislocated constituent prior to its right dislocation. If CLLD is movement-based, then CLLD occurs before RD takes place as in the derivation in (181). I assume that the preposition di on the right-dislocated phrase is inserted after right dislocation has taken place in order to satisfy the case filter.
In summary, Villalba and López’s arguments for a clause-internal analysis do not apply to Italian, which does not show the same asymmetry found in Catalan. Extraction of RD from CLLD and vice versa is possible, consistent with the analysis being proposed.29
Clause-external analyses maintainining that RD is base-generated cannot account for the presence of movement and the related evidence examined in Section 4.4. These analyses include De Cat (2007) where CLLD and RD are respectively left and right adjoined to a maximal projection above T, Frascarelli (2004) where RD and CLLD share the same position but RD involves TP‑inversion as explained later in this section, and Cardinaletti (2002) where RD is generated as the complement of a projection that contains the non-dislocated part of the sentence in its specifier, as illustrated in (182) and the corresponding structure (183).
With respect to Cardinaletti’s analysis, López (2009: 261) and Samek-Lodovici (2006: 859) also point out that it incorrectly predicts the availability of wh‑extraction from clitic-doubled right-dislocated phrases, since the latter occur in a complement and head-governed position. López (2009: 259) also remarks that this analysis fails to account for the binding relations available between distinct right-dislocated phrases, since multiple right-dislocated phrases would neither c‑command each other nor be able to reconstruct back into the main TP as they are not generated there. (p.151)
Vallduví (1992: 103) and Zubizarreta (1994a) provide a clause-external analysis similar to De Cat (2007) but involving movement. Under Vallduví’s, CLLD and RD left- and right-adjoin to IP, while in Zubizarreta’s CLLD left-adjoins to TP and RD right-adjoins to CP. As pointed out in Cecchetto (1999), mirror analyses of this kind where CLLD and RD occur as opposite left and right adjuncts incorrectly predict that CLLD and RD phrases will share the same syntactic properties, thus failing to account for the significant discrepancies between them. Although not all of Cecchetto’s described discrepancies apply—Sections 4.3.4 and 184.108.40.206.3 showed that reconstruction behaves similarly in both constructions and Sections 4.3.5 and 220.127.116.11.1 showed that RD is not clause-bound—Cecchetto’s observation remains valid for all the discrepancies discussed in Section 4.4, such as the differences in NE-cliticization, the effect of tenseness on extraction, the obligatoriness of clitic doubling in CLLD and its absence in RD–, the impossibility of wh-extraction from CLLD phrases and its availability under RD–. Also problematic for mirror analyses is Villalba’s (2000: 188) observation that CLLD can create island effects that are absent under RD.30 Furthermore, analysing RD in terms of right-adjunction predicts an incorrect distribution of clitic doubling under left-peripheral focus (Samek-Lodovici 2009: 351).
As Cecchetto (1999) and Villalba (2000) point out, similar difficulties affect the Kaynian double-topicalization structure described in Cecchetto (1999) and adopted here when assuming that RD and CLLD phrases share the same position as maintained in Samek-Lodovici (2006). The problem is not the structure proposed for RD, which is necessary to account for its properties as argued in Sections 4.3 and 4.4 and is not affected by Cecchetto’s critical remarks as already explained in Section 18.104.22.168, but rather the claim that CLLD corresponds to the same structure prior to the final remnant movement. Under this hypothesis, CLLD and RD share the same specifier position, incorrectly predicting a uniform set of syntactic properties across the two.
Building on Cinque (1990), Frascarelli (2004) views CLLD31 as base-generated in a left-peripheral topic projection, see (184) and the corresponding structure (185). Right-dislocated phrases are claimed to generate in the exact same position and to occur clause-rightmost due to the inversion of the lower TP (or ‘IP’ in the original analysis), which moves to a higher functional projection ‘GP’ at the top of the structure; see (186) and the corresponding structure (187). A similar analysis is also proposed in Frascarelli and Hinterhölzl (2007) for RD+, see their discussion of right-dislocated topics generated in the projection FamP located above TP.
As already pointed out, the analysis cannot account for the movement nature of right dislocation discussed in Section 4.4, or for the divergences between RD and CLLD. López (2009: 229) also notes that the anti-reconstruction data supporting a base-generated analysis are ambiguous between a CLLD and a hanging topic representation, and hence inconclusive, while De Cat (2007) discusses some important interpretational differences between CLLD and RD that cannot be derived if CLLD and RD share the same specifier.
(p.153) Here, I wish to further examine two specific observations concerning binding and relativized minimality proposed as evidence for the above analysis that at first might appear to raise an issue for the alternative movement-based analysis proposed in this book.
The first observation concerns the alleged divergence in the reconstruction properties of CLLD and RD. According to Frascarelli (2004), the subject Leo of the CLLD topic in (188)(a) can corefer with the pro subject of the main clause, whereas the same is not true when the same topic is right-dislocated, as in (189)(a) (I find both sentences grammatical, hence the parentheses around the star in the second example). Under Frascarelli’s analysis, (188) is grammatical because the base-generated CLLD prevents Leo from c-commanding the following pro. In contrast, in (189), the matrix pro is claimed to c-command Leo, and thus violates condition C, because under a strict antisymmetric analysis à la Kayne (1994: 23) the highest specifier of the specifier of a projection XP c‑commands into XP, and therefore pro in (189) c-commands Leo.
A first problem arises with respect to the adjunct/argument asymmetry. As discussed in Section 4.3.4, testing across a larger set of informants shows that sentences involving late-inserted adjuncts are more acceptable than corresponding sentences involving arguments. The same contrast applies in this case: sentence (189)(a) is significantly more acceptable than the corresponding sentence involving an argument in (190). This systematic contrast is predicted to be absent by Frascarelli’s analysis, because the binding relation making (189) ungrammatical should hold independently of the adjunct/argument nature of the phrase containing the bound noun.32 (p.154)
A second potential problem emerges from the analysis’ reliance on Kayne’s (1994) definition of c‑command where, as mentioned, the highest specifier of the specifier of a projection c‑commands into that projection. Under this definition, the claimed contrast between CLLD and RD in (188) and (189) should disappear whenever the binding pro subject is not placed highest in the dislocated TP. For example, sentence (191), where pro follows the adverb probabilmente (probably), should contrast with (189) because pro no longers c‑commands Leo. Personally, I cannot find such a contrast. Yet its presence is a distinctive prediction of the analysis under discussion.
Frascarelli (2004) also claims that CLLD and RD constituents are always generated in the left peripheral topic projection of their own clause, and, consequently, that CLLD and RD constituents in higher clauses must have moved there. This allows Frascarelli to explain the ungrammaticality of a sentence like (192) in terms of relativized minimality. Due to the crossing movements in structure (193), the raised lower topic Maria becomes the closest c‑commanding topic for the trace of the higher topic a Gianni, making the structure ungrammatical. Note that the analysis crucially relies on the assumption that CLLD phrases are base-generated before any movement takes place, since otherwise the topic a Gianni could be base-generated above the raised Maria and avoid the violation at issue; this assumption will play a critical role below. (p.155)
The first issue concerns the grammatical status of these sentences. As shown by examples (194) and (195), they improve significantly when the CLLD topics are not both proper names, suggesting that processing complexity rather than relativized minimality might be the cause for the unacceptability of (192).
Furthermore, the relativized minimality analysis becomes problematic when extended to RD. It incorrectly predicts that very natural and grammatical sentences like (196) are ungrammatical.
To see this, consider the possible derivations available under the above analysis. Following Frascarelli’s assumptions, the initial structure for the sentence in (196) must be as in (197), with il vino generated as the CLLD topic of the lower complement, while a Marco is generated as the CLLD topic of the matrix clause. To get to sentence (196) we must raise the lower topic il vino and right-dislocate the higher topic a Marco. (p.156)
If we raise the lower topic il vino and then right-dislocate the higher one a Marco, we obtain the derivation in (198). First, il vino raises from the embedded clause to the front of the entire structure since it eventually precedes the matrix verb. Then, a Marco moves above il vino in order to enable the final step where a Marco is right-dislocated by raising the rest of the clause to a higher specifier. The final structure in (199), however, leaves the raised topic il vino as the closest c‑commanding topic for the trace ‘tk’ of a Marco, hence violating relativized minimality and incorrectly predicting the sentence to be ungrammatical. Nor is it possible to avoid the crossing movements of steps 2 and 3 by raising il vino first and then base-generate a Marco above it, since as we saw above in the discussion of (192) CLLD phrases must be generated before any movement takes places.
Executing right-dislocation before CLLD movement, is equally unsuccessful. This derivation starts again with structure (197), then right-dislocates the topic a Marco by moving the entire matrix TP to specGP as in step 1 of (200). We must now raise il (p.157) vino above the matrix verb but this is not possible. Movement to a top TopP projection, shown in step 2 of (200), involves extraction from an unselected specifier and is therefore unavailable. Raising il vino only within its TP violates Chomsky’s root extension condition (1995: 248) because it does not target the root.
In conclusion, whichever the order in which right dislocation and CLLD are applied, a derivation consistent with the grammaticality of (196) is not possible, showing that Frascarelli’s base-generated analysis cannot be combined with relativized minimality to provide a successful account of the word order effects highlighted in Frascarelli (2004) and hence it is not supported by their existence.
4.6 Crosslinguistic variation
As the previous sections have shown, recent studies of RD across different languages show that the position of right dislocation, its moved vs. base-generated nature, and, if moved, even the A'- vs. A-movement status of the movement involved might vary from language to language. Here, I briefly consider some additional differences between Italian, Catalan, and French concerning the final position of right-dislocated constituents and the presence of movement.33
4.6.1 Variation in position
The position of right-dislocated phrases appears to vary, occurring clause-externally in Italian and French (for Italian see Section 4.3 and also Cardinaletti 2001, 2002; Frascarelli 2000, 2004; Samek-Lodovici 2006. For French see De Cat 2007), but lower than T in Catalan (Villalba 2000; López 2009; Feldhausen 2008, 2010).
Villalba (2000: 190) and Feldhausen (2008: 148), for example, note that Catalan allows for the licensing of right-dislocated NPIs. Some of Villalba’s examples follow, with the NPIs licensed by the neg-marker no shown in bold. These examples show that Catalan right-dislocated phrases unlike Italian ones are c‑commanded by the preceding neg-marker and thus must occur clause-internally in a position lower than T (gloss and translation have been slightly modified). (p.158)
Villalba’s examples contrast sharply with their Italian direct counterparts, which are strongly ungrammatical, consistently with the evidence discussed in Section 4.3.3 (the third Catalan example cannot be replicated because Italian lacks the counterpart for gaire).
Both Villalba (2000) and Feldhausen (2008) also observe that Catalan CLLD phrases show anti-reconstruction effects. Feldhausen (2008: 150) applies to Catalan the same tests of condition C for dislocated arguments and adjuncts examined in Section 4.3.4 finding that in this language they support a clause-internal analysis of right dislocation. Villalba (2000: 190) compares sentence (203)(a), which is grammatical despite the fact that reconstruction of the initial DP would violate condition C, with the ungrammatical (203)(b), where condition C is expected and indeed appears to be violated as predicted by the low position attributed to right-dislocated phrases in this language. The corresponding Italian sentences in (204) are instead both grammatical, confirming the clause-external position of RD advocated in this book.
(p.159) Further evidence for the low position of Catalan RD is provided in López (2009) with respect to the adjunct/argument binding asymmetry discussed in Section 4.3.4. The same asymmetry is absent in Catalan, where both arguments and adjuncts violate condition C. This follows from the low position of right-dislocated phrases, since in this case even late-inserted adjuncts remain c-commanded by a binder in specTP.34 An example is provided in (205), repeated from López (2009) with the original format and gloss.
López tested his results with four Catalan linguists through an email questionnaire consisting of 24 sentences and including wh- and CLLD-constructions, as well as distracters. Unfortunately, the original sentences of the emailed questionnaire and the related judgements are not included in López (2009), preventing an accurate assessment of the proposed evidence. In particular, two aspects of the reported data are slightly problematic. First, the two sentences reported in López (2009), listed in (205), do not mark the position of stress. If stress was also left unmarked in the original questionnaire sentences, the grammaticality assessments might have been provided under a different intonation than the one intended, potentially affecting the reliability of the test.35 Second, the test’s reliability might have been adversely affected (p.160) by the distinct grammatical roles assigned to the referent Joan across the two sentences, which prevent them from forming a minimal pair. These potential problems notwithstanding, López’s evidence converges with Villalba’s observations in supporting a low position for Catalan RD and hence also for the presence of genuine variation in the analysis of RD across Catalan and Italian.
4.6.2 Variation with respect to movement
Italian and Catalan right-dislocated phrases are the product of movement. The evidence supporting this conclusion for Italian was examined in Section 4.4. The same conclusion is reached for Catalan by Villalba (2000) on the basis of connectedness and island sensitivity effects, and by López (2009: 215) on the basis of the empirical properties distinguishing RD from base-generated hanging topics.
According to De Cat (2007), the same does not hold in French, where right-dislocated phrases show properties typical of base-generated phrases. While some of the proposed evidence is open to interpretation—e.g. the non-licensing of parasitic gaps, which Villalba (2000) attributes to the non-quantificational nature of RD—De Cat’s data show a clear contrast with their Italian counterparts, suggesting a genuine difference between the two grammars.
In particular, French RD shows no reconstruction effects and it is insensitive to strong islands (De Cat 2007). The first property is illustrated by the following French sentence, where according to De Cat the pronoun ses ‘his’ cannot be interpreted in the scope of the quantified subject, thus disallowing an interpretation where each master dismisses one of his own disciples. As (207) shows, reconstruction is instead possible in Italian, where a focused quantified subject may bind the dislocated pronoun. The same holds with unfocused preverbal quantified subjects, provided an appropriate context is supplied as is the case in (208). If right-dislocated phrases were base-generated higher than (Q-raised) preverbal subjects, as claimed by De Cat for French, then (208)(a) should be infelicitous under the suggested distributive interpretation.
The insensitivity of French RD to strong islands is illustrated in (209), where sa fille ‘his daughter’ follows the matrix-related dislocated pronoun moi ‘me’, showing that dislocation across complex-NP islands is possible. The same does not hold in Italian, as shown by the corresponding sentence in (210). Further examples of island sensitivity for Italian RD are provided in section 22.214.171.124.3.
In so far as the above distinctions will hold up to further investigations, they suggest that the base-generated vs. moved nature of RD is a parametric property.
The variability of empirical tests used for RD across distinct studies makes it difficult to compare like with like and reach an accurate picture of its crosslinguistic properties. A rigorous crosslinguistic comparative study that also distinguishes RD– from RD+ is even harder to achieve because the great majority of published RD analyses do not consider marginalization and consequently do not disambiguate between marginalization and RD–, thus potentially assigning to RD– the properties of marginalization and vice versa. Nevertheless, the existence of clear alternations with respect to specific properties like those illustrated for Italian, French, and Catalan in the previous sections strongly suggests that the position of right dislocation and its movement- vs. base-generated nature vary across distinct languages.
The syntactic properties of right-dislocated phrases differ from language to language. As far as Italian is concerned, RD may occur with or without clitic doubling, and constituents targeted by right dislocation are located above TP and reach this position via movement. The analysis proposed here locates right-dislocated items in the specifier of an RP projection located above the TP that originally hosted them, which in turn remnant moves to the specifier of an additional projection XP taking RP as complement. Under this analysis, the items within the remnant moved TP and the right-dislocated items do not c-command each other. This analysis successfully accounts for the many properties examined in this chapter, including NE-cliticization, right-roof violations, wh-extraction from dislocated constituents, reconstruction, and the differences between right dislocation and CLLD.
(p.162) An important consequence of these results concerns the status of unstressed phrases following focus when clitic doubling is absent. Since right dislocation without clitic doubling is possible, these phrases are ambiguous between a right dislocation and a marginalization analysis and care should be taken to determine their status through the empirical tests discussed in this and in the previous chapters. Such disambiguation is essential, because as this chapter has shown the structural representation and properties of marginalized and right-dislocated phrases are radically different.
Only once marginalization and right dislocation are clearly distinguished, does it become possible to determine the syntactic position of focus relative to discourse-given phrases following at its right, which is necessary if we wish to reach an appropriate understanding of its distribution and structural properties. We already considered the relation between focus and marginalized phrases in Chapter 3, showing that when right dislocation is absent focus occurs in situ. As the next chapter will show, the same holds for focused constituents followed by independent right-dislocated phrases not overlapping with focus. When focus is generated within a larger constituent targeted by right dislocation, however, focus moves leftwards in order to evacuate the hosting constituent and enable its right dislocation. This interesting interaction, which is responsible for the occurrence of left-peripheral foci, is addressed in the second part of the next chapter.
(1) Only for clitic-doubled RD. When clitic doubling is absent, Zubizarreta (1994a) claims that the affected phrases right-adjoin to an aspectual projection between T and VP. The evidence provided for this latter case, however, does not distinguish between marginalized and genuinely dislocated phrases.
(2) This analysis was presented by Kayne in his 1995 Harvard lectures. The analysis is described in Cecchetto (1999) and there is no other reference for it. It crucially differs from the analysis in Kayne (1994) in that it involves movement of the right-dislocating constituents to the position of CLLD constituents, followed by remnant movement of the remaining of the clause.
(3) Only for clitic-doubled RD. When clitic doubling is absent, Frascarelli and Hinterhölzl maintain that the dislocated constituents move to a position above TP followed by TP inversion (their analysis uses IP instead than TP). The proposed reconstruction-related evidence does not control for marginalization, which provides an alternative possible cause for the observed reconstruction cases.
(4) For some native speakers of Italian, sembrare is not a raising verb: they judge (i) as ungrammatical and replace it with the finite complement in (ii). These speakers inevitably assess example (h) and (i) in the main text as ungrammatical, but as we just saw the reason is unrelated to right dislocation.
(5) Additional cases of obligatory clitic-less right dislocation are examined in Samek-Lodovici (2010), which argues for the presence of right dislocation within DPs. Since DPs lack clitic-hosting functional heads, right dislocation is always clitic-less. The presence of right dislocation can nevertheless be detected through other diagnostics, such as scope. For example, in (i) where the entire sentence provides new information, the indefinite in bold is not dislocated and therefore still interpretable in the scope of the preposition ‘senza’ (without). Contrast this with (ii), which should be read as an answer to (i) with contrastive focus on the adjective immediato ‘immediate’. The indefinite in bold is now right-dislocated and consequently takes scope over ‘senza’, forcing the odd interpretation where the felling concerns a single specific tree, rather than any tree. The change in scope properties follows from the right dislocation of the indefinite, which places it in a DP‑external position outside the scope of ‘senza’ (for further discussion see Samek-Lodovici, 2010).
(6) Following a strict minimalist analysis, Abels (2003: 114) claims that TPs cannot move when occurring as the immediate complements of the phase head C. The issue is whether this theoretical claim remains valid given Rizzi’s (1997, 2004) decomposition of the CP projection in multiple sub-projections, with TP no longer the immediate complement of finite C. While this book questions the presence of a FocusP projection, Rizzi’s results about the distinct positions of finite complementizers, wh-phrases, and non-finite complementizers remain valid. Abels (2003) does allow for TP-movement when TP is not the immediate complement of phasal C.
(7) The following examples, slightly adapted from Benincà (1988: 137), show how indirect objects can be clitic-doubled when an object clitic or other complement-related clitic is also present. When this is not the case, even indirect objects cannot be clitic-doubled when located clause-internally; see (i)a and (ii)a.
(8) In colloquial Italian, optional clitic doubling of a focused subject might at first appear possible. The following utterance was attributed to a Fiat worker in Turin in an article of the daily newspaper Corriere della Sera (Roncone 2009). Examples like (ii), however, where the same second person clitic is present despite the presence of a first person subject and thus cannot be attributed to clitic doubling, show that the clitic is better analysed as an independently available benefactive clitic referring to the hearer.
(9) When the negative phrase is an object, rather than a subject, subject and object can occur in either order. This is unsurprising, because non-negative subjects are ambiguous between a marginalized and a right-dislocated reading and will occur before the object when marginalilzed and after the object when right-dislocated. See the following examples, where marginalized and right-dislocated status is respectively marked via the ‘M’ and ‘R’ subscripts.
(10) An alternative way to make the same point is to consider the parallelism between (i) and the equally ungrammatical left dislocation of a negative quantifier in (ii). Rizzi (1997) maintains that (ii) is ungrammatical because the negative quantifier, which is assumed to be generated in a position above TP, cannot build a well-formed operator-variable chain at LF. If it does not move, the quantifier will have no variable to bind; if it moves, the chain will be ill-formed because the variable will not be in an A-position (Rizzi 1997: 295).
Rizzi’s account, however, cannot extend to the sentences in (48)–(53) in the main text where the negative quantifiers are included in a larger clause. Its validity for sentences like (i) is also in doubt, since the account crucially relies on the base-generated status of nessuno in (i). As discussed in Section 4.4, right-dislocated phrases are instead extracted and can reconstruct in their base-generated position, which in turn makes a well-formed quantifier-variable chain possible.
Although the corresponding judgements confirmed the claimed asymmetry, this pair has been omitted from the results reported in the main text in view of comments by Kayne (p.c.) and van de Koot (p.c.), who remarked that when the dislocated noun is deverbal, as is the case with promesse ‘promises’, the argument/adjunct asymmetry might be determined by the null agent implicit in the noun, thus making the pair irrelevant for testing binding by the initial null subject. The same problem does not affect the two sentence pairs discussed in the main text, because the dislocated nouns prove and dimostrazione are non-agentive under the interpretation required in these sentences. Furthermore, the following argument/adjunct pairs, involving the nouns voci ‘rumours’ and accuse ‘allegations’, which disallow coreference with the initial pro subject under the intended interpretations, confirm the results discussed so far.
(12) In De Cat’s (2007) French examples, a right-dislocated item from the higher clause—rather than a focused one—intervenes between the lower clause and the item right-dislocated from the lower clause. Of the corresponding Italian translations, shown here, I find the first, involving dislocation from an untensed clause, grammatical, whereas the second, involving a tensed clause, is marginal.
(13) An anonymous reviewer wonders why the following sentence is ungrammatical when the infinitival adjunct per vedere il direttore is not right-dislocated (hence the lack of a comma after ROMA). The sentence is provided in its original form, which lacked focus subscripts. Presumably the stressed phrase a ROMA is focused.
We need to distinguish two cases, depending on the discourse status of the infinitival complement per vedere il direttore. If it is not discourse-given, then the cause of the ungrammaticality is the stress pattern. Italian main stress falls rightmost and should thus fall on the rightmost non-dislocated item, namely direttore. Indeed, sentence (i) becomes grammatical when stress falls on direttore, see (ii).
The second case occurs when the adjunct per vedere il direttore is discourse-given and marginalized in situ. Marginalized phrases always require the presence of a context sentence that must be duly read, possibly aloud, before grammaticality can be assessed. This is because unlike right-dislocated phrases, marginalized phrases do not seem to allow for accommodation: their discourse-givenness is not naturally inferred from their position and prosody. When such context is provided, sentence (i) is grammatical, see (iii).
The same reviewer also wonders whether the analysis proposed for sentence (iv) also predicts the grammaticality of sentence (v), which s/he finds ungrammatical (personally, I find (v) grammatical, hence the parentheses around the star symbol).
As shown in (vi), the two sentences are structurally identical except for the indirect object intervening between the verb and the focused subject in (v). The indirect object is likely to be the cause of the ungrammatical assessment, as some native speakers of Italian appear to disallow discourse-given material from occurring between a verb and a post-verbal focused subject, possibly due to an adjacency constraint on case-assignment.
Indeed, sentence (v) becomes grammatical again when the indirect object is expressed through a clitic and the other constituents are made prosodically less heavy. This is shown in (vii), which is grammatical and involves the same ditransitive verb and right dislocated object of (v).
Whatever the final analysis of (v), it is worth recalling that the sentences used to show that right dislocation can move constituent outside a clause is (iv), i.e. the type of sentence deemed grammatical by all native speakers, reviewer included. The ungrammaticality of (v) for some speakers is undoubtedly worth studying, but it does not undermine the point being made nor the structural analysis being provided.
(14) A similar, but not identical, argument applies with respect to De Cat’s examples of right roof violation. Consider sentence (1). The corresponding derivation under a clause-internal analysis of RD would have to raise the lower indirect object to the RD-position in the higher clause, as shown in step 2 of derivation (ii). Then the subject of the higher clause would have to dislocate to an RD-position preceding the previous dislocated item in order to match the final word order. Finally, the sentential adjunct would have to shift to an even higher position in the mid-field of the higher clause, again to match the final word order. This latter movement operation is the most questionable. It cannot be an instance of right dislocation because the moved phrase is both stressed and focused. Nor can it be movement to a clause-internal focus projection, since the CP does not express the entire focus of the sentence and clause-internal focus raising is unavailable as explained in Chapter 3.
In contrast, the clause-external analysis of right dislocation derives the sentence at issue with no need to stipulate any new operation, as shown in (iii). The indirect object dislocates first, followed by the subject, followed by remnant movement of the entire clause, yielding the correct order.
(15) As for why ne-cliticization is impossible under CLLD, Cinque (1990: 71) argues that any of the structures potentially involving it is excluded. Structure (i) is excluded because ne has no source. Structure (ii) is excluded because the trace ‘ti’ does not qualify as any of the legitimate nominal empty categories. It cannot be pro because it is not identified, nor PRO because it is governed, nor an NP trace because it is not A-bound in its governing category, nor a variable because the construction does not involve the movement of an operator (Cinque 1990: 73). Finally, extracting the quantifier on its own, as in (iii), is also not possible because the quantifier is not referential and cannot enter into a binding relation with its trace, nor a government chain due to the absence of a moved operator in CLLD constructions.
(16) Cinque’s test allows for further interesting results not directly related to RD. The first one concerns the syntactic status of preverbal subjects. As the following examples show, they do not allow for ne-cliticization even when supposedly extracted from object position, as in the following passive and unaccusative clauses. These data support the analysis of Italian preverbal subjects as base-generated CLLD topics, as argued—ignoring minor differences—in Frascarelli (2007) and Alexiadou and Anagnastopoulou (1998).
Furthermore, the observation that the wh-extracted subjects of passives and unaccusatives require ne-cliticization, as discussed in the main text, supports Rizzi’s (1986) claim that Italian subjects are wh-extracted directly from their VP-position rather than from their derived preverbal position. In the latter case, we would expect ne-cliticization to be absent on a par with the above data in (i)(a) and (ii)(a).
Under his analysis, events determine locality domains for wh-extraction. Wh-chains are subject to the Single Event Condition (SEC), which requires that the minimal constituent containing the entire chain asserts the existence of a single event in the actual world. In sentences involving adjuncts, like those in (i), the event variable of the matrix and that of the adjunct may form a single macro-event that satisfies the SEC, provided that the adjunct event variable is unbound and hence free to identify with the event of the matrix clause. Extraction is possible from the non-finite adjunct in (i)a but not from the tensed adjunct in (i)b because finite tense existentially quantifies over event variables (Higginbotham 1985). In the tensed adjunct (i)b the event variable of the adjunct is bound, which in turn blocks identification with the matrix event variable and therefore also the construction of a macro-event. Consequently, wh-extraction spans across two distinct events, violating the SEC and resulting in the sentence ungrammatical status.
The alternation is also sensitive to the type of verb in the main clause. Under Truswell’s analysis, factives behave like tensed adjuncts in that they presuppose the existence of the event expressed by their complement. When the complement is non-finite, event identification enables the creation of a single macro event. Wh‑extraction then satisfies the SEC and the sentence is fine. When the complement is tensed, instead, its event variable gets existentially bound and the entire sentence expresses two disjoint events. Consequently, wh‑extraction violates the SEC, determining an ungrammatical sentence.
In contrast, bridge verbs neither presuppose, nor assert, nor deny the event expressed by their complement. Therefore, the only event asserted to exist in the actual world—and hence the only event relevant for the SEC—is the event expressed by the bridge verb itself (e.g. the saying, the thinking). The event expressed by the complement pertains to the belief word of the bridge verb agent, not to the actual world inspected by the SEC. The SEC is thus satisfied, and wh‑extraction is possible independently of the tense specification of the complement, as shown by the two examples in (ii).
Truswell’s analysis extends to other type of adjuncts, such as purpose clauses and prepositional participials. It also accounts for other empirical properties, explaining, for example, why matrix clauses involving predicates expressing accomplishments and achievements are more likely to allow extraction from untensed adjuncts than other predicates.
(18) Unlike wh-extraction, the sensitivity of RD-extraction to tenseness extends beyond factive verbs. Compare the examples in (i) and (ii), which hold independently from the presence of clitic-doubling (i.e. for both RD– and RD+). Using Truswell’s analysis, I speculate that the examples (i) and (ii) encourage a discourse-given interpretation of the complement, hence presupposing the associated event, which eventually leads to the presence of two distinct events and prevents movement as in Truswell’s analysis of factives (see footnote 7).
The contrast with CLLD extends to these cases as well; see the grammatical CLLD sentences in (iii) involving tensed complement clauses.
Surprisingly, right dislocation to a higher clause is also sensitive to the type of focused constituent being used to delimit the matrix right edge. Focused subjects are fine, as we already saw. Adjuncts are possible too, as shown by (iv), where the focused adverb is interpreted as modifying the matrix clause. In contrast, dislocation past the focused indirect object of the main clause appears degraded, see (v).
Fascinating as they are, the investigation of these properties goes beyond the goals of this work and is left to further research.
(19) Under this analysis, the pronominal or anaphoric nature of the object clitic is irrelevant, predicting that co-occurrence with ne will extend to reflexive clitic as well. As (i) shows, the prediction is borne out.
(20) RD– and RD+ become indistinguishable when examining dislocated quantified subjects of passives and unaccusative verbs, since they are generated in object position but being subjects do not involve clitic doubling by object clitics. As we already saw in Section 4.4.1, in these cases ne-cliticization is obligatorily absent with CLLD and obligatorily present with RD–. As for the RD+ cases, since clitic-doubled right-dislocated subjects are doubled by pro in specTP (Cardinaletti 2001, 2002), no overt object clitic is present, and therefore extraction of ne becomes possible. The outcome is thus identical to the RD– cases, requiring obligatory ne-cliticization as shown in (i)a–b. The derivation of (i)b follows in (ii).
(21) The extraction restrictions just discussed predict that sentences like (i) cannot involve object extraction across both clauses. The only possible analysis has the entire lower clause right-dislocated, and the object i ladri (the thieves) right-dislocated only with respect to the lower clause, as schematized in (ii).
That this latter analysis is indeed possible is confirmed by the grammaticality of (iii), where the dislocated status of the lower clause is made explicit by the corresponding singular object clitic lo in the main clause, which contrasts with the plural object clitic li required for doubling the plural object i ladri. The corresponding structure is sketched in (iv).
(22) If López (2009) is correct in maintaining that CLLD, too, is derived via movement, then the absence of wh‑extraction under CLLD could be accounted for through a similar analysis whenever clitic doubling is present.
(23) Many CLLD examples in Cecchetto (1999) involve a left-peripheral object DP and are therefore technically ambiguous between a CLLD and a hanging topic (HT) analysis (Benincà 2001). This raises the issue of which of the properties discussed by Cecchetto pertain to CLLD and which to HTs. This ambiguity does not affect the discussion of RD in this section. For a review of the properties distinguishing CLLD and HT see the comparison between CLLD and LD in Cinque (1990: 57), where ‘LD’ stands for Benincà’s HTs.
(24) Note that the initial clause che gliela presti cannot be analysed as a raised subject, as subject raising is case-driven and clauses resist case (Stowell 1981). Furthermore, even preverbal subjects have recently been shown to constitute left-dislocated constituents binding a null pronoun in specTP (Frascarelli 2007, see also Barbosa 1995, Alexiadou and Anagnostopoulou 1998, Cardinaletti 2004, Rizzi 2004).
(25) Incidentally, note that the ungrammaticality of (158)(b), showing quantifier binding into a CLLD example, is unexpected if CLLD is movement-based as claimed by Villalba (2000) and López (2009), since reconstruction should enable it.
Like López, Frascarelli and Hinterhölzl interpret the above sentence as evidence supporting the absence of reconstruction for RD (or, more precisely, RD+. The cliticless counterpart of (i), ambiguous between RD– and marginalization, is uninformative because the distributive interpretation becomes possible but the postfocus indirect object could be marginalized in situ rather than right dislocated).
The data examined in (159)–(162), however, show that the bound interpretation associated with reconstruction is present with right-dislocated objects. It is also available for right-dislocated indirect object involving the anaphoric possessive proprio (own); see (ii) and (iii), where a postverbal subject successfully binds a possessive pronoun within a right-dislocated indirect object. Note that it is reconstruction rather than QR that is responsible for the bound interpretation, since binding under QR in (iv) yields an ungrammatical weak crossover interpretation.
While the ungrammaticality of sentence (i) needs to be eventually accounted for, this evidence shows that reconstruction in RD+ does extend to indirect objects.
(27) As in the original derivation in Villalba (2000: 211), the proposed derivation shows the object clitic ho doubling the CLLD complement que se l’havia comprat only at step 4. Steps 1–3 lack it, probably because the clitic is licensed by the CLLD phrase. Consequently, these steps show the full form em for the first person dative pronoun rather than the reduced form m visible in step 4 and in the final sentence.
(28) Without the evacuation step, the RD phrase would have to move to its final position before CLLD takes place, yielding the two ungrammatical derivations in (i) and (ii). In (i), CLLD precedes remnant movement and consequently cannot target the root node, violating Chomsky’s root condition. In (ii), CLLD follows remnant movement, thus involving extraction from an unselected specifier.
(29) It may at first appear surprising that extraction from CLLD and RD is possible at all, considering that wh-extraction from a clitic-doubled subordinate clause is not possible, as seen in Section 126.96.36.199. The examples in this section, however, involve the clitic ne (of them), and wh-extraction from right-dislocated phrases involving this clitic are actually possible. See for example sentence (i), where wh‑extraction under ne-cliticization is grammatical. If the clitic ne is base-generated in T (or a related position), rather than heading the big DP of RD+, then the pattern is accounted for, since wh-extraction would occur from a theta-assigned position.
When RD involves an object clitic other than ne, as in (ii), wh-extraction is blocked because the clitic creates an island to extraction as explained in Section 188.8.131.52. As expected, the corresponding case of extraction of CLLD from RD is also ungrammatical, see (iii).
Extraction of RD from CLLD shows the pattern just observed for extraction of CLLD from RD, suggesting that CLLD shares a similar analysis in this respect: wh-extraction from CLLD is possible under ne-cliticization in (vi), but not with another clitic in (vii). Extraction of RD from CLLD is heavily marginal with a clitic other than ne in (viii), and fully ungrammatical when involving PPs blocking wh-extraction in (ix). Note that (viii) is ambiguous between a CLLD and hanging topic (HT) analysis.
(31) In Frascarelli’s paper, CLLD is referred to as ‘LD’, for ‘left dislocation’. Her explicit references to the CLLD-related evidence in Cinque (1990) and her data, which include clitic left-dislocated PPs consistent only with a CLLD analysis, clearly identify CLLD as the relevant construction.
(32) The same problem affects the claim that CLLD is base-generated in Frascarelli and Hinterhölzl (2007). They observe that in (i) if the CLLD object containing the R-expression Leo could reconstruct due to its moved nature, it would violate condition C whenever bound by the pronominal subject lui ‘he’. Since the sentence is grammatical, reconstruction, and hence movement, must be excluded.
The phrase con Leo ‘with Leo’, however, is an adjunct modifier allowing for late insertion. When the adjunct is replaced by an argument as in (ii), the resulting sentence is indeed sharply ungrammatical, confirming the presence of reconstruction.
(33) Fernández (2012) shows that other aspects of right dislocation might vary too. In particular, he shows that right dislocation in English may only consist of a right-dislocated hanging topic. This, in turn, accounts for the properties that distinguish English right dislocation from Italian, such as the obligatory presence of a resumptive pronoun, the fact that dislocated items must be nominal, and the impossibility of multiple right-dislocated constituents.
(34) López (2009: 253) observes that the adjunct/argument asymmetry is also absent under CLLD, whereas Italian CLLD shows it (Cecchetto 1999). This raises a potential problem for López’s analysis of CLLD, at least in so far it is intended to provide a uniform account across Italian and Catalan.
(35) Controlling for the intended intonation is essential to any test of RD because the position of stress determines what structure the informants are actually assessing. Consider for example sentence (i). This sentence is ungrammatical when stress falls rightmost, as in (ii), and grammatical when stress falls on the indirect object, as in (iii). It is ungrammatical in (ii) because this intonation is only consistent with a structural analysis where the object occurs in situ. As a result, clitic doubling violates condition C. The intonation in (iii) is instead grammatical because it allows a structural analysis where the object is right-dislocated clause-externally, thus also enabling clitic doubling. Marking the position of stress is thus essential for ensuring that informants assess the intended structures.