South Slavic Clitic Placement is Still Syntactic
			Steven Franks

	There has been much recent debate about whether special clitic 
placement in general, and in the South Slavic (SSl) languages, in 
particular, can be handled exclusively through the exploitation of 
familiar syntactic categories and movement mechanisms, or whether some 
special phonological reordering is required, such as HalpernUs 
(1992/1995) RProsodic InversionS (PI). This debate has primarily 
concentrated on Serbian/Croatian (SC) and, to a lesser extent, Bulgarian 
(Bg). In this paper I champion the syntactic approach, reviewing existing 
arguments why SC clitic placement demands syntactic means and presenting 
some novel ones. I also extend the analysis to Slovenian (Sln) and 
Macedonian (Mac), showing how these languages differ from SC and Bg, 
respectively, in one minor respect: clitics are prosodically neutral 
rather than enclitic. I assume Greed and argue that SSl special clitics 
require syntactic support in the form of head adjunction, and that when 
no head independently moves up to them to check its own features, the 
clitics themselves move, raising, or when that option is unavailable, 
actually lowering overtly to a host head.
	Adopting the proposal that SSl clitics are all functional heads 
in the extended projection of V, I first summarize arguments from Franks 
and Progovac (1994), avar and Wilder (1994), and Progovac (1996), which 
all present evidence that independently necessary syntactic principles 
regulate regulate clitic placement and that PI encounters serious 
difficulties. The ellipsis data in (1), due to Stjepanovi
 (in progress), show that different clitics occupy distinct head 
positions; the elided material is presumably VP (or AuxP) in (1a), AgrOP 
in (1b), AgrIOP in (1c), and no XP in (1d). Note that this cannot be 
handled in a system such as Anderson (1995), in which the clitics are all 
inserted simultaneously.
(1)	a.	Ona	mu	ga je 		dala,	a	i	ja	sam	mu	ga	[dala]
		she     him.dat	it.acc	aux.3sg	gave	and	also	I	aux.1g	him.dat it.acc	gave
		TShe gave it to him and I did too.U
	b.	Ona mu ga je dala, a i ja sam mu [ga dala]	
	c.	..., a i ja sam [mu ga dala]
	d.	?*Ona mu ga je dala, a i ja sam [mu] ga [dala]
	The program of placing SSl clitics syntactically faces a 
challenge in that there are some obvious phonological effects which need 
to be built into the system. I defend a RfilteringS approach, in which 
the results of strictly syntactic movements are modulated by the 
phonology. Thus, the syntax will generate SC (2) but this crashes at PF 
because the clitics are not supported prosodically.
(2)	*Mi	je  napisao		taj	pesnik		knjigu.
	me.dat	aux.3sg	wrote		that	poet		book
	TThat poet wrote me a book.U
In colloquial Sln, however, clitics can appear initially when 
contextually determined, as in (3). Sln clitics, I claim, appear in the 
same positions as SC ones, but are prosodically neutral.
(3)	a.	(Ali)	mu	ga		dajes`?									b.	Ga			pelje	kot	otroka.
		Q 		him.dat	it.acc 	give												him.acc	leads	like	child		
		T(Are) you giving him it ?U													
T(She) leads him like a child.U
	Bg pronominal clitics are verb-adjacent rather than in second 
position, but like SC they can only be enclitic. Consider the paradigm of 
possible permutations in (4).
(4)	a.	Vera	mi			go		dade	vc`era.						b.	
Vc`era Vera mi go dade.
		Vera	me.dat	it.acc	gave	yesterday
		TVera gave me it yesterdayU
	c.	*Vera mi go vc`era dade.												d.	
*Mi go dade Vera vc`era.
The only unacceptable variants are (4c), in which the clitics are not 
adjacent to the verb, and (4d), in which there is nothing to the left of 
the clitic cluster. Note that anything can satisfy the enclitic 
requirement; typically the verb will raise past the clitics, but even the 
conjunction i TandU is adequate. Both are illustrated in the sentence in (5).
(5)	[Sres`tnax	Ivan.]	Sprjax				go,				i		mu				
kazax ...
	met.1sg		Ivan	stopped.1sg	him.acc	and	him.dat	told.1sg
	T(I) met Ivan. (I) stopped him, and (I) told him ...U
Further confirmation of this can be seen in (6), where the clitics 
precede the imperative if and only if there is somethingQanythingQfor 
them to be enclitic on:
(6)	a.	Donesi			mi			go!												b.	Ne	mi			go		nosi!		
		bring.impv	me.dat	it.acc													neg	me.dat	it.acc	bring.impv
		TBring it to me!U																			
TDonUt bring it to me!U
	c.	*Ne nosi mi go!
One might imagine handling this paradigm by having the verb move to the 
left of the clitics to avoid a PF crash. But this is false under standard 
minimalist assumptions for two reasons: such movement would neither be 
feature driven nor would it respect Greed.
	More serious is the problem presented by Macedonian, in which 
(4d) is acceptable, the relevant difference between Mac and Bg being once 
again that clitics are prosodically neutral only in the former. In Mac, 
on the other hand, clitics follow imperative forms so that (6c) rather 
than (6b) is grammatical; see Alexander (1994). V movement past the 
clitics therefore cannot be for PF reasons; the impossibility of invoking 
PI here thus sheds grave doubt on the PI account of comparable phenomena 
in SC and Bg. I claim that the finite V moves to  AgrS to check features 
and, since these are weak, must wait until LF everything else being 
equal. In Bg, but not Mac, the finite verb thus excorporates from the 
clitic cluster and adjoins to AgrS.  Following Watanabe (1993) and Bodkovi
 (1995), excorporation out of an adjunction structure obligatorily occurs 
when the features of the adjoined elements have already been checked.) 
This movement ordinarily takes place at LF, under procrastinate, but can 
be forced to occur overtly in order for the derivation to converge at PF. 
Exactly the same reasoning pertains to the imperative in Bg, except that 
the relevant features are those of the Imperative head. In Macedonian, on 
the other hand, the Imperative features are strong, deriving the order in 
(6c) overtly. Clearly the problem cannot be any prosodic deficiency, 
since these clitics in Mac are prosodically neutral and otherwise happily 
appear as proclitics, i.e. before all finite verbs, even in initial position.
	Another context in which initial clitics are unacceptable in Mac 
is with dative possessive clitics in DP, which I treat as the head of 
AgrIO, definite DP optionally selecting for AgrIOP.
(8)	a.	z`ena mi  Tmy wifeU															b.	
pomladiot ni sin Tour younger sonU
	c.	ne mnogu postarite  deca Ther not much older childrenU
Movement of the clitics rather than the host is required to accommodate 
(8), in which the dative AgrIO clitic appears immediately after the 
highest head in a definite DP (taking AP to dominate NP, and definiteness 
to be inflectional). There is no phonological motivation here, hence I 
propose that clitic movement is necessary to avoid violation of the 
syntactic requirement that special clitics need to be syntactically 
supported (through adjunction). Since no head has any independent reason 
to move up to adjoin to AgrIO, it is stranded and, assuming Greed, has to 
move itself to be supported. Various ways of effecting this are 
considered, the simplest being to lower AgrIO to the next head down, 
namely X in (9).
(9)	[DP D[+def] [AgrIOP tAgrIO [XP ...X+AgrIO... ]]]