Slavic languages – “ SVO ” languages without SVO qualities?

: Slavic languages are commonly classified as SVO languages, with an exceptional property, though, namely an atypically extensive variability of word order. A systematic comparison of Slavic languages with uncontroversial SVO languages reveals, however, that exceptional properties are the rule. Slavic languages are ‘ exceptional ’ in so many syntactic respects that SVO appears to be a typological misnomer. This fact invites a fresh look. Upon closer scrutiny, it turns out that these languages are not exceptional, but regular members of a different type. They are representative of a yet unrecognised type of clause structure orga-nisation. The dichotomy of ‘ head-final ’ and ‘ head-initial ’ does not exhaustively cover the system space of the make-up of phrases. In addition, there arguably exists a third option (T3). This is the type of phrasal architecture in which the head of the verb phrase is directionally unconstrained. It may precede, as in VO, it may follow, as in OV, and it may be sandwiched by its arguments within the phrase. From this viewpoint, the Slavic languages cease to be exceptional. They are regular representatives of the latter type, and, crucially, their collateral syntactic properties predictably match the properties of this type.


Introduction
There is a remarkable consensus in the literature 1 that Slavic languages deserve their place in the SVO type of the Greenbergian word order typology, although this verdict is based first and foremost on a forced choice among Greenberg's (1963: 61) three major types, 2 namely SOV, SVO, and VSO. For Slavic, SVO seems to be the least inappropriate type, notwithstanding the consensus that Slavic languages are exceptional when compared with other Indo-European SVO languages, such as North-Germanic, Romance, or English. Siewierska and Uhlířová (2010: 109) put it as follows. "Apart from the location of clitics there are virtually no syntactic constraints on the ordering of phrases in main declarative clauses. Thus in each of the Slavic languages all twenty-four possible combinations of a subject, direct object, indirect object and verb occur as grammatical declarative orders." This characterization may be too strong in details, but not with respect to the order variations of subject, object, and verb across styles and contexts.
If Slavic languages are assigned to the least inappropriate class of the Greenbergian word order types, the predictive gain with respect to syntactic properties of Slavic languages is little but the loss of predictive accuracy for typology with respect to the SVO type is high. This should raise concerns. Dixon (2011: 183) makes the following point: "More of the world's languages are like Russian than are like English." If taxonomic considerations coerce a grouping of Slavic languages together with English, North-Germanic, or Romance, then Slavic languages have to be acknowledged as highly exceptional in comparison to uncontroversial SVO languages. This indicates that such a type assignment would miss essential generalizations.
The fact that the word order in (1a) and (1b) happens to be identical is not a sufficient justification for assigning English and Russian to the same clause structure type. The shared S-V-O linear order in a minimal sentence such as (1a) is a partial overlap only since English lacks all the other word order variants, e.g. (1c, d). It is a general property of Slavic languages that depending on the information structure setting, any of the six permutations of the three major constituents yields an neighboring languages with respect to the order of subjects and objects, Mayo's (1993) assessment of Belarusian seems to be appropriate for Russian and Ukrainian, too. "Belarusian and Russian are genetically very close and structurally very similar languages. […] The structural distance between Belarusian and Russian is of the kind prototypically acknowledged for different dialects of one language" (Hentschel 2014: 93-94). Even Sorbianpresently predominantly verb-finalis not strictly SOV, as (i) illustrates (Scholze 2015: 206). More than a century ago, Liebsch (1884: 227) described the word order as free, with SV and VS, or VO and OV, as well-formed orders.
acceptable sentence, three of which are (1a, c, d). This is clearly not true for English and other [S [VO]] 3 languages. And there are additional, predictable, systematic differences (see Table 1, below). It is worth emphasizing that these differences are not language specific but systematic differences between Slavic languages and typical [S [VO]] languages.
Already in the early days of generative grammar, Ross (1967Ross ( : 75, 1970: 251) coined the term "scrambling" for Russian word order variations. In generative grammar, from then on, Slavic languages are regarded as SVO languages with the special property of excessive scrambling. If this were an accurate characterization, Slavic languages would justly be expected to share relevant syntactic properties of uncontroversial [S [VO]] languages except for those which scrambling accounts for. However, this is exactly not the case, as summarized in Table 1: Slavic languages do not share grammatically predictable structural properties of languages with an [S [VO]] clausal architecture. They share properties found in SOV languages to a large extent. Hence, if Slavic languages are assigned to the [S [VO]] type, this covers 1/8th of the properties listed in Table 1, namely the least significant one (i), and only by begging the question. If the criterion were rephrased ("as the grammatically mandatory order"), Slavic languages would drop out. The contrasts listed in Table 1 are contrasts between types of phrasal architectures. They follow from the grammatical conditions that characterize the respective clause structures. On the one hand, there is the structurally highly constrained type that consists of head-initial phrases only, namely the [S [VO]] setting. Since a VP is at the base of the clausal architecture, an SVO clausal architecture reflects properties of head-initial VPs, plus the unique structural positioning of the subject. On the other hand, and for principled reasons to be explicated in Section 3, the SOV architecture is structurally less 'tightly knit', and so is theyet to be characterizedtype that Slavic languages are argued to belong to. What SOV and Slavic languages share is the absence of constraints that apply to a strictly head-initial phrasal architecture with a VP-external, obligatory subject position. In other words, they do not share the [S [VO]] clause architecture. It is the absence of particular constraints that is responsible for the parallels between SOV languages and Slavic languages in the preceding Table 1.
Given the systematic absence of parallels between [S [VO]] and Slavic languages in Table 1, it seems unjustified to characterize them as an "exceptional" subset of [S [VO]]. This would be tantamount to the contention that Slavic languages are [S [VO]] languages without the defining properties of [S [VO]] languages.
In syntactic typology, "SVO" is understood ambiguously, at different levels of analytic depth. Greenberg's original notion is the phenomenological one. A language is assigned the label SVO if in a simple clause, subject-verb-object is an acceptable and common word order for a minimal declarative clause in the given language, consisting of a verb plus subject and object. In this purely descriptive sense, which takes the order S-V-O as a type marker, Polish (2a), English (2b), and German (2c) would count as SVO languages, 4 which is highly uninformative, of course. For German, this would neglect the fact that the order (2c) is just one option within the range of possibilities of a verb-second language, that is, with respect to the choice of the clause-initial phrase. For Polish (2a), the SVO attribution would obscure the fact that the word order variants of a simple declarative clause differ significantly from the English word order (2b). In fact, any one of the 24 permutations of the four words of (2a) yields a grammatical Polish sentence and is acceptable in an information-structurally adequate context (see Leszkowicz 2015: 121;Zabrocki 2016: 140). 5 (2) a. Marek dał Ewie kwiaty.

Polish
Marek Nom gave Eve Dat flowers b. Mark gave Eve flowers. c. Markus gab Eva Blumen. German The classification used by Dryer and Haspelmath (2013) in WALS is essentially phenomenological, too. In chapter 81 (Dryer 2013a(Dryer , 2013b) -"Order of subject, object and verb"), Russian is characterized as follows: "Russian is an example of a language with flexible word order in which SVO order can be considered dominant, so Russian is shown on the map as SVO." The criterion for this assignment in WALS is the fact that SVO is "the order that is more frequently used." 6 Reliable typological criteria cannot be safely based on such a weak criterion. Theoretically adequate criteria need to be structure-based (Haider 2020).
In some languages, the basic word-order type -VO or OVis 'masked' by superimposed syntactic properties, as for instance the V2-property of Germanic languages. In typology, the verb-second property is known (but not respected) at least since Mallinson and Blake (1981: 129). 7 Dutch, German, or Frisian are OV languages, that is, languages with a head-final VP. On top of this, these languages are 'V-second'. In a declarative main clause, the finite verb figures in a VP-external, clause-initial position, preceded by a single slot for an arbitrary constituent. Whenever a subject is put into the clause-initial position and the clause contains only a single verb plus an object, the resulting order happens to be S-V-O. However, the clause-initial position is not reserved for the subject. 8 Slavic languages in general, and Polish in particular, do not neatly fit into the defining schemes for VSO, SOV, or [S[VO]] languages, although each of these word orders is a grammatically admissible sequence for a Polish declarative clause, 9 as illustrated by examples such as (3a-c), from Leszkowicz (2015: 121) and Zabrocki (2016: 140). Evidently, Polish could not be filed as an [S[VO]] language that is 6 Frequency counts are not presented. 7 "The order used for a stylistically unmarked version of John saw Mary in German would be SVO, too, but to simply call German an SVO language would disguise the verb-second nature of its word order." 8 Here is the result of a search in the DWDS corpus (www.dwds.de, Referenz-und Zeitungskorpora) timespan 1900-2018 and a Google search filtered for books (accessed Dec. 16, 2020): (i) "Damit hat er" ('with-it has he' …): 2147 DWDS, 72.400 (Google) (ii) "Er hat damit" ('he has with-it' …): 877 DWDS, 72.000 (Google).
Note that both, damit ('with it') and er ('he'), are anaphoric. So, both 'contend' for an early position in an utterance. The clause-initial nonsubjects outnumber the clause-initial subjects. 9 Polish is not exceptional but representative of Slavic languages. The very same variations in Russian, for instance, are discussed in Dyakonova (2009); see chapter 3.1.
"SVO" languages without SVO qualities simultaneously a VSO, an SVO, and an SOV language. If someone insist that (3a) is an object scrambling variant of (3c), and (3b) a subject-postponing variant of (3c), this is just an ad-hoc reaction 10 to counterevidence for the SVO conjecture, as long as this conjecture is systematically contradicted in other relevant aspects (see Table 1).
SOV  1996) has stressed that the Celtic VSO languages do not admit subject expletives. As for SOV languages, there is no language known that requires an obligatory subject expletive in otherwise subjectless clauses. German (4c) is a clear case of an SOV language that does not tolerate an expletive subject (Haider 2010: 11). In fact, no SOV language is known that requires an expletive subject with passivized intransitives.
Swedish then dance Pass EXPL whole night Def c. Heute wird (*es) nicht gearbeitet. German today is (*EXPL) not worked d. dat niet wordt gewerkt in veendijken Dutch 12 that not is worked in peat-dikes 'that work is not carried out in peat dikes' 11 In Generative Grammar, "the requirement that a clause have a subject position" (Chomsky 1982: 10) has been wrongly overgeneralized as a universal requirement. Chomsky's (1981: 40) original formulation is the correct one: "The subject of a clause is obligatory in English and similar languages." 12 https://www.raadvanstate.nl/uitspraken/@123349/202001473-1-r1/ (accessed Feb. 4th, 2021).

"SVO" languages without SVO qualities
In [S [VO]] languages such as Norwegian and Swedish (4a, b) (cf. Åfarli 1992: 85;Falk 1993: 106), the subject expletive is mandatory. In German (4c), which is SOV, a subject expletive (in the alleged subject position) is ungrammatical. In fact, there is no structural space for it, since the structurally obligatory subject position is an [S [VO]] property. An immediate consequence is the following. English is the only Germanic language in which intransitive verbs cannot be passivized, the reason being that there is no suitable subject expletive available for satisfying the mandatory subject position; see Haider (2019, Sect. 5.1). The same is true for Romance null-subject languages.
In pro-drop [S [VO]] languages, and consequently in all Romance null-subject languages, too, the standard passive applied to intransitive verbs is ungrammatical. Spanish (5a) and Italian (5b, c) are representative. In French (5d, e) (Rivière 1981:42;Gaatone 1998: 124), on the other hand, this is not so, since French is not pro-drop and employs a pronoun (the counterpart of which falls prey to the null-subject property in pro-drop languages) as expletive subject.
(5) a. * Fue trabajado duro aquí. Spanish was worked hard here b. * È stato dormito in questo letto Italian has been slept well in this bed c. * È stato tossito per il fumo has been coughed because-of the smoke d.
Il aété dormi dans ce lit French EXPL has been slept in this bed e.
Il a beaucoupété fumé dans cette salle it has much been smoked in this room The Romance facts are particularly instructive since they testify against the oxymoronic concept of "null expletives" (e.g. Biberauer et al. 2010;Safir 1985;Sternefeld 1985). According to this still widely adopted assumption, there is the same kind of subject position in an SOV clause structure, but in SOV, it is deemed to be 'filled' with an empty expletive [sic!], as for instance in (4c, d). This assumption cannot be upheld if tested seriously. If a null expletive were an option, (5a-c) would be the perfect nullexpletive counterparts of the French expletive il in (5d, e). However, such constructions are ungrammatical in any Romance null-subject language. 13 Romance 13 In Vèneto, the vernacular of the Italian province Veneto, intransitives can be passivized, but only in the presence of an obligatory expletive of the 'there'-type. We are grateful to Cecilia Poletto, for her native speaker confirmation; see also Brunelli (2007: 27). Obviously, the 'there'-type of expletives is not affected by pro-drop.
(i) Z'è stà parlà de ti  (5), with the essential difference that there is no expletive subject attested in Russian, Bulgarian, and Polish, which are representative for Slavic languages in this respect. In any [S [VO]] languageprodrop or notthe pattern represented by (6) would be ungrammatical, as (5a-c) and the Scandinavian languages (4a, b) have shown. 15 (6) a. V komnate bylo nakureno Russian in room was smoked b. Obètom bylo napisano v gazete About this was written in newspaper c. Bylo dobavleno saxaru v smes' was added sugar Gen into mixture d. V stajat e vlizano Bulgarian in room is entered e. Po trevata e xodeno on grass Def is walked Impf-ppp f. W tym pokoju było już sprzątane Polish in this room is already cleaned g. czy było rano sprzątane whether was in-the-morning cleaned Perlmutter and Moore (2002, Sect. 3) explicitly adopt Sobin's (1985) "Silent-Expletive Hypothesis" according to which "impersonal clauses have a silent expletive (dummy) as subject." So, in (7b), the subject is deemed to be a "silent expletive". This interpretation is not appropriate, however. There is a null subject in (7b), but it is not a null expletive. It is the null variant of a semantically empty subject argument of the verb. 16 The construction in (7b) is active, with a direct argument with object case (viz, accusative, replaced by a genitive in the domain of negation in Russian). Blevins (2003Blevins ( , 2006 has shown in detail that "subjectless" and "impersonal" are grammatically different concepts. The passive of an intransitive verb is subjectless. The Russian impersonal construction is an active construction with a null subject. The semantically emptied subject argument is represented by a pronoun that is subject to pro-drop in Russian. Remember that the ungrammaticality of the passive of intransitives in Romance null-subject languages clearly demonstrates that "silent expletives" do not exist, see Haider (1987Haider ( , 2019 and Alexiadou and Anagnostopoulou (1998).
beyond border NEG were Neut published Neut such articles Gen 'There weren't any such articles published abroad.' In an earlier paper, Moore and Perlmutter (2000) had tried to demonstrate that Russian admits 'quirky subjects', that is, dative noun phrases in a subject position that function as subjects. Hitherto, this phenomenon had been attested only for Icelandic, which is an uncontroversial [S[VO]] language. If correct, this would be a noteworthy indication of an obligatory structural subject position in Russian. However, Sigurðsson (2002: 697), who evaluates their arguments for subjecthood in detail in comparison with Icelandic dative-subject constructions, concludes that their arguments are untenable because the alleged dative subjects in Russian lack the subject properties that have been attested for Icelandic oblique subjects.

Restrictions on interrogative subjects in SVO
Multiple wh-constructions are infrequent enough to be free of normative regula-

tions. Hence they provide direct insight into grammatical restrictions of [S[VO]]
16 In German impersonal versions, the semantically empty subject shows as non-referential es 'it' in (i): (i) Wann hat es den Schornstein vom Dach geweht? (impersonal construction) when has it the chimney Acc off-the roof blown ('When did it Indef blow the chimney off the roof?') (ii) Wann wurde der Schornstein vom Dach geweht? (passive construction) when was the chimney Nom off-the roof blown languages that are absent in OV languages. The two characteristic restrictions of [S [VO]] languages with respect to multiple-wh constructions are the restriction against bare 17 wh-subjects in situ and the restriction against in-situ wh-adverbials of a higher semantic type (see Szabolcsi and Zwarts 1993), that is, why and how. English (8a, b) is representative also for SVO-Germanic 18 and Romance SVO languages for the restriction against wh-subjects in-situ. 17 Wh-pronouns embedded in a subject noun phrase do not fall under this restriction, the reason being that in the case of a bare wh-pronoun as subject, the wh-pronoun is a wh-operator as subject.
If the wh-item is part of a subject noun phrase, the noun phrase is the subject; for details see Haider "SVO" languages without SVO qualities who determines how long who the chance gets REFL what-for how to qualify.
The crucial issue is this. In prototypical [S [VO]] languages, the subject is VP-external. It is assigned a unique structural position reserved for the subject. 21 In OV, the subject remains in its original position within the directionally unconstrained licensing domain of the verbal head, viz. the VP. Hence the subject shares structural properties with the other VP-internal arguments, namely those staying in their base positions. Slavic languages allow for multiple fronting of wh-words; see, among many others, Rudin (1988), Bošković (1997Bošković ( , 1998, Liakin and Ndayiragije (2001), and Meyer (2003Meyer ( , 2004, and the state-of-the-art report of Błaszczak and Fischer (2001). The examples in (10) [(10a) : Meyer 2004: 253] are taken from Czech since it is a Slavic language in which multiple wh-phrases clearly cannot be assumed to form a single constituent that is fronted to the sentence initial position. 22 Czech canonically fronts all interrogative phrases to the left, 23 but allows for material intervening between the initial wh-phrase and a second one (10c, d) (Toman 1981: 298 As illustrated by (10b, d), there is no English-like restriction on an in-situ whsubject in Czech. Just like in German, an interrogative phrase may precede an interrogative subject in a multi-interrogative question. With inanimate objects, there is not even a preferred order. With an animate object wh-phrase, koho 'whom', instead of co 'what' in (10a, b), there is a preference for the order subject > object, but the order object > subject is by no means ungrammatical (cf. Meyer 21 In English, the head position is the position of the finite auxiliary or of the expletive auxiliary do in (i):
Bosnian/Croatian/Serbian who is whom seen 'Who has seen whom?' b. Koga je ko vidio?
Each one of these variants is grammatical but of course they are not equivalent in their information structure effects. The order in (13) presents kogo ('whom') as the sorting key for the <object, subject> pairs asked for. This presupposes a discourse setting in which object-participant before subject-participant is an adequate order in terms of the discourse structure.
(13) Kogo j kto i ty xočeš', čtoby e i pobil e j ? Russian whom who you want thatbeat -This difference in information-structure fitting is reflected in comparative judgments with a slight preference for the pattern (11a, 12a) over (11b, 12b), respectively, when such sentences are judged in isolation, that is, out of a context that favors the other order.
Another source of preferences among otherwise equally grammatical variants is processing-based. Even if the grammar of a Slavic or an SOV language such as German does not block fronting of a wh-object across a wh-subject, the computation of the filler-gap structures faces different demands by different serializations, as illustrated by German (14a, b) in contrast with English (15a, b) see Haider (2010: 126-128 In (14a), the parser meets a second wh-item before the gap for the first item can been identified lower in the tree. This extra load for the buffer 25 would not arise in the equally grammatical variant of (14a) in which wen is fronted and was remains in situ. In (14b), such an effect does not materialize since the gap comes right after the in-situ wh-subject. In an SVO language such as English, however, it is the grammar that is at stake. (15a, b) are not merely degraded for processing reasons, they are ungrammatical and fully unacceptable. A pronominal whsubject is ungrammatical in-situ in multiple wh-constructions.

Restrictions on adverbial wh-items in the preverbal position
In English, in Romance, and in the Germanic [S[VO]] languages, why and how are the odd balls among adverbial interrogatives, as (16a) illustrates. In OV-languages, this contrast is unknown (16b, c) [(16c): Saito 1994: 195].
(16) a. Who has answered this question when/where/*why/*how? b. Wer hat diese Frage wann/wo/weshalb/wie beantwortet? German who has this question when/where/why/how answered c. Dare-ga naze soko-ni itta no? Japanese who Nom why there-to went Q-PARTICLE Semantically, why and how are higher-order quantifiers. Who, what, when, where ask for individual-type values such as objects, points of time, points of location. Why and how, however, ask for properties of events. They do not quantify over individual-type variables but over sets. In other words, they quantify over properties of the denotation of the phrase headed by the verb. So, they have to c-command it. The domain of higher order quantifiers must contain the base position of the (finite) verb. Consequently, the wh-adverbial has to precede the VP and its in-situ position should be the position of the corresponding adverbials. (17) illustrates the dilemma of SVO clauses. Although adverbials may precede the VP 25 If the two involved wh-items are formally identical, this leads to a processing break-down, as expected.
whom Acc did you whom Acc ask [to the party to invite] (17a, b), their wh-counterparts are ungrammatical in this position (17c, d). The structural reason that rules out the pre-VP position in VO is explicated in Haider (in press).
(17) a. He has very carefully/therefore answered the question. b.
He has very often/therefore asked the same question c. *Which question has he how (carefully)/why answered? d. *Which question has he how (often)/why asked?
Eventually, the only position left for this kind of wh-adverbials is the clause-initial one (18a). However, there is a competition for this position in case the subject is a wh-phrase as well. This leads into a no-win impasse (18b). For (18b), there is no acceptable alternative. If who is placed first, there is no licit position left for how often, and if how often is fronted, who is ill-placed.
How often has he asked which question? b. *How often has who asked this question?
For English, this set of facts is without any exception. In an aggregated 5.8-billionword corpus, that is BNC, CocA, and NOW, taken together, there is not a single sequence attested for has how often, has how carefully, or has why, followed by a verb. English is representative of the Germanic SVO languages 26 in this respect. Romance languages confirm the picture. If Slavic languages were SVO languages, they should pattern like the other SVO languages in this respect, but they don't. Russian is representative and the examples (11d) and (11e) above show that it patterns with the OV languages. Russian is an apt test case also since it does not obligatorily front multiple wh-items; see Meyer (2004) and Dyakonova (2009) for ample evidence. So, one can check the position of the wh-element that is left behind. If it is a member of the 'special' class of adverbials it must precede the finite verb.
Mne interesno, kakoj fil'm Boris kak často smotrel Russian me interests which film Boris how often saw 26 Fanselow (2004: 85) reports that three of his five informants rated (i) as acceptable. (ii) was not checked. Internet searches of Swedish do not produce a single sentence with a fronted wh-item followed by varför (ii) or hur ofta ('how often'). This contrasts with facts from OV-Germanic. Google searches for German produce these results: wer wie oft ('who how often'): 2,120 (Books), 939 (News). wer warum ('who why'): 6,820 (B), 2,150 (N).
it plays no role who that laughed why (ii) *Det spelar ingen roll vem som varför skrattade.
Mne interesno, kakuju poezdku Maša kak dolgo planirovala me interests which journey Mary how long planned d. *Mne interesno, kakuju poezdku Maša planirovala kak dolgo In Russian as well as in other Slavic languages, how may precede or follow an interrogative subject, and, as Russian illustrates, the second interrogative does not need not to be in the clause initial position, but it must precede the finite verb. This, however is exactly the position where an in-situ interrogative item may not be placed in an uncontroversial [S[VO]] language.

Left-adjoined adjuncts
Let us continue in this area of grammar but switch the focus to another syntactic property of adjuncts. Left adjuncts of left-headed phrases, that is, head-initial phrases, are constrained in yet another way (see Haider in press): The head of the left-adjoined adjunct of a head-initial phrase must be adjacent to the phrase it is adjoined to. Crucially, this constraint does not hold for left adjuncts of rightheaded, that is, head-final, phrases.
This constraint is operative in Germanic as well as Romance languages and arguably it is a cross-linguistically operative constraint. For brevity's sake, let us call it LLC, that is, the left-left-constraint. In both sub-families -Germanic and Romancenoun phrases are head-initial, and in Romance and the North-Germanic group plus English, the VP is head-initial, too. Hence, in both groups, the LLC applies to adjuncts of noun phrases, and in the Romance, English, and North-Germanic, the LLC constrains adjuncts of VPs as well. (20)  If the position of the verb in (22) were the position of the head of a head-initial VP, then the LLC would apply, but it does not. Preverbal adverbial phrases in Slavic behave just like preverbal adverbial phrases in OV languages. If they get lengthy, they tend to be postponed, but not because they would be ungrammatical otherwise. In Slavic languages, not only VPs but also NPs evade the LLC, as the following examples from Bulgarian, Russian, and Polish demonstrate (23a-c). What this implies is that an NP does not count as a strictly head-initial phrase in these languages either. But, this is apparently not a cross-Slavic property. Bosnian/Croatian/Serbian (23e) and Czech do not admit this pattern (Siewierska and Uhlířová 2010: 116

Fillers of gaps in left branches (left-branch extractions -LBE)
The following examples (24) illustrate a well-known property of Slavic languages; 27 see Grosu (1973) and Roberts (1997: 189). Adnominal attributes may be fronted and thereby get separated from their NP. This is common for questions (24a), and also for the corresponding answers (24b). This kind of fronting results in a filler-gap constellation, according to accepted analyses of the construction for Russian (Bailyn 2012: 62-64) and other Slavic languages (Bošković 2005(Bošković , 2014(Bošković , 2017 and therefore, this construction is a perfect testing ground for the SVO thesis. It is a cross-linguistically robust property of [S [VO]] languages that preverbal phrases are grammatically illicit domains for gaps of fronted fillers. As (25) illustrates, a filler may relate to a gap in a postverbal phrase but not to gaps in a preverbal phrase. Given this cross-linguistically valid constraint, the prediction for SVO languages is evident. Whenever a phrase ends up in a position preceding the main verb in its VP-internal position, the phrase in its derived position is opaque for filler-gap relations terminating within this phrase. Let us call it the 'gap-phrase'.
27 Only Bulgarian and Macedonian do not share this property. Bošković (2005Bošković ( , 2014 correlates this with another exceptional fact of their noun phrase syntax, namely the so-called article system of these two languages. Bulgarian and Macedonian are languages with definiteness markers suffixed to the noun. 28 Note that PPs may be fronted in English, as the following example from the BNC illustrates: (i) that on such a small surface, large impacts would shatter the surrounding terrain, causing cracks.
Here comes the crucial prediction. If Slavic languages are SVO languages, they are expected to show the typical pre-versus postverbal asymmetry for extractions. Left-branch extractions are predicted to be acceptable only for gap-phrases in postverbal positions but excluded when a gap-phrase is in a preverbal position. Bošković (2005: 37) explicitly states that "it is worth noting in this respect that LBE constructions actually sound best when the remnant of LBE precedes the verb." In sum, the SVO-based prediction turns out to be wrong (26) (26), the gap-containing phrase is preverbal. Nevertheless, each construction is acceptable, given an appropriate context for the information structure effect of the particular word order with a preverbal object. The respective grammars of these languages do not rule out such a construction. 30 Gaps in clauses adjoined to head-final phrases are SOV counterparts of gaps in left-branches. Germanic NPs are head-initial, so their syntactic properties are identical with the syntactic properties of NPs in SVO languages. VPs and APs are head-final in continental West Germanic. The examples in (27) are taken from German. First, left-branch extractions comparable to (26) areas predictedlicit for scrambled phrases (27a). In (27a), the infinitival object clause is scrambled across the subject. In an SVO setting, a filler-gap relation into such a scrambled clause would be ruled out, under any analysis of scrambling. If scrambling adjoins the scrambled phrase to a head-initial VP, the gap is in a left branch and therefore it should be ungrammatical. If, on the other hand, the scrambled phrase is deemed to end up in a 29 The examples in (26a, c, d) have been counter-checked with five native speakers per language. The word order is information-structurally selective and therefore requires appropriate contexts of utterance, one of them being focus, cf. the Polish example (the underlined capital letters indicating contrastive intonation): (Ale) jaki Paweł saMOchód kupił swojej żonie, to radzę ci zobaczyć. '(But) what a CAR Paul bought for his wife, I suggest you to have a look.' 30 Note that an uncontroversial SVO language like English shows that information structure fitting would not 'overrule' grammar. No focus variant makes (i) acceptable: (i) which i this girl stroked [i cat] yesterday. pre-VP functional specifier position, the gap would be in an inaccessible position, too. In each case, the predicted result would be 'ungrammatical'. However, in SOV, filler-gap relations terminating in a scrambled clause are grammatical and acceptable (Haider 2010: 155-157). In (27b), an adjunct of an AP is questioned, and it is a left branch of the AP "[um wie viel] teurer als Kupfer" (for how much more expensive than copper). for how much is it more-expensive than copper What (27) re-confirms is the fact that gaps contained in left branches of head-final phrases are accessible. The left-branch constraint is a constraint on left branches of head-initial constituents.

Rigid word order in head-initial phrases
If Slavic languages are [S[VO]] languages, their VPs are head-initial. A characteristic collateral property of head-initial phrases is the rigid word order of headinitial phrases. It is a popular legend that free word order directly correlates with overt morphological markers for case. This is easy to falsify. On the one hand, there are languages with free word order in the absence of morphological marking, and on the other hand, there are languages with morphologically distinct case marking that do not permit word order variation at all. Bulgarian (28)  Bill gave dolls to children b. *Bill gave to children dolls In Icelandic, the word order is rigid in spite of rich case marking by distinctive paradigms. In (30), dative and accusative are distinctively marked on the nouns. This notwithstanding, Dehé (2004: 94) reports that "the inverted order was rejected", that is, the order (30b) was rejected by all her informants, without exception.
Icelandic They showed parents Def-Dat kids Def-Acc b. *Þau sýndu krakkana foreldrunum Dutch (and the other OV-Germanic languages) contributes another facet to this picture. In Dutch, as well as German (Haider 2013: 207-210), Frisian or Afrikaans, prepositional objects may be 'scrambled' within a head-final phrase such as the VP (Geerts et al. 1984: 989f.), but the very same phrases crucially may not get scrambled in a head-initial phrase, such as an NP. This can be demonstrated in a minimal-pair context. VPs are head-final, NPs are head-initial, and a verb can be converted into a noun, as a nominalized infinitive. This yields the closest possible context for comparisons. The facts show that 'scrambling' is not a holistic property of a given language. It is a property of phrase structure. Head-final phrases facilitate scrambling; head-initial phrases impede it; for details see Haider (2015).
(31) a. Toen hebben [de autoriteiten het kind aan de moeder teruggegeven] VP then have the authorities the child to the mother back-given c. Toen hebben de autoriteiten aan de moeder i het kindi teruggegeven d. het [teruggeven van het kind aan de moeder] NP the back-give of the child to the mother b. *het [teruggeven aan de moeder i van het kindi ] NP If Slavic languages were filed as structural SVO languages, their word order variability would appear to be truly exceptional in more than one respect. First, they allow for VP-internal word order variation in an allegedly head-initial VP, that is, variable order of postverbal objects. Second, they allow for word order variation across the boundary of a VP, that is, to positions preceding the allegedly headinitial verbal head of the VP. In other words, they would have to be acknowledged as SVO languages that freely scramble within the VP and out of the VP. This serialization freedom for objects is absent and ungrammatical in uncontroversial [S [VO]] languages in other subfamilies, that is, Germanic languages, such as English and North-Germanic as well as Romance languages.

Rigid word order of auxiliaries
The final property to be called up is once more an invariant property in SVO languages. If a simple clause contains more than one verb, the verbs are serialized in an invariant relative order. In other words, auxiliaries and quasi-auxiliaries, such as modals or causative verbs, are invariably serialized in such a sequence that the morphosyntactically dependent verb follows the verb it depends on. In English, modals select a bare infinitive, the tense auxiliary have selects a participle, and the auxiliary be selects the "-ing"-form when coding the durative aspect, or a participle when coding the passive. In each case, the selected form invariably follows the selecting verb. Evidence from OV languages shows that invariable ordering is a type-dependent property. In almost all 31 Germanic OV languages, there is order variation among the verbs in a simple clause, and this is of course independent of the fronting of the finite verb due to the V2-property of Germanic languages (Bouma and van Noord 1998;Haider 2003). (33) illustrates this property for Dutch and (34) for German. These sequences are all clause-final sequences of verbs. The numbers are the Google search hits with the filter 'News', followed by those with the filter 'Books', followed by the number of unfiltered hits. All variants are synonymous.
31 Only Frisian does not permit order variation among the verbs. In nonstandardized varieties of German, for instance in Swiss-German regional varieties, the variation is even more extensive (see Wurmbrand 2004) in the sense, that the auxiliaries may precede objects. This is rare in standard German, but it is attested (Thomas Mann, Buddenbrooks): This kind of verb order variation is absent in those SOV languages in which verbs are confined to their base position. In these languagesfor example Japanese or Frisianthe relative order is the mirror image of the relative order in VO languages. The dependent verbs precede the verbs they depend on. The crucial difference between the OV and the VO situation is this: With VO, not a single language is attested that would allow for verb order variation in the sense that in a simple clause, a dependent verb may follow or precede the verb it depends on. If there is word order variation in the relative order of the verbs of a simple clause, the language cannot be an [S [VO]] language. The Slavic languages do not offer as ample testing opportunities as the Germanic languages provide. This is due to the fact that, first, a lot of Slavic languages either have a very restricted set of auxiliaries (e.g. Russian) or the auxiliaries in most cases appear in an enclitic form (e.g. B/C/S, Czech) or as what has been labelled "mobile inflection" in Polish by Embick (1995). Second, a lot of Slavic languages lack modal verbs altogether. However, in those cases in which one can observe modal verbs or nonclitic auxiliaries, Slavic languages rather pattern with OV languages than with VO languages in terms of observable verb order variations [(35g): Schürcks 2009: 31].
The sequence of verbs and auxiliaries may be interrupted by adverbials. This is an indication that the verbs do not form a cluster of the kind known from OV languages.
3 There is more than head-initial and head-final The above survey of syntactic characteristics of Slavic languages has produced two general results. First, the Slavic language family shares a set of structural properties (with differences in details, of course). Second, this set of properties does not match the syntactic profile of typical [S [VO]] languages. In such languages, that is, strictly head-initial languages with an obligatory VP-external subject position, these properties are ungrammatical.
In this situation, the obvious question to ask is the following. Is this a problem of Slavic languages or is it a problem of a taxonomy that files Slavic languages as SVO languages? It is a problem for a taxonomy that does not provide adequate space for languages such as the Slavic languages, and there are a lot more languages that closely resemble Slavic languages cross-linguistically. Dryer (2013a) lists 181 languages of his sample as languages "lacking a dominant word order" and this list needs to be extended by those languages that are misclassified as SVO, such as the Slavic language. The crucial difficulty is the definition of "dominant word order": "This means that it is either the only order possible or the order that is more frequently used" (Dryer 2013b). Obviously, this definition joins two disjoint properties, namely a strict word-order property (viz. "the only order") with a variable-word-order property. In the strict-order type, variation is ungrammatical; in the variable-order-type, non-syntactic factors, such as information structure, guide the choice of a variant from the pool of grammatical variants. The grammars behind these types are obviously different and incompatible. Nevertheless, such languages are grouped together in spite of their incompatible properties. In Slavic languages, S-V-O is regarded as the neutral word order, but 'neutral' is a concept of pragmatics, namely information structuring. S-V-O is just one of the grammatically admissibly serializations, and it happens to be the order that is not associated with specific information structure effects of the given versus new, or focus versus background organization of an utterance (see Haider 2020). If "dominant" is construed as grammatically determined (rather than preferred for pragmatic or language processing reasons), Slavic languages are languages with no grammatically determined (= dominant) order.

Taxonomic space for Slavic languages
The taxonomy of clause structures is part of the taxonomy of cross-linguistically attested phrase structuring. A hitherto unquestioned assumption in phrase structure theory is the peripherality generalization. 32 According to this axiom, the head of a phrase is assumed to either precede or follow its arguments. (36) illustrates the word order of the respective VPs. Each arrow indicates the parametric directionality property of the head (or its projection, viz. V′) that structurally licenses the arguments and thereby determines whether they follow or precede. In (36a) and (36c), the subject is VP-internal. The evidence from Slavic and languages similar to Slavic points to the conclusion that the structures (36a-c) do not exhaustively cover the system space of VP structuring in natural languages. The peripherality generalization is arguably based on too narrow a sample of languages, namely languages with a fixed directionality for phrasal heads (viz. phrase-final or phrase-initial). Let us therefore assume that the peripherality constraint is not universal but characteristic only of the large class of languages with a specified directionality of the head of the phrase (see Haider 2015). Under a parametrical approach, the directionality of heads permits, as standardly assumed, a head-initial or a head-final setting. But crucially, there is also a third option, namely the underspecified setting. This amounts to a third type, in addition to the types with a parametrically specified directionality. 34 In this third type -"T3" or type with variable head positioninga head may surface in 32 Kornai and Pullum (1990: 34) refer to Stowell's (1981: 70) wrap up of X-bar theory, in terms of a list of "plausible and potentially very powerful restrictions on possible phrase structure configurations". One of the restrictions is the peripherality restriction for heads. 33 Intermediate head-positions in complex VPs are admitted only as empty positions (see Sect. 3. 2). 34 In language typology, 'flexible order' has been recognised: "There are many other languages in which all six orders [of S, O, and V] HH are grammatical. Such languages can be said to have flexible order. Flexible order languages are sometimes described as having "free" word order, though this is misleading, since there are often pragmatic factors governing the choice of word order" (Dryer 2013a). alternatively available positions within its phrase since it is able to license its arguments in either direction (37), because no specific direction is imposed by the grammar.
(37) (C) [ VP … ← [V / …]] ←, /: ambidirectional licensing ("T3"), as in Slavic Consequently, T3 admits the SVO-like order (38a), the SOV-like order (38b), the VSO-like order (38c). In addition, T3 allows for a pattern that is excluded in both OV and VO, in which the verbal head is sandwiched by its objects (38d). In other words, any position of the verbal head relative to its arguments is a grammatically licit serialization. In each case, the relative order of subject and objects remains unaffected. Scrambling may rearrange this order. This is true for main clauses as well as for embedded clauses, as the bracketed complementizer indicates. There are numerous languages that resemble Slavic languages in their word order characteristics. They are likely to outnumber [S[VO]]-languages proper, since, as in the case of Slavic languages, in word order typologies, such languages tend to be grouped with [S[VO]] languages. After all, a pattern such as (38a) is a frequent pattern in these languages, for independent reasons, though. Information structuring as well as language processing preferences account for unequal frequencies of the variants.
Phrase structure properties are grammatically determined. They are not servants of information structuring (see Fanselow and Lenertová 2011). The relation between syntax and information structure is asymmetric. Whenever syntactic structuring allows for word order variation, this variation space will be exploited by information structuring. On the other hand, if a syntactic structure imposes strict word order, information structuring is not able to unfasten the grammatical ties. Head-final "Languages with highly flexible word order are themselves a linguistic type" (Dryer 2007: 113). Gell-Mann and Ruhlen (2011:17291) regard FWO (free word order language) as a separate category. 35 The variable V-positioning may serve as a demarcation point, for instance in information structuring, by separating the area of backgrounded information; see Junghanns (2001). structures ('SOV') as well as structures with variable head-positioning ('T3') provide more headroom for information structuring (cf. Junghanns and Zybatow 1997;Kučerová 2007) than an SVO architecture.
As for scrambling proper, that is, variation of the relative order of arguments, the Slavic languages admit the same scrambling potential as head-final phrases, and in particular as head-final VPs, in addition to their T3 verb position alternations, with the arguments in their base order. In general, local scrambling presupposes that the scrambled position is a position within the directionality domain of the head. This condition is met by languages with head-final as well as for languages with T3 phrases, but not for languages with head-initial phrases. 36 In the following subsection, it will be briefly pointed out how T3 structures neatly fit into the already existing theoretical setting that accounts for head-final and head-initial structures and their specific properties. It will become clear why T3 and SOV share the properties they have been shown to share in the preceding section.

Grammar-theoretical space for Slavic languages
The original implementation of the peripherality generalization (39b) for head-initial VPs in combination with the binary-branching principle (Chomsky 1981: 171), Barss and Lasnik (1986) showed to be empirically indefensible. If the initial and the final position of the head of a phrase were equally admissible foot positions, that is, the lowest structural positions, the respective phrase structures would have to be either right-or left-branching, 37 as in (39a)  In German, VP is head-final and therefore a scrambling domain. NPs are head-initial and therefore scrambling is not admitted, even for the very same constituents (Haider 2010(Haider , 2015: (i) Geld auf ein anderes Konto übertragen auf ein anderes Konto Geld übertragen money to an other account transfer to an other account money transfer (ii) das Übertragen des Geldes auf ein anderes Konto -*das Übertragen auf ein anderes Konto des Geldes the transfer (of) the money to an other account the transfer to an other account (of) the money 37 A projection is right-branching, if every branching node on the projection line of the head is a node on the righthand side, that is, a node that follows its sister node. Conversely, in a leftbranching structure the branching node on the projection line precedes its sister node.
Head-final phrases have a right-branching structure (39a), that is, the branching node is the right node on the projection line. Head-initial structures, on the other hand, would have a left-branching structure (39b) in the X-bar model. As an immediate consequence, all terminals in these structures would have to come in the mirror image order of (39a), cross-linguistically. If a verbal head takes YP as its innermost complement, then Obj 2 , and finally Obj 1 , the resulting serialization is Obj 1 ←Obj 2 ←YP←V for head-final VPs. For head-initial VPs, it should be the mirror image order, namely V/YP/Obj 2 /Obj 1 , as in (39b). This is empirically incorrect, however. The serialization (39a) is invariant across OV and VO, as a comparison between OV-and VO-Germanic demonstrates (40).
(40) a. jemandem Obj1 etwas Obj2 in seine Tasche YP stecken somebody Dat something Acc into his bag put b. put somebody Obj1 something Obj2 into his bag YP In fact, the structure (39b) is empirically inadequate in many more respects. The syntactic predictions entailed by such a left-branching structure turn out to be wrong, as Barss and Lasnik (1986) showed for English. In reaction to their findings, Larson (1988) suggested a derivational conversion of the structure (39b) into a right-branching "VP-shell" structure as an empirically more adequate structure. However, this is a curative approach. Why should an English VP start out with a structure that has to be repaired derivationally? Why could it not start with a grammatical structure right away? It can, but this structure is more complex for head-initial phrases than for head-final ones. What follows is merely a sketch. For details please consult Haider (1992Haider ( /2000Haider ( , 2010Haider ( , 2013Haider ( , 2015.
Arguably, in no language would complex phrases be left-branching, that is, structured like (39b). The internal build-up of phrases is universally rightbranching (Haider 1992(Haider /2000. They may be right-headed or left-headed, but their structure is nevertheless universally right-branching. It is the clash between the universal directionality of branching and the specific directionality of the head that is responsible for the VO-specific syntactic restrictions. In the head-final setting, the directionality under which the head of the phrase accepts its complements is congruent with the universal directionality of the phrase structure, as indicated by the arrow-signs in (41). On each constituent level, the respective complement is a sister node of a node on the projection line on the canonical side, which is determined by the head of the phrase. It is a sister of the head or of one of its projections.
The complex structure of head-initial phrases is the immediate result of a directionality mismatch of the head and the general branching directionality. In (41), each object of the verb has as sister node a node of the projection line of the verbal head (i.e. a V′-node) and the object meets the directionality requirement. It precedes.
In the head-initial setting (42), the objects above the foot position of the verb are directionally 'misplaced'. In order to meet the directionality requirement of the head, they would have to follow the head or their respective sister node as a node on the projection line of the head. However, they precede, due to the universal branching requirement. In this case, the only admissible option is a reinstantiation of the head. The grammatical outcome would be (42a). Since there is only one lexical verb for three verb positions, the surface position is the highest position in the structure (42b). An immediate empirical confirmation for the verb slots comes from VO-languages with particle verbs. Particles of particle verbs may be left behind, that is, stranded, when there is more than one position available for the verb. This is typically the case when a finite verb is fronted in a Germanic V2 language (43a). In VO languages such as English (43b-d) [(43b)/(43c): CocA; (43d): Dehé 2002: 3], however, there is another stranding option. A particle may get stranded in one of the verb positions in a structure such as (42b). In (43b-d), the particle position is one of the empty verb positions of (42b).
The patterns in (44) are an aggregate of the patterns admissible in a head-initial setting (44c, d), the head-final pattern (44a), plus the pattern with a sandwiched verb (44b), which is diagnostic of the ambidirectional setting of T3 languages. It is absent both in SVO and in SOV languages.
In a diachronic perspective, SOV and SVO typically are diachronic outcomes of T3, due to the change from 'unspecified' to 'specified' directionality of heads, with one of the two familiar options, namely head-final or head-initial. The Romance and Germanic languages can serve as a good example of such a diachronic development. They all started out with a T3 grammar. The contemporary Romance language is head-initial. The Germanic languages split into an SVO and an SOV group when the directionality value of the head got fixed, with the V2-property as an 'enzymatic' factor of the split; see Haider (2014) for details. In general, patterns such as (44a) provide the link to the SOV-type with a head-final specification. A head-initial specification finds its link in (44c) and (44d), as the VSO and the SVO type. As for the diachronic dynamics that lead to SVO, it is easy to point out a general fact. No language is known that has developed (without external influences such as strong bilingualism) from a language like English into a language like Latin, Russian, or Sanskrit. However, many languages are known to have developed in the other direction.

Recapthe syntactic properties of Slavic languages as T3 properties
This subsection reviews the properties discussed in Section 2 in the order of presentation and relates them to the theoretical background sketched in the preceding Section 3. The specific syntactic profile of the Slavic clausal architecture is the profile of languages in which the position of the head of the VP is variable. Variable head-positioning is not restricted to the VP in Slavic languages. NPs, too, show effects of this property in the majority of Slavic languages (see Section 2.5 on leftbranch extractions and Section 2.4 on the LLC effect).

4.1.
An obligatory preverbal structural subject position is the hallmark of an [S [VO]] clausal architecture, with its obligatory, VP-external structural position for the subject. Obligatory expletive subjects are an immediate correlate of this property (see Section 2.1). In SVO, the VP-internal position of the subject argument precedes the verbal head and is therefore not in its directionality domain. As a consequence, it is raised to a VP-external functional spec-position (45a). In subjectless constructions, this obligatory spec position is 'plugged' by an expletive.  (45c), any argument of a verb remains within the directionality domain of the verbal head or a projection of it, whence the absence of the particular subject-related functional projection in the clause structures of these languages. The functional projection in (45a) provides a directionally licensing head for the preverbal, VP-internal subject, namely the functional head. Moreover, it is the trigger of raising the subject-XP to the spec-position of the licensing functional head. This is an effect of a general licensing condition which is explicated in detail in Haider (2015: 84). The licenser and the licensee must c-command each other under canonical directionality. In (45a), F°c-commands the VP-internal subject position and the subject in its derived spec-position c-commands F°by virtue of being raised to the spec position. The very same relation holds VP-internally and triggers the VP-shell structure 38 of complex, head-initial phrases (see Haider 2015: 85). In SOV and T3 grammars, any argument of a head finds its well-formed position already within the directionality domain of its head (45b, c  ]]. The external position is a functional spec position. A wh-item in a functional spec position qualifies as a whoperator. In-situ wh-elements in base positions are linked to wh-operators (Haider 2004(Haider , 2010. In SVO, the wh-subject is necessarily in an operator position, in SOV and T3 it is not. An operator wh-subject cannot be dependent on a preceding wh-item. This excludes dependent wh-subjects in SVO but allows such subjects for SOV and T3. 4.3. The restriction against higher-order adverbial wh-phrases in-situ in immediately pre-VP positions is a sequel of the LLC constraint (see next paragraph), that is, the constraint for adjuncts adjoined on the directionally noncanonical side of a phrase. However, this very position would be required for higher order wh-adverbials because their scope requirements are met only in a pre-VP position. Consequently, higher order adverbials are obligatorily fronted to the clauses initial position in SVO languages. This leads to an impasse, if the second wh-phrase in the clause is the subject. In this case, there is no grammatical outcome available. Slavic languages confirm the diagnostics of the scopal requirements of why and how. If the Slavic counterparts remain in-situ, they are in a preverbal and never in a post-verbal position. In Slavic languages, however, a preverbal position is a position within the (ambidirectional) directionality domain of the verb. So, in-situ why and how may occur in these positions in T3, as well as in OV languages, but not in SVO.

4.4.
Adjuncts of head-initial phrases are adjuncts on the non-canonical side of a phrase. Since they are adjoined outside of the directional licensing domain of the head of the phrase, an adjacency condition (i.e. LLC, Sect. 2.4) applies (Haider in press). In strict [S [VO]] languages, this restriction applies to preverbal adverbials as well as to prenominal attributes. In the Germanic OV-languages, adverbial phrases are exempt since the preverbal side is the canonical side in OV. However, in all Germanic languages, noun phrases are head-initial, so LLC applies in all these languages. In T3 phrases, prenominal as well as preverbal adjuncts are within the (ambidirectional) licensing domain of the respective head, whence the absence of an LLC effect for VP and NP adjuncts. It is this property that also explains the following Slavic property, namely left-branch extraction.

4.5.
Left-branch-extraction (LBE) is illicit for left branches on the noncanonical side of a phrase, that is, from left branches of head-initial phrases. In [S[VO]] languages, this constraint applies to all kinds of phrases simply because any phrase is headinitial. In the Germanic SOV languages, it applies to left-branches of NPs (46a), since these are head-initial phrases, but not to left branches of VPs (46b).
(46) a. *Bessere i braucht man dafür [i Theorien] NP German better Acc-Pl needs one for-it theories 'One needs better theories for this' b. So eine Theorie i wird [i zu verteidigen] wohl kaum jemand bereit sein such a theory shall [to defend] PRT hardly anyone ready be 'Hardly anyone shall be ready to defend such a theory' In (46b), the infinitival object clause is scrambled across the subject. If scrambling is modeled by adjunction, then the scrambled clause is a left branch of the head-final VP. If, on the other hand, scrambling was modeled as fronting into a specifier position, it would be ruled out by the constraint that rules out any filler-gap relation terminating in a gap within a preverbal specifier position. In T3 languages, both configurations in (46) are well-formed, since in each case the phrase that contains the gap is within the directionality domain of the head of either the NP or the VP.
4.6. The characteristic and illustrious word order freedom of Slavic languages is the overall result of several interacting factors, namely the scrambling potential of phrases within the directionality domain of the head in its base position, the wider range of filler-gap dependencies (see the factor discussed above) due to wider directionality domains, plus the variable positioning of the ambidirectional head in a given phrase as a T3 option, and eventually, the serialization variation for auxiliaries (see property 4.7 below). Let us assume that V°in (47) is one of the many ditransitive verbs with an agentive subject, an experiencer as indirect object, and a direct object. The arguments in the lexical argument structure are hierarchically organized and the projection onto phrase structure conserves this hierarchy. Under these premises, (47a-d) are results of the alternatively available head-positioning for the verb. Consequently, the argument positions in (47a-d) are base positions.
(47) a. Subj V°Obj 1 Obj 2 b. Subj Obj 1 V°Obj 2 c. Subj Obj 1 Obj 2 V°d . V°Subj Obj 1 Obj 2 The relative order of the arguments in (47a-d) is identical. Next, scrambling may apply. The defining property of scrambling is a change of the base order of arguments. (48) lists a few scrambling variants of (47).
(48) a. Subj V°Obj 2 Obj 1 (scrambling variant of 47a) b. Subj Obj 2 V°Obj 1 (scrambling variant of 47a) c. Subj Obj 2 Obj 1 V°(scrambling variant of 47b or 47c) d. Obj 2 V°Subj Obj 1 (scrambling variant of 47d or A-bar topicalization) e. Obj 2 Subj V°Obj 1 (scrambling variant of 47a, 47b or A-bar topicalization) Whenever the clause-initial argument is not the highest ranking argument of the lexical argument structure of the verbal head, there are two sources for this serialization. The order can be the result of scrambling or the result of topicalization, that is, the type of fronting that applies to interrogative phrases too. In technical terms, an argumental phrase in a topicalized position is in a non-argument position which is in a filler-gap relation to an argument position. Clause-internal scrambling, however, is filler-gap relation between two argument positions. In other words, the filler is in a (potential) A-position, and the gap is in the base position of the filler item. This characterization of word order variants is in accordance with the grammatical properties associated with these word order variants, especially with respect to their binding and scopal properties, as documented and discussed in the literature, for instance Bailyn (2003aBailyn ( , 2003bBailyn ( , 2004Bailyn ( , 2012, Junghanns andZybatow (2009), andTitov (2013). Titov (2013Titov ( : 35, 2018, for instance, argues that the preverbal object in a Russian clause with OVS order is in an A-position rather than in an A-bar-position. This, together with a postverbal subject, is hard to reconcile with the clause structure of an SVO language, but it is fully compatible with a T3 clause-structure. The base order of the arguments is subject before object. The verb is positioned in one of the admissible positions, and in this case in the V-S-O order. The object is scrambled across the subject and the verb to the left edge of the "V-S-O"-ordered VP and remains within the directionally unconstrained licensing domain of the verb. Thus, the scrambled object does not leave the argument domain. In an SVO structure, any position of an object preceding the finite verb and the subject is necessarily an A-bar position since it is a position outside the directionality domain of the verb. In sum, the word order patterns of argumental expressions in Slavic languages are expected patterns of T3 languages rather than unexpected, highly exceptional patterns of [S [VO]] languages. In a T3 architecture, they are part and parcel of the T3 system's potential. Under an SVO perspective, Dixon's question (2011:183) "Why impose word order and then dis-impose it?" would remain unanswered since prototypical SVO languages do not scramble.
The correlation with information structure effects is not the cause but the effect. Whenever grammar admits variation, pragmatics and other components take advantage of it. Information structuring employs the syntactic freedom as a vehicle for partitioning this set of variants in terms of information structuring. Language processing is another component. For example, if grammar admits, embedded sentences or PPs and clauses tend to be postponed, unless the language is strictly head-final (= grammatically head-final).
For T3 languages, V-positioning is a means of clause-partitioning and scrambling allows for congruency between syntactic and information structure domains. Given this background, it need not come as a surprise that Prague School syntax has always regarded syntax from an information structuring vantage point. Unlike in Romance and North-Germanic, there has been little chance of detecting immediate evidence for an articulate SVO clause structure. 4.7. The variable serialization of auxiliaries is a direct consequence of the ambidirectional licensing capacity of a T3 head, too. An auxiliary or quasi-auxiliary (e.g. modal, causative, epistemic verbs) selects the form of the dependent verbal head. In Indo-European languages, this is typically an infinitive, a supine (participle) or an aspectual form, such as the English durative marked by "-ing".
In VO, that is, in the head-initial setting, directional selection entails that the phrase with the selected head follows. In OV, the head-final setting, the selected phrase precedes. The T3 setting is ambidirectional. So, in principle, an auxiliary may follow or precede. This, in combination with the verb order variation within the selected VP permitted by the T3 quality of the VP, accounts for the variability of verb orders within a simple clause.

Summary
Filing Slavic languages as [S [VO]] languages is syntactically unjustified. Slavic languages are languages of a type that has not been sufficiently recognized yet in grammar theory. In this type ("T3"), unlike in the head-initial versus head-final types, the positioning of the head in the (verb) phrase is not constrained directionally. Slavic languages are representative of the unspecified directionality value (at least for the VP) whichwhen valuedyields the well-known OV, VO, and VSO grammars. What appears to be highly exceptional from an SVO vantage point is completely regular in T3. Slavic languages ought to be recognized as what they are, namely regular members of a well sized type of T3 language, rather than highly exceptional SVO languages. This will also enhance the predictive power, descriptive precision, and the theoretical accuracy of fit of the respective type assignments.