The Syntactic Integration of Appositives: Evidence from Fragments and Ellipsis

Abstract

The goal of this squib is to explain an intriguing data set involving appositives and fragment answers, thereby providing support for the idea that appositives, and by extension parentheses more generally, are related to the host clause in syntax, via parenthetical coordination.Appositives include appositions and appositive relative clauses (ARCs), such as those illustrated in (1a) and (1b), respectively. In both examples, the anchor is John.Various approaches to these construction types abound in the literature (see De Vries 2006, Dehé and Kavalova 2007, Heringa 2011:122–139, and Kluck 2012 for overviews). We split these lines of thought asunder. What is of special interest is the question of whether and how an appositive is related to its anchor. Orphanage approaches maintain that appositives are syntactically isolated from their anchor; integration approaches do assume a structural relationship, albeit possibly of a special kind. For instance, Haegeman (1991) claims that parentheticals are derived separately from their host clause, as syntactic ‘‘orphans’’; they are only interpreted as related to their host when ‘‘contextualized’’ post-LF. Espinal (1991) argues that parentheticals and the host clause lie on different planes in a three-dimensional space. Hence, they are syntactically unrelated to the host but intersect with it at the terminal string, as is somewhat suggestively depicted in our example (2).Although the abovementioned approaches account for the well-known scopal independence of parentheses (see below), we do not think the word string can be treated as a given; rather, it should be derivable from the syntactic structure ( possibly in combination with some PF rule system). Therefore, let us turn to a more wholehearted integration approach, which we adopt below. De Vries (2009, 2012) argues that the appositive is ‘‘parenthetically coordinated’’ to the anchor, mediated by a syntactic functional head that can be dubbed Par.1 Like a coordination phrase, ParP is inserted in the position of the relevant noun phrase, which then occupies its specifier (3).2To be clear, we discard a plain right-adjunction analysis of ARCs as in, for example, Jackendoff 1977 because it does not account for scopal independence, among other things. To illustrate, in (4) a quantifier in the host fails to bind a pronoun inside the ARC (see, e.g., Espinal 1991, De Vries 2007, and Kluck 2011 for more discussion).In our approach, Parenthetical Merge effectively blocks c-command (see De Vries 2012 for details).Adopting Par has some immediate benefits. First, the Par approach trivially explains why the anchor and the appositive are linearly adjacent, as they are phrase-mates. Note that this is much less clear on an orphanage approach (more generally, the unlinearizability of orphans or otherwise multirooted structures seems to be a major problem). Second, on a more abstract level—again presupposing that syntax mediates between sound and meaning—the head Par provides the necessary syntactic locus for Potts’s (2005) ‘‘comma feature,’’ the semantic operator that renders the information appositives convey as secondary to the proposition denoted by the host clause. Finally, a syntactic integration approach such as the Par account predicts that parentheticals can be recursively construed with respect to each other. They obviously can, as (5) illustrates.Thus, there are clear indications for structural integration.We cannot discuss the internal structure of appositives in this squib (see Cardoso and De Vries 2010, Heringa 2011, and the references there for extensive discussion). What interests us here is the connection between parenthetical material and the host (or a constituent thereof). Below, we discuss data involving fragment answers whose distribution receives a natural explanation on the Par approach. Such an account does not seem to be available in any straightforward manner on an orphanage approach, and therefore the data presented constitute (yet another) challenge to this line of thinking (see Arnold 2007).Fragment responses are nonclausal utterances that can be seen as the remnant of some ellipsis process. Here, we will subscribe to—and indirectly provide support for—the gist of a PF deletion approach, which implies a complete syntactic structure combined with ellipsis at the PF interface (see Ross 1969, Lasnik 2001, Merchant 2001, 2004, Watanabe 2004, Griffiths and Lipta´k, to appear). As in sluicing constructions, the eventual remnant is focus-moved to the left periphery of the clause, and ellipsis targets roughly C′.3 This is illustrated in (6).Presupposing that grammatical processes apply to constituents only, we expect not only the moved fragment but also the ellipsis site to be a complete phrase (here, the entire C′, which is possible because the fragment fronts out of it). This will become relevant later on. Furthermore, we will assume that ellipsis involves given material, in the sense that it is recoverable from the context (see Tancredi 1992, Schwarzschild 1999, Merchant 2001). Therefore, no difference in interpretation should obtain between an elliptical clause and its nonelliptical counterpart.With these prerequisites in place, let us return to appositive constructions. Observe that an ARC may be anchored to a nominal constituent (7a) or a clause (7b).In the context of a question, we can test whether the ARC can be retained with or as part of the remnant. Unsurprisingly, if the question triggers a direct object fragment answer, as in (8), the object-related ARC may surface; see (8Bi). However, a clause-related ARC is also fine; see (8Bii). If the question involves the subject, as in (9), the object-related ARC is completely unacceptable, as in (9Bi), but the clause-related ARC is still fine, as in (9Bii). This is the case not only for clausal ellipsis: VP-ellipsis produces the same results. We indicate this by placing the auxiliary did in parentheses. Moreover, the observations in (8) and (9) extend to languages other than English, including Dutch. It is fair to note that data similar to those in (8) and (9) are discussed by Arnold and Borsley (2008), albeit from a different perspective, but whose main conclusion we share (see footnote 9 for a few remarks).In short, an ARC can only surface next to an anchor that is at least partially overt. Given that ellipsis constructions are semantically identical to their nonelliptical counterparts, this is a mystery that needs to be resolved.First, notice that the pattern in (9) is confirmed by additional data. Depending on lexical choices, an ARC can be ambiguous between two readings. This is shown in (10), where which is awful relates either to the object Mary’s computer or to the proposition as a whole, involving a stealing event. In the context of an object fragment answer, (11), this remains the same. Interestingly, however, the ambiguity disappears if the elided phrase includes the object, (12); here, only the propositional reading survives.A similar pattern is observed if sentential adverbs like probably and unfortunately are employed as fragmentary responses (note that these elements can escape ellipsis, as they are base-generated above TP; see Cinque 1999). For instance, the ambiguity present in (13) is lost in (14B), where only the ‘‘clausal anchor’’ reading is retained.Again, a constituent-related ARC can only surface if its anchor is overt.We will now show that the syntactic integration approach to appositives, in combination with the PF deletion account of ellipsis, provides a natural explanation of the above data. First consider the propositional reading. In such examples, the ARC takes the full CP as its anchor.4 The analyses in (15) and (16) repeat the relevant (Bii) examples from (8) and (9). In both cases, ellipsis targets a constituent, the C′ (residing within Spec,ParP), and the derivation is syntactically and semantically convergent. Observe that the elliptical parts in the answers are fully recoverable from the preceding questions. In the diagrams, the ovals with gray type represent the ellipsis sites.5Next, let us turn to the regular constituent reading, where the ARC takes a noun phrase (here, the direct object) as its antecedent. The analysis in (17) repeats example (8Bi). Here, the entire ParP is focusmoved, 6 and consequently elision of the C′ leaves the direct object in combination with the ARC as a fragment answer.Finally, the analysis in (18) represents the impossible case, repeated from (9Bi). Although the ellipsis site in (18B) would be recoverable from (18A), ellipsis necessarily targets a nonconstituent here, which is excluded. Notice that if the ARC were not paratactically coordinated but some kind of orphan, nothing would have prevented the derivation of this very sentence.7To complete the argument, let us briefly return to the examples in (11), (12), and (14), repeated in (19), (20), and (21). Recall that (19) has two readings. The structure of the constituent reading (19Bi) equals the one in (17), the propositional reading (19Bii) the one in (15). By contrast, (20) only has the propositional reading, like the example represented in (16). The impossible constituent reading would have to correspond to (18), which is excluded because it would involve ellipsis of a nonconstituent. For the same reason, the derivation of (21) only converges if the ARC modifies the entire clause.Thus, the Par approach straightforwardly explains why ellipsis leads to disambiguation in (20) and (21) but not in (19). Since ellipsis sites, by definition, are semantically interpretable (as they must be recoverable from the context), an appeal to ‘‘uninterpretability’’ is excluded.In the explanation advanced above, we adopted a mediating functional projection ParP in the structure. A potential alternative might be to extend Hornstein and Nunes’s (2008) approach to ‘‘non-c-commanded adjuncts’’ to parentheses. From this perspective, an ARC would be ‘‘concatenated’’ with the anchor, but the projection would not be ‘‘labeled,’’ which effectively leads to a multirooted structure. According to Larson (2012), who applies this strategy to elliptical sentences involving swiping (Merchant 2002), the additional structure does not count as a constituent part of the host; this in turn allows clausal ellipsis to take place such that the relevant phrase survives.8 When we apply these ideas to our crucial example (18), the structure becomes as shown in (22).However, as is evident from this representation, the elliptical sentence is incorrectly ruled in, similarly to the situation in true orphanage approaches. For this and other reasons, it seems to us that this is not the way to go.To explain the unacceptability of (18B)/(22B), one might fancy a rule that bans anchors that are phonologically null. However—apart from the stipulative nature of such a constraint—it runs into difficulties with respect to the following data:(23B)’s unacceptability cannot be due to the incompleteness of the DP anchor, as incomplete DPs—like incomplete CPs—make for acceptable anchors in other environments, as (24B) illustrates.The difference in acceptability between (23B) and (24B) follows naturally from the Par approach. In (24B), the partial covertness of the anchor is derived by NP-ellipsis, licensed by the genitive -’s (see Lobeck 1995), in combination with clausal ellipsis; see (25).There is no NP-ellipsis in (23B), and consequently, the derivation of (23B) would require subextraction as well as nonconstituent deletion, which is banned; see (26).Thus, we have shown that quite intricate patterns involving complex noun phrases can be explained on the same set of very general assumptions we used before.We provided evidence from the distribution of appositives in ellipsis environments to demonstrate that they form a constituent with their anchor, mediated by a functional parenthetical phrase. The findings here thus provide further support for the general idea that parenthetical constructions are integrated at the level of syntax with the host clause into which they interpolate,9 and indirectly for the PF deletion approach to ellipsis, which maintains the presence of syntactic structure.