Closest Conjunct Agreement in Serbo-Croatian: A Rule-Ordering Account Zorica Puškar and Andrew Murphy Claim: Closest Conjunct Agreement (CCA) poses a problem for standard theories of Agree as it seems to be sensitive to linear proximity rather than c-command. Some recent works complicate the Agree mechanism to circumvent violating Minimality (Boškovi´c 2009) or to make reference to linearity (Bhatt & Walkow 2013, Marušiˇc et al. to appear). Based on new empirical data, we propose that all observed patterns of conjunct agreement in Serbo-Croatian (SC) are derived in syntax from the order in which the basic operations Agree, Merge and Move apply at &P, and subsequently, TP. Data: Our account is based new data from SC we obtained in a questionnaire study with 60 native speakers. We identified the following patterns in SC: (1) Resolved Agreement (RA): Full agreement with both conjuncts usually resulting in default agreement (masculine), (2) Last Conjunct Agreement (LCA): Agreement with the NP of a preverbal &P linearly closest to the verb, (3) (postverbal) First Conjunct Agreement (FCA): Agreement with the linearly closest NP of an &P in postverbal position, (4) (preverbal) First Conjunct Agreement: Agreement with the structurally highest NP of an &P in preverbal position (previously unattested for SC, and problematic for previous accounts of conjunct agreement in SC (cf. Boškovi´c 2009). Furthermore, LCA in postverbal position is entirely unattested (5). Background: We assume that the head of an asymmetric conjunct phrase & agrees in gender with its complements and projects these values to the &P node. Additional assumptions: (i) Syntactic computation is driven by four basic operations: External Merge (M ERGE), Internal Merge (M OVE), Downward (Head-Complement) Agree ( ↓AGR↓) and Upward (Spec-Head) Agree ( ↑AGR↑) (we follow Abels 2012 in claiming that a head can carry out both Agree operations). (ii) The order of operations is in principle free, but not all of them result in a convergent derivation. (iii) Uniform Order of Operations: Once an order of operations is chosen, it is maintained at each cycle of the derivation. The order of operations on the T head must not diverge from the order inside the &P. (iv) M OVE is an optional operation that applies as early as possible if it does not have a negative effect. (v) Gender agreement on & is fallible (cf. movement-conditioned participle agreement in French; Kayne 1989). Conversely, φ-agreement is infallible and therefore its failure leads to a derivation crash. Proposal: The four operations above can interact in various ways: if M ERGE precedes both Agree operations, it will feed them allowing them to apply. If Agree were to apply before M ERGE, there would not yet be anything for the head to agree with and Agree would be counterfed. Thus, it is possible for the & head to agree with both, one or none of its arguments depending on the order in which operations are carried out, and assuming (v) above, counterfeeding of Agree for gender at &P does not lead to a crash. The order at &P must be maintained at TP. We will show that assuming (iii) will allow us to derive whether the &P moves to Spec-TP (the EPP property). If ↑AGR↑ is ordered before ↓AGR↓, M OVE has to apply to feed Agree, enforcing the &P movement to Spec-TP. Conversely, if ↓AGR↓ is ordered before ↑AGR↑, M OVE cannot apply, otherwise it bleeds ↓AGR↓, thus the &P stays in a postverbal position. Thus, the patterns of conjunct agreement identified in SC can be derived by the orders in (6). Resolved Agreement (1): At &P, the (a) order in (6) applies resulting in feeding of both Agree operations since M ERGE applies first. Thus the & agrees with both conjuncts. Whether the &P moves to Spec-TP depends on whether ↑AGR↑ or ↓AGR↓ applies first. If ↑AGR↑ applies first, M OVE must precede it to prevent a crash. This is ruled out in the converse case. Last Conjunct Agreement (2): The (b) order in (6) results in counterfeeding ↑AGR↑ since it applies too early to be fed by M ERGE. As a result, the &P only projects the value of the second conjunct (7). At TP, ↑AGR↑ precedes ↓AGR↓, thus M OVE must apply in order to feed it (8). If it did not apply, it would result in a crash since ↑AGR↑ would not find a goal (9). This derives the fact LCA can only happen preverbally, and excludes postverbal LCA (5). This pattern is problematic for other accounts as the fact that T seems to look past the structurally higher first conjunct looks like a Minimality violation. First Conjunct Agreement (3): (6)(c) leads to counterfeeding of ↓AGR↓ since it applies before M ERGE. The &P will therefore only bear the features of the first conjunct (10). M OVE cannot apply at T since it would bleed ↓AGR↓ and result in a crash (11). Absence of M OVE will result postverbal FCA. Additionally, we will show that postverbal FCA can be

derived by counterfeeding of both Agree operations by the order in (6) (d). Here, the &P’s gender feature will remain unvalued and T will find the structurally higher first NP (12). The final order (6)(e) derives the previously unattested pattern of preverbal FCA (attested in Slovenian; Marušiˇc et al., and predicted to be impossible by Boškovi´c 2009). In sum, CCA, which has long been assumed to be a linear phenomenon, can be adequately captured without making reference to linearity under the assumption that there order in which syntactic operations apply can fail to apply. We show that all possible orders arising from the factorial typology of four operations can derive all attested patterns of conjunct agreement in Serbo-Croatian and rule out the one unattested order. Thus, the contribution of this paper is both empirical and theoretical as it shows that what look like Minimality violations or linearity effects on the surface can in fact be accounted for by the interaction of standard syntactic operations. (1)

(2)

(3)

[ &P [ NP1 Otac] i [ NP2 sin]] su gledali utakmicu. father.MSG and son.MSG are watch.PRT. MPL game ‘Father and son watched the game.’ [ &P [ NP1 Sva odela] i [ NP2 sve haljine]] su juˇce prodate. all suit.NPL and all dress.FPL are yesterday sell.PRT. FPL ‘All suits and all dresses were sold yesterday.’ Juˇce su prodata [ &P [ NP1 sva odela] i [ NP2 sve haljine]]. yesterday are sell.PRT. NPL all suit.NPL and all dress.FPL

(RA)

(preverbal LCA) (postverbal FCA)

(4)

[ &P [ NP1 Krave] i [ NP2 telad]] su mirno pasle po polju. cow.FPL and calf.NPL are peacefully graze.PRT. FPL across field ‘Cows and calves grazed peacefully in the field.’ (preverbal FCA)

(5)

*Juˇce su prodate [ &P [ NP1 sva odela] i [ NP2 sve haljine]. yesterday are sell.PRT. FPL all suit.NPL and all dress.FPL ‘All suits and all dresses were sold yesterday.’

(6)

(7)

Orders deriving patterns of CCA in Serbo-Croatian: a. (M OVE) > M ERGE > { ↑AGR↑ , b. *(M OVE) > ↑AGR↑ > M ERGE > c. (*M OVE) > ↓AGR↓ > M ERGE > d. (*M OVE) > ↓AGR↓ > ↑AGR↑ > e. (M OVE) > ↑AGR↑ > ↓AGR↓ > LCA (&P) (M OVE ) > ↑AGR↑ > M ERGE > ↓AGR↓: (8) [ &PF NPN ➋ [ &′ &F ➋ NPF ] ] ✘ ➊

↓AGR↓} ↓AGR↓ ↑AGR↑ M ERGE M ERGE

(*postverbal LCA)

→ Resolved Agreement → LCA (preverbal) → FCA (postverbal) → FCA (postverbal) → FCA (preverbal)

LCA (TP) M OVE > (M ERGE ) > ↑AGR↑ > ↓AGR↓: [ TP [ &PF NP [ &′ & NP ] ] TF [ vP t&P [ VP V NP]]] ➋

➌ ➊

(9)

(10) LCA (TP) (M ERGE ) > ↑AGR↑ > ↓AGR↓: *[TP T [ vP [ &PF NP1 [ &′ & NP ] ] [ VP V NP]]]

FCA (&P) (M OVE ) ↓AGR↓ > M ERGE > ↑AGR↑: [ &PN NPN ➋ [ &′ & ➋ NPF ] ]

✘ ➊ (11)

➌

FCA (TP) M OVE > ↓AGR↓ > M ERGE > ↑AGR↑: (12) *[TP [ &PN NP1 [ &′ & NP ] ] T [ vP t&P [ VP V NP]]] ✘➋

✘ ➊

FCA2 (TP) (*M OVE ) > ↓AGR↓ > ↑AGR↑ : [ TP TN [ vP [ &P_ NPN [ &′ & NPF ] ] [ VP V NP]]] ✘

➊

Selected References: Boškovi´c, Ž. (2009). Unifying first and last conjunct agreement. NLLT 27/3: 455– 496. • Bhatt, R. & M. Walkow. (2013)‘Locating agreement in grammar: an argument from agreement in conjunctions’. NLLT 31/4: 951–1013. • Marušiˇc, F., A. Nevins, & B, Badecker. ‘The Grammars of Conjunction Agreement in Slovenian’. (to appear) Syntax.