Asymmetric DOM in Coordination: A Problem for Movement-Based Approaches
Laura Kalin, Philipp Weisser · Linguistic Inquiry · 2018
Differential object marking (DOM) is a common crosslinguistic phenomenon whereby overt case marking on objects surfaces only on a subset of objects, namely, those high in definiteness, specificity, and/ or animacy (e.g., Comrie 1979, Croft 1988, Bossong 1991, Enç 1991, Aissen 2003, de Swart 2007). In Spanish, for example, simplifying somewhat, overt case marking of objects (boldfaced throughout the squib) is required when the object is specific and animate and banned when the object is nonspecific or inanimate.In this squib, we set out to (a) introduce new findings revealing that many DOM languages allow asymmetric marking in coordinations when conjuncts are mismatched in terms of animacy/definiteness, and (b) show that these findings are extremely problematic for many popular (broadly) Minimalist accounts of DOM, namely, those that derive DOM via movement (e.g., de Hoop 1996, Torrego 1998, Woolford 1999, Bhatt 2007, Rodríguez-Mondoñedo 2007, Baker and Vinokurova 2010, Richards 2010, López 2012, Ormazabal and Romero 2013), at least insofar as these accounts are intended to be general accounts of DOM and/or apply to the languages that allow asymmetric marking.The squib is organized as follows. In section 2, we outline two movement analyses of DOM. In section 3, we discuss why such accounts predict asymmetric DOM in coordinations to be impossible, and in section 4, we show that many DOM languages do in fact allow asymmetric DOM. In section 5, we explore whether movement analyses can be salvaged (for languages that allow asymmetric DOM), and we argue that they cannot. In section 6, we offer preliminary conclusions.Movement-based accounts of DOM take raising of the object out of VP as in (2) to be a necessary (though perhaps not sufficient) ingredient of DOM.Here we lay out two specific accounts, which stand in as instantiations of more general types of accounts: (a) accounts in which movement of the object is to a Case position (e.g., Bhatt 2007, Rodríguez-Mondoñedo 2007, López 2012, Ormazabal and Romero 2013), and (b) accounts in which raising of the object feeds case competition with the subject (e.g., Baker and Vinokurova 2010, Baker 2014a, Levin and Preminger 2015). Across movement-based accounts, a common component of the motivation for movement is that the object must raise out of VP to escape existential closure (Diesing 1992).The first type of account is exemplified by Rodríguez-Mondoñedo 2007. Rodríguez-Mondoñedo argues that transitive v in Spanish can only check [number] features, and so can only assign Case to an object that is ϕ-incomplete, in other words, one with only [number]. Case assigned by v has a null spell-out. If an object has a [person] feature (carried by animate specific nominals), then the object is ϕ-complete and cannot have its Case checked by v. Such an object thus needs to raise (ultimately, to Spec,DatP) to check its Case. Since the projection it checks Case with is DatP, the marking is dative a.The second type of account is exemplified by Baker and Vinokurova 2010. Looking at the Turkic language Sakha, Baker and Vinokurova argue that DOM is derived by movement out of VP, which is a phase, into the higher CP phase. Since the subject is also in this higher phase (and is so far caseless), the object enters into case competition with the subject (Marantz 1991) and so, per the case assignment rules of Sakha, receives dependent accusative case. Unlike in Spanish, this case is not syncretic with dative; rather, its exponent is a unique accusative form, -(n)I. Objects that do not raise remain caseless.A preliminary problem with taking movement to be a general property of DOM is that not all DOM languages have (at least obvious) syntactic movement of the marked object; see, for example, Hebrew (Shlonsky 1997), Kannada (Lidz 2006), and Northeastern Neo-Aramaic languages (Kalin 2018). However, it might be that there is movement in these languages but it is not detectable with the normal tests or is covert, so this is not a fatal blow to movement-based accounts of DOM. In the following sections, we present coordination data that we take to more clearly show that movement is neither a necessary ingredient of DOM nor a general property of DOM in all languages.While the accounts cited above apply various tests to establish the higher position of marked objects (e.g., adverb placement, binding), some of the most reliable tests of syntactic movement are typically not applied, namely, tests involving islands. Islands are syntactic configurations that prohibit movement out of them. If an alleged movement-derived effect fails in such a configuration, this suggests movement is involved. If, on the other hand, the effect is not blocked, this suggests that movement is not involved.In the domain of DOM, many islands cannot be tested since the alleged movement step is very short. Fortunately, one of the most crosslinguistically robust islands is applicable. As is well-known since Ross’s (1967) discovery of the Coordinate Structure Constraint (CSC), it is not possible to move one whole conjunct out of a coordination.While theoretical accounts of coordination islands vary, the data are clear.In order to apply this test to the DOM cases at hand, we simply need to conjoin one element that is supposed to undergo movement (a marked object, DP1 in (4)) and one element that is supposed to stay in situ (an unmarked object, DP2).If a marked object and an unmarked one cannot be conjoined (i.e., if (4) is ungrammatical), then this indicates that there is indeed something wrong with the configuration, plausibly because it is ruled out by the corresponding CSC island violation schematized in (5).1If, however, (4)-like sentences are grammatical, and one conjunct is marked while the other is unmarked (or the conjuncts bear different markers), then this suggests that there is not a crosslinguistically necessary connection between DOM and movement, nor is a movement analysis feasible for the languages in which (4) is grammatical.2Here, we test the configuration in (4)/(5) against a sample of eleven DOM languages from five language families. Of the eleven languages, nine allow one conjunct to be marked while the other remains unmarked; further, for both languages that disallow asymmetric marking, there is a closely related language that allows it. These results strongly suggest that a movement analysis cannot be maintained for most of the languages in our sample; thus, either movement is not the source for DOM in any language, or there simply is not a unified source for DOM crosslinguistically.A preliminary note about our examples: For each language, if the language’s object marker precedes the object, we use a coordination in which the marked conjunct is the second one, and if the language’s object marker follows the object, we use a coordination in which the marked conjunct is the first one. In doing so, we ensure that the DOM marker takes scope over only one conjunct and not the whole coordination phrase. In all the languages we tested apart from Tamil and Spanish, if the conjuncts are reversed, marking of just one conjunct is also grammatical ( just not unambiguously asymmetric).Let us start with Romance languages. Recall from (1) that in Spanish, objects that are specific and animate bear the marker a. If we conjoin an animate nonspecific object and an animate specific object, we see that asymmetric DOM is possible.Preliminarily, then, even in a language for which a movement-based account has been specifically motivated (e.g., Torrego 1998, Rodríguez-Mondoñedo 2007, López 2012), asymmetric DOM inside coordinations is possible.3 Note also the agreeing adjective juntas, which shows that this is indeed a case of DP-coordination.DOM is also found in most southern Italian dialects (D’Alessandro 2017). All have in common that first or second person pronouns are marked, (7b), while nonhuman objects are not, (7a). As in Spanish, DOM appears in the form of the dative marker a, and asymmetric DOM is allowed, (8). Data here come from the Neapolitan dialect.In yet another Romance language, Romanian, the main trigger for DOM is animacy (Dobrovie-Sorin 1994). Animate objects are marked with the preposition pe while inanimates are not, (9). Again, asymmetric marking in coordinations is possible, (10).Romance languages seem to consistently allow asymmetric DOM.Turning to another branch of Indo-European, Indo-Iranian, we find mixed evidence. In Nepali, dative -laai is also used to mark animate and specific direct objects (Schikowski 2013), (11), and mismatched objects can be conjoined, (12), as in Romance.In Hindi, dative -ko marks specific direct objects (see, e.g., Mahajan 1990, Bhatt and Anagnostopoulou 1996), (13), but notably does not allow asymmetric marking, (14); any combination of a marked object with an unmarked one is ungrammatical.In contrast to all the languages exemplified earlier, Hindi disallows asymmetric DOM.4Moving on to another language family, we take a look at Finnish. In Finnish, nonpronominal objects bear genitive case, (15a), while pronominal objects bear accusative, (15b).5 It is possible to conjoin a pronominal object with a nonpronominal one, (16).We thus have evidence from Uralic for the acceptability of asymmetric DOM.The next family we consider is Turkic. In Turkish, specific objects are marked with a unique accusative case while nonspecific objects are unmarked (Enç 1991, Kornfilt 1997), (17). Like Hindi, Turkish does not allow conjunction of objects with asymmetric marking, (18). (Similar judgments are reported in Kornfilt 1997.)However, Caucasian Urum, a related Turkic language spoken by ethnic Greeks in Georgia, exhibits a DOM system that looks nearly identical to the Turkish system on the surface, (19), but does allow asymmetric DOM, (20).Turkic languages, then, differ as to whether they allow asymmetric DOM.Next we turn to Semitic languages, many of which exhibit DOM. In Hebrew (e.g., Danon 2006), for example, definite objects are case-marked by the proclitic et whereas indefinite objects are not, (21). Asymmetric marking of conjuncts is possible, (22).In Amharic, a Western Semitic language, the accusative marker -n attaches only to definite objects, while indefinite objects remain unmarked, (23). As in most of the languages we have looked at, it is possible to conjoin marked and unmarked objects, (24).Semitic thus seems to consistently allow asymmetric DOM.Finally, speakers of Tamil, a Dravidian language, mark definite objects with accusative, while other objects are usually unmarked (see the discussion in Lehmann 1989), (25). In Tamil, it is also possible to conjoin objects with different case markers, (26).6While DOM in each of these languages has many complexities that we cannot discuss here, it is clear that many (if not most) DOM languages allow asymmetries in case marking with conjoined objects. Nine of the eleven languages from five different language families allow conjunction of marked objects with unmarked ones, while only two (Turkish and Hindi) do not. Since movement is prohibited out of coordinations, these data strongly suggest that there must be some non-movement-related mechanism that is behind DOM, at least in nine of our languages.In this section, we discuss three possible approaches that could be pursued to maintain a movement analysis of DOM, and present arguments against each.The first potential challenge to our argument is that possibly the coordination island is not as robust as we make it out to be. Exceptions to the CSC have occasionally been attested. For example, Bošković (2009) has shown that some speakers of Serbo-Croatian allow extraction of the left conjunct.There are three reasons why a solution along these lines does not go through. First, violations of the CSC are very infrequent crosslinguistically. To our knowledge, no such exception has been reported for any of the eleven languages in our sample. Second, even in languages like Serbo-Croatian, only the leftmost conjunct can be extracted from a coordination phrase; this would predict that only the leftmost conjunct could ever bear DOM. Crucially, this prediction is wrong: the examples from Spanish, Italian, Romanian, and Hebrew above clearly show that the rightmost conjunct can be the only one that bears DOM, and for the most part, there are not linear restrictions on asymmetric DOM. (Again, Tamil and Spanish are exceptions here.)Third, we can simply test whether short movement allows for violations of the CSC, and we can see that it does not. Tamil, for example, allows short scrambling of the direct object over the indirect object (see Sarma 2003, Baker 2014b), (27). However, Tamil does not allow scrambling of just one conjunct, (28).We therefore reject the idea that some languages lack coordination islands and that this is responsible for the availability of asymmetric DOM in coordinations.The second objection to our account might be that movement is in fact symmetric but case assignment is not. In other words, in a configuration where only one of the conjuncts is high in definiteness/animacy, it might be that the whole &P raises (dashed line), but in this higher position, only one of the conjuncts receives case (solid line).There are two reasons why this proposal cannot save a movement account. First, Weisser (2017) argues that (non-DOM) case marking in coordinations is always symmetric. On the basis of fifteen case studies, Weisser shows that, once ellipsis and allomorphy are controlled for, case is always distributed evenly among all of the conjuncts in nominal conjunction. If (29) were the right kind of analysis for asymmetric DOM, then this would entail that the grammar does not allow case assigners to reach inside a coordination and target just one conjunct, except in the case of DOM.The second argument against (29) is the same as one in section 5.1. If case assignment could target a specific conjunct inside an &P, then we would expect to find ordering/hierarchy effects, with only the highest or linearly closest conjunct able to receive DOM, as this is what is found with agreement into coordinations (e.g., Marušič, Nevins, and Badecker 2015). But again, our data do not confirm this prediction. Spanish, Romanian, Italian, and Hebrew are all head-initial, and thus we would expect that the DOM case assigner should be able to pick out only the left conjunct because it is (a) structurally higher and (b) linearly closer to the case assigner. However, in all of these languages, we have shown that the second conjunct can be marked while the first one remains unmarked.A final alternative for reconciling a movement-based account with the data at hand is to analyze our examples of asymmetric DOM as involving conjunction of a bigger category (e.g., vPs) plus subsequent gapping, rather than DP-conjunction. One possible structure would be something like (30). Under this analysis, the movement that results in DOM could be internal to one of the conjuncts and thus not violate the CSC.There are a number of arguments against such analyses. First, gapping in many languages, among others in Romance (Repp 2009), comes with a specific intonation, namely, a pause in the position of the elided verb and a high boundary tone on the edge of the first conjunct. Our consultants produce asymmetric DOM examples without gapping intonation. Second, in some languages, we can tell from the morphology that we are dealing with nominal conjunction. For example, in Tamil, the coordination strategy that marks both conjuncts with -um conjoins DPs and PPs only (as in (26)), while verbal and clausal conjunction employs a different strategy. Thus, a derivation in terms of ellipsis in this language is not an option. Third, in languages like Spanish, (31), and Hebrew, (32), relative clauses and small clauses show that there must be a plural conjoined DP even in asymmetric DOM cases.Under a VP/TP/CP-coordination approach, (31)–(32) are hard to account for because there would simply be no plural constituent to adjoin the relative clause to or to predicate the small clause of. We therefore conclude that, for at least some of the languages we discuss, conjunction reduction is not a plausible analysis.We have shown that many DOM languages allow asymmetric DOM in coordinations, a finding that is problematic for movement-based accounts of DOM.7 It is important to note that our claim is that across languages movement is not a necessary ingredient for the phenomenon of DOM. It may very well be, however, that within a particular language, movement is indeed necessary for DOM. For Hindi, for example, movement has been argued to be required for DOM (e.g., Bhatt and Anagnostopoulou 1996); this analysis is in fact supported by our findings, as Hindi disallows asymmetric DOM in coordinations. For Spanish, on the other hand, while many movement accounts have been put forward (e.g., Rodríguez-Mondoñedo 2007, López 2012), these accounts are not supported by our findings; while it may be that marked objects raise when possible in Spanish (and thus movement correlates with DOM), it must also be possible for objects to receive DOM in situ, namely, in asymmetric coordinations.8 A larger, more areally diverse sample of languages should be gathered, in order to better understand why some languages allow asymmetric DOM and how common this really is.At this point, one might wonder what sorts of accounts can deal with asymmetric DOM. First, purely morphological accounts that derive case alternations by means of impoverishment/feature freezing (e.g., Keine and Müller 2008, Glushan 2010) could have these postsyntactic operations locally target just one conjunct in a coordination. Second, analyses of DOM that take the fundamental ingredient to be different structural sizes of marked and unmarked objects (e.g., Danon 2006, Lyutikova and Pereltsvaig 2015) could appeal to the idea that the whole coordination receives case, but only one of the conjuncts is of the right size to morphologically host case. Finally, accounts that appeal to last-resort rescue strategies for deriving DOM (e.g., Kalin 2018) could explain single-conjunct DOM as a highly local rescue. We leave this question open for future research.Authors are listed in alphabetical order. We would like to thank Mark Baker, András Bárány, Jonathan Bobaljik, Stefanie Böhm, Thomas Graf, Julie Legate, Jason Merchant, and the audience at NELS 48 for helpful discussions of this work, as well as all of our language consultants, as cited in the squib. This work was supported by the NSF (Research Grant BCS-1451098, PI: Bobaljik) as well as the Alexander-von-Humboldt Foundation (Projects: “Case and Coordination” and “Consequences of the SOCIC Generalization”).