Getting in the first word : Prosody and predicate initial sentences in Serbian

Second position clitics in Serbian are informally characterized in terms of their hosts which correspond to either the first word or the first phrase within some clitic positioning domain. Serbian is a free constituent order language, with a default SVO order, but with predicate initial orders in the presence of pro-drop or postposed subjects. Earlier proposals range from those in which first phrase positioning is taken to be syntactic and first word to be prosodic, with the initial prosodic word hosting the clitics (Zec and Inkelas 1990; Halpern 1995), to those which analyze both FirstPhrase and FirstWord as a unified syntactic phenomenon (Franks and Progovac 1994; Bošković 2001). While we argue for the relevance of prosody in first word placement, we shed new light on the scope of its role.


Background
Second position clitics have long attracted attention as a window onto the interfaces between prosody and syntax.In this paper, we take as our empirical base second position clitics in Serbian.In Serbian, clitics are informally characterized in terms of their hosts, which correspond to either the first word or the first phrase (within some clitic positioning domain; Klavans 1982;Anderson 2005; among others).Serbian is a free constituent order language with a default SVO order, but with predicate initial orders in the presence of pro-drop or postposed subjects (Predolac 2011 gives a more detailed account of the principles underlying the word order possibilities).
We will be focusing on three major issues.The first of these is the choice of clitic host -what determines FirstWord vs. FirstPhrase clitic placement?The second question is that of the impact of the initial constituent -predicate vs. argument -on clitic placement in general, and the choice of FirstWord vs. FirstPhrase in particular.Finally, what is the role of prosody in clitic placement?In considering the actual placement possibilities (or cases) it is important to distinguish between simply focusing on the clitic host (as in (1a) below), and taking into account the initial constituent (as in (1b)).
(1) a. Word/phrase placement FirstWord FirstPhrase b.Context-rooted word/phrase placement Pred FirstWord Pred FirstPhrase Arg FirstWord Arg FirstPhrase Glossa general linguistics a journal of In other words, we claim that the more fine-grained classification in (1b) is essential, and that the standard two-way split in (1a) is inadequate to capture the data.The four cases are illustrated in (2) and ( 3): ( In the remainder of this paper, we turn our attention to the question of the role of prosody in clitic placement.In section 2 we present evidence for the relevance of FirstWord clitic hosts, arguing against proposals which question the relevance of this host type.In section 3, which focuses on the role of prosody in FirstWord clitic placement, we show an asymmetry in this respect between argument and predicate initial sentences.In section 4 we examine topicalized PPs, with a particular emphasis on those headed by prepositions which have dual status -functioning either as clitics (as is usual for prepositions) or as prosodic words, showing that the asymmetry established in section 3 extends to this case as well.In section 5 we present our analysis of clitic placement in Serbian, and in section 6 we close with concluding remarks.

The status of FirstWord clitic hosts
In this section we present empirical evidence for the existence of FirstWord hosts which, as we argue, present a robust category.This perspective is based on the Diesing, Filipović-Đurđević and Zec study (2009;henceforth DFZ) which provides corpus and experimental evidence for the four-way division in (1b).In the corpus study, the proportions of the four cases were assessed in two corpora, a corpus of the daily press, and a corpus of literary prose.1 Similar studies were conducted by Pereltsvaig (2008) for Russian, and Pancheva (2005) for Old Bulgarian.The DFZ corpus study demonstrated that FirstWord placement is far more common when the initial constituent is a predicative phrase.When the initial constituent is an argument, FirstPhrase placement is more common.The corpus study thus showed essential asymmetries among the four factors: the two initial constituent types, and the two placement options.
The experimental part of the DFZ study included production and acceptability judgment tasks, with both experiments using the same sets of stimuli.As shown in Table 1, these consisted of 120 sentences with equal numbers of the argument and predicate cases.Within the argument initial case, there were the NP (subject and object) and PP categorial possibilities, with a further division between those with a demonstrative or possessive determiner, or with an adjectival modifier.For the predicate initial case, three options were presented: VP, NP, and AP.These also had various branching possibilities, with AP branching into specifier + A, NP branching into adjective + N, and VP branching either into V + object or adverb + V.
In the production experiment, which utilized a paper and pencil questionnaire, 38 participants were asked to choose between FirstWord and FirstPhrase clitic hosts.In each sentence presented to the participants the critical clitic was omitted, and the two positions of clitics (after the first word and after the first phrase) were replaced with a line, i.e. a blank to be filled in, as shown in ( 4 As can be seen in the chart in Figure 1, the results revealed that 97.59% of participants chose FirstWord placement in predicate initial sentences and only 7.02% chose FirstWord placement in argument initial sentences.Logistic regression performed on participants' responses, with the dependent variable being the placement of a clitic in one of the two possible positions for each of the two sentence categories, revealed that the observed difference was significant: χ 2 (1) = 1557.16,p<0.0001.
In a second experiment focusing on acceptance rates, the stimuli were presented on a computer display, and the participants (48 in this case) were instructed to judge, in a binary fashion, whether or not the sentence appearing on the screen was acceptable in their language.Here too an asymmetry was revealed.The acceptance of argument initial sentences was 97% in the FirstPhrase position, 92% for FirstWord.For predicate initial sentences acceptance of the FirstPhrase position was 72% and 97% in FirstWord.This can be seen in Figure 2.
While both placement options were judged acceptable for both the argument and predicate sentence types, in each case there is a clear, distinct, and statistically significant preference for one of the clitic placements for each sentence type.Logistic regression of yes/no answers in the sentence acceptability task revealed a significant main effect of sentence type: χ 2 (2) = 232.65,p<0.0001, a significant main effect of clitic position: χ 2 (2) = 228.12,p<0.0001, and a significant interaction between the two: χ 2 (1) = 181.24,p<0.0001.Reaction times taken in the acceptability task conformed to this pattern, with FirstPhrase placement in argument initial sentences being processed faster than FirstWord placement, and FirstWord placement being processed faster than FirstPhrase placement in predicate initial sentences (as reported in the DFZ study).
It should be noted that, in predicate initial sentences, somewhat degraded results in both the production and acceptability judgment tasks are due to relatively low scores in the VP case, when compared with the AP and NP cases.(This result of the DFZ experiment was not reported in the 2009 paper.)In the production task, FirstPhrase was selected in 6.31% for NPs, 0.92% for APs, and only 0.13% for VPs.Acceptance rates for the FirstPhrase option were 89.04% for NPs, 80.23% for APs, and 47.61% for VPs.Thus VPs score fairly low in both production and acceptability judgments, which strongly suggests that they are highly disfavored as FirstPhrase hosts.
The DFZ study not only provides evidence for the classification in four-way (1b), but also establishes a pattern of preferences within that classification.This is summarized in the table below.
We attribute this pattern of preferences to the information structure associated with each of the four clitic placements.The two preferred types of clitic hosts are consistent with discourse neutral interpretations.However, FirstWord hosts in argument initial sentences and FirstPhrase hosts in predicate initial sentences call for special discourse conditions that we associate with a [+ contrast] feature specification.In particular, sentences with  these less preferred clitic placements are fully acceptable if interpreted as a contrastive topic or a contrastive focus (Diesing 2010;Diesing & Zec 2011).2Overall lower preference for these clitic hosts may well reflect the intricacy of accommodating for a [+contrast] interpretation in the absence of any context.We close this discussion with further examples of the four types of cases that the DFZ study has established as essential for any approach to the second position clitics in Serbian; all are taken from the DFZ set of stimuli.Argument initial sentences are exemplified in (5)-( 7), and predicate initial sentences in ( 8)-( 11).Only cases with adjectives as specifiers are exemplified in argument initial sentences.In predicate initial sentences, both types of VP cases are presented, with the indication that (10b) and (11b) are considerably degraded.
( In the remainder of this paper, we turn our attention to the question of the role of prosody in clitic placement.

Prosody in FirstWord clitic placement: The argument/predicate asymmetry
As follows from the discussion above, we will be focusing primarily on FirstWord placement.But before discussing our perspective in more detail, we place it in the context of earlier work.While the DFZ study provides strong support for differentiating the FirstWord and FirstPhrase clitic placements, the status of this distinction in previous accounts of second position clitics in Serbian is less clear.It is important to note that earlier proposals focus primarily on argument initial cases, such as those in ( 5)-( 7), essentially ignoring the predicate initial cases in (8)-( 11).The detailed description of the facts in Browne (1975) is a notable exception.
Working with this fairly restricted set of data, the previous analyses differ from each other largely in whether the placement of clitics is syntactic (Franks & Progovac 1994;Bošković 2001), prosodic (Radanović-Kocić 1996), or accomplished by some combination of means (Halpern 1995;Zec & Inkelas 1990).Within a purely syntactic approach to clitic placement, the distinction between FirstWord and FirstPhrase placement falls away -both are cases of clitics attached to phrases, in parallel to analyses of left branch movement; there is no formal distinction between the two types.Clitic placement in Serbian is thus represented as a unified phenomenon.Proponents of the purely syntactic approach reserve all reordering operations to the domain of syntax, with no PF adjustments being allowed (Bošković 2000).Various alternative approaches (e.g.Zec & Inkelas 1990;Halpern 1995) permit a mixed view, in which the FirstWord/FirstPhrase distinction is manifest theoretically -FirstWord placement is prosodic, and FirstPhrase placement is syntactic.The key assumption underlying this bifurcation is that FirstWord clitic hosts are prosodic words (Zec & Inkelas 1990;Zec 2005), an assumption that presents a challenge to the exclusively syntactic accounts.But, as already noted, all previous accounts rely on a limited database, primarily the argument initial sentences, and it is in these cases that the DFZ study finds that FirstWord placement is the less preferred option.
With this background, we now turn to exploring the role of prosody in clitic placement.In particular, we will argue for the importance of prosody in FirstWord placement both in predicate initial and argument initial sentences.Though some earlier proposals have claimed a role for prosody in clitic placement, we substantially expand the scope of that role.In particular, we have considerably expanded the database by introducing predicate initial FirstWord cases along with the argument initial cases already well known in the literature.We further claim that the placement of second position clitics is not uniform, and that factors other than prosody, such as syntax and information structure, play a significant role.These further factors shed light both on the bifurcation into FirstWord and FirstPhrase clitic hosts, and on the pattern of preferences shown in Table 2.We thus claim that the expanded set of data established by the DFZ study cannot be accommodated in a uniform fashion, either syntactic or prosodic, and crucially calls for an interface approach.
We begin with the proposal that second position clitics prosodically subcategorize for prosodic words, which in the case of FirstWord placement also serve as their sole hosts (following Zec & Inkelas 1990; as well as Zec 2005 and the references therein).Any lexical element, as well as a small set of designated functional elements, maps into a prosodic word, which is prosodically independent as a bearer of stress and pitch.This is illustrated in (12a), where the FirstWord host corresponds to an adjective in the specifier position that maps into a prosodic word, with the clitic attaching to it recursively and forming another prosodic word layer.The formal mechanism for this attachment is the clitic's lexically encoded subcategorization frame, which insures that a clitic both attaches to a prosodic word host, and forms with its host a new prosodic word (following Inkelas 1990;Zec & Inkelas 1990;Chung 2003; among others).The subcategorization frame in (12b) encodes both the type of prosodic host, and the directionality of attachment (for arguments, see Zec 2005;Zec & Filipović-Đurđević 2016).In this case, it mandates that, in the prosodic component, the clitic should be immediately preceded by a prosodic word host.An important question to ask is whether prosody alone is responsible for FirstWord clitic placement in both the argument and predicate cases, under the broadly held view that syntactic information is not readily available in the prosodic component (Selkirk 1980;1986;Nespor & Vogel 1986;Truckenbrodt 1995;Chung 2003;among others).That is, can any sentence initial prosodic word serve as a FirstWord clitic host, regardless of the role it plays in the syntax, and regardless of its larger syntactic context?This question was addressed in an experimental study by Zec and Filipović-Đurđević (2016), which follows the DFZ design.Expanding on the range of data used in the DFZ study, they presented a variety of initial constituents in order to vary the configurations of the FirstWord

FirstPhrase FirstWord
Argument Preferred Predicate Preferred  3. The Spec position below admits adjectives as well as possessive and demonstrative determiners which we assume, following Despić (2013), to be adjoined to NP. Examples for all cases in Table 3 are given in ( 18)-( 21), annotated with the acceptability values that emerged from the experiment.The Zec and Filipović-Đurđević results reveal a pronounced asymmetry between argument initial and predicate initial sentences.In the predicate initial context, any initial prosodic word can serve as a FirstWord clitic host, while in the argument initial context only initial prosodic words that do not correspond to syntactic heads can readily assume this role.These results hold for both production and acceptance tasks.
The production task consisted of a completion task using a pencil and paper questionnaire.The participants were asked to complete sentences in which the clitic had been omitted.The chart in Figure 3 summarizes the results for all sentence types.
While in predicate initial sentences participants overwhelmingly placed clitics after the FirstWord hosts, in argument initial sentences FirstWord placement was much less frequent, and reached relatively significant numbers only in the pre-head case.
This asymmetric pattern was also borne out in the acceptance task.In predicate initial sentences FirstWord hosts were highly acceptable for all three head types, as shown in the chart in Figure 4.But in argument initial sentences, acceptance rates are high for FirstWord hosts of the pre-head type, but are considerably degraded for the two head initial types, as shown below in Figure 5.
Taking into account both production and acceptability results, we note a prominent presence of FirstWord hosts only in the predicate case, but not in the argument case.The experimental results strongly suggest that in the predicate case, all three types of FirstWord hosts figure robustly in the participants' performance.In the argument case, however, only FirstWord hosts of the pre-head type show a significant presence, while FirstWord hosts of the post-head and double head types are practically absent from production and are accepted at considerably degraded rates.This leads to the conclusion that all types of FirstWord hosts are available in predicate initial sentences, while only FirstWord hosts of the pre-head type robustly figure in argument initial sentences (for detailed discussion, see Zec & Filipović-Đurđević 2016).This asymmetry in the availability of FirstWord hosts is presented in Table 4:  Next, we present cases of each FirstWord head type in both structural positions.The examples are all taken from the stimuli of the experiment whose results we just reported (Zec & Filipović-Đurđević 2016).All acceptability patterns are those established by the reported experiments.

Predicate initial
In predicate initial sentences, the clitic can occur after a FirstWord host which is in a specifier, head, or following the first of two conjoined heads.Thus, as shown in ( 14) and ( 15), all head types are available in predicate initial sentences, both with initial APs and NPs.Moreover, in situations where the clitic follows the head, the type of constituent following the clitic is of no consequence, as can be seen in (14b, c) and (15b, c). ( 14) Initial AP a. Neobično su darežljivi prema amaterskim pozorištima.Pre-head unusually are-cl generous towards amateur theaters 'They are unusually generous towards amateur theaters.'b.Željni su slobode, a roditelji to ne desiring are-cl of freedom, but parents that not razumeju.
Post-head (NP, gen) understand 'They crave for freedom, but their parents do not understand that.' c.Osetljiv je na prigovore, i zato treba da ćutimo.Post-head sensitive is-cl to criticism and therefore (we) should be silent 'He is sensitive to criticism and we should therefore keep silent.'d.Srećna je i uzbuđena što ide na takmičenje.Double-head happy is-cl and excited because (she) goes to competition 'She is happy and excited because she will be participating in the competition.'

Argument initial
In argument initial sentences, however, only pre-head FirstWord hosts are available both with initial subject and with initial object NPs, as shown in ( 16) and ( 17) below.Clitics can readily be placed following a FirstWord host that corresponds to a specifier or other left branch element, as in (16a) and (17a).Clitic placement following any FirstWord host corresponding to a syntactic head is practically unavailable.The sentences in ( 16c, d, e) and (17b, c, d) become felicitous if the clitic is placed after the FirstPhrase, the standard clitic placement in the argument case.As already noted, FirstWord hosts in the argument case bear the feature [+contrast], indicating that the referent of the initial constituent is contrasted with alternatives in a set (usually that denoted by the head noun).

Minimal distinctions
The asymmetry between the predicate and argument cases is brought into relief with examples of minimal distinctions in the availability of FirstWord hosts.Thus, the First-Word host of the post-head type is available in (18a), a predicate initial sentence, but not in the argument initial sentence in (18b).( 18 Note that all these sentences are perfectly acceptable with the clitic placed after the First-Phrase; it is only FirstWord placement that imposes this constraint. What are the implications of the pattern summarized in Table 4 for the prosodic status of FirstWord hosts?In the predicate case, any initial prosodic word can host clitics, regardless of its syntactic collocation.It is precisely such lack of sensitivity to syntactic configurations that is the hallmark of the prosodic scenario (e.g., Chung 2003).In the argument case, however, only initial prosodic words that correspond to a pre-head element can serve as FirstWord hosts, while those corresponding to a constituent head are less than felicitous in this role.Such sensitivity to syntactic configurations is inconsistent with a prosodic scenario.One possible interpretation is that FirstWord hosts are selected prosodically in the predicate initial case, and are characterized syntactically in the argument initial case.However, this move does not accommodate the prosodic word status of FirstWord hosts in both the predicate and argument cases, which is further accentuated with initial prepositional phrases (PPs), to be addressed in section 4.
The hypothesis we will pursue here is that FirstWord hosts are selected in the prosodic component for both initial predicates and initial arguments.But in the latter case we will have to allow for an interference coming from another component of the grammar.This, we argue, is the information structure.The feature [+contrast], already posited for FirstWord hosts with initial arguments, is inconsistent with elements serving as syntactic heads.In sum, in the predicate case, prosody alone is responsible for the selection of FirstWord hosts, while in the argument case, prosody interfaces with syntax and the information structure in the selection of FirstWord hosts.

Topicalized PPs: argument/predicate asymmetry persists
The proposed asymmetry in the availability of FirstWord hosts in predicate initial and argument initial sentences has important implications for sentences introduced by prepositional phrases (PP).As function words, prepositions do not routinely map into prosodic words, and in the case of Serbian, predominantly assume the role of proclitics, that is, they prosodically attach to their host prosodic word at its left edge.
Of interest here is a subclass of prepositions that optionally map into prosodic words, and thus vary between clitic and prosodic word status.These prepositions with dual prosodic status will provide further evidence for the asymmetry between FirstWord hosts in argument initial and predicate initial sentences.The prediction is that PPs headed by prepositions with dual prosodic status should provide multiple options for FirstWord clitic hosts (note that dual prosodic status is also evidenced in some conjunctions and auxiliaries, as discussed in Zec 2005, but this is not relevant to our claims here).Crucial for us is the question whether this phenomenon exhibits the type of asymmetry we have already established for other types of FirstWord hosts.Stated more specifically, we predict that these prepositions, being syntactic heads, should occur freely as clitic hosts in contexts with PP predicate fronting.But where PP arguments are sentence-initial, the prepositions will be disallowed as clitic hosts.
In what follows, we present initial PPs in the predicate case in section 4.1, and in the argument case, in section 4.2.In both cases, we examine the behavior of both prosodically deficient prepositions as well as those with dual prosodic status.In section 4.3, we present yet another instance of the incompatibility of FirstWord hosts with constituent heads in argument initial sentences.

Initial PPs in the predicate context
In this section we present two cases of initial predicate PPs, varying in their prosodic status.The first case deals with those prepositions that have the prosodic status of clitics, and the second case with a set of prepositions that have dual status, functioning either as clitics or as prosodic words.

PPs headed by prepositions with the prosodic status of clitics
Focusing on the predicate initial case, we first present sentences with topicalized PPs headed by prepositions which function prosodically as clitics.As such, they are attached to a host prosodic word, recursively forming another prosodic word layer, as in ( 20).The initial prosodic word in (21) which serves as FirstWord host has the prosodic structure as in ( 23).As shown in ( 24), the preposition u, which being a clitic is prosodically deficient, cannot serve as a FirstWord host.The ill-formed (24) clearly demonstrates the prosodic condition on the host of the clitic.Without attaching to a prosodic word host, the preposition, being prosodically deficient, cannot itself host a clitic.With this in mind, in the next section, we turn to the prepositions that have dual prosodic status.

PPs headed by prepositions with dual prosodic status
In addition to the prepositions that invariably have the status of clitics, attaching to the following prosodic word, as illustrated in ( 22), there is also a class of prepositions with dual prosodic status.Below we provide examples for three such prepositions: protiv 'against' in ( 25)-( 26), van 'out of' in ( 27)-( 28), and ispred 'in front of' in ( 29)-( 30).
The preposition protiv can either act as a proclitic, joining the following prosodic word and forming with it a FirstWord host protiv svake, as in (25), or it can form a prosodic word on its own which acts as a FirstWord host, as in (26).Note that, in (25), the first prosodic word protiv svake 'against every' is a non-constituent, and is part of the larger phrase protiv svake ekonomske logike 'against every economic logic'.In ( 26), but not in (25), the preposition exhibits standard properties of phonological words, bearing stress and pitch accent.(In these and following examples, context sentences are translated but not glossed.)(25) Context: 'Prices of raspberries are too low.'Protiv svake su ekonomske logike.against every are-cl economic logic 'They go against any economic logic.' (26) Context: 'They are so very radical.' Protiv su poreskog sistema u zemlji, kao takvog.against are-cl tax system in (the) country as such 'They are against the system of taxation as such.' Likewise, the preposition van can either form the FirstWord host jointly with the following prosodic word, as in van životne in (27a) and van okvira in (27b); or on its own, as in the two cases in (28).( 27 When the subcategorization frame is invoked, the preposition acts as a clitic, and when it is suppressed, the preposition maps into a prosodic word and is thus capable of serving as a FirstWord clitic host.

Initial PPs in the argument initial context
We now turn to the behavior of the two classes of prepositions in argument initial sentences.We first present argument initial PPs headed by prepositions with the prosodic status of clitics, and then we consider topicalized PPs headed by prepositions of dual status.

PPs headed by prepositions with the prosodic status of clitics
Topicalized PPs in the argument case are like those in the predicate case in at least one respect: in PPs headed by a preposition that invariably corresponds to a clitic, the First-Word clitic host includes the preposition prosodically attached to the following phonological word, just as in the predicate case.This is shown in (34): the string na novim constitutes an initial prosodic word that serves as a FirstWord clitic host, (34a), while the string od moje performs this role in (34b).
(34) a. Na novim se uredjajima primećuju mnogi propusti.on new se-cl equipment noticed many omissions 'On the new equipment one can notice many omissions.'b.Od moje je odluke zavisilo dalje poslovanje firme. of my is-cl decision depended further activity (of) company 'Future activity of the company depended on my decision.'Again, as in the predicate case, the initial preposition, being itself a clitic, cannot host the second position clitic.

PPs headed by prepositions with dual prosodic status
Turning now to prepositions with dual prosodic status, we see that they behave differently in predicate initial and argument initial PPs.In the argument initial sentences these prepositions act only as clitics, as shown below in (35a) and (36a); they cannot form FirstWord hosts on their own, even though this option is available in predicate initial cases.Thus, in the argument initial case only one of the two options is realized, while in the predicate initial case both options are possible.As shown in section 3.2, only prosodic words sponsored by non-head syntactic elements can serve as FirstWord hosts in the argument case.Dual status prepositions, which serve as PP heads, follow this pattern.Although they can sponsor prosodic words, their head status prevents them from serving as FirstWord clitic hosts in the argument case.As syntactic heads, these prepositions are incompatible with the feature [+contrast], a required discourse attribute for FirstWord hosts with initial arguments.The examples in ( 35)-( 36) show minimal pairs illustrating this asymmetry.In the first of each pair, the preposition functions as a clitic, and adjoins to a left branch element in the phrase.The result is a prosodic word which can function as a FirstWord clitic host.In the (b) examples, where the clitic functions as a prosodic word on its own, its syntactic head status is inconsistent with its FirstWord role in an argument initial sentence, as predicted.

Prepositions form a FirstWord host only with a non-head
The head/non-head asymmetry with initial arguments is manifested in yet another interesting way.Our finding that FirstWord placement in the argument initial case is only possible with non-head prosodic word sponsors extends to PP arguments, both those headed by prepositions that correspond to clitics and those with a dual status.Namely, while a preposition may prosodically attach to a prosodic word sponsored by any morphosyntactic element, this prosodic word is a possible FirstWord host only when its sponsor is a syntactic non-head.This is illustrated in (37)-(38).In (37a), the clitic preposition o 'about' attaches to a prosodic word sponsored by the possessor Brankovoj 'Branko's'.As the sponsor is a non-head, the resulting prosodic word o Brankovoj can act as a FirstWord host.In (46b), the prosodic word sponsor is the nominal head poeziji 'poetry ' and, due to (38), where the PP is headed by a dual status preposition.In (38a) the preposition attaches to a prosodic word sponsored by a demonstrative (occupying a left branch of the NP; again see Despić 2013), which makes ispred ove an eligible FirstWord host.The example (38b) shows that if the preposition attaches to a prosodic word sponsored by the head noun, the resulting prosodic word ispred zgrade is not a licit FirstWord clitic host.Note that this asymmetry does not extend to the predicate case.In (39), the prosodic word sponsor corresponds to a syntactic head.The preposition attaches to a prosodic word sponsored by the NP head, leaving out its complement, and yielding na dnu as the First-Word host.In (40) (repeat of (27b)), the initial PP is headed by a dual status preposition which, in this case acts as a clitic, and attaches to the following prosodic word sponsored by the NP head, yielding van okvira as the FirstWord host.(Again, the preceding contexts are translated but not glossed.)(39) Context: 'How did they rank in the competition?'Na dnu su lestvice.on bottom are-cl scale 'They are at the bottom of the list.' (40) Context: 'This question is very complicated,' I van okvira je ovog članka.and outside frame is-cl this article 'and is outside the scope of this article.'Thus unlike in the argument case, in the predicate case any initial prosodic word can act as a FirstWord host, regardless of whether its sponsor is a syntactic head or a syntactic non-head.

Serbian clitics in the grammar
Among earlier proposals, most influential were the accounts which assumed that clitic placement in Serbian is uniform, and that what descriptively appears to be a bifurcation into FirstPhrase and FirstWord clitic hosts can be reduced to a FirstPhrase case and addressed solely in the syntactic component.With only argument initial sentences taken into account, this assumption was reasonably tenable although not devoid of problems.
However, we have presented a broader set of data, which includes not only argument initial but also predicate initial cases, demonstrating the formal significance of FirstWord clitic hosts, which cannot be subsumed under the FirstPhrase case.We therefore pursue a non-uniform approach to clitic placement in Serbian, proposing a formal account which employs different mechanisms for the two cases.However, we propose here a revision to the account we presented in our earlier work (Diesing & Zec 2011).Our claim is that, in both argument initial and predicate initial sentences, FirstPhrase hosts are selected in the syntax while FirstWord hosts are selected in the prosodic component.
We further argue that both FirstWord and FirstPhrase clitic placement are non-uniform, but in distinct ways.In the case of FirstWord hosts, predicate initial sentences and argument initial sentences differ in that the latter clearly show the influence of discourse factors on clitic placement.For argument initial sentences, this type of clitic placement is sensitive to information structure: FirstWord hosts can only be elements that are marked [+contrast] (of the type proposed in Kiss 1998;Féry & Samek-Lodovici 2006).We relate to this our finding that FirstWord hosts are incompatible with the host being a syntactic head, and further assume that constituent heads repel the feature [+contrast].In predicate initial sentences, FirstWord placement is uniformly prosodic and is insensitive to information structure.It is the preferred placement, and has an unmarked, presentational interpretation.
FirstPhrase hosts are selected in the syntactic component in both cases, but also differ in information structural import.For argument initial sentences, FirstPhrase placement is the preferred case, and this placement applies quite generally.FirstPhrase placement in predicate initial sentences is a contrastive option, and is only possible with non-verbal predicates -i.e. in copular sentences.
What all four cases have in common is that the clitics are associated with the highest functional head, as convincingly argued in Stjepanović (1998) (see also Bošković 2001;Progovac 2005).With this crucial assumption, we begin by considering the two preferred types of clitic hosts, argument initial FirstPhrase placement and predicate initial FirstWord placement.We propose that the first of these involves instances of topic or focus fronting (we assume some version of an expanded left periphery, as in Rizzi 1997;2004;and others).
For illustration purposes we will consider the case of a clause-initial topic.A sentenceinitial subject NP is a default topic.All other XPs may appear sentence-initially via scrambling, and are marked Top/Foc, depending on discourse structure.Clitics are in the highest functional head, namely the head of the functional projection hosting the topic XP.The clitics attach to XP (and prosodically, to the closest prosodic word on the left), and the lower copies of the XP and clitic(s) are deleted (assuming a copy theory of movement).The example in (41) illustrates an object topic; we show only the relevant movement of the object NP.In the case of predicate initial sentences, the preferred FirstWord placement is prosodic, rather than syntactic.Recall that VP-initial orders arise particularly in the presence of pro-drop or postposed subjects.
Furthermore, predicate fronting in Serbian is subject to certain discourse conditions comparable to those in English (Ward 1990;Landau 2006;2007).In particular, a fronted predicate may invoke a meaning already in the discourse, providing some sort of affirmation (Ward 1990).They are thus anaphoric in a certain sense.(There is also the possibility of a contrastive meaning, we will address that option below in the discussion of the predicate initial sentences with first phrase placement.)This is particularly the case for the Serbian examples with postposed subjects, which we illustrate in (42).We propose to represent the anaphoric link with the preceding discourse required by the fronted predicate with an empty operator in Spec,CP.The clitic(s) occupy the head C, and the predicate (whether a verb phrase, or the predicate of a copular sentence) is in Spec,TopP.Thus, the "initial" predicate actually follows a null operator in Spec,CP.Clitic(s) in C require a host (in particular, the clitic requires a host on its left, a subcategorization condition that cannot be met when it appears in initial position; see also Chung 2003 on Chamorro) and therefore lowers to the right of the first word by prosodic inversion (Halpern 1995;Condoravdi & Kiparsky 2001;  Sentences with pro-drop differ from those with postposed subjects in that they do not require a discourse antecedent for the sentence-initial predicate.Indeed, it is not clear that in these cases any fronting has taken place at all.With the pro subject in Spec,TP, and the clitic in T, the clitic is here also in a position requiring a prosodically induced rightward adjustment to find a host, as shown in ( 43 Thus, in the case of predicate initial sentences, the FirstWord placement is prosodic, rather than syntactic.In particular, the clitic requires a host on its left, a condition that cannot be met when it appears in initial position, calling for prosodic inversion. Turning now to the discourse marked cases, in argument initial sentences the left branch element of the XP is marked [+contrast].As with predicate initial FirstWord hosts, these sentences have an empty operator, but in this case the empty operator is a Contrast operator rather than semantically empty.There is an initial scrambling of the NP to TopP to check the Topic feature.If there is no further movement of the NP, the NP checks its Contrast feature via Agree.The clitic in C which, having no overt host to its left is stranded, must then undergo prosodic inversion.Example (44) shows a fronted object NP. (As in the earlier example, we show only the movement of the object NP in the tree below.)FirstWord cases within topicalized PP arguments, as discussed in section 4, shed further light the interplay between prosodic and syntactic conditions on host selection: prosodic words eligible for FirstWord hosts are those sponsored by a left branch, even if that element belongs to the NP complement.Thus the morphosyntactic element sponsoring the initial prosodic word has to be a syntactic non-head, but does not have to be at the highest level of syntactic constituency, leading to potential mismatches between syntactic and prosodic constituencies.
To summarize, in both cases of FirstWord clitic hosts, prosodic inversion is a consequence of the fact that the syntax does not provide a prosodically realized host in initial position -the initial specifier is occupied by an empty element in both cases.The argument initial case is distinguished from the predicate initial case in that the initial operator is associated with the [+contrast] feature.
Finally, FirstPhrase placement in predicate initial sentences involves fronting of a PredP to a Topic position on the left periphery.As mentioned above, the fronting of the predicate induces a meaning in which the predicate invokes a set of meanings with which it is contrasted (Landau 2007).These are not expletive constructions, and the clitic (or clitics) is in the head of TopP, that is, the projection hosting PredP, as in (45).There is no vP/VP fronting to this left-peripheral position, only PredP may undergo Topic fronting -this construction is only possible with copular sentences.Similar restrictions on VP-fronting have been observed for South Slavic (Bulgarian, Czech, Slovak, and Slovenian) and Romance (Old Spanish, early forms of European Portuguese, and modern Romanian) by Rivero (1991) and Lema and Rivero (1989), but see Wilder and Ćavar (1994) for a claim that such VP-fronting is allowed in Croatian.Earlier accounts expressed the restriction in terms of the ECP -auxiliaries are not able to license VP traces. 4We simply stipulate that VPs cannot be contrastive Topics in Serbian.As in the case of non-predicate fronting, the clitic attaches to PredP, and prosodically, to the preceding prosodic word.

Concluding remarks
We have proposed an account of clitic placement in Serbian in which FirstWord clitic placement is prosodic and FirstPhrase clitic placement is syntactic.While this may seem more complex than previous unified accounts, our analysis analysis is based on a substantially expanded data set, which raises empirical objections to the validity of a uniform approach.
We are, however, left with the puzzle of why FirstWord placement after the head in the argument initial case is less than felicitous.This issue was in fact addressed in Halpern (1995), who posited a prosodically based constraint to eliminate FirstWord hosts that correspond to constituent heads in a set of contexts he called fortresses.As in most previous work, his data included only argument initial cases.In addition, stating this constraint in prosodic terms makes it overly restrictive once predicate initial cases are included in the picture.
As characterized in our analysis, FirstWord placement in the predicate and argument cases differ in terms of information structure.In the argument case FirstWord placement is only possible after a specifier or adjunct on the left branch which is marked [+contrast].We claim that this is due to the following constraint: (46) The head constraint: The head of an argument phrase cannot bear structurally induced contrast.
This constraint is reminiscent of conditions on the Nuclear Stress Rule, which avoids placing highest prominence on the head (Cinque 1993;Arregi 2002), or conditions on focus projection, with focus marking a non-head, but projecting out of the head (Selkirk 1995).
In Serbian, a related condition is manifest in the local domain circumscribed by the clitic and its FirstWord host, stated here as (46).However, more work will be required to understand the full import of this condition.

Figure 2 :
Figure 2: Acceptance rates: mean acceptance rates for the argument (left), and predicate sentences (right) with a clitic positioned after the first word (light gray), and after the first phrase (dark gray).

Figure 1 :
Figure 1: Production results: percent of participants placing a clitic after the first word (light gray), and after the first phrase (dark gray) when completing argument (left), and predicate sentences (right).

Figure 3 :
Figure 3: Production results: percent of sentences completed with clitic placed after the first word.

Figure 4 :
Figure 4: Acceptance rates for predicate initial sentences.

Figure 5 :
Figure 5: Acceptance rates for argument initial sentences.

FirstWord
that refl-cl song pleased both (the) audience and (the) jury 'That song was favored by both the audience and the jury.' b.Loše ćemo igrače izbaciti iz prve ekipe.bad will-cl players kick out from first team 'We will kick out bad players from the first team.'c. *Članovi su kluba dobili nove knjižice.Post-head (NP, Gen) members are-clof-club received new membership cards intended 'Members of the club received new membership cards.' thieves refl-cl and crooks always somehow manage intended 'Thieves and crooks always somehow manage.' this are-cl decision reluctantly accepted 'They reluctantly accepted this decision.'*Ministra su prosvete zasuli brojnim minister are-cl of-education flooded with numerous pitanjima.Post-head (NP, Gen) questions intended 'They flooded the minister of education with numerous questions.' b. *Studente su iz Beograda nagradili prošle godine.Post-head (PP) students are-cl from Belgrade (they) awarded last year intended 'They gave an award to the students from Belgrade last year.intended'Fairies and witches have been expelled from modern children's stories.' host ] PW ] PW As already noted, the resulting prosodic word need not correspond to a syntactic constituent.This can be seen in the examples presented in (21)-(22): the preposition forms a prosodic word with the modifier, yielding u dobroj in (21) and na lošem in (22), each serving as a FirstWord clitic host.(21)U dobroj smo fizičkoj kondiciji.in good are-cl physical shape.'We are in a good physical shape.' reputation because of frequent misconduct 'He has a bad reputation because of frequent misconduct.' in a good physical shape.' heard of the organization that promotes young talents.' a. Protiv te se organizacije vodi sudski postupak.against that se-cl organization file law suit 'A law suit has been filed against that organization.'b. *Protiv se te organizacije vodi sudski postupak.against se-cl that organization file law suit (36) a. Ispred ove je zgrade postavio saobraćajni znak. in front of this is-cl building placed traffic sign 'In front of this bulding he placed a traffic sign.' b. *Ispred je ove zgrade postavio saobraćajni znak. in front of is-cl this building placed traffic sign this is-cl building placed traffic sign 'In front of this bulding he placed a traffic sign.' b. *Ispred zgrade je opštinskog suda postavio saobraćajni znak. in front of building is-cl (of) municipal court placed traffic sign task 'This task is very important.'b.

Table 1 :
Types of sentences used in the psycholinguistic experiments.

Table 2 :
Clitic placement preferences.elements.The types of stimuli are presented in (13).Only the cases in (13a), those with an element preceding the head, were covered in DFZ.Head initial phrases, as in (13b) and (13c), are a new type of case introduced in this study.3

Table 3 :
Structure of stimuli.
cl of-club received new membership cards intended 'Members of the club received new membership cards.'Likewise, a FirstWord host of the double head type is available with an initial predicate, as in (19a), but not with an initial argument, as in (19b).
Finally, the preposition ispred forms a FirstWord host either by virtue of combining with the following prosodic word, as in (29), or on its own, as in (30).
) a. Context: 'According to the doctor on duty, the patient is recovered.' this, the resulting prosodic word o poeziji cannot serve as a FirstWord clitic host.