Mismatches and Mitigation at CS-PF Interface: The Curious Case of li

Abstract

Taking the domain of polar questions in Bosnian/Croatian/Montenegrin/Serbian (BCMS) as the empirical background, the paper probes into the syntax–phonology (CS-PF) interface and discusses insertion and movement as PF-repair strategies mitigating against the lack of convergence at PF. Contra previous accounts, the analysis treats li (lexicalization of Q) as a ‘run-of-the-mill’ 2P clitic in BCMS, whose host cannot always be provided by syntax. I provide evidence against Prosodic Inversion—‘the usual suspect’ for post-syntactic movement in Slavic—thus adding to the body of evidence that Prosodic Inversion does not take place in BCMS. I argue that the PF Movement in such cases has to be raising and adopt Local Dislocation to account for them. Probing into the interaction between Future I and polar questions provides further insights into the ordering of PF Movement operations in BCMS.

1. Introduction

Due to its suitability to inform issues at the syntax–phonology interface, the nature of the cliticization has been one of the central subjects of inquiry in the literature on Slavic clitics. Over the last two decades, persuasive evidence has been presented in favor of the so-called ‘weak phonology approach’ to cliticization in Serbo-Croatian (SC) or Bosnian/Croatian/Montenegrin/Serbian (BCMS) (see Bošković, 1995 et seq.) Syntax provides well-formed syntactic structures to phonology, and the role of phonology is to pick out the convergent one(s) among them. Though the brunt of BCMS data (see Bošković, 1995, 2001; Franks, 1998/2010; Stjepanović, 1998, 1999, among others) can be accounted for under this approach, recent data arrived through large-scale experiments (see Marelj et al., 2024) suggest that there are configurations in which post-syntactic operations need to take place in BCMS as well.

The paper is organized as follows: Section 2 provides the necessary background regarding the nature of clitics, in general, and BCMS clitics and the cliticization process, in particular. Section 3 focuses on the long-noted puzzling behavior of clitic li in l-participle-li construction in BCMS and provides an account for it. Taking the experimental data that suggest that judgements about the relevant data are more graded than standardly assumed (see Marelj et al., 2024) as a starting point, Section 4 discusses their implications for CS-PF interface in BSCM. Section 4 further probes into the interaction between Future I and polar questions, thus providing insights into the ordering of PF operations in BCMS. Section 5 concludes the paper.

2. Light Preliminaries

2.1. CS-PF Interface and the Nature of Clitics

Clitics, standardly understood as syntactically independent and phonologically dependent—‘prosodically deficient’—elements (in the way that the comparable non-affixal forms/elements are not) are a ‘poster child’ for CS-PF mismatches, responsible for the lack of isomorphism between syntactic and phonological structures involving them.1

To mitigate against convergence problems at PF, language uses various insertion, movement, and deletion strategies. Being prosodically weak, i.e., unable to serve as an independent prosodic domain, clitics must become a part of an adjacent domain for stress assignment purposes. To ensure convergence (i.e., pronounceability) at PF, a PF Movement strategy called Prosodic Inversion (Halpern, 1992/1995) is often argued to apply in such cases in Slavic (see Izvorski et al., 1997; Embick & Izvorski, 1997; Rudnickaya, 2000, among others). After attaching itself to its host, the clitic forms a prosodic word with it. Though dependent on their host prosodically, clitics are, typically, nonselective with respect to the category of their hosts.2

2.2. BCMS 2P Clitics in General and li in Particular

The main protagonist of the paper is the interrogative clitic li, a default lexicalization of Q in polar questions in BCMS. As illustrated in (1), the clitic li clusters with the canonical (auxiliary and pronominal) second position (2P) clitics:3^,4

1.	Kupujete	li	mu	ga?
	buy.PRS.2PL	Q	him.DAT	him.ACC
	‘Are you buying it for him?’

As further illustrated in (2), BCMS 2P clitics appear in a stringent linear order, with li heading the cluster, followed by auxiliary clitics (apart from AUX.3SG je ‘is’), in turn, followed by pronominal clitics:5

2.	li ⪧ AUX[−3SG] ⪧ DATφ ⪧ ACCφ ⪧ GENφ ⪧ AUX[3SG]φ

The linear order of clitics matches the hierarchical arrangement of projections, but, as we will see directly, in BCMS, they cluster together only prosodically, not syntactically.

3.	[CP discourse-related clitics [TP auxiliary clitics [vP/VP pronominal clitics]]]

As the contrast between (4a) and (4b) shows, rather than second in their clause, BCMS clitics occur in the second position of their Intonational Phrase (I-phrase—#), the behavior originally noted and elaborated by Browne (1975/2004) and Radanović-Kocić (1988). If the first element of the clause forms a separate I-phrase, BCMS 2P clitics do not follow it directly, but must appear second in their own I-phrase. Since prosodically ‘heavy’ elements get parsed as their own I-phrases, clitics do not appear directly following them (4a), even if that means that they occur after the 3rd, 7th, or nth word in a clause (4b).

4.	a.	#Ja#	#tvoja	mama#,	#sam	ti	kupila	prsten.
		I	your	mother	AUX1SG	you.DAT	buy.PRT.3SG.F	ring.ACC
	b.	#Ja#	#tvoja	mama#	kupila	sam	ti	prsten.
		I	your	mother	buy.PRCT.3S.F	AUX1SG	you.DAT	ring.ACC
		‘I, your mother bought you a ring.’

The pattern with appositives (4) is due to Radanović-Kocić (1988, 1996). The contrast further obtains with clausal subjects, parentheticals, and fronted heavy constituents (see Radanović-Kocić, 1988, 1996; Bošković, 2001, for discussion and elaboration).

The 2Phood of BCMS clitics is, thus, computed prosodically and a single host is sufficient for the entire cluster (5). Provided, however, each of clitics can satisfy its 2Phood appropriately, even clause-mate clitics need not cluster together. This is illustrated in (6), where both the ‘neutral’ (6a) and the ‘marked’ variant (6b), illustrating the clause-mate-clitics split, are grammatical:6

5.	a.	[X]ω CL CL CL
	b.	kupujete li mu ga (cf. (1))

6.	a.	Ivan	i	Stipe	su	ga	dali	Mariji.
		Ivan	i	Stipe	AUX.3PL	it.ACC	give.PRCT.3PL	Mary.DAT
		‘Ivan and Stipe gave it to Mary.’
	b.	%Dali	ga	Mariji	su	Ivan	i	Stipe.
		Give.PRT.3PL	it.ACC	Mary.DAT	AUX.3PL	Ivan	and	Stipe
		‘Give it to Mary, Ivan and Stipe did.’					Wilder and Ćavar (2002)

Further, in BCMS, their prosodic requirement can be computed either with respect to the first prosodic word (W1) or the entire constituent phrase (P1):7

7.	a.	[Taj	pesnik]	mi	čita	knjigu.	(P1)
		that.DEM	poet	me.DAT	read.PRS.3SG	book.ACC
		‘That poet reads a book to me.’
	b.	[Taj	mi	pesnik]	čita	knjigu.	(W1)
		that.DEM	me.DAT	poet	read.PRS.3SG	book.ACC
		‘That poet is reading a book to me.’

The most striking piece of evidence that syntax is responsible for the clitic placement (see Ćavar & Wilder, 1994, 1996; Franks & Progovac, 1994; Progovac, 1996; Progovac, 2000; Franks & King, 2000; Stjepanović, 1998; and especially Bošković, 1995 et seq., particularly Bošković, 2001: pp. 11–36) are the so-called ‘syntactic fortresses’ (Halpern, 1992/1995). As illustrated in (8), if one of the possibilities for calculating 2Phood within the I-phrase (W1 or P1, as in (7)) violates the rules of syntax, it is avoided since it leads to ungrammaticality. In (8b), the clitics split the head of the phrase from its complement, which leads to a syntactic violation (see Schütze, 1994; Progovac, 1996; Franks, 1998/2010, among others for elaborate discussion), but is otherwise unproblematic in terms of PF (studenti ‘students’ is the first prosodic word in the I-phrase of the cluster, which makes it a perfect host for the clitic). Clearly, we must be dealing with a syntactic restriction, since PF is not able to tell apart complements from non-complements (compare the ungrammatical (8b) with the grammatical (7b)).

8.	a.	[Studenti	istorije]	su	mu	ga	kupili.
		students	history.GEN	AUX.3PL	him.DAT	him.ACC	buy.PRTC.3PL
	b.	*[Studenti	su	mu	ga	istorije]	kupili.
		students	AUX.3PL	him.DAT	him.ACC	history.GEN	buy.3PL
		‘Students of history bought it for him.’

To sum, though 2Phood is computed prosodically, clitic placement is accomplished in syntax (hence, the ordering in (2) becomes less of a mystery) and as expected, it follows syntactic rules (8). Provided no syntactic violations occur, the host of the clitics can be of any category: X⁰, XP, an argument, or an adjunct. For completeness, I add (9a) with an X⁰ and (9b) with an adjunct XP host, respectively.

9.	a.	Kupuju	mu	ga	studenti	istorije.
		buy.PRS.1PL	him.DAT	him.ACC	students	history.GEN
		‘Students of history are buying it for him.’
	b.	Danas	mu	ga	studenti	istorije
		today	him.DAT	him.ACC	students	history.GEN
		kupili.
		buy.PRS.1PL
		‘Students of history are buying it for him today.’

3. Puzz-Li-ing

3.1. Data

Originally observed by Rivero (1993) and first discussed at length in Bošković (1995), (10) captures the contrast relevant to this paper. Whereas a finite verb ljubi (‘kisses’) makes a perfect host for the clitic li (10a), the corresponding l-participle poljubio (‘kissed’) is argued to be unable to host it (10b).8

This behavior of li seems truly puzzling; recall that, cross-linguistically, nonselectivity with respect to the category of the host is generally taken to be the hallmark characteristic of clitics, a trait that is further exhibited by all the other 2P clitics in BCMS, since they are regularly hosted by l-participle (see (6b), for instance)

10.	a.	Ljubi	li	Mariju?
		kiss.PRS.3SG	Q	Marija.ACC
		‘Does he kiss Marija?’
	b.	*Poljubio	li	je	Mariju?
		kiss.PRCT.SG.M	Q	AUX.3SG	Marija.ACC
		intented: ‘Did he kiss Marija?’

3.2. Previous Accounts

The common denomination of previous accounts for the contrast in (10) is that li is unlike any other clitic in BCMS, either with respect to its syntactic or its phonological behavior. Specifically, li is uniquely ‘choosy’ in syntax with respect to the syntactic category of its host, thus allowing only X⁰ to host it (see Migdalski, 2006, 2009 for discussion and elaboration) or li forces Prosodic Inversion (PI) across two phonological boundaries, which is illicit (see Schwabe, 2004 for discussion and elaboration).

Either line of reasoning faces empirical problems of either undergeneration (11a) or overgeneration (11b).

11.	a.	Šta/koji poklon	li	si	mi	kupio?
		what/which present	Q	AUX.2SG.	me.DAT	buy.PRCT.M.SG
		‘(I wonder) what/what kind of present did you buy me?’
	b.	*?Poljubio	li	bejaše	Mariju?
		kiss.PRCT.M.SG	Q	AUX.IMPF.3SG	Marija.ACC
		intended: ‘Had he kissed Mary?’

In (11a), li is hosted by a phrase and the sentence (though requiring contextual furnishing) is perfectly grammatical in BCMS, thus challenging the main prediction of Migdalski’s account. On the other hand, an example like (11b), predicted to be grammatical under Schwabe’s account, is out. Namely, li is the only clitic in (11b) i.e., there is no PI process involving other clitics—(11b) features a non-clitic auxiliary bejaše—yet the sentence is ungrammatical.9

3.3. Current Account

3.3.1. Bag of Assumptions: ‘Run-of-the-Mill’ Clitic and Participle

The account I present here (see also Marelj et al., 2024) treats li as a ‘run-of-the mill’ BCMS clitic. For the purposes of the current discussion, it would suffice to treat it is an operator clitic found in the left periphery and as standardly assumed, base generated/externally merged in C⁰ (Rivero, 1993; Bošković, 1995; Progovac, 2005, among others). As for the l-participle, it is a ‘run-of-the-mill’ participle in that it is [-finite].10 In languages in which V-to-C movement takes place in finite clauses, only finite verbs move to C. Non-finite verbs in such languages never do (see Roberts, 1991; Zwart, 1995; Den Dikken & Hoekstra, 1997, among others). Indeed, for (10), the core of the account boils down to [+/-finite] distinction; the ability of verbs and the inability of participles to move to C (see Bošković (1995) for elaborate discussion).11

Unlike in canonical verb-second languages (German, Dutch, Afrikaans, Frisian, Danish, Norwegian, and Swedish), the standard tests for I/T-C movement give less straightforward results in BCMS for a variety of reasons, some of which (e.g., the negation head being a proclitic) have nothing to do with the verb movement. Putting aside interrogatives, the evidence that finite verbs can move to C is, however, sufficiently supported in BCMS. For instance, under ellipsis, in Fragment Answer, rather than the presence of a raised finite AUX or Do-support, as it is the case in English, in BCMS, it is finite verbs themselves that survive ellipsis; kupujem (‘buy.PRS.1SG’) in (12) and mislim (think.PRS.3SG) in (13) (see also Progovac, 2005, et seq. for discussion and elaboration):12

12.	Q:	Da	li	mi	nešto	kupuješ?
		PRT	Q	me.DAT	something.ACC	buy.PRCT.M.SG
		‘Are you buying me something?’
	A:	Kupujem.
		buy.PRS.1SG
		‘I am.’

13.	Q:	Misli	li	ponekad	na	Mariju?
		think.PRS.3SG	Q	sometimes	on	Marija.ACC
		‘Does he sometimes think about Marija?’
	A:	Misli.
		think.PRS.3SG
		‘He/she does.’

3.3.2. Analysis

Polar Questions at CS-PF Interface: [+finite]V⁰-li

The analysis is couched in terms of Rizzi (1997, 2001, 2004) and Rizzi and Bocci (2017) finely articulated left periphery, which, for ease of exposition, I briefly present here. Rizzi proposes a finely articulated CP-domain, with several discourse-related phenomena formalized in terms of hierarchically ordered phrases. The domain hosts topicalization, focalization, scope, discourse related elements, clause typing and (non)finiteness, the highest and the lowest heads, respectively, in this domain (14c). The clause type (declarative, interrogative, etc.) is expressed by Force. Since complementizers show a degree of sensitivity to finiteness, FinP is found as the lowest head within the articulated CP-domain. In English, for instance, Force in finite declarative is lexicalized as that and Force in non-finite declarative as for (14), but not all positions need to be lexicalized, of course. Since we are dealing with polar questions, next to ForceP, Interrogative Phrase (IntP) is of interest for our discussion. For English, Rizzi (2001) and Rizzi and Bocci (2017) argue that IntP hosts complementizers like if, both in root and embedded questions.13

14.	a.	I believe that Mary will win the election.	that	[+finite, +declarative]
	b.	I would prefer for Mary to win the election.	for	[−finite, +declarative]
	c.	[Force [Int [Top[Foc [Fin]]]]]

It is standardly assumed in the literature that clitic li licenses polar question. Couched in terms of finely articulated CP domain, I argue that li is located in F⁰ or, more precisely, Force in polar questions is lexicalized as li.

Quite like other clitics, though a fully-fledged autonomous syntactic element, to ensure convergence at PF, li must find a host. Hosts can, in principle, be of any category, but the cliticization (i.e., incorporation of the clitic into an adjacent prosodic domain for stress purposes) can occur only if the host is placed by syntax into a position that allows PF Merger (see Bošković, 2001 et seq.) to take place. This means that since li is in F, the highest head in the CP-domain, a legitimate host must either be externally merged in F or move/undergo internal merge to F in syntax.14

In (10a), repeated here as (15a) being [+finite], the verb is a legitimate host for the clitic in F⁰. As illustrated in (15a) and (quite like obligatorily in polar questions in Germanic), the movement of finite verbs takes place. The trigger for V-to-C movement has never been answered satisfactorily (see Zwart, 2020 for elaborate discussion). For the purposes of our discussion, we can treat it as the illocutionary Force feature—a [+Q] on F. Importantly, it is a syntactically motivated movement that correlates with [+/-finiteness], so that only finite verbs are able to check the relevant feature (be it via Move/Internal Merge or Agree).15 Consequently, the verb moves to F, left adjoining to li (15b). It is to this syntactically well-formed structure that the process of PF Merger of Bošković (1995, 2001) applies to, thus satisfying the prosodic requirement of li. The clitic and the host are adjacent, and the role of PF Merger is simply to put the clitic (i.e., elements lacking word-level prosodic structure) with an adjacent host to form a phonological word (ω) (15c). Hence, PF Merger—a version of Morphological Merger of Marantz (1988), adopted and adapted by Bošković (1995 et seq.)—is all that is needed for convergence at PF.16

15.	a.	Ljubi	li	Mariju?
		kiss.PRS.3SG	Q	Marija.ACC
		’Does he kiss Mary?’
	b.	CS: [ForceP [F ljubi +li]….. [FinP [Fin ljubi ][TP [T ljubi] [vP/VP ljubi Mariju]]]]]]
	c.	PF: [[ljubi] ω li]ω

Polar Questions at CS-PF Interface: l-Participle-li

The l-participle in BCMS does not seem to be as low as its English counterpart (see Progovac, 2005 for discussion and elaboration). As an illustration, observe that in Fragment Answer cases illustrated in (16), whereas l-participle in BCMS survives ellipsis, its English counterpart does not, and the raised AUX or Do-support is needed.

16.	Q:	Da	li	si	mi	nešto
		PRT	Q	AUX.2SG	me.DAT	something
		kupio?
		buy.PRCT.SG.M
		‘Have you bought me something?’
	A:	Kupio	sam.
		buy.PRCT.SG.M	AUX.1SG
		‘I have’.

Though it is able to escape vP/VP ellipsis, thus clearly finding itself in a position higher than its English counterpart, the participle in BCMS cannot move as high as a finite verb in BCMS. For one thing, participles cannot move past TP-adverbs (see Bošković, 1995, et seq., a.o.) Note first that adverbs like pravilno (‘correctly’) have two possible attachment sites, yielding two different readings. One possibility for pravilno is to be vP/VP-adjoined, yielding the manner reading. Its sentential reading, on the other hand, arises when it is TP-adjoined. As illustrated in (18a), provided the participle odgovorio (‘answered’) stays in situ/within the vP-domain, both the manner and the sentential readings are available, since both attachment sites are higher than the in situ position of the participle. In (17b), the participle must move overtly to host the clitic (17b). Clearly, this movement has to be lower than T, since the only available reading in (17b) is still the manner reading. It also has to be higher than vP, since the participle hosts auxiliary. The reader is referred to Bošković (1995) who convincingly argues that l-participles is, and the auxiliary clitics can, be found lower than T.

17.	a.	Jovan	je	pravilno	odgovorio	Mileni.
		Jovan	is	correctly	answered	Milena.DAT
		‘Jovan gave Milena a correct answer.’				(manner reading)
		‘Jovan did the right thing in answering Milena.’				(sentential reading)
	b.	Odgovorio	je	pravilno	Milena.
		‘He gave Milena a correct answer.’
		*‘He did the right thing in answering Milena.’				(Bošković, 1995: p. 249)

It is standardly assumed in the literature on BCMS that participles—externally merged within vP/VP—consequently move to a position in which AUXs are externally merged (the AuxP (‘lower’ AuxP)—see Progovac, 1996; Ćavar & Wilder, 1994; Bošković, 1995 among others):

18.	[Aux odgovorio je [VP odgovorio Mariji]]

Finally, note that when the non-clitic version of AUX is used in Fragment Answer, only the AUX survives the ellipsis (19). Given auxiliary clitics are ‘special clitics’—a type of clitic that exhibit syntactic positions often different from the positions of their non-clitic counterparts—and given they need a host, the conclusion is that whatever the position of odgovorio je is (17b), it must be lower than that of the full form auxiliary jesam in (19) as the presence of the participle in (19) results in ungrammaticality.

19.	Q:	Da	li	si	mi	nešto	kupio?
		PRT	Q	AUX.2SG	me.DAT	something.ACC	buy.PRCT.SG.M
		‘Have you bought me something?’
	A:	Jesam.		*Jesam	kupio	*Kupio	jesam
		AUX.1SG		AUX.1SG	buy.PRCT.SG.M	buy.PRCT.SG.M.	AUX.1SG
		‘I have.’		Intended: ‘I have.’		Intended: ‘I have.’

Returning to the contrast in (10) and the ungrammaticality of the example (10b), repeated here as (20a), note that since l-participles in Slavic do not move as high as C in syntax (see Bošković, 1995 et seq.; Embick & Izvorski, 1997, among others), they are not legitimate candidates to host li, and the ungrammaticality of (20a) in BCMS is expected. Since the prosodic requirements of li cannot be met (20c), an otherwise syntactically well-formed output (20b) is filtered out by PF.

20.	a.	*Poljubio	li	je	Mariju?
		kiss.PRCT.SG.M	Q	AUX.3SG.	Marija.ACC
		intented:’Has he kissed Maria?’
	b.	CS: [C/F li [ T [AUX poljubio+je [vP/VP Marija
	c.	PF: # li…poljubio..je

3.3.3. Predictions

Under the account presented here, only elements that undergo external merge in C/Force or are independently moved/undergo internal merge to C/F in syntax can host li. It follows then, that, in contrast with [+finite] verbs, not only l-participles, but other [-finite] forms are expected to be poor hosts for li, since none of them are able to move to the C/F in syntax. As illustrated in (21a) and (22a), this prediction is borne out:17

21.	a.	*Otići	li	želite?		[infinitive]
		leave.INF	Q	want.PRS.2PL
	b.	Želite	li	otići?	[finite V movement to C/F]
		want.PRS.2PL	Q	leave.INF.
		‘Would you like to leave?’

22.	a.	*?Pojedene		li	su	jagode?
		eat.PASS.PRCT.PL.F		Q	AUX.3PL	strawberries
		[passive participle]
	b.	Da	li	su	pojedene		jagode?
		PRT	Q	AUX.3PL	eaten.PASS.PRCT.PL.F		strawberries
		[da-insertion in C/F]
		‘Have the strawberries been eaten?’

Recall that language uses different strategies to mitigate against convergence problems at PF, be it insertion, movement, or deletion. BCMS is no exception here. Both insertion and movement are illustrated in (21)–(22). Since the finite verb (želite ‘want’) moves in syntax (21b), PF Merger forms a prosodic word from the host and the clitic, while in cases like (22b), da-insertion ensures that the clitic li satisfies its prosodic requirement.

To sum, the data we have looked at so far support the ‘weak phonology approach’ to clitics in BCMS, as proposed and particularly developed by Bošković (1995) et seq., and as best summed-up in slogan form as ‘syntax proposes, phonology disposes’ (Franks, 1998/2010), meaning that, to a large extent, it is syntax that is responsible for the ordering of clitics, with phonology having but a filtering role on its outputs. The empirical data seem to suggest that clitics of the left periphery like li can be supported by finite, but not non-finite verbal forms in BCMS, since only finite verbal forms can move as high as C/F in syntax.

4. A Second Look at Polar Questions at CS-PF Interface

4.1. Larger Data Sets—More Complexities

A large-scale online truth-value judgment/acceptability task experiment (see Marelj et al., 2024) shows, however, that rather than in a binary fashion (grammatical vs. ungrammatical), the l-participle-li construction is judged as graded. Namely, though both finite verbs and XPs were significantly more likely to be judged as acceptable hosts for the clitic li, l-participle was not completely rejected. Rather, the likelihood of saying ‘yes’ to sentences like (10b)/(20a) was 36% (see Marelj et al., 2024 for details and elaboration).

Hence, though a subset of native speakers shows a clear contrast between data like (10a)/(15a) and (10b)/(20a), in terms of grammaticality judgments, the gradient judgments of others seem to suggest that in the absence of ‘proposals by syntax’, ‘phonology’ sometimes does more than just ‘filtering’ alone.

4.2. Post-Syntactic Movement: Lowering or Raising?

Since non-finite forms do not move to C/F in syntax, the acceptability of (23) suggests that rather than just putting adjacent elements together, we must allow for a version of PF Merger that, crucially, allows some reordering/rearrangement of elements at PF, in line with the original formulation of Morphological Merger (Marantz, 1988).

23.	% Poljubio li je Mariju?

This being the case, the next question that arises here is whether the process that reorders l-participle and li so that the prosodic requirement of li can be satisfied is that of lowering (of li) or raising (of l-participle).

Recall that Prosodic Inversion (PI) is a ‘usual suspect’ in Slavic (see Halpern, 1992/1995; Izvorski et al., 1997; Embick & Izvorski, 1997; Rudnickaya, 2000, among others) when it comes to repairing prosodically ill-formed strings. PI (Halpern, 1992/1995)—a last-resort option to satisfy the licensing requirements of phonological material—attaches a clitic to the right of the edge of the ω, at the closest relevant prosodic boundary (24).

24.	Prosodic Inversion (PI):
	[ [ ]ω ___ ]ω attach an enclitic to the right edge of a phonological word to form
	another phonological word

As elaborately argued in Franks (2008, 1998/2010), in the case of negative polar questions in Bulgarian, PI seems to be an elegant solution. For brevity, I briefly present the data and the core of his analysis here. In Bulgarian, proclitic ne (NEG) is postaccenting i.e., it not only forms a prosodic word with the element to its right, but if that element is itself not a prosodic word, ne shifts the stress to it and then procliticizes onto it. As a results, if it is another clitic that follows ne directly, the clitic will become stressed. This situation is illustrated in (25), with pronominal clitic me (me) bearing stress.

25.	Ne	MÉ	pitaxa.
	NEG	me	ask.AOR.IMP.3P
	‘They didn’t ask me.’			(Vakareliyska, 2023: p. 9)

Quite like in BCMS, clitic li is a 2P enclitic in Bulgarian, while, unlike in BCMS, auxiliary and pronominal clitics are V-adjacent and normally not separable. Further, predictably, li (Q) is located higher than auxiliary and pronominal clitics in Bulgarian as well. Quite like in BCMS, the order ‘li > aux/pronominal clitics’ is normally rigidly retained at PF. Negative polar questions present a unique environment in Bulgarian where this rigid ordering does not apply. In such questions, li (which is otherwise an enclitic in C⁰) appears not only lower than the pronominal clitics, but further splits the otherwise inseparable cluster, as exemplified in (26).

As elaborately argued by Franks (2008, 1998/2010), the way to straightforwardly account for the surface position of li in cases like (26) is via PI. If the syntax leaves li (an enclitic in C⁰) with no host to its left, at PF, it gets to be lowered and attaches at right edge of the prosodic word. Now, given ne is postaccenting, it is ne MÚ that becomes a prosodic word that li attaches to. Since PI entails lowering of li and attaching it at the right edge of the first prosodic word, the process ends up splitting, otherwise not separable sequence of V-adjacent clitics. Thus, li ends up as an enclitic on ne MÚ, while the rest of the V-adjacent clitics—gi e—in (26) procliticize, as expected, on pokazvala.

26.	9li [ω ne mú] [ω gi e pokázvala] → [[ω ne mú] + li] [ω gi e pokázvala]

In BCMS, PI does not, however, seem to be the right answer for l-participle-li cases.18 In BCMS, PI overgenerates by predicting both the acceptable clitic ordering and the unacceptable ones. Specifically, for a case like (27a), PI would predict both the order in (27b) and in (27c), whereas native speakers’ judgements are consistent only with (27b).

27.	a.	#li	poljubio	je	Mariju?
		Q	kiss.PRCT.SG.M	AUX.3SG.	Marija.ACC
	b.	#[poljubio]ω li ω] je] ω Mariju
	c.	#[poljubio]ω je ω] li] ω Mariju

In case of l-participle li string in BCMS, PI would lower li and attach it to the right of the edge of the ω at the closest relevant prosodic boundary, which could either be ‘[poljubio]ω’ alone or ‘[ [ poljubio]ω je]ω’.

Rather than ‘lowering’, it is the process of ‘raising’ that is required here. There are two possibilities. Either Local Dislocation (Embick & Noyer, 2001; Embick, 2007, 2010) or a kind of process that applies to clitics with purely prosodic dependency and which they conveniently label leaning on a host (Embick & Noyer, 2001, fn.6). Crucially, if it is leaning on a host kind of process, unlike in Embick and Noyer, it has to involve raising of the participle, rather than lowering of li (PI).

Assume, for the moment, that it is Local Dislocation (LD) that takes place in polar li-question in BCMS. LD takes place under adjacency (28) and correctly predicts the only acceptable order at PF is (28).

28.	F[Q]   ⌢   PRCT → [[PRCT] [F[Q]]]

The proposal raises two immediate issues that need to be addressed. Namely, given Embick and Noyer’s conception of LD, ‘ontological demotion’ of l-participle from M(orphological)-Word to Subword is also expected to take place as a result of it. Clearly, this, however, is not the case. In cases like (23), l-participle remains an M-Word. The fact that l-participle is an M-Word, to begin with, leads to the second issue. Namely, whereas LD can apply to both M-words and Subwords, it cannot target ‘ontologically’ mismatched objects, which, given Embick and Noyer’s treatment of clitics and affixes as syntactically on a par (i.e., Subwords), would exactly be the case here.

Across different languages and within the general architecture of Distributed Morphology (DM), other researchers argue that clitics are an independent syntactic category (see, for instance, Kramer, 2014; Preminger, 2014; Despić, 2017, and, most elaborately, Arregi & Nevins, 2012). Specifically, for BCMS, couched in DM, Despić (2017) offers an elegant account of the availability of ellipsis in cases of clitics and unavailability of ‘suspended affixation’ in BCMS, the crux of which is that the former are M-Words, while affixes are Subwords. As illustrated in (29), quite like other clitics, li can be easily elided, and as typical of ellipsis, this always happens in the second conjunct.

29.	Voliš	li	je	i	ceniš	li	je
	Love.PRES.2SG	Q	AUX.3SG	and	respect	Q	AUX.3SG
	‘Do you love her and respect her?’

As for the lack of ‘ontological demotion’, recall that participles in BCMS (Slavic) are complex, always carrying a sub-set of φ-features (number and gender). Considering this, it is not implausible that they remain ‘unaffected’ by LD because ‘ontological demotion’ would lead to the loss of information (see also Despić, 2017, who reasons along the same lines).

To sum, the proposal here draws a parallel between the finite verb and its corresponding participle in polar questions in BCMS. Both finite verbs and participles involve Movement. However, while finite verbs move in syntax, participles move at PF, undergoing a variant of Morphological Merger of Marantz, 1988 (as understood in Harley & Noyer, 1999; Embick & Noyer, 2001).19

4.3. Ordering of PF Operations: Future I and PQ

The way Polarity Questions (PQ) strategies have been discussed so far might suggest that the inventory of PQ strategies in BCMS is rather sparse. This, however, is not the case. BCMS has a rich inventory of PQ strategies. As illustrated in (30), apart from da-insertion (22b) there is je-insertion, and both da and je have contracted forms, da l’ and je l’ (30b) and (30c), respectively. Further, there is zar, a non-clitic lexicalization of Q (30d). As it only appears in two specific biased contexts (see Todorović, 2024 for discussion and elaboration), zar seems to function as a pragmatically conditioned allomorph of li.20

30.	a.	Je	li	su	nestale?
		PRT	Q	AUX.3PL	disappear.PRCT.3PL.F
		‘Have they disappeared?’
	b.	Je l’ su nestale?
	c.	Da l’ su nestale?
	d.	Zar	nisu	nestale?
		Q	NEG.AUX.#PL	disappear.PRCT.3PL.F
		‘Haven’t they disappeared?’

Note now that all of these additional strategies take place at Vocabulary Insertion (VI), an operation at PF. Distributed Morphology is a late insertion model; only abstract features are present at the level of Syntax, while all vocabulary insertion takes place post-syntactically at VI. In Section 4.2, it has been argued that the process by which l-participle-li construction is derived takes place at PF, and involves post-syntactic movement (raising) of participle to ensure convergence i.e., pronounceability at PF and it was proposed that the process in question is Local Dislocation (Embick & Noyer, 2001 et seq.)

Embick and Noyer (2001) and Embick (2007, 2010) provide evidence that Local Dislocation applies after Vocabulary Insertion. There is evidence in BCMS polar questions that also suggest that the PF Movement in l-participle-li questions can only take place after VI. Moreover, in what follows, I will show that in Future I polar questions, l-participle PF Movement takes place only after another instance of LD has taken place.

Before presenting and discussing the data, a brief excursion into Despić’s (2017) analysis of Future I is needed. Examples in (31) illustrate Future I in BSCM, formed from the 2P future clitic and an infinitive (31d). Despić (2017) proposes that whenever the future auxiliary appears at the I-phrase boundary and directly precedes an infinitive (31a), Local Dislocation (31b) takes place in order for the 2P requirement of FUT to be met. This post-syntactic re-ordering is followed by the final -i reduction in Standard Croatian and the full ti-reduction (truncation of the infinitive) in Standard Serbian. The full reduction—‘ontological demotion’ of INF—is further followed by place assimilation typical of stem-affix combinations (see Despić, 2017, for details and elaboration), as in (31c). It seems reasonable to conclude that these phonological changes the clitic and the infinitive undergo can only be explained if the movement occurs post-syntactically. Note, further, that when there is an element that can independently satisfy 2P requirement of FUT, available for insertion at VI, no post-syntactic reordering occurs (compare (31d) with (31e)).

31.	a.	#će	jesti
		FUT	eat.INF
	b.	T[fut]  ⌢  V → [[V] [T[fut]]]
	c.	pro	Ješćeš		(/s/changes to/ʃ /in front of/t͡ɕ/)
		eat.AUX.FUT.2SG
		‘You will eat.’
	d.	Ti	ćeš	jesti?
		you	AUX.FUT.2SG	eat
		You will eat.’
	e.	?*Ti		ješćeš.
		You	eat-AUX.FUT.2SG

This small digression allows us to take a deeper look into the ordering of post-syntactic operations in BCMS by exploring the interaction between Future I and polar questions.

First, Future I polar questions (32) demonstrate that BCMS has to allow for some (well-defined) cases of PF Movement. Namely, regardless of one’s particular account, it seems reasonable to conclude that the newly ‘minted’ M-word ješćeš can only be created at PF. This being the case, the operation that is further responsible for its raising to support li (32) must also occur at PF. Hence, we can conclude that the two PF movements are ordered with LD of INF and FUT taking place first, followed by the LD of the synthetic future form ješćeš and li.

32.	Ješćeš	li	jagode?
	eat.AUX.FUT.2SG	Q	strawberries
	‘Will you eat strawberries?’

Crucially, both instances of PF Movements in BCMS occur after VI, which, further, provides an additional argument that neither of them can be PI. Namely, if phonological material (e.g., da) is inserted at VI, thus being able to satisfy the 2P requirement of clitics (li and ćeš in 33a), post-syntactic reordering—LD of FUT and INF will not take place (compare (33a) with (33b)), since, recall, a single host (da in (34))—inserted at VI—is sufficient to take over the 2P requirement of all the clitics.

33.	a.	Da	li	ćeš	jesti	jagode?
		PRT	Q	FUT.2SG.	eat. INF	strawberries
		‘Will you eat strawberries?’
	b.	?*Da	li	ješćeš	jagode?
		PRT	Q	eat.FUT.2SG	strawberries

The behavior of synthetic future is quite different from finite verbs, because the ‘finiteness’ of synthetic future is a ‘mirage’, arising from PF reordering of the infinitive and the future clitic. To that extent, note the contrast between (33b) and sentences in (34). Sentences in (34) feature a finite verb and both are, as expected, perfectly fine, since the host is supplied either via movement in syntax or via insertion at VI.

34.	a.	Da	li	jedeš	jagode?
		da	Q	eat.PRS.2SG	strawberries
	b.	Jedeš		li	jagode?
		Eat.PRS.2SG		Q	strawberries
		‘Are you eating strawberries?’

To sum up, among the PF processes that take place in BCMS, insertion at VI takes place before PF Movement attested in the language. It then seems reasonable to conclude that if 2P requirement of clitics can be satisfied at VI, that is the preferred option. Viewed from this perspective, it is, perhaps, not surprising that the default strategy in BCMS is argued to be that of insertion (see Todorović, 2024; Šimík, forthcoming, for discussion and elaboration). The PF Movement in the case of l-participle-li is truly the last-last resort option when it comes to polar questions in BCMS. Though last-last resort, its application is still to be preferable since the syntactically well-formed structure results in an unpronounceable string at PF. Namely, given that something like (35) holds, an example like (23), repeated here as (36b), should be understood as an instance of a PF repair that allows integration of the clitic into the prosodic structure.

35.	Full Interpretation:
	In order to be well-formed at PF (i.e., pronounced), phonetic content has to be
	incorporated into prosodic structure.		(Anderson, 2005: p. 39)

36.	a.	li {poljubio}ω	→{{poljubio} ω li}
	b.	%Poljubio li je Mariju?

In other words, PF convergence—the need for every prosodic word to be pronounced—forces this last-last resort, post-syntactic mechanism to apply here. Otherwise, if the PF Movement were not to take place, li would not be integrated into the prosodic structure and the string, though grammatical, would remain unpronounceable.21

5. Conclusions

By exploring the behavior of 2P clitics in the left periphery of the clause, this paper probes into the nature of CS-PF interface in BCMS, a language where the positioning of clitics is argued to be accomplished in syntax and where the role of PF is restricted to filtering out the outputs provided by it. Though the brunt of BCMS data (see Bošković, 1995, 2001; Franks, 1998/2010; Stjepanović, 1999, among others) can be accounted for under this approach, recent data on polar questions, arrived through large-scale experiments (see Marelj et al., 2024) as well as data on polar Future I questions suggest that there are well-defined configurations in which post-syntactic operations need to take place in BCMS as well. Contra previous analyses, the account presented here treats li (lexicalization of Q) as a ‘run-of-the-mill’ 2P clitic in BCMS, whose host, when illicit to be provided by movement in syntax, can be provided by post-syntactic movement. I provide evidence against Prosodic Inversion—‘the usual suspect’ for post-syntactic movement in Slavic—thus adding to the body of evidence that PI does not take place in BCMS. I argue that the PF Movement in such cases has to be one of raising and adopt Local Dislocation (Embick & Noyer, 2001; Embick, 2007, 2010) to account for them. Probing into the interaction between Future I and polar questions provides further insights into the ordering of the PF Movement operations in BCMS.