Clitic Placement, Prosody, and the Bulgarian Verbal Complex

Abstract

This paper compares competing ways of understanding the fact that clitics but nothing else freely and necessarily intervene between the two verbal heads in Bulgarian compound tenses of the type [participle + (clitics +) auxiliary]. These involve a participle fronted for focus reasons. The problem is then how the clitics get in the middle. I argue that prosodic and morphological approaches are not adequate, nor is any PF-filtering necessitated. Instead, the complex head structure [[participle + clitics] + [auxiliary]] must be created syntactically, with the participle adjoining to the clitics before the resulting complex adjoins to the auxiliary.

1. The Participle-Fronting Problem

This article explores conceptual and analytical issues surrounding the possibility of interpolating clitics between a focally fronted verbal participle and an auxiliary verb in Bulgarian (Bg), as shown in (1).¹

1. 1. Procela mu ja beše.

   read_FEM him_DAT it_ACC had_3SG

   'She had READ it to him (and not sung it).'

The existence of this construction is used as a springboard for consideration and comparison of various mechanisms of clitic placement that have been proposed in the literature. While the syntactic operation of participle fronting has been well studied, the fact that clitics intervene in examples such as (1) has not been fully appreciated.² Do they appear after the participle because of a PF requirement, presumably prosodically motivated, that prevents the clitics from being initial, or do they follow the participle because that is where they are left by the syntax? In this paper I examine the data and conclude that the latter kind of approach is the correct one. The former alternative would be to position the clitics between the participle and the auxiliary through some kind of non-syntactically driven movement. I will argue that (1) is best handled not through PF-movement, either in the Prosodic Inversion sense put forward in Halpern 1992 or in the Morphological Merger sense articulated in Embick and Noyer 2001, but rather through successive syntactic head movements. I therefore ascribe (1) to a syntactically created complex head structure, as in (2a) or (2b); I will ultimately argue for (2b) over (2a).

• 2.

  1. a. [pročela + [[mu ja] + [beše]]]

  2. b. [[pročela + [mu ja]]+ [beše]]

The first half of the paper considers more general issues of clitic placement in Bg, providing a context for the eventual reexamination of (1) in more depth in section 5.

2. The Tobler-Mussafia Effect

Within a clause, clitics in Bulgarian ordinarily appear immediately before the verb, regardless of how much material precedes them.³

1. 3. Včera v gradinata Olga sigurno mu dade snimkite.

   yesterday in garden_DEF Olga surely him_DAT gave pictures_DEF

   'Yesterday, in the garden, Olga surely gave him the pictures.'

It is worth noting that this distribution distinguishes them from second position (or Wackernagel) clitics, which are found for example in the neighboring South Slavic language Bosnian/Croatian/Serbian (BCS). Moreover, as in other languages, when there is more than one clitic they appear in a fixed order. Consider some simple Bg sentences with several clitics, such as (4):⁴

• 4.

  1. a. Ti si mu gi pokazvala.

     you AUX_2SGhim_DAT them_ACC shown_FEM

     'You have shown him them.'

  2. b. Tja mu gi e pokazvala.

     she him_DAT them_ACC AUX_3SG shown_FEM

     'She has shown him them.'

Descriptively, clitic auxiliaries precede the pronominal clitics, except for third singular e, which follows them, and, among the pronominals, the dative clitic precedes the accusative one, as schematized in (5).

1. 5. AUX (except e) >> DAT >> ACC >> e

When, however, no overt element precedes the clitic group, they must follow the verb instead. In the Romance linguistics tradition, I will refer to this phenomenon as the Tobler-Mussafia (TM) effect. Compare the examples in (6) with those in (4):

• 6.

  1. a. Pokazvala si mu gi.

     shown_FEM AUX_2SGhim_DAT them_ACC

     'You have shown him them.'

     (*Si mu gi pokazvala.)

  2. b. Pokazvala mu gi e.

     shown_FEM him_DAT them_ACC AUX_3SG

     'She has shown him them.'

     (*Mu gi e pokazvala.)

Since the generalization is that the clitics are pre-verbal unless this would leave them in absolute initial position within the clause, a promising way of understanding (6) is to derive it through some kind of post-syntactic local reordering operation.⁵ That is, the syntax would produce an output in which the clitics precede the verb, and this order would be adjusted on the PF-side to comply with the prohibition in Bg against initial clitics, as in (7).⁶

• 7.

  1. a. // [si mu gi] [pokazvala] → // [pokazvala] [si mu gi]

  2. b. // [mu gi e] [pokazvala] → // [pokazvala] [mu gi e]

3. PF-Side Approaches

In this section, two types of PF-side operations that likely “reorder” Bg clitics are evaluated and rejected as candidates for producing (1).⁷ I assume that the syntax proper makes no statements about linear order per se. Linearization is instead a property imposed on language by virtue of the temporal exigencies of articulation, and ordering arises only as part of the Spell-Out process mapping from syntax to PF.⁸As Embick and Noyer (2001: 556–57) put it: “Syntax generates and moves terminals according to its own principles and is oblivious to morphophonological processes. PF takes the output of the syntax and resolves morphophonological dependencies according to its own principles.”⁹ Given this, the question raised by (1) is whether the word order simply reflects the output of the syntax, as depicted in (2), or whether it embodies a mismatch between syntax and morphology which can only be derived through some post-syntactic mechanism that reorders the clitics, along the general lines of (8).

1. 8. [mu ja] [pročela] [beše] → [pročela] [mu ja] [beše]

3.1. Prosodic Inversion and li

A long standing debate in the clitic literature concerns the need for prosodically motivated reordering of clitics with respect to their host. Those who adopt Prosodic Inversion (PI) argue that whenever the syntax leaves an enclitic stranded in initial position, the clitic is moved to the right edge of the adjacent prosodic word to its right, thereby saving the structure. Since the relevant elements are defined in prosodic terms, this inversion has traditionally been understood as a last-resort PF-side operation, as in the studies of clitics by Halpern (1992), Percus (1993), and Schütze (1994). These works concentrate on the applicability of PI as the explanation for why clitics can split nominal expressions (typically NPs, but also PPs where the P is proclitic) in BCS. Since the consensus of subsequent research is, I believe, that there is no real need to invoke PI to accommodate the splitting facts in BCS,¹⁰ to illustrate the effects of PI I turn instead to a far more persuasive case for it in Bg.

In Bg, as in all the Slavic languages which have it, the question particle li is unequivocally enclitic: it is always pronounced at the right edge of a host prosodic word.¹¹ In neutral Yes/No questions, this will be the verb:

1. 9. Dade li Ana knigata na Petko?

   gave Q Ana book_DEF to Petko

   'Did Ana give the book to Petko?'

Also, as in the other languages, the proclitic negative particle ne forms a prosodic word with the element to its right:

1. 10. Ne dade li Ana knigata na Petko?

   NEG gave Q Ana book_DEF to Petko

   'Didn't Ana give the book to Petko?'

However, since ne is post-stressing in Bg, if the following element is a clitic, then ne and the now stressed clitic constitute an independent prosodic word and provides a host for li, in (11) from Rivero 1993.¹²

1. 11. Ne mú li ispratix kniga?

   NEG him_DAT Q sent_1SG book

   'Didn't I send him a book?'

This is true regardless of what the first clitic in the sequence happens to be or how many there are, as shown by (12).

• 12.

  1. a. Ne mú li ja dade Ana?

     NEG him_DAT Q it_ACC gave Ana

     'Didn't Ana give it to him?'

  2. b. Ne si li mu gi pokazvala?

     NEG AUX_2SGQ him_DAT them_ACC given

     'Haven't you shown them to him?'

  3. c. Ne mú li gi e pokazvala?

     NEG him_DAT Q them_ACC AUX_3SGshowed_FEM

     'Hasn't she shown them to him?'

This striking property of li is most straightforwardly handled through Prosodic Inversion. If the syntax leaves li, which is a simple enclitic, in C⁰ with no host to its left, at Spell-Out it is linearized at the right edge of the prosodic word to its right.¹³ The result is that, when the prosodic conditions so dictate, li ends up splitting clitic sequences that are otherwise never separable.¹⁴ Thus (12c) has the prosodic structure in (13), where ω indicates a prosodic word.

1. 13. li [_ω ne mú] [_ω gi e pokázvala] → [[_ω ne mú] + li] [_ω gi epokázvala]

That is, li is enclitic on ne mu and gi e is proclitic on pokazvala. The participle, then, does not move to C⁰, but rather is left lower by the syntax, immediately to the right of li.¹⁵

The fact that li splits the two pronominal clitics indicates that its placement must be purely prosodic, since even aspectual adverbials such as veče 'already' or particles like už, maj 'maybe' that sometimes break up the clitics + verb sequence can never do this.¹⁶

• 14.

  1. a. Az sâm už ti go dala.

     I AUX_1SG maybe you_DATit_ACC given_FEM

     'I have maybe already given it to you.'

  2. b. *Az sâm ti už go dala.

  3. c. Az sâm ti go už dala.

In sum, the grammar of Bg appears to countenance PI for li. We should therefore ask whether the TM effect is similarly amenable to a prosodically driven account.

3.2. The Status of Clitic Non-Initiality in Bulgarian and Macedonian

It is a relatively straightforward matter to dismiss this same kind of PI as the force behind whatever causes the clitics to follow the verb in (6), repeated as (15).

• 15.

  1. a. Pokazvala si mu gi.

     shown_FEM AUX_2SGhim_DAT them_ACC

     'You have shown him them.'

  2. b. Pokazvala mu gi e.

     shown_FEM him_DAT them_ACC AUX_3SG

     'She has shown him them.'

Although most accounts motivate it with the assumption that the pronominal and auxiliary clitics are prosodically enclitic, there is clear evidence that this cannot be at the root of the TM reordering phenomenon. For one thing, as discussed in Franks and Bošković 2001 inter alia, elements such as i 'and' which cannot serve as prosodic hosts nonetheless obviate TM:¹⁷

1. 16. Kupix na Ivan edna kniga i mu ja pročetox.

   bought_1SG for Ivan one book i him_DAT it_ACC read_1SG

   'I bought a book for Ivan and read it to him.'

As observed in Pancheva 2005: 114, the clitics can be supported in either direction and, in fact, are generally prosodically parsed as proclitic on what follows them. Even more telling are the examples in (17), where the clitics remain in front of the verb even when preceded by syntactic units which necessarily induce an Intonational Phrase (I-phrase) boundary; cf. Nespor and Vogel 1986:¹⁸

• 17.

  1. a. Petko, edin moj prijatel, mi go dade.

     Peter a my friend me_DAT him_ACC gave

     'Petko, a friend of mine, gave it to me.'

  2. b. Ako njakoj, kojto e s men, mi dade njakolko

     if someone who is with me me_DAT gives several

     čipa...

     chips

     'If someone who is with me gives me some chips...'

  3. c. Ivan, spored Marija,go viždam vseki den.

     Ivan according Maria him_ACC see_1SG every day

     '(Talking about) Ivan, according to Maria, I see him every day.'

  4. d. Sled kato si živjal dosta s edna žena, ja

     after as AUX_2SG lived enough with a woman her_ACC

     poznavaš mnogo dobre.

     know_2SG very well

     'After you have lived with a woman enough, you know her very well.'

This telling fact, sometimes noted in passing (e.g., by Bošković (2001: 218, fn 7), by Pancheva (2005: 114, fn 7)) but typically overlooked, might suggest that the appropriate generalization should not be couched in prosodic terms. The examples in (17), in which the clitics are initial in their I-phrases, show that TM is neither a consequence of immutable lexical enclitic status, as is typically claimed, nor is it a consequence of some need for the clitics to be “second” in their I-phrase, as Pancheva (2005) proposes. The TM effect in Bg is apparently sensitive to initiality in some other sense, which has not previously been fully appreciated. I argue that this initiality should nonetheless still be couched in terms of prosodic terms.

One traditional insight about “second position effects,” often capitalized upon in Optimality Theoretic accounts of clitic placement, is that these are a minimal displacement from absolute initiality within some domain.¹⁹ The grammaticality of (17), therefore, can only be taken as evidence that the I-phrase is not the appropriate domain, and not that the domain should be defined in syntactic rather than prosodic terms. There are of course other prosodic domains. In their seminal workNespor and Vogel (1986) treat a range of prosodic domains, which, in accordance with the Strict Layer Hypothesis are nested (rather than overlapping).²⁰ And the availability of additional domains opens the door also to expressing in prosodic terms the relevant notion for Bg clitic non-initiality. Indeed, the digression in the previous section on the behavior of li drew attention to the fact that the domain of non-initiality for li is the Prosodic Word, while for BCS second position enclitics it is the traditional I-phrase. Returning, then, to the question of the domain of non-initiality for the Bg clitics, my proposal is that the correct domain is the next larger unit of prosodic organization, the Utterance.²¹ Hence the presence of an I-phrase boundary, as in the Bg examples in (17), does not induce the relevant kind of non-initiality. The clitics are still non-initial with their Utterances.

An argument for this kind of approach can be based on a comparison between Bulgarian and closely related Macedonian (Mac). In Franks (in press) I argue that Mac differs from Bg in that there is no such prohibition against Utterance initiality. Thus, the Mac versions of Bg (6)/(15) are as follows, essentially the opposite of Bg:

• 18.

  1. a. Si mu gi pokazvala. [Mac]

     AUX_2SG him_DAT them_ACC showed_FEM

     '(They say that) you showed them to him.'

     (*Pokazvala si mu gi.)

  2. b. Mu gi pokazvala.

     him_DAT them_ACC showed_FEM

     '(They say that) she showed them to him.'

     (*Pokazvala mu gi.)

I take the absence of the TM effect in Mac to mean that Bg obeys a restriction which Mac lacks. The operation in (7), which in Bg (re)linearizes the sequence with the clitic group after pokazvala, is thus not needed in Mac. In short, the clitic syntaxes of Bg and Mac are similar here, but the prohibition against Utterance initial clitics—operative in the former but not the latter—leads to a PF-side adjustment in Bg.²² Hence, whenever the two languages differ with respect to clitic placement, a PF-side adjustment has taken place in Bg that is absent in Mac. That is all that the Tobler-Mussafia effect is. As a consequence, the derivation of the Bg sentence inevitably passes through a Mac-like stage whenever there is a difference with respect to clitic placement. Following this reasoning, Bg (6)/(15) must look like Mac (18) before the adjustment necessitated by the need not to pronounce the clitics Utterance initially.

3.3. Some PF-Side Derivations and Operations

Let us return in this light to the li facts discussed in section 3.1. We saw that the negation element ne in Bg creates a separate prosodic word with a following clitic.²³ In the absence of negation, however, the order is as in (19):

1. 19. Dade li mu ja Ana? [good in Bg but not in Mac]

   gave Q him_DAT it_ACC Ana

   'Did Ana give it to him?'

   (*Mu ja dade li Ana?) [good in Mac but not in Bg]

Thus, when the verb is followed by clitics, li separates them from the verb. Note that while this is true of Bg, in Mac the clitics precede the verb, hence it is the starred order in (19) that is grammatical in this language. Extending the approach taken in the preceding paragraph, these data indicate that li precedes (mu ja) dade and then PF-side adjustments apply as needed.

In Franks 2006b, I argue for a strongly derivational account of the word order in (19) as follows: (a) the clitics, as always, are linearized to precede the verb, (b) li, since it merges in C⁰, precedes them; (c) li is then linearized with respect to the prosodic word mu ja dade; (d) TM applies to Utterance initial mu ja. These derivational steps, where the intonational units are ω for Prosodic Word; CG for Clitic Group;²⁴ # for Intonational Phrase; and υ for Utterance, are as follows:

• 20.

  1. a. [[mu ja]cg[dade]_ω] [Ana] _ω [output with mu ja linearized]

  2. b. li [[mu ja]cg [dade]_ω ] [Ana]_ω [li merged but not linearized]

  3. c. [[[mu ja]cg [dade]_ω ] li] [Ana]_ω [li linearized; Mac final output]

  4. d. υ #[[[ dade]_ω li] [mu go]cg] [Ana]_ω# υ [clitics undergo TM; Bg final output]

The general idea, as discussed below, is that all heads are first linearized to the left, following Kayne's Linear Correspondence Axiom, but the pronominal clitics are not considered Utterance-initial until after li has been merged and is itself linearized, as shown in (20c). Subsequently, only once the entire clause is sent to PF as an Utterance does the Clitic Group violate the Bg prosodic prohibition against Utterance initiality. Hence its linearization with respect to its host is erased and reevaluated to follow dade li, as in (20d). In Mac, on the other hand, since this language lacks the Utterance-initial constraint, this last TM step never applies. Hence, in Mac the final output remains as in (20c), without the TM effect.

TM is a process with the following properties. First, it it applies obligatorily when needed but never otherwise. Second, it never disturbs the ordering of the clitics among themselves. Third, in order for it to apply the clitic group cannot contain any intervening material but must instead be treated as a seamless whole. To see this, compare ungrammatical (21b) with (14a), repeated as (21a).

• 21.

  1. a. Az sâm už ti go dala.

     I AUX_1SG maybe you_DAT it_ACC given_FEM

     'I have maybe already given it to you.'

  2. b. *Dala sâm už ti go.

  3. c. Dala sâm ti go.

These properties suggest that TM is a last-resort PF-side reordering process, as depicted in (7). This process operates in Bg but not Mac: it targets the clitics as a group, erases the linearization when it violates their lexical non-initiality requirement, and recombines them with the adjacent verb so that they follow rather than precede this verb. This can be handled in the morphology by rebracketing the clitics as a Clitic Group and subjecting them to the process of Local Dislocation, as follows:²⁵

• 22.

  1. a. [_cg si mu gi] [_v pokazvala... → [_v pokazvala + [_cg si mu gi]]...

  2. b. [_cg mu gi e] [v pokazvala]... → [_v pokazvala + [_cg mu gi e]]...

Embick and Noyer (2001) describe Local Dislocation as a type of Merger which applies after Vocabulary Insertion to manipulate already linearized strings. To avoid literal movement on the PF-side, contra Embick and Noyer, I think of this slightly differently: linear precedence is defined derivationally, as successively larger units are sent to Spell-Out. When precedence violates prosodic requirements, such as non-initiality, it can be erased and relinearized. This should not be taken as comparable to traditional feature-driven and hierarchically defined syntactic movement.

The following scenario seems reasonable. As will be argued in section 4.3, the output of the syntax does not in fact leave the clitics in a single, large head position. Specifically, it will be claimed that clitic auxiliaries (sâm, si, etc.) are in Agr_S⁰, third singular clitic e is really not an agreement marker, but rather in T⁰ (the next head down), and the pronominal subcluster (e.g., mu gi) is left-adjoined to T⁰. Adjacent clitics are, however, rebracketed into a Clitic Group, and it is this entity that is targeted by Vocabulary Insertion.²⁶ It is only subsequent to Vocabulary Insertion that a Local Dislocation process applies when it is necessary to avoid the clitics being Utterance initial, as in (22).

We then want to ask whether this is what accounts for the ordering of the clitics in the superficially similar compound verb construction in (1), repeated here.

• 23.

  1. a. Pročela mu ja beše.

     read_FEM him_DAT it_ACC had_3SG

     'She had READ it to him (and not sung it).'

Does Morphological Merger position the clitics to follow the participle after the syntax has left them in initial position? In order to address this question, one important concern for the subsequent discussion will be the principle in (24), from Embick and Noyer 2001, their (47).

1. 24. If a Merger operation moves an element A to a target B, then A and the head of B are either both MWds or both SWds.

2. 25.

   1. a. MWd = a maximal X⁰

   2. b. SWd = a terminal X⁰ contained in another X⁰

This principle of morphology prevents Local Dislocation operations from attaching a subword (SWd) to a morphosyntactic word (MWd), where the former is defined as a terminal X⁰ contained in another X⁰ and the latter is defined as a maximal X⁰. It also prevents the morphology from situating a MWd internal to another MWd. It is this particular aspect of (24) which I will argue creates a problem for treating clitic placement in Bg compound verb construction (23) in terms of Local Dislocation. I will further show that the order in (23) obtains even when TM should not be invoked, implying that it is never the reason why clitics split participles from auxiliary verbs.

4. Syntactic Approaches

Approaches that make use of Morphological Merger are fundamentally syntactic in that they can only readjust what the syntax provides them with in the first place. It is for this reason important to have a clearer picture of how the syntax might treat clitics before resolving the question posed by (23). In this section we therefore consider some general issues and specific proposals about the syntax of clitic placement.²⁷

4.1. lterated Head Movement

A fairly popular view, deriving in part from the theory of phrase structure in Chomsky 1995, is that clitics are non-branching phrases.²⁸ Boškovic (2002) argues that the clitics are generated as non-branching XPs and move as such (presumably for checking reasons) to specifier positions. Only once the verb has raised past them to the next higher head position can the clitics—now in their X⁰ guise—undergo head movement themselves, adjoining to the left of the verb. For Bošković, left-adjunction is the consequence of Kayne's (1994) Linear Correspondence Axiom (LCA).²⁹ This process of head adjunction takes place successively, as target heads appear above the clitics in the course of building up the structure. Thus, given a base (or intermediate) structure as in (26a), the eventual result is a complex head with the general structure in (26b), Bošković's (16a) and (16b), respectively.³⁰

• 26.

  1. a. [vp si [v' [a_griop mu [a_grio' [a_grdop gi [a_grdo'[vp dal]]]]]]]

  2. b. [si_n + [mu₁ + [gi_j + dal_i]_k ]_m] [vp t_n [v' t_m [agriop t₁ [agrio' t_k

     [agrdop t_j [agrdo' t_i [vp t_i]]]]]]]

In this system of iterated head movement, the clitics together with the verb constitute one large right-branching X⁰ by the end of the derivation.³¹ In section 4.3, although I reject the kind of highly articulated structure in (26b), I too will exploit the LCA.

4.2. PF-Filtering and Scattered Deletion

The iterated head movement approach immediately raises two types of concerns: Does it really provide the right internal structure? and How are orders which deviate from (26) produced? I turn now to the second concern, postponing a more careful consideration of the first until the next section. The TM effect is an obvious issue here. Bošković (2001) and Franks and Bošković (2001) analyze TM in terms of scattered deletion of subwords (SWds).³² Under this account, for example, (6)/(15) would have to be treated as follows, with pronounced copies underlined and in boldface:

• 27.

  1. a. [ si + [ mu + [ gi + pokazvala ]]] [ si mu gi pokazvala ...[=(6a)/(15a)]

  2. b. [ e + [ mu + [ gi + pokazvala]]] [ e mu gi pokazvalal ... [ e mu gi pokazvala ...[= (6b)/(15b)]

This is Bošković alternative to the kind of Local Dislocation approach, schematized in (22) and repeated in (28).

• 28.

  1. a. [cg si mu gi] [v pokazvala... → [v pokazvala + [cg si mu gi]]...

  2. b. [cg mu gi e] [v pokazvala]... → [v pokazvala + [cg mu gi e]]...

For him, PF-side reordering mechanisms do not exist, so there could never be any need to invoke them.³³ Instead, the correct copies are pronounced through the complex interaction of PF requirements that filter out the offending copies. In order to produce (28), Bošković needs to assume (i) that the auxiliary and pronominal clitics are prosodically enclitic and (ii) that third person singular e must be last in the sequence of clitics. To fully appreciate the power of this system, consider how it derives (12c), repeated as (29a):

• 29.

  1. a. Ne mu li gi e pokazvala?

     NEG him_DAT Q them_ACC AUX_3SGshowed_FEM

     'Hasn't she shown them to him?'

  2. b. [ ne + [ e + [ mu + [ gi + pokazvala ]]]] + li [ ne e mu gi pokazvala ] ... [ ne e mu gi pokazvala ...

It is by virtue of yet a third PF stipulation that (29b) obtains, namely, that li must be second in its intonational phrase. The result is a highly modulated scattered deletion of pieces of complex X⁰s (which can be considered MWds under the definition in (25a)), so that the copies of the five clitics (ne, mu, li, gi, and e) in (29) that are actually pronounced each belong to different parts of the structure.

4.3. A More Finely Grained Structure

This section presents arguments for distinguishing the structural position of the auxiliary clitics and the pronominal clitics, with an excursus on the special nature of third singular (j)e.

4.3.1. Scattered Deletion

I first argue against Bošković's (2002) use of scattered deletion to obtain these facts and then show how the ordering of clitics in (5), repeated as (30), can be handled in purely syntactic terms.³⁴

1. 30. AUX (except e) >> DAT >> ACC >> e

I do not see scattered deletion as a particularly elegant or explanatory solution, and I list some reservations about invoking this mechanism so cavalierly. First, it is not clear that extending the copy and delete approach to SWds is either necessary or desirable. Its application, like that of ellipsis in general, might be restricted to syntactic constituents and prohibited from looking inside of complex X⁰s. Second, also on the conceptual side, I note that it requires the prosodic properties of specific vocabulary items to determine which copy to pronounce, both of itself and of other items. Intuitively, this seems incorrect, because if the syntax leaves undetermined which copy is to be overt, this should be resolved before any actual morphology or phonology can take place. Third, the scattered-deletion account ascribes the same sequence of clitics completely different constituencies depending on whether the sequence appears preverbally or postverbally (as the result of TM), although the actual clitic order internal to the clitic group remains immutable. Fourth, Bošković's approach to my mind amounts to a sophisticated variant of templatic solutions to clitic ordering.³⁵ Fifth, as shown by the contrast in (21) and the comparable minimal pair in (31), it is only when the clitics are grouped together that they can follow the verb, which indicates that TM targets the clitic group as some kind of unit. As discussed in Franks (2006a), aspectual adverbs can marginally intervene at certain points when the clitics precede the verb, but this is absolutely impossible when they follow:³⁶

• 31.

  1. a. ^?Az sâm veče ti ja dala?!

     I AUX_1SG already you_DAT it_ACC given_FEM

     'I have already given it to you?!'

  2. b. *Dala sâm veče ti ja.?!

     given_FEM AUX_1SG already you_DAT it_ACC

Sentence (31a) is imaginable in a context which appropriately emphasizes the adverb. For example, imagine that I gave you a book but then I forgot that I had already done so; later, you come and ask about the book, and I am surprised that I had gaven it to you already. In this case, I could certainly respond as in (31a). But, if I choose to leave out the subject pronoun az 'I' and thereby invoke TM, I could never say (31b). Sixth, there is reason even to doubt the empirical correctness of the PF stipulations Bošković needs in order to modulate the deletions. For example, when li is used to mark contrastive focus, it may freely follow multiple prosodic words, as in (32a–b), (from Bošković 2001: 232; 236), or (32c).³⁷

• 32.

  1. a. Novata kola li prodade?

     new_DEF car FOC sold_2/3SG

     'Was it the new car that she/he/you sold?'

  2. b. Sâvsem nova li roklja noseše?

     completely new FOC dress wore_2/3SG

     'Was it a completely new dress that she/he/you were wearing?'

  3. c. Koj kakvo li mi e kupil?

     who what FOC you_DAT AUX_3SG bought

     '(I wonder) who has bought me what?'

Finally, simply requiring that e appear last obscures some likely larger generalizations about its special status, generalizations which lead to a finer structure of where in fact the various clitics go. I now turn to this problem.

4.3.2. On Idiosyncrasies of (j)e

In following rather than preceding the pronominal subcluster, the third singular auxiliary clitic behaves differently from the other auxiliary clitics in virtually all the South Slavic languages, as well as in West Slavic Upper and Lower Sorbian. This suggests that this element, which I represent generically as (j)e, in fact occupies a different position from the other auxiliary clitics and that two head positions are involved, one before the pronominal clitics and one after.³⁸ Elsewhere in West Slavic (as well as in South Slavic Macedonian), the third-person auxiliary is null, indicating that there is in fact no morphological exponent of third person agreement and that, where it appears, it is marking something other than subject agreement. I suggest, therefore, that although the auxiliary clitics are person agreement markers in Agr_s, (j)e is actually inserted in T, one head down, where it seems to have a more or less expletive status.³⁹ Next, since the pronominal clitics precede (j)e but follow the other auxiliary clitics, these would have to be adjoined to T⁰. The structure I argue for is thus as in (33).

• 33.

  1. a. [a_grs' [a_grs sâm] [tp [t [ti go] + T⁰] ...

  2. b. [agrs- [agrs agr⁰] [tp [t [ti go] + e] ...

One (small) reason to implicate tense as the locus of (j)e is that in Slovenian (Slvn) it is precisely the future tense auxiliary clitic bom 'will.lSG' which along with je comes after rather than before the pronouns.⁴⁰ There is however variation among the languages in whether the auxiliary clitic appears in Agr_S or T. Interestingly, not just (j)e but the other clitic auxiliaries also once came last, implying a historical change from the auxiliary clitics surfacing in T to their surfacing in Agr_S, as well as their synchronic T-to-Agr_S movement.⁴¹

Morphological evidence from BCS lends further support for this approach. In that language the tonic forms of the auxiliaries consist of the morpheme je, which I take to originate in T⁰, plus an Agr_S portion identical to the clitic:

• 34.

  1. a. jesam'(I)am' = je + sam 'AUX_1SG' [BCS]

  2. b. jesi '(you) are' = je + si'AUX_2SG'

  3. c. jesmo '(we) are' =je + smo 'AUX_1PL'

  4. d. jeste '(you) are' = je + ste 'AUX_2pl

  5. e. jesu '(they) are' = je + su 'AUX_3PL'

These forms can be derived by raising T to adjoin to the left of Agr_S, in keeping with the LCA. The one morphological exception is in the third singular, which is just a tonic je, This follows from the assumption that the Agr_S portion is null for the third singular. It thus seems that at one time Agr_S lowered to T —as a morphological operation following Embick and Noyer's (2001) account of Def(initeness) lowering in the Bg DP—but now T raises to Agr_S. This change is apparently complete in some Croatian dialects where even je comes first.

As an anonymous reviewer points out, the details of this proposal require considerable fleshing out. It seems that in moving to Agr T⁰ must excorporate from the pronominal clitics to prevent them from preceding the auxiliary and to allow material such as už in (14a)/(37a) to intervene. On the other hand, je must be inserted so as to follow the pronominal clitics, i.e., just in case T⁰ does not raise to Agr. Now, if I am correct in characterizing third singular Agr in these languages as having no features and it is the features of Agr that attract those of T,then it stands to reason that T⁰ will not raise just in case Agr is (featureless) third singular.

Be that as it may, the empirical idiosyncrasies find no expression in a scattered deletion account, which treats the fact that (j)e is pronounced last as accidental. Instead they point to an approach that attempts to read fixed aspects of word order off hierarchical structure.

Other idiosyncrasies of BCS je are that it participates in various well-known dissimilatory processes that the other auxiliaries do not. It forces selection of ju over je for the third singular feminine accusative pronoun, resulting in the sequence ju je and supporting the idea that the pronominal clitics plus je form a complex head. It can also be deleted (or not inserted) altogether, as when it would follow se (i.e., se instead of se je in the literary language) and also after pronouns ending in -e, namely, me 'me' and te 'you' (i.e., me or te instead of me je or te je colloquially). It can also occur initially in Yes/No questions, supporting li, as in (35), from Bošković 2001: 121, fn 33:

• 35.

  1. a. Je li (je) on istukao Petra.

     DUMMY Q AUX_3SGhe beat Peter

     'Did he beat Peter?'

This je is not a clitic. Bošković calls it “a non-clitic counterpart of the clitic li,” but I prefer to think of it as a pleonastic element inserted to provide support for li. In this regard it is very similar to BCS C⁰ da. Note that, as (35) shows, it can cooccur with auxiliary/T⁰ je, which I suggest is also pleonastic, as well as with finite verbs when they themselves do not front:

• 36.

  1. a. Je li vole Mariju.

     DUMMY Q love_3PL Maria

     'Do they love Maria?'

4.3.3. The Argument from Interpolation

Returning to the structures in (33), these make certain predictions about the potential for interpolation of particles and adverbs.⁴² These predications are indeed borne out. Recall the paradigm in (14) above, repeated as (37).

• 37.

  1. a. Az sâm už ti go dala.

     I AUX_1SG maybe you_DAT it_ACC given_FEM

     'I have maybe already given it to you.'

  2. b. *Az sâm ti už go dala.

  3. c. Az sâm ti go už dala.

Assuming the structure in (33a), už can adjoin to TP in (37a) separating the auxiliary clitic from the pronominal ones. It can, as in (37c), also of course adjoin to the phrase below this (νP or some functional projection right above υP, given arguments in Izvorski 1993 and Lambova 2003, 2005, among others, that the participle undergoes short movement to outside of νP). It cannot, however, split the pronominal clitics in (37b), since these constitute a complex X⁰. Consider now (38), in the third person:

• 38.

  1. a. *Tja ti už go e dala.

     she you_DAT maybe it_ACC AUX_3SG given_FEM

     'She has maybe already given it to you.'

  2. b. *Tja ti go už e dala.

  3. c. Tja ti go e už dala.

As expected, (38a) is ungrammatical and (38c) is grammatical. What is surprising here is the impossibility of (38b). This follows, however, from the structure in (33b), since ti go is adjoined to e in T⁰.Franks and Rudin (2005) and Franks (2006a) report similar facts for aspectual adverbs. Consider where veče 'already' can go in (39):

• 39.

  1. a. *Tja ti veče ja e dala.

     she you_DAT already it_ACC AUX_3SG given

     'She has already given it to you.'

  2. b. Tja ti ja veče e dala.

  3. c. Tja ti ja e veče dala.

Veče cannot appear anywhere inside of the complex head [t [ti ja] e]. And once again, the impossibility of separating the pronominal clitics from the auxiliary clitic only holds for e in (39b). As we saw in (31a), repeated as (40), this contrasts with the far more acceptable attempt to interpolate an aspectual adverb on the other side of the pronominals:

1. 40. ^?Az sâm veče ti ja dala?!

   I AUX_1SG already you_DATit_ACC given_FEM '

   I have already given it to you?!'

I thus conclude that a structure as in (33), where the clitics do not all occupy the same functional head, is warranted for Bg.

4.4. A Note on Negation and Participle Fronting

As pointed out by Pancheva (in press), among others (e.g., Lema and Rivero 1989, Embick and Izvorski 1995, or Bošković 1995) participle fronting does not cooccur with negation in the modern South Slavic languages, including Bg. Thus, whereas (41a) is perfectly normal, (41b) is ill-formed:

• 41.

  1. a. (Tja) ne beše pročela knigata.

     she NEG had_3SG read_FEM book_DEF

     'She had not read the book.'

  2. b. *(Tja) ne pročela beše knigata.

She argues that the negation marker ne and the finite verb form a complex syntactic head. Indeed, negation is not separable from the conjugated auxiliary or verb, except when the verb is immediately preceded by pronominal clitics (in Bg and Mac only, not in second position BCS or Slvn). This suggests to me that ne is right above Agr and proclitic on Agr (or on T, if Agr is phonologically empty). Pancheva (in press) considers various ways of blocking (41b), which need not concern us here although they certainly warrant more careful investigation. The interesting point she makes, however, is that Bg is in this regard different from Old Church Slavonic (OCS). In OCS the order ne + participle + auxiliary was clearly possible. Pancheva (in press) cites (42) and in the abstract for the talk on which the paper was based she offers (43):⁴³

1. 42. Ne moglЪ bi tvoriti ničesože. [OCS]

   NEG could be_COND.3SG do_INF nothing_GEN

   'He couldn't do anything.' (John 9.33)

2. 43. Česo radi muro se ne prodano bystЪ? [OCS]

   why because perfume this NEG sold be_PAST.3SG

   'Why wasn't this perfume sold?' (John 12.5)

These examples form part of her general argument, contra Willis 2000, that OCS had basic T-final structures (alongside T-initial ones), whereas modern Bg is only T-initial. Thus they do not involve participle fronting in OCS, whereas in Bg that would be the only way to derive them, and the participle cannot front to separate ne from the agreeing auxiliary.

5. Analysis of Compound Verb Constructions

This section examines the status of (1), repeated as (44), in light of the preceding discussion.

1. 44. Pročela mu ja beše.

   read_FEM him_DAT it_ACC had_3SG

   'She had READ it to him (and not sung it).'

It is argued that the clitics here are not positioned by any type of TM mechanism but rather through successive syntactic adjunctions. But before turning to (44) it is necessary to examine compound verb constructions with the auxiliary before the participle, in order to show that in this order the two do not form a syntactic constituent.

5.1. When the Auxiliary Precedes the Participle

Bg has several compound tenses with tonic auxiliaries, as in pluperfect (45).⁴⁴

1. 45. Ivana beše pročela knigata.

   Ivana had_3SG read_FEM book_DEF

   'Ivana had read the book.'

It is clear that the auxiliary and the participle do not form a complex X⁰ head in (45), since certain material is able to split them. Consider the following examples:⁴⁵

• 46.

  1. a. Ivana beše veče pročela knigata.

     Ivana had_3SG already read book_DEF

     'Ivana had already read the book.'

  2. b. Marija beše napâlno zabravila za sreštata.

     Maria had_3SG completely forgotten about meeting_DEF

     'Maria had completely forgotten about the meeting.'

  3. c. Studentite bjaxa vsički pročeli knigite.

     students_DEF had_3PL all read books_DEF

     'The students had all read the books.'

  4. d. Ivana beše nabârzo pročela knigite.

     Ivana had_3SG quickly read books_DEF

     'Ivana had quickly read the books.'

  5. e. Ivana beše vinagi čela ljubovni istorii.

     Ivana had_3SG always read love stories

     'Ivana had always read love stories.'

  6. f. ^?Ivana beše vinagi mnogo razkazvala za svoja život.

     Ivana had_3SG always much told about her_DEF life

     'Ivana had always told the story of her life at length.'

Example (46c) is of particular interest because Q-float indicates the availability of a specifier position between bjaxa 'had' and pročeli 'read' through which the subject studentite 'the students' has passed. While subjects are generally postverbal in questions, under certain conditions even they can appear between the auxiliary and the main verb, as in (47), again from Krapova 1997:⁴⁶

• 47.

  1. a. Beše li Ivan pročel knigata?

     had_3SG Q Ivan read_MASC book_DEF

     'Had Ivan read the book?'

  2. b. Kakvo beše Ivan pročel?

     what had_3SG Ivan read_MASC

     'What had Ivan read?'

Once we conclude that beše 'had' and pročela 'read' occupy distinct head positions, it is of course hardly surprising that pronominal clitics can also intervene between the auxiliary and the participle when called for by the TM effect, as in (48).

1. 48. Bese mu ja pročela.

   had_3SG him_DAT it_ACC read_FEM

   'She had read it to him.'

While the order in (48) is the only one possible, we do not actually know whether this is the order provided by the syntax or the consequence of post-syntactic Local Dislocation, which repositions the clitics mu ja after beše. That is, it could simply be that the syntax provides something like (49), where beše is merged in a high Aux position.⁴⁷

1. 49. [a_uxbeše] [t mu ja + T⁰] [a_sp pročela... [υp... [a_sp mu ja...

Alternatively, it could be that beše is merged in an Aux position just below T, so that the syntax produces something like (50). When initial, the clitic group will subsequently undergo Morphological Merger with beše, as shown in (50b), to produce the desired surface order in (48).⁴⁸

• 50.

  1. a. [t mu ja + T⁰] [a_ux beše] [a_sp pročela... [υp... [a_gro mu ja ...

  2. b. [cg mu ja] [beše] [pročela] → [beše + [cg mu ja]] [pročela]

Surprisingly, it turns out that both (49) and (50) may be correct.^49,50 When anything is added in front of (48), obviating the TM effect, two grammatical orders emerge, (51a) and (51b), with no special difference in meaning.⁵¹

• 51.

  1. a. Tja beše mu ja pročela.

     she had_3SG him_DAT it_ACC read_FEM

     'She had read it to him.'

  2. b. Tja mu ja beše pročela.

The possibility of (51a) alongside expected (51b) suggests that something akin to structure (49) must be available. This is corroborated by application of the adverb interpolation test to (51a), since the aspectual adverb veče 'already' can appear between any of the postulated X⁰s:

1. 52. Tja ( veče ) beše ( veče ) mu ja ( veče ) pročela.

The claim that there is no compelling reason for the clitics to form a single complex X⁰ head with the verb in Bg leads us to expect that veče should also be able to intervene in (51b). This is indeed possible (for speakers who tolerate adverb interpolation), as shown by (53):

1. 53. Tja mu ja veče beše pročela.

This approach implies that the clitics do not move just in order to be adjoined to some X⁰; rather, pronominal clitics necessarily move to T, whereas the auxiliary clitics which show person agreement are in Agr_S. While Bošković (2001, 2002) contends that the clitics in Bg differ from those in BCS in forming a single complex head together with the verb, the adverb interpolation evidence discussed here and in Franks 2006a argues against this. Bošković applies various tests which show that the clitics occupy distinct head positions in BCS, but one cannot conclude from the failure of these tests when extended to Bg that other factors do not complicate the data. For one thing, the tests are much debated, as noted by Progovac (1996, 2000), and, for another, they consistently also fail in very closely related Slvn, as meticulously demonstrated byGolden and Milojević Sheppard (2000), despite the fact that, as shown by Bošković (2001) and Franks and King (2000), clitics in Slvn have a syntax (but crucially not a phonology) that is virtually identical to BCS. To take just one example, consider the following (near) minimal pair, from Bošković (2002: 331, fn 5):

• 54.

  1. a. Oni su, kao što sam vam rekla, [BCS]

     they AUX_3PLas that AUX_1SGyou_DAT said

     predstavili se Petru.

     introduced REFL_ACC Peter_DAT

     'They, as I told you, introduced themselves to Peter.'

  2. b. *Te sa, kakto ti kazax, predstavili se [Bg]

     they AUX_3PL as you_DAT said_1SG introduced REFL_ACC

     na Petâr.

     to Peter

     'They, as I told you, introduced themselves to Peter.'

One problem with (54b), however, is that Bošković has applied otiose TM to Bg, under the impression that this is prosodically driven. With the expected order, in fact, the sentence improves somewhat:

1. 55. ^??Te sa, kakto ti kazax, se predstavili na Petâr.

It improves considerably more if a pronominal rather than reflexive direct object clitic is employed:⁵²

1. 56. ^?Te sa, kakto ti kazax, gi predstavili na Petâr.

   they AUX_3PLas you_DAT said_1SG them_ACC introduced to Peter

   'They, as I told you, introduced them to Peter.'

It thus seems to me that many aspects of clustering in Bg (as in BCS if Bošković is correct) are on the PF side of the grammar, having to do with Morphological Merger and/or prosodic phrasing rather than with syntactic-head adjunction.

5.2. Participle Fronting is Adjunction

We now turn to the crucial construction, in which the participle precedes the auxiliary. As discussed by Lambova (2003, 2004b, 2005), among others, this has a focusing effect:

1. 57. Pročela beše knigata.

   read_FEM had_3SG book_FEM

   'She had READ the book.'

Crucially, here no splitting is possible, and any attempt to interpolate anything comparable to the examples in (46) is strongly deviant:⁵³

• 58.

  1. a. Pročela (* veče ) beše knigata.

     read already had_3SG book_DEF

     'She had (*already) READ the book.'

  2. b. *Zabravila (* napâlno) beše za sreštata.

     forgotten completely had_3SG about meeting_DEF

     'She had completely FORGOTTEN about the meeting.'

  3. c. Pročeli (* vsički ) bjaxa knigite.

     read all had_3PL books_DEF

     'They had (*all) READ the books.'

  4. d. Pročela (* nabârzo ) beše knigite.

     read quickly had_3SG books_DEF

     'She had (*quickly) READ the books.'

  5. e. *Čela (* vinagi ) beše ljubovni istorii.

     read always had_3SG love stories

     'She had (*always) READ love stories.'

  6. f. *Razkazvala (* vinagi mnogo ) beše za svoja život.

     told always much had_3SG about her_DEF life

     'She had (*always) TOLD the story of her life (*at length).'

This can be taken as evidence that participle fronting in Bg involves left-adjunction of the participle to the auxiliary, so that they form a complex maximal X⁰, i.e., a MWd. For this reason, no phrasal material or anything adjoined to an XP can interrupt them. This adjunction is driven by a Focus Projection (FocP), which is situated immediately below CP. Foe attracts the participle whenever it bears focus features. On its way up to Foc, the participle adjoins to the auxiliary, creating a complex head, and the two continue to raise together.

Similarly, whatever their judgments about (47), all speakers find a clear contrast between (59a) and (59b):

• 59.

  1. a. Beše li (^??Marija) kupila (Marija) kâšta?

     had_3SG Q Maria bought_FEM Maria house

     'Had Maria bought a house?'

  2. b. Kupila li (**Marija) beše (Marija) kâšta?

     bought_FEM Q Maria had_3SG Maria house

     'Had Maria BOUGHT a house?'

It will be noted that the question particle li can intervene in (59b), just as it can in (59a). However, since I take this to be a consequence of PI as discussed in section 3.1, the fact that li can split the participle + auxiliary complex does not contradict its MWd status.

Much more troublesome than splitting by li, on the other hand, is the fact that the pronominal and verbal auxiliary clitics can also appear between the participle and the auxiliary. Thus, parallel to (48), we have (1), repeated again as (60):

1. 60. Pročela mu ja beše.

   read_FEM him_DAT it_ACC had_3SG

   ‘She had READ it to him (and not sung it).’

If the TM effect involves Morphological Merger, and if pročela beše constitute a complex X⁰, as indicated by their otherwise impenetrable cohesion, then the principle in (24) faces a dilemma: there is no credible sense in which the clitic cluster is analyzable as a SWd. Local Dislocation of SWds, but not of MWds, places them internal to a MWd. This is clearly inconceivable with Bg clitics, as demonstrated by any attempt for the clitics to target the prefix pro-, the participial suffix -l-, or the agreement inflection -a of pročela in (61):

• 61.

  1. a. *pro-mu ja-čela

  2. b. *proče-mu ja-la

  3. c. *pročel-mu ja-a

I therefore conclude that (60) cannot be analyzed in terms of post-syntactic Local Dislocation of the clitics onto the following verbal head.

5.3. Getting the Right Order in Compound Tenses

In this final section I sketch the form a solution might take that does not come up against the Distributed Morphology principle in (24). Since mu ja cannot undergo Local Dislocation to be placed inside the complex head pročela beše, the only way to construct this order is to create it through successive adjunctions in the syntax, creating one large X⁰. As expressed in (2), there are only two possible alternatives. These structures are given again in (62), assuming movement is ultimately to Foc and the LCA as always gives rise to left-adjunction upon linearization.

• 62.

  1. a. [f_oc pročela + [[Cg mu ja] + [beše]]]

  2. b. [f_oc [pročela + [cg mu ja]] + [beše]]

That is, either, as in (62a), the clitics adjoin to beše before the participle pročela does, or else, as in (62b), the participle first adjoins to the clitics and then this unit adjoins to beše.

Recall that on the basis of (51) we were led to posit two base positions for the auxiliary, one high and the other low with respect to T. Since the pronominal clitics have been argued to be in T, fronting of the participle should interact with the posited structures in specific ways. Consider what happens if the high position is used for beše, as in (49). This produces the structure in (62b), with each adjunction indicated as follows:

• 63.

  1. a. [a_ux beše] [t mu ja + T⁰] [a_sp pročela] →

  2. b. [a_ux beše] [t [a_sp pročela] + [t mu ja + T⁰]] →

  3. c. [a_ux [t [a_sp pročela] + [t mu ja + T⁰]] [a_ux beše]]

Now we take the low position for beše, as in (50), again indicating the adjunction steps:

• 64.

  1. a. [t mu ja + T⁰] [a_ux beše] [a_sp pročela] →

  2. b. [t mu ja + T⁰] [[a_sp pročela] [a_ux beše]] →

  3. c. [[[a_sp pročela] [a_ux beše]] [t mu ja + T⁰]]

Interestingly, this does not result in splitting at all. It turns out, however, that contrastive focusing of the l-participle can also lead to the (admittedly marked) order in (65).

1. 65. ^?Pročela beše mu ja, (a ne razkazala).

   read_FEM had3_SG him_DAT it_ACC and NEG told

   ‘She had READ it to him (and not related it).’

This is predicted by the derivation in (64). We thus find support for the complex X⁰ structure in (62b) over that in (62a).⁵⁴ Moreover, since I argue that only the participle + auxiliary order involves adjunction, there is no way to derive the impossible (66):⁵⁵

1. 66. *Beše pročela mu ja.

   had_3SG read_FEM him_DAT it_ACC

I conclude that (1) is indeed created in the syntax, with a structure as in (62a).

The claim that the order of elements in (1) is the output of syntax, which means that the kind of mismatch between morphological and syntactic structures elsewhere observed in Bg is not at play here, makes one further interesting prediction. Since there is no syntactic representation in which the clitics precede the participle, hence no reason to invoke Local Dislocation, we would not expect an order with the clitics preceding the participle to reemerge in constructions where the complex is non-initial. That is, TM as a last-resort operation implies “anti-TM” effects as well, just in case its conditioning factors are not met. But this never happens with the participle + auxiliary order, a quite surprising fact given the usual behavior of Bg clitics. While no other focus can precede the participle, other non-focus material is acceptable in various contexts. Crucially, as shown in (67), this material invariably fails to provide support for the clitics. Instead, the only way these sentences can emerge is with the participle preceding the clitics:

• 67.

  1. a. *...če mu ja pročela beše.

     that him_DAT it_ACC read_FEM had_3SG

     ‘...that she had READ it to him.’

     (...če pročela mu ja beše.)

  2. b. *...i mu ja pročela beše.

     and him_DAT it_ACC read_FEM had_3SG

     ‘.. .and she had READ it to him.’

     (...i pročela mu ja beše.)

  3. c. . *Tâkmo mu ja pročela beše.

     just him_DAT it_ACC read_FEM had_3SG

     ‘She had just READ it to him.’

     (Tâkmo pročela mu ja beše.)

Since we know that there is absolutely nothing wrong on the PF-side with the orders in (67), it must be that they do not arise for the simple reason that the syntax never produces them in the first place. On the other hand, the orders in (68), with the auxiliary first hence no focusing, are generated by the syntax and are, as expected, perfectly grammatical:

• 68.

  1. a. ...če mu ja beše pročela.

     ‘...that she had read it to him.’

  2. b. ...i mu ja beše pročela.

     ‘...and she had read it to him.’

  3. c. Tâkmo mu ja beše pročela.

     ‘She had just read it to him.’

These facts corroborate my claim that the reason the participle precedes the clitics in (1) is because it adjoins to them in the syntax and not through any type of Local Dislocation operation. Thus, the problem of coming up against the Distributed Morphology principle in (24) never even arises.

Macedonian also displays anti-TM effects that support a syntactic analysis for ordering the participle before the clitics. Recall that Mac lacks the Bg prohibition against Utterance initial clitics, hence is not expected to have any TM effects at all. There are nonetheless constructions in which clitics fail to be in initial position and, predictably, participle fronting is one of them. While the particular pluperfect construction considered in this paper is lacking in Mac, Olga Mišeska Tomić (p.c.) provides the following comparable paradigm:

• 32.

  1. a. Go imam kupeno. [Mac]

     it_ACC have_1SG bought

     ‘I have bought it.’

     (*Imam go kupeno.)

  2. b. Kupeno go imam (ama ne go imam pročitano).

     bought it_ACC have_1SG but NEG it_ACC have_1SG read

     ‘I have BOUGHT it (but not read it).’

     (*Go kupeno imam...)

The clitic is, as expected, initial in (69a), but when the participle fronts the clitic must follow it. Since there is no PF-side reason why the ungrammatical order go kupeno in (69b) should not be acceptable, it must be that the syntax simply fails to produce it.

6. Conclusion

In this paper, I have assumed, along withEmbick and Noyer (2001: 560), that “morphological structure is, unless further operations apply at PF, simply syntactic structure.” In this context, I first considered two classes of approach to positioning clitics within the clause in Bulgarian, specifically with regard to the Tobler-Mussafia effect, and noted problems for each. One kind of approach relies exclusively on head-adjunction in the syntax and another makes use of additional post-syntactic Local Dislocation mechanisms. I argued against an integrated head-adjunction approach, which, although it eschews PF-movement, still requires heavily modulated PF-filtering in the guise of word-internal scattered deletion. Morphological Merger, although it is likely to be needed elsewhere for resolving mismatches between prosodic or morphological requirements and the output of the syntax, cannot straightforwardly accommodate placement of the clitics between a fronted participle and its auxiliary. I have proposed a solution that instead creates the necessary structure directly in the syntax, crucially by fronting and adjoining the participle to the clitics, and I have shown how such an approach might work in a variety of constructions.