Despić (2013) argues that Serbian binding data provide support for the Parameterized DP Hypothesis (e.g., Fukui 1988, Corver 1992, Zlatić 1997, Bošković 2003, 2005, 2008). His key claim is that the differences in binding possibilities observed between English and Serbian result from the presence of a DP layer in the former vs. its absence in the latter.1 Apparent determiners in Serbian are claimed to be NP-adjoined elements (with category left unspecified).2 In this squib, I show that Despić's analysis makes incorrect predictions about Serbian and closely related Slavic languages: Bulgarian and Macedonian.
1 Binding Data and Despić’s Proposal
Despić (2011, 2013) observes that English and Serbian prenominal possessives differ in binding possibilities. English prenominal possessives (e.g., his, John’s) can be coreferential with an R-expression (1a) or a pronoun (1b) elsewhere in the clause, whereas their Serbian coun- terparts cannot (2a–b).3
Following Kayne (1994), Despić (2011, 2013) assumes that specifiers are adjuncts and adopts the definition of c-command in (3), according to which phrases c-command out of the category they are adjoined to or are specifiers of.
To accommodate the facts in (1), which indicate that English possessives do not c-command out of their nominals, Despić (2011, 2013) follows Kayne (1994) and Szabolcsi (1983) and locates English possessives not in Spec,DP but in a Spec,POSSP position within DP, as in (4) (adapted from Despić 2013:244, (9)).6
Here, his fails to c-command John owing to the presence of DP; hence, Principle C does not prevent their co-construal.
To accommodate the facts in (2), which suggest that Serbian possessives do c-command out of their nominals, Despić (2011, 2013) proposes the structure in (5), in which DP is absent and prenominal possessives are NP-adjoined.7
Without a superordinate DP node, the NP-adjoined prenominal possessive njegov ‘his’ c-commands out of the subject NP under (3). Principle C then prevents its co-construal with the R-expression Marka ‘Marko’ in VP.
In a further development of his proposal, Despić (2011) observes that coreferential interpretations of the relevant elements in Serbian remain unacceptable even if demonstratives or other agreeing8 determiners appear in the structure.
*Ovajnjegovi drug smatra Markai veoma
this his friend considers Marko very
‘This friend of hisi considers Markoi very intelligent.’
??MnogiDejanovii prijatelji su njegai kritikovali.
many Dejan.POSS friends AUX him criticized
‘Many of Dejani’s friends criticized himi.’
Since the italicized items do not interfere with the c-command relation between the prenominal possessive and the R-expression or pronoun, Despić (2011, 2013) takes them also to be NP-adjoined.9 The three boxed NP nodes in (7) count as instantiations of the same node. Hence, the two XPs adjoined to [NPdrug] c-command out of the subject.
Note that if ovaj ‘this’ were a D head instead of an adjunct, the relevant c-command relation between njegov ‘his’ and Marka ‘Marko’ would be disrupted by ovaj’s DP projection, incorrectly predicting coreferential readings to be possible (cf. (4)).
MnogoKusturičinihi prijatelja je kritikovalo
many Kusturica.POSS friends AUX criticized
‘Many of Kusturicai’s friends criticized himi.’
(Despić 2011:71, (82))
Mnogonjegovihi prijatelja je kritikovalo Kusturicui.
many his friends AUX criticized Kusturica
‘Many of hisi friends criticized Kusturicai.’
(Despić 2011:72, (84))
2 Two Predictions
Despić’s (2011, 2013) account of binding with possessives makes two simple predictions that Despić does not discuss but that can be directly tested. First, the account predicts that Serbian possessive nominals, when further embedded within the subject as complements of some head X (10a), should show relaxation of the constraints on coreference observed in the unembedded cases. Second, the account predicts that counterpart examples in Bulgarian and Macedonian, two related Slavic languages held to project DP by advocates of the Parameterized DP Hypothesis (Bošković 2005, 2008, Despić 2011), should parallel English with respect to binding, and not Serbian (10b).11
The source of these predictions is the same in both cases. The presence of a higher projection (XP/DP) of which the possessive is neither a specifier nor an adjunct is expected to block c-command under (3) and hence permit coreference elsewhere in the clause. As I show, both of these predictions are incorrect.
2.1 NP Complements and Coreference in Serbian
Noun complementation is a candidate for the structure in (10a). Relational nouns are standardly recognized as nouns that take N complements (e.g., Barker 1995, Partee 1997, Partee and Borschev 2000).12 In particular, Bošković (2013) argues that in Serbian postnominal possessives like (11a), the relational N head ( prijatelj ‘friend’) selects the NP that follows (njegove majke ‘his mother’) as a complement. Then, consider the larger structure (11b).
Since njegove ‘his’ is neither an adjunct nor a specifier of the larger NP, the definition of c-command in (3) blocks the prenominal possessive from c-commanding out of the subject. The relevant coreferential readings are thus predicted to become available since Conditions B and C will not be violated. This prediction is not correct, however. The prenominal possessive cannot be coindexed with the R-expression or the pronoun, as shown in (12). The binding possibilities are the same as when the larger NP is absent (cf. (2)).
*[NP Prijatelj [NPnjegovei majke]] je zagrlio
friend his mother AUX hugged
‘Hisi mother’s friend hugged Markoi.’
*[NP Prijatelj [NPMarkovei majke]] je zagrlio
friend Marko.POSS mother AUX hugged
‘A friend of Markoi’s mother hugged himi.’
2.2 Bulgarian and Macedonian Binding Data
The Parameterized DP Hypothesis explicitly links the presence of DP in a language to the presence of overt definite articles (e.g., Bošković 2005, 2009, Despić 2011). Serbian classifies as ‘‘DP-less’’ by this criterion since it lacks definite articles. Two Slavic languages do have definite articles, however—namely, Bulgarian and Macedonian. These are widely accepted as ‘‘DP languages’’ by proponents of the Parameterized DP Hypothesis. As such, they are predicted to pattern together with English as far as binding potential of prenominal possessives is concerned. That is, prenominal possessives should not be able to c-command out of their DPs, rendering relevant coreferential readings acceptable. This prediction is not correct, however, as shown in (13) for Bulgarian and (14) for Macedonian.13
*Negovijati papagal uhapa Ivanivčera.
his.the parrot bit Ivan yesterday
‘Hisi parrot bit Ivani yesterday.’
*Ivanovijati papagal negoi uhapa včera.
Ivan.POSS.the parrot him bit yesterday
‘Ivani’s parrot bit himi yesterday.’
*Negovioti papagal goi grizna Jovani včera.14
his.the parrot him.CL bit Jovan yesterday
‘Hisi parrot bit Jovani yesterday.’
*Jovanovioti papagal goi grizna negoi včera.
Jovan.POSS.the parrot him.CL bit him yesterday
‘Jovani’s parrot bit himi yesterday.’
Furthermore, in both languages, as in Serbian, coreference between the prenominal possessive and the R-expression or pronoun remains unacceptable when a demonstrative (15a)/(16a) or a quantifier (15b)/(16b) is added in the structure. Under the Parameterized DP Hypothesis, these elements are expected to project DPs in a DP language and hence would be expected to enable coreference under (3).15
*Tozinegovi papagal uhapa Ivani včera.
this his parrot bit Ivan yesterday
‘This parrot of hisi bit Ivani yesterday.’
*VsičkiteIvanovii papagali negoi uhapaha včera.
all.the Ivan.POSS parrots him bit yesterday
‘All Ivani’s parrots bit himi yesterday.’
*Tojnegovi papagal goi grizna Jovani včera.
that his parrot him.CL bit Jovan yesterday
‘That parrot of hisi bit Jovani yesterday.’
*MnoguJovanovii papagali goi griznaa negoi
many Jovan.POSS parrots him.CL bit him včera.
‘Many of Jovani’s parrots bit himi yesterday.’
These results are clearly problematic for Despić’s (2013) account of the contrast between English (1a–b) and Serbian (2a–b). Bulgarian and Macedonian, which should pattern identically to English on Despić’s accountinsofar asthey resembleitin termsof therelevantparameter (presence of DP), do not do so. Instead, in sharp contrast to expectations, they pattern identically with Serbian.
Note, however, that one could argue that DP languages differ with regard to the position of prenominal possessives. If so, such a variation must be explained somehow. A reviewer suggests that Despić’s (2011, 2013) analysis in fact predicts that the Serbian binding possibilities would be available in a DP language if the possessives were adjoined to DP; and Bulgarian and Macedonian might be such languages. The evidence for such a proposal, the reviewer continues, is the word order of the possessive and the definiteness marking. While English definite articles are freestanding prenominal elements, Bulgarian and Macedonian definite articles are postnominal suffixes. The latter linearly follow possessives: Jovanov-iot ‘Jovan.POSS-the’. To derive such a word order, the reviewer suggests that the possessives undergo adjunction to the left edge of DP. The proposal hence tracks the binding possibilities in Bulgarian and Macedonian without jeopardizing Despić’s (2011, 2013) theory.
This is certainly a possibility. However, such a proposal raises a number of questions. First, why would Bulgarian and Macedonian possessives (including referential elements) be required to adjoin to DP rather than remain in situ? And, what features would drive this?
Second, how does the learner derive this analysis? Could acquisition depend on the postnominal clitic nature of the definite article? If so, Romanian (for instance) would be expected to pattern with Bulgarian and Macedonian, which does not seem to be the case. Even though native speakers vary in their acceptability judgments of the relevant examples, there is unanimous agreement that there is a possessive form that always allows coreference, contrary to what is found in Bulgarian and Macedonian.
Third, this proposal leads us to expect that uncontroversial DP-adjoined elements should show the same behavior as possessives. ‘Only’ and ‘even’ are widely held to adjoin to DP. They are clearly adverbs in all their other occurrences in English, (17c–e).
Only the president knows the answer.
Even the president knows the answer.
John only tasted the stew.
John speaks only to Mary.
John only recently started working at home.
Under the DP-adjunction analysis, then, we should expect to see the definite article suffix attach to ‘only’/‘even’ in Bulgarian and Macedonian, as with possessives. This is not the case, however.
Samo prezidentat znae otgovora.
only president.the knows answer.the
‘Only the president knows the answer.’
Daze i prezidentat znae otgovora.
even and president.the knows answer.the
‘Even the president knows the answer.’
Samo pretsedatelot go znae odgovorot.
only president.the it knows answer.the
‘Only the president knows the answer.’
Duri i pretsedatelot go znae odgovorot.
even and president.the it knows answer.the
‘Even the president knows the answer.’
Finally, note that under the proposed DP-adjunction analysis of possessives in Bulgarian and Macedonian, the word order in (15b) is unexpected: the definite article precedes the possessive (cf. (13)): vsički-teIvanovi papagali ‘all-the Ivan.POSS parrots’.16
3 The Special Behavior of Prenominal Possessives
The above results strongly imply that the different binding potential of English and Serbian prenominal possessives cannot be attributed to the presence vs. absence of DP in the two languages, as proposed by Despić (2013). But if this is not the source of the crosslinguistic difference, what is? Triplets like (20a–c) suggest a potential answer. We have seen that a Serbian prenominal possessive cannot be coreferential with an R-expression elsewhere in the clause, whether the former is a possessor in the subject phrase (20a) or embedded within it (20b). At the same time, (20c) shows that c-command remains relevant since a comparable pronoun in object position admits coreference with a possessive embedded in PP as expected under (3).17
*[Markovai majka] je zagrlila njegai.
Marko.POSS mother AUX hugged him
‘Markoi’s mother hugged himi.’
*[Prijatelj [Markovei majke]] je zagrlio njegai.
friend Marko.POSS mother AUX hugged him
‘A friend of Markoi’s mother hugged himi.’
[Prijatelj [PP od Marka Markovićai]] je zagrlio
friend from Marko Markovic AUX hugged
‘A friend of Marko Markovici hugged himi.’
These facts suggest that it is some property of Serbian (Bulgarian and Macedonian) prenominal possessives in particular that is responsible for the noncoreference facts in (20a–b). Roughly speaking, prenominal possessives seem to behave ‘‘as if’’ they occupied a very high c-commanding position in the nominal, even when their surface position is embedded. This point naturally recalls Szabolcsi’s (1983) analysis of Hungarian dative possessors, which can raise to an edge position within their nominals and even escape them altogether under certain conditions. Perhaps Serbian (Bulgarian and Macedonian) prenominal possessors uniformly raise at LF to the edge of their largest containing nominal (21), a position from which they might c-command the remainder of the clause under (3), with binding potentials constrained appropriately.
Markovei [prijatelj [
Markoveimajke]] je zagrlio
Marko.POSSmother AUX hugged
‘Markoi’s mother’s friend hugged himi.’
Normal object pronouns (20c) would, by contrast, remain in situ and hence would behave like normal embedded phrases with respect to c-command. I will not attempt to pursue this proposal further, but simply observe that it seems warranted given the results discussed above and given the clear problems with the proposal in Despić 2013.
I would like to thank John Bailyn, Dan Finer, Norbert Hornstein, and the anonymous reviewers for their constructive and very careful comments and suggestions. They have helped me to sharpen the squib significantly. Special thanks go to my advisor Richard Larson for his constructive criticism and constant encouragement.
1 See Despić 2011 for additional assumptions the author makes to account for the relevant data.
2 In the works of Bošković (e.g., 2005, 2008), Serbian determiners are argued to be adjectives or adjective-like elements. This proposal is based on the claim that Serbian determiners morphologically and syntactically behave like adjectives in every respect. See the works cited for details of this proposal.
3 A reviewer points out that native speakers do not unanimously share the reported acceptability judgments for Serbian examples throughout the squib and suggests that intonation structure might be an interfering factor. A con- trolled acceptability study of the relevant data, which would manipulate intona- tion structure, is needed to settle the issue. Note that the reported acceptability judgments of Serbian examples in section 1 are taken directly from Despić's work. The data introduced in section 2 were collected in written form from 12 native speakers of Serbian with no prior linguistic training. The new data as- sume neutral intonation.
4 A reviewer suggests avoiding examples with the verb smatrati ‘consider’ for binding facts since they might involve a small clause, which in turn may affect Condition B. I use examples with this verb to illustrate the facts that Despić (2011, 2013) builds his analysis on while keeping English and Serbian examples as similar as possible. Despić presents examples with other verbs as well, with no difference in acceptability. To avoid the potential interference of the small clause structure with the relevant binding possibilities, I will use the latter type of examples in section 2, where I introduce new data.
5 The pronoun used in the Serbian example (2b) is a strong/full pronoun, njega ‘him’. It has a corresponding weak/clitic form, ga ‘him’. Despić (2009: 22n4) claims that the use of the strong vs. weak form of a pronoun does not interfere with the acceptability of the relevant coreference. In particular, he says that with the use of a clitic, ‘‘the sentence somehow ‘improves’ (but still stays ungrammatical).’’
6Kayne (1994) also assumes that the Spec position of the DP dominating PossP contains an operator responsible for operator-variable binding of pronouns.
8 The term agreeing refers to agreement of a determiner and a noun it modifies in case and φ-features (Despić 2011:72n21).
9Despić (2013) also takes these data to dismiss the weaker version of the Universal DP Hypothesis, according to which DP is present in Serbian only when there is an overtly realized item in Spec,DP, that is, a demonstrative.
Despić (2013) also shows that the adjunct-based approach he is advocating is superior to the Cinqueian cartographic approach (Cinque and Rizzi 2009) since no binding differences are found when an ordinary adjective cooccurs with a possessive.
10 Nonagreeing quantifiers are elements that do not agree in case and φ-features with a noun. They appear with nouns in genitive case and are hence referred to as ‘‘genitives of quantification.’’ See Bošković 2012 for a proposed structure for these elements.
11 (10b) uses PossP for Bulgarian in parallel to (4) in English; nothing hinges on this, however. PossP could be replaced with NP with no change in binding predictions under (3); compare (10a). Note that irrelevant structure in (10b) is suppressed. The full structure is [ PossP-ov [NPpapagal]], on a par with [PossP’s [NPfather]] in (4).
12 A reviewer points out that Despić (2011, 2013) does not advocate such an analysis in his work and therefore cannot be held responsible for claims made by other authors. The analysis shown above is a standard analysis of relational nouns across languages: the N complement supplies the value of the y variable for a relational noun (e.g., friend (x,y)). One could argue for the alternative, namely, that relational nouns take adjuncts rather than complements, as the reviewer suggests. However, such a proposal needs to be verified.
13 Bulgarian data are from Boris Harizanov, Adrian Krastev, Angelina Markova, Radostina Petkova, and Petya Osenova (pers. comm.). Macedonian data are from Andrijana Pavlova, Kiril Ribarov, and Ilina Stojanovska (pers. comm.).
Note that one native speaker of Bulgarian reports different acceptability judgments. In particular, this speaker claims that coreference between the prenominal possessive and the R-expression is acceptable whereas coreference with the pronoun is not. This variation in acceptability judgments calls for a controlled acceptability judgment study that would settle the issue.
14 Macedonian requires object clitic doubling when the object is definite; hence, there is an object clitic in these examples.
15 In Bulgarian, I use the quantifier vsički ‘all’ because the quantifier mnogo ‘many’ requires a prepositional phrase: mnogo ot X ‘many of X’.
16 See Mladenova 2007 for arguments that the universal quantifier has a definite article suffix.
17 Note that in (20c), I use the phrase Marko Markovic. I do this to avoid a morphological restriction interfering with the acceptability of the example. Namely, if a possessive is a one-word element, the preference is to use a prenominal form (Markov prijatelj ‘Marko’s friend’) rather than a postnominal one (prijatelj (od) Marka ‘a friend of Marko’); however, if a possessive is a multiword element, it can only appear postnominally (with or without a preposition). See Corbett 1987 and Zlatić 1997, among other works, for details on this morphological restriction.