Portal de Periódicos da CAPES

A Dilemma with Accounts of Right Node Raising

2012; The MIT Press; Volume: 43; Issue: 1 Linguagem: Inglês

10.1162/ling_a_00078

1530-9150

Bradley Larson,

Historical Linguistics and Language Studies

There is a dilemma in current studies of right node raising (RNR): the main approaches to the construction make fundamentally contradictory predictions that account for overlapping sets of data. In this squib, I argue that no single current analysis can account for the range of data, and I argue against the possibility that the analyses work in concert to account for the data. That is, given that each current analysis accounts for some but not all of the documented data, there are two logical possibilities: either none of the analyses are correct, or more than one analysis is correct, each in its limited purview, and duties are shared such that all the data are accounted for. I argue for the former conclusion.Under the second option just mentioned, RNR is derived either by means of one particular operation or by means of a different one. That is, the term right node raising is better viewed as a surface-level description for a family of derivations: some stem from applying the first operation, the others stem from applying the second (as argued by Barros and Vicente (2011)). This view departs sharply from the assumptions of most work on RNR and requires critical investigation. Such investigation reveals no motivation to analyze RNR as being derived in two separate ways. This being the case, the RNR dilemma remains.Much work has been done on the derivation of RNR, and theories fall fairly neatly into three groups: ellipsis, multidominance, and movement. All previous work argues that RNR is derived via multidominance rather than ellipsis (or movement) (or vice versa).1 Ellipsis accounts of RNR (see, e.g., Wilder 1997, Giannakidou and Merchant 1998, Hartmann 2000, Bošković 2004, Ha 2008) maintain that an RNR sentence like (1) has the form in (2) at some earlier stage in the derivation. The italicized string in the first conjunct is elided phonologically (see (3)). This produces the effect of having a single string of words interpreted in both conjuncts while only appearing in one.The advantage to such a view rests in the fact that the elided material in the first conjunct shows effects found in traditional VPellipsis. For example, both VP-ellipsis and RNR display vehicle change effects (see Fiengo and May 1994). The index of an R-expression can “change vehicles” and become associated with a pronoun in the same syntactic position if the R-expression is elided. Constructions that would otherwise run afoul of Principle C can be rescued by ellipsis (and vehicle change in turn) in (4) and by hypothesis in (5). This fact has, as yet, found no explanation in multidominance theories of RNR. Take it as a premise that vehicle change effects arise solely out of ellipsis.Multidominance accounts of RNR (see, e.g., McCawley 1982, Wilder 1999, De Vos and Vicente 2005, Bachrach and Katzir 2008, Grosz 2009, Larson 2009) propose that the “shared material” is literally introduced (i.e., merged) into both conjuncts simultaneously, as in (6). Under one of many linearization schemes (see Wilder 1999 or Bachrach and Katzir 2008), the shared material appears in the rightmost position while being interpreted in both conjuncts.This approach has the advantage of plausibly accounting for cumulative agreement in RNR. Cumulative agreement (7) is found when subjects from separate conjuncts function together to effect plural agreement on a shared verb.2 This fact has not yet found explanation in ellipsis theories of RNR. Take it as a second premise that cumulative agreement is a result of multidominance alone (see Grosz 2009 for details).In short, there is a dilemma. There are RNR sentences that are amenable to ellipsis analyses though not multidominance ones, and there are RNR sentences that are amenable to multidominance analyses though not ellipsis ones. The common tack in the RNR literature is to suppose that a single theory is correct and to work on its empirical shortcomings: an “exclusivist” hypothesis. But this is not the only logical possibility. There is another approach in which both theories are correct, though limited in scope: an “eclectic” hypothesis. I will argue that the eclectic hypothesis is descriptively inadequate; an exclusivist approach is more parsimonious and should be assumed.It is difficult to determine how one could attack or defend the eclectic hypothesis. Barros and Vicente (2011) have devised a particularly ingenious test. I adopt it here, though I come to quite different conclusions than they do. In what follows, I present a simple example test of this type.First, Barros and Vicente present a specific RNR construction that can, by hypothesis, only be derived by ellipsis. Such a sentence will have a signature “prompt” that unambiguously marks it as such. For example, what one may dub morphological mismatch can be derived only via ellipsis, not multidominance. The mismatch in question here is the difference in person between my and her. In (8), even though the overt possessive pronoun in the second conjunct is different in person from the one interpreted in the first conjunct, the phrase in the first can be elided.This sentence can be interpreted such that the speaker did not pass his or her own exam and Alice did not pass her own exam. This is possible in ellipsis generally (9). That this interpretation is possible will be a prompt for ellipsis-derived sentences.Multidominance accounts predict that for (8) the possessive pronoun her should have the same interpretation in each conjunct, because there is simply no elided pronoun to relate to the other nominal. As a result, multidominance cannot be the source of (8).Second, Barros and Vicente (2011) present an RNR sentence with a prompt that can only be derived by multidominance. For example, sentence (7) displaying cumulative agreement (repeated here as (10)) cannot be derived by ellipsis because such an account would predict singular agreement with the verb as in (11). But through the simultaneous Merge of two subjects with the T head, cumulative agreement can be achieved via multidominance (again, see Grosz 2009 for details).Finally, Barros and Vicente construct examples that have the prompts of both ellipsis- and multidominance-derived RNR sentences like (12). If these combination sentences are unacceptable, it follows that neither the ellipsis nor the multidominance account can reduce to the other (i.e., neither account can produce both prompts). In effect, both are needed to account for the data. If the combination sentence is acceptable, then either one of the analyses is superfluous, or both are.In (12), since there is cumulative agreement, the sentence must have been derived via multidominance and that makes the morphological mismatch reading not possible. On the surface, that (12) is unacceptable under the relevant interpretation suggests that the eclectic hypothesis is correct. But as we will see, this is deceiving and the exclusivist hypothesis is superior.In what follows, I investigate arguments for the eclectic hypothesis. The arguments are presented as suites of constructions: an ellipsis exemplar, a multidominance exemplar, and a test combination sentence. I show that when appropriate test conditions are applied, the sentences remain acceptable and arguments for the eclectic hypothesis fail to hold.From examples (10)–(12), we saw that morphological mismatch is generally possible in RNR—just not when cumulative agreement is present. The sentence (8) showing morphological mismatch, however, is not minimally different from the test sentence in (12).3 If these sentences are to be true minimal pairs, the ellipsis exemplar should have the same pregap structure as the test sentence. In (13), the reading with a bound pronoun in the first conjunct is unavailable.Here, even when multidominance is not necessary to derive the sentence, morphological mismatch of this sort is not licensed.4 Since the mismatch is not allowed in sentences like (13), the fact that morphological mismatch is also not allowed in (12) (i.e., a sentence like (13)) does not show that cumulative agreement is what blocks it.Another suite of examples involves a similar ellipsis-induced morphological mismatch paired with a multidominance-derived interpretation of different. The example of morphological mismatch is shown in (14). Here, the verbal morphology is mismatched (like the VP-ellipsis examples discussed in Lasnik 1995). The verb in the shared element is in its infinitival form even though it is interpreted in the first conjunct in its past participial form.This sentence is paired with (15), in which different can have a so-called internal reading—namely, that Alice and Iris each work on separate topics. Bachrach and Katzir (2008), following Abels (2004), claim that this interpretation arises from multidominance and not from ellipsis.When a sentence has the morphological mismatch prompt, the internal reading of different is not blocked (16). Following the logic in section 3.1, it cannot be the case that both ellipsis and multidominance are required. In fact, the logic suggests that neither approach is correct, given the conceptual incompatibility discussed earlier.This fact is not predicted under the eclectic account. Prompts from both multidominance and ellipsis are present in a single sentence, and the sentence is acceptable under the intended reading.In another set of examples, vehicle change is the prompt for ellipsis (17) and is paired this time with cumulative agreement (18).When both of these prompts are present in the same sentence, the result is ungrammatical with the vehicle change reading.But there is an important difference between the test sentence (19) and the model ellipsis sentence (17). Note that in (19), the R-expression Claire is separated from its antecedent by another R-expression. This is quite different from what we find with (17). When that sentence is altered so as to be more similar to the test sentence, the new version (20) disallows the vehicle change reading as well. Given this, there is little reason to blame cumulative agreement for the lack of vehicle change.5In another suite of examples, ellipsis-induced vehicle change (21) is paired with the multidominance-induced internal reading of different (22).When both of these prompts are introduced in the same sentence, both the vehicle change reading and the internal reading of different can be licit simultaneously, contrary to the prediction of the eclectic hypothesis.In a final suite of sentences, I introduce a new prompt for multidominance accounts and pair it with verbal morphological mismatch. As Sabbagh (2007) notes, a universally quantified NP that is shared between the conjuncts can take scope over indefinites outside the shared material (24). This could not have been derived by ellipsis, assuming that this type of scope taking depends on covert quantifier raising. It would require covert across-the-board movement, which Bošković and Franks (2000) have argued does not exist.When (24) is altered such that there is an ostensibly ellipsis-derived morphological mismatch, the new sentence retains an interpretation in which the universal QP has high scope, again suggesting that the eclectic approach is incorrect.While a logical possibility, the eclectic hypothesis about RNR is not empirically supported. In fact, in this squib the eclectic approach has been shown to make exactly the wrong predictions when the test conditions are controlled sufficiently, and it should thus be discarded. We are unfortunately left with the dilemma we started with. The study of RNR is at an impasse: the current analyses cannot account for the facts on their own, and as shown here they also cannot account for them when working together.