Slavic languages are Type 3 languages: replies

To emphasize that we are extremely grateful to Paola Merlo and Giuseppe Samo (M&S) for having devised and run computational experimental tests of two central predictions of our theoretical claim is an understatement. What they did is unique. Their experimental tests have been carried out so promptly that they are jointly published with the paper. This was Wundt’s (1888) aim when he laid the scientific foundations of psychology. Here is his trendsetting advice for downgrading and replacing introspection as a valid data source.


Paola Merlo & Giuseppe Samo -"Calculemus!"
To emphasize that we are extremely grateful to Paola Merlo and Giuseppe Samo (M&S) for having devised and run computational experimental tests of two central predictions of our theoretical claim is an understatement. What they did is unique. Their experimental tests have been carried out so promptly that they are jointly published with the paper. This was Wundt's (1888) aim when he laid the scientific foundations of psychology. Here is his trendsetting advice for downgrading and replacing introspection as a valid data source.
"It is totally in the hands of psychologists to take care that these defects disappear more and more. The only thing they have to do is to seize the experimental method." 1 And he adds: "Presently, there are two circumstances that stand in its way. One is arrogance. There are still people who consider experimenting a philistine art which one must not deal with, if one doesn't want to risk losing the privilege of residing in the pure ether of thought." Wundt (1888: 308), tranlated H&S . In Wundt's time, linguistics was a model of methodological rigor, 2 whereas psychology had just been founded as an academic discipline. Today, the will be similar to the outcomes for Romance, modulo the VO/OV split in the Germanic family. 5

Matthew Dryer -inclusion versus exclusion
We greatly appreciate Matthew Dryer's (MD) contribution to the debate of type assignment concerning (his) criteria of inclusion and (ours of) exclusion. Typologists standardly ground their decisions on inclusion, that is, shared properties. In the target paper, we intentionally employ only criteria of exclusion, that is, properties that languages of a particular type do not display. At first glance, such an alternative appears to be but different perspectives on the same issue. Upon second thought however, it turns out that the alternative perspectives demonstrably lead to contradicting results. It is this apparent inconsistency that deserves attention and needs to be resolved. MD argues that Russian qualifies as SVO because it shares what seems to be SVO core properties. We, on the other hand, argue that Russian must not be classified as an [S[VO]] language because quite a few grammatical properties are inconsistent with other core properties of this very type. Let us therefore start with the inclusion relation (1).
Let L i be a natural language, P 1−n a conjunction of n grammatical properties, and W j a word order type. An inclusion-based type assignment rests on the implication (1a) or its equivalent form (1b). (1) Inclusion: a. If P 1−n (L i ) then L i ∈ W j b. If L i ∉ W j , then ¬P 1-n (L i ) If (1) is to serve as the premise of a valid inference (by modus ponens), the premise first of all must be empirically true. This, however, is not guaranteed in the context we are dealing with. Languages of a given type may happen to partially share properties of other types. This invalidates inferences based on the inclusion condition. Here is an example. Afrikaans, Amharic, Dutch, Frisian, German, Kurdish, or Persian, to name but a few languages, are head-final for verbal heads of phrases (SOV), but noun phrases and PPs are head-initial. Genitives follow the noun. Relative clauses follow the noun in most of these languages. Thanks to Dryer and Haspelmath (2013), these facts can be retrieved from the WALS database. What these languages illustrate is a so-called disharmonic phrase structure organization. In Dutch, moreover, the most frequent and in many cases the only grammatical relative order of auxiliaries ('dominant') is a relative order as in English. So, according to an inclusion criterion based on the four properties adduced by MD, Dutch would qualify as an SVO language. Descriptive grammarians however insist that it is an SOV language with a superimposed V2-property of the finite verb in main clauses. Why would grammarians insist on Dutch being SOV? They factorize the web of properties, subtract the interfering irrelevant ones (e.g. the V2-property) and identify the bare SOV structure. 6 It surfaces most clearly in non-finite structures. No unquestionable [S [VO]] language obligatorily places nominal objects in front of the non-finite main verb, and no unquestionable SOV language places them after the base position of the verb.
(2) a. iemand de oplossing (te) vertellen someone IO  In sum, an inclusion criterion is fully reliable only if the adduced criteria are necessary and sufficient for the given type, or in other words, if no other type is compatible with the selected criteria. This is not the case for the set of four properties 7 adduced by MD. Such indeterminacies reduce the discriminatory as well as the predictive power of typological classifications. "Dominant word order" is a case in point. The way SVO is defined in typological surveys is not selective enough: 8 "Where a language is shown […] as having a particular order as the dominant order in the language, this means that it is either the only order possible or the order that is more frequently used." (Dryer 2013b). An either-or condition ties together two mutually exclusive properties, namely languages with word order variation and languages with strict word order. So, only because of such a definition, English ends up as a member of the same type as Russian, although most of the word order variants of Russian are ungrammatical in English, and although the English word order pattern is just one of the many grammatical patterns of Russian. The two grammars are obviously disjoint but the languages allegedly belong to the same word order type.
T3 languages are wholly compatible with MD's four properties and this is what we claim for Slavic languages, too. We base our type assignment of Slavic languages exclusively on exclusion criteria. The set of properties we adduce is not shared by the crucial type, that is [S [VO]]. Hence the exclusion condition (3) is factually valid and an inference is logically sound. According to the set of criteria we adduced, Russian cannot be assigned to the [S [VO]] type since each property is a property that disqualifies a language for this type. (3) Exclusion a. If P 1−n (L i ), then L i ∉ W j b. If L i ∈ W j , then non ¬P 1−n (L i ) In the final paragraph, MD concludes that "there is a way to explain why Russian exhibits the VO characteristics noted without saying that the language is syntactically SVO." We totally agree since we think that this is exactly what we have tried to achieve. T3 languages subsume grammatical properties shared by VO languages as a subset of their properties. This is also the reason why T3 languages have hitherto not been generally acknowledged as a type of its own.
3 Jacek Witkośon binding & scope, and an invitation for testing a prediction "It is a pleasure to comment on" the comment of Jacek Witkoś (JW). His discussion of binding and scoping effects associated with fronting of binders or bindees not only shows that Polish and German behave alike in this respect and unlike the showcase for SVO, namely English. This is what Frey (1993) has studied and theoretically modelled in detail by contrasting German with English. 9 It is summarized in Haider (2010: 150-152). In a well-formed binding relation, a displaced binder is always the chain head viz. the highest element of a movement chain. Scoping, on the other hand, can operate on every chain link (Frey 1993). A quantifier Q can attain a wide scope reading with respect to an expression E, if Q c-commands at least one link of the chain of E; see also Aoun and Li's (1993)  In (5), scrambling produces alternative scope domains. On the one hand, the cardinal quantifier may be assigned wide scope since the scrambled direct object c-commands the dative DP with the universal quantifier. On the other hand, the universal quantifier may be assigned wide scope because the dative object c-commands the trace of the scrambled accusative. In other words, the base position (= trace position) of the scrambled object is in the c-command domain of the dative object and therefore, the cardinal quantifier is in the scope domain of the universal quantifier. It is this set of circumstances that offers a possibility of directly testing the T3 V-positioning hypothesis against the DP-fronting ('scrambling') hypothesis of the standard Generative framework because the predictions for scopal ambiguities differ in a crucial case, namely (6a). In the SVO & scrambling analysis of preverbal objects in Slavic languages (6b), the preverbal objects are in derived positions. So, each DP in (6b) c-commands the trace of the other DP. Consequently, two readings are possible. Each quantifier alternatively qualifies for wide-scope.
In the T3 analysis (6c), however, the two objects are in base order and in base positions, with the verb in the foot position of the VP. So, the predicted scope relation for (6a) is the same as for (4), namely unambiguous. For scope-taking, it is irrelevant that the verb in (4) is in a higher position in the T3 V-chain than in (6c). VP (T3 structure: objects in base positions) In sum, if the SVO-based 'scrambling' analysis (6b) is correct, (6a) will be just as scope ambiguous as (4), which we doubt, however. We trust in our theory and predict that adequately chosen instances of (6a)ceteris paribus 10will turn out as not scope ambiguous. In our paper, the focus is narrowed to what we take to be the limiting case. A limiting case is a constellation with as little as possible interference. In our case, the limiting case is the passive of a verb with a single argument, i.e. its subject. When such a verb is passivized, the subject argument gets syntactically neutralized. The result is a clause without any syntactically available argument. This is the limiting case, but of course and by far not the only context for identifying obligatory expletive subjects. Here is once more a selection of data illustrating the predicted contrasts between SVO and SOV, with a null-subject property (7c, d) and without (7e, f). The author of (7f) is Jean de La Fontaine (Contes I.I, 518).
(7) a. dass gelacht wurde German that laughed was (SOV: no expletive) b. Waar niet wordt gelachen, wordt niet geleerd 11 Dutch where not is laughed, is not taught (SOV: no expletive) c. *Quiè stato lavorato troppo poco Italian here is worked too little (SVO, no expletive available) 10 The test items should be free of scope-biasing factors such as specific information structure conditions like focus on one of the two objects, and the situational content should not bias a reading with a particular scope. 11 https://ikwerkaanwerkplezier.nl/professional/blog/stress-burnout-en-overspanning/drietips-om-leren-leuker-te-maken-voor-jezelf d. *that (it)/(there) was laughed English (SVO, no expletive available) e. For så bliver det snydt Danish because then is it cheated 'because then there is cheating' (SVO: obligatory expletive) f. Il fut dansé, sauté, ballé.
French it was danced, jumped, danced. (SVO: obligatory expletive) German and Dutch are OV. The SOV clause structure neither needs nor admits a VP-external subject position since all arguments find their place within the verb projection in the same directionality domain, namely in preverbal positions. Consequently, there is no room for an expletive that takes care of an otherwise empty obligatory subject position (Haider 2014(Haider , 2019. Danish and French are testimonies for SVO languages with an obligatory subject expletive. In both cases the expletive is homonymous with a third person singular pronoun. Such a pronoun would be turned into a null pronoun in prodrop languages. But Romance pro-drop languages, represented by Italian (7c), supply evidence 12 sufficient for debunking the oxymoronic idea of an "empty expletive" as a theoretical fiction that does not materialize in languages. Overt expletive subjects, however, must not be confounded with semantically empty subject arguments (for details see Haider 2019) and their null versions, as in the Italian example (8a).
Italian (si passivante) not REFL of speak in-fact 'In fact, one does not speak about it.' b. Om det pratas det mycket.
Swedish (s-passive) about it speak Passive EXPL much c. Darüber spricht es sich leichter.
The examples in (8a-c) are grammatically closely related. Both, the Italian impersonal si-construction (8a) and the Swedish s-passive (8b) are diachronic successors of a middle construction illustrated by German (8c). The semantically empty subject es ('it') is obligatory in German intransitive middles (8c) but es is ungrammatical as an expletive in subjectless passives (8d). In Italian, the semantically empty subject is a null subject due to the null-subject property (prodrop). This is the key for understanding the contrast between the passive (9a) and the si-construction (9b) in a Romance pro-drop language.
(9) a. * È stato parlato di te. Italian has been talked about you b. Si e parlato di te.
If the standard passive (viz. be-auxiliary plus perfect participle) is applied to an intransitive verb in a pro-drop language, as in (9a), a pronominal expletive would be phonetically null, because of the pro-drop property. However, null expletives would be unidentifiable. A semantically empty subject as in (9b), on the other hand, can be identified by virtue of being related to an argument slot of the verb. Arguments are specified in the lexical entry of the verbal head. So in (9b), the counterpart of the German es ('it') in (8c) is the null-subject of middle constructions and its successor construction in Italian.
It is crucial to distinguish between expletives and semantically empty subjects in Slavic languages, too. Perlmutter and Moore (2002) as well as AA, who follows them, do not seem to honour this difference. The subject of impersonals such as (10a) is a null pronoun representing the semantically empty subject argument. This must not be confused with a syntactically argumentless intransitive passive, as illustrated in the target paper by example (10b).
Polish 13 whether was cleaned in-the-morning 'whether cleaning took place in the morning.' c. Die Boote hat es in Stücke gerissen.
German the boats Acc has Sg it Nom in pieces ripped 'The boats broke in pieces.' d. Bátana hefur brotið í spón.
Icelandic boats Def-Acc has Sg broken in pieces (10a) involves a null subject, but (10b) is subjectless. The accusative on the object in (10a) is the indicator of the presence of a null subject. 14 In undisputed SVO languages, the subject position of (10b) would have to be filled with an expletive or the sentence would be ungrammatical. The null subject in (10a) is the null variant of the semantically empty subject es ('it') in the impersonal variant of many transitive verbs in German (10c). (10d) (Zaenen and Maling 1990: 145) is the Icelandic counterpart of (10c), with a null pronoun for the semantically empty subject, as with weather verbs. Both in Czech as well as in other Slavic languages, and in Icelandic, too, weather verbs are the simplest instance of verbs with a semantically empty null subject in these languages. AA's own examples (8) are not representative of null expletives but of the semantically empty subject argument of meteorological verbs in a null-subject option. 15 Having explicated this often neglected but indispensable distinction once moreviz. overt expletive subject versus overt semantically empty subject argument, which, in pro-drop languages, is a null subject (see Haider 2019), we are ready to delve into the debate proper.
AA cites Pitteroff and Schäfer as authorities for the ungrammaticality of "strict" intransitive passives in French. According to them, a bare intransitive passive such as (7f) is ungrammatical, which, by the bye, implies an imperfect command of French by La Fontaine. Anyway, we do not feel compelled to accept their rating. Instead, we prefer the evidence of grammars and large corpora. As for grammars, Rowlett (2009: 46) explicitly states "This conclusion is further supported by the existence of impersonal passives based on intransitives, as in (87)." His example (87) is AA's starred example, viz. Il aété dansé. Boer (1954: 55, §68) 16 and Hériau (1989: 164) are earlier testimonies for the very same example, and Miclosich (1883: 81) is an even earlier one. They all refer to the very specimen as grammatically well-formed French. Corpus searches readily produce tokens of 14 In nominative-accusative case systems, accusative assignment presupposes nominative assignment. In the absence of a nominative, the object is assigned nominative, but not accusative; see Haider (2000b), and Szucsich (2006)  (11) a. Il aété dansé et chanté. 17 EXPL was danced and sung 'There was dancing and singing' b. Il aété marché, piétiné, martelé par une foultitude de clients. 18 EXPL was walked, trampled, hammered by a multitude of customers Helland (2000: 88) states "Au contraire, les verbes intransitifs (ou inergatifs) comme courir, marcher, nager, tousser, etc., qui possèdent un argument externe, se prêtent à la passivation." His examples are examples of a bare or "strict" (see above) passive of intransitives.
AA endorses Pitteroff and Schäfer's conjecture "that the overt element in Spec,TP in French or its covert counterpart in Russian is not a real expletive and needs to associate with an argument." This is contrary to the facts. First, there is no compelling evidence but counterevidence for the assumption that il associates with something in (11) and (7f). Syntactically, il, unlike English there, does not behave like an element that needs an associate, as the agreement contrast between the parallel constructions in English and French in (12a, c) documents. There needs an associate from which to copy agreement features (12a, b). Il does not associate with a DP (Haider 2019: 28). In (12c) and in contrast with (12a, b), the finite verb agrees with the pronominal expletive and not with the postverbal argument, as number agreement reveals.
(12) a. There have Pl emerged some profound questions Pl b. There has Sg emerged a profound question Sg c. Il est/*sont apparu/*s des disparités. French it is Sg /have Pl appeared/ Pl-Agr some disparities d. Des disparités sont/*est apparues. some problems have Pl /is Sg appeared The fact that there is associated with a post-verbal argument disqualifies it, by the way, as an expletive in the English intransitive passive (Haider 2019: 35-36). However, if we look more closely, it is not the item there but the structural subject position that is associated with the donor of agreement features, and there cannot provide such features by itself. This is evident from PP-inversion data that can easily be retrieved from corpora. 19 In French, the expletive is a pronoun, and pronouns provide agreement features, namely third sg. in the case of il. Second, if "a covert counterpart" of a lexical expletive were a grammatically available option, this would entail that Romance null-subject languages passivize intransitive verbs, which is contrary to the facts, as AA admits (see fn. 12 above). The same considerations that apply to French apply to Russian. An "empty expletive" is a fictitious concept. Let us turn now to Icelandic as the oddball among North Germanic languages with respect to subject expletives. As for Icelandic impersonal passive, there is a clear contrast and it supports our theory. In embedded clauses, Icelandic employs an expletive, see (13a) (Thráinsson 2007: 355), (13b) (Maling and Zaenen 1978: 491), and (13c), 20 while German forbids it, as the direct counterpart (13d) illustrates.
she said that EXPL has been danced yesterday c. Enég man [að það var dansað].
but I remember EXPL was danced d. Er sagte dass (*es) getanzt wurde auf der Hochzeit they say that (EXPL) danced was at the wedding Second, there is an essential difference between Icelandic and other SVO languages, such as English or Romance languages, which accounts for the apparent absence of a subject expletive in V2 clauses. The difference concerns the syntactic properties associated with the structural subject position. In Icelandic, the subject argument is not bound to end up in the VP-external structural subject position. Nominative may be assigned VP-internally and the subject may remain in-situ, while a non-subject argument is placed into the structural subject position. This is the renowned and well-studied "quirky subject" property of Icelandic. What this shows is that in Icelandic, the structural subject position is not obligatorily involved in case-management. It is a structurally obligatory, VP-external position that is open for the subject, or, alternatively, in case the subject is a lower ranked argument, for the higher ranked argument, as in (14a) (Thráinsson 2007: 292 (14b) (Thráinsson 2007: 276), the object is not fronted and an expletive serves as the dummy subject for the subject position and has been raised to Spec-C.
have Pl him Dat been-forgiven Pl all crimes Def-Nom.Pl b. Það var barið barn.
there was hit Sg.Neut child Nom/Acc.Sg.Neut The remarkable property of Icelandic in this respect is the fact that even a trace of virtually anything that can be fronted to Spec-C can also serve as a filler for the spec-T position. 21 This is a peculiarity, but Icelandic is not the alleged exception to the structural subject requirement of SVO languages. It is because other items and especially the trace of fronted items can serve as fillers as well that Icelandic does not always present an expletive in the otherwise empty structural subject position. Given this fact, we wonder how sustainable the final message in AA's comments will be, which states that "it is uncontroversial that Greek is a VO language." T3 seems to be at least a plausible alternative.

Eric Fuß -Early Germanic is not Slavic
Readers should not be perplexed finding themselves confronted with a treatise on Old High German in a volume devoted to Slavic languages. The subject is wellchosen. Arguably, T3 is the clause type of early stages of Germanic languages, too. So, we are grateful to Eric Fuß for contributing his viewpoint. Since readers are not 21 Thráinsson (2007: 330) notes "similarities between Stylistic Fronting (SF) and the overt expletive" and "All this may seem to suggest that the elements fronted by SF and the overt expletive have a similar role" (p. 331). One instance remains unaccounted for, namely the absence of an expletive in the position following the finite verb in subjectless V1 questions; see Thráinsson (2007: 312).
assisted by background information in the target paper, a few remarks may be of help.
Let us do away with a potential misunderstanding first of all. The presupposition of the title question -"Early German = Slavic"is not what we claim. Early German and modern Slavic languages aren't grammatical twins, but as proposed in Haider (2013Haider ( , ch. 5.4, 2014, Slavic languages and the Old Germanic languages, too, are T3 languages. Such languages can be as diverse as the class of languages with an [S[VO]] clause structure, but they share the core properties of the respective type. Relevant is the following. If a given type assignment is empirically adequate, languages of a given type are expected and predicted to share core properties of that type unless independently ascertained factors interfere, not more and not less. Let us start with an easily accessible feature, namely word order variation. It is commonplace that classical languages, viz. Ancient Greek, Latin, and Sanskrit, display this property beyond doubt. Furthermore, it is handbook wisdom that Old Germanic languages display ample word order freedom. Faarlund (1994: 54, 63) reports that "Old Scandinavian generally has a rather free word order. This is true also at the phrasal level" and "The sentence in Old Scandinavian is characterized by […] a relatively free word order, in the sense that the relative order of phrases in terms of grammatical function is variable. The order of elements in terms of discourse function, on the other hand, is rather fixed." According to Van der Wal and Quak (1994: 105) "Word order in Old High German and Old Saxon was rather free. The modern SVO or verb-second order in unmarked declarative sentences is by no means obligatory in the earliest texts." The period of word order freedom extends well into the so-called 'middle' periods of each Germanic language. In these periods, the directionality of the verbal heads became fixed. This can be studied best in the Germanic family, with their split into an OV and a VO group of languages (Haider 2014).
As for the theoretical coverage, it seems the first author has not been successful in expounding the sharp difference between competing accounts in Section 3 of Haider (2014), otherwise EF could not have misunderstood it when writing that the "approach in terms of Variable Head Positioning (VHP) seems to be very similar in spirit to the Double Base Hypothesis (DBH) proposed by Susan Pintzuk" and "moreover, the approach by Pintzuk is embedded under a general theory of language change in terms of Grammar Competition (Kroch 1989), which captures quantitative and variationist aspects of word order change." In fact, the claim presented in Haider's paper is the very opposite. First, grammar competition 22 between VO and OV in a stage of English is implausible since it is not restricted to English. The very same patterns that motivated the grammar-competition conjecture for Middle English are found in any other Germanic language from the Old to Middle Germanic periods. So each and every speaker of a Germanic language would have had to be bi-grammatical, constantly switching between two grammars. Whoever is happy with such a theory is free to endorse it. We don't, since it is too far from plausible; see the discussion in Haider (2014).
Second, even the Old-to-Middle English data fail to fit Kroch's and Pintzuk's double-base hypothesis. If they fitted, we would be able to neatly factorize the data into two complementing subsets, namely SVO and SOV structures. But, this is not the linguistic reality. What we see is the complete array of T3 structures (16). The proponents tend to overlook the attested word order patterns that are ungrammatical both in SOV and in SVO, but licit in T3, viz. (16a, b). This is the crucial piece of evidence, and it is discussed at some length in Haider (2013Haider ( , ch. 5, 2014 (16a, b) remain unaccounted for. They do not fit since they are neither SOV nor SVO orders. The "general theory" of grammar change that, according to EF, is missing, is the theory presented in Haider (2014), which characterizes Germanic SOV and SVO as successors of T3, with a single change, namely the fixing of the directionality of the verbal head. There are only two possible options for instantiating a directionality value, namely either V / O or O ← V. In the Germanic family, each option has found its implementation, with VO in the North-Germanic group and OV in the continental West-Germanic group.
The "general theory of change" rests on a "general theory of structure" which has been developed in the past three decades, starting with Haider 1992Haider [2000. For a commented research bibliography please consult the preface of Haider (2013: ix-x). A succinct exposition is given in Haider (2015). The bare fundamentals are presented in the target paper in Section 3.2, in Haider (2014) What we observe in the majority of languages are lexical heads with fixed directionality. This property is parametric, with two alternative values. Let us refer to them as "left" or "right." In an OV-language, the verb licenses its arguments to the left (17a). In a VO language, it licenses to the right (17b). In this case, the universal condition (A1), viz. universal right branching, cannot be instantiated in complex projections unless the verb is re-instantiated. 24 The result is the so-called shell structure of complex head-initial VPs. For details, see Haider (2015 T3 is the option with an un(der)specified value for the directionality parameter. This means that the directionality of licensing is directionally unconstrained. In diachronic terms this often turns out as the precursor of not yet constrained. In T3 structures, the verb may occupy any of the structurally available positions, that is, in the foot-position, as in SOV (17a), and in any higher position, as in SVO (17b) or VSO (17d). In other words, the T3 languages display the cumulative serialization patterns of SOV (17a) and SVO (17b), and in addition (17c). It is this pattern that immediately betrays a T3 structure. The other patterns, especially the frequent pattern (17b), have led to confounds. It is due to this pattern that T3 languages have been misclassified as SVO, which is a main topic of the target paper. The theory of grammar change based on the tripartite typing invites the prediction that diachronically, a period with fixed directionality is typically preceded by a T3 period. The Indo-European languages are testimonies for this claim. On the other hand, languages with fixed directionality are unlikely to drop directionality again but they may re-value it occasionally. We are especially grateful to Merlo & Samo for having tested the prediction for the development of Romance languages. It is an example of a diachronic development, starting with a T3 language (Vulgar Latin) and resulting in a family of SVO languages as offspring.
In present day versions of Generative Grammar, the SVO sentence structure of English is the prototype of the universal clause structure. SOV is a 'distortion' of it, derived by the obligatory fronting of noun phrases. 25 This is bluntly contradicted by what is known about grammar changes. SVO languages are a point of no return, as Gell-Mann and Ruhlen (2011: 17291) summarize it in their Figure 1. An SOV language may change by drift or natural selection into SVO, but not vice versa (see Haider 2021). So, how would SOV languages come into being if the centre of grammatical gravitation is SVO? Why engage in all kinds of movements that distort an SVO base structure if a grammar without such movements is a simple and more perfect system? This is the question. 26 EF mainly adduces data from Old High German. These texts, however, are notorious for an inescapable drawback. They primarily teach us how monks translated Latin texts into local idioms. The only text that is likely to come close to a spoken variety is a 4 page fragment of a travel guide, viz. the "Parisian conversations" ("Pariser Gespräche"). EF excerpts "Isidor," which is a collection of translated texts of Isidor of Sevilla, and "Notker," a collection of translations 27 by a Benedictine monk of this name. 25 Haegeman (2001) discusses various attempts of implementing this assumption in the respective version. 26 The various approaches in main stream Generative frameworks that presume a universal SVO base order are unable to answer it satisfactorily since diachronically, these theories put the cart before the horse. 27 For the use in convent schools, he translated classical literature and Bible texts; notably the Psalter.
Slavic languages are Type 3 languages Braune (2004: §la) emphasizes that the "Old High German language is accessible only in a very small opening. As monks and clerics, the writers of OHG glosses and texts were members of the social upper class. What they wrote belongs to highly specialized text types. Everyday language is not at our disposal, except for the Paris conversations." (translation HH ) Nevertheless, the academic opinion and EF, too, favor SOV, apparently as the least inappropriate type for OHG. Frequent VO orders are explained away either by invoking extraposition, although extraposition of noun phrases is ungrammatical in any modern Germanic OV language, 28 or by assuming that VO "simply mimics the ordering found in the corresponding Latin clause." On the other hand, a small percentage of divergences between the Latin text and its OHG translation (7% in the Tatian corpus, according to EF's Table 1), where the Latin order is VO and the translation is OV, is adduced as evidence for SOV. Evidently, this is flawed reasoning. It is by no means excluded that the parallel orders show that in OHG and in Latin, the very same order is acceptable while in a minority of cases, the translator prefers fronting some items, for whatever reasons. Who could tell exactly when the translating monk 'mimics' and when he abides by the grammar? All we see is that there are about fifty-fifty OV and VO orders, 29 just as one would expect to see in a T3 language. This hypothesis straightforwardly accounts for them, without invoking unmotivated extraposition (partly of unextraposable elements) or alleged grammatically deviant "mimicking." Finally, we are short of any clue as to why EF reopens a discussion on ungrammatical V-Obj-Aux orders in Section 2.2. The universal ungrammaticality of V-Object-Aux orders is a subcase of the universal restriction (A1), viz. the "Basic Branching Condition" (Haider 1992(Haider [2000(Haider ], 2010(Haider : 339, 2013. This condition subsumes the FOFC 30 cases he refers to, since it rules out any left-branching (extended) projections. FOFC is just a descriptive generalization and it immediately follows from (A1). This explanation has been published more than a decade before FOFC has been 28 See Schallert (2007: 71) for cases of allegedly extraposed non-extraposable items in OHG. 29 In addition, OVO-orders occur, just like in OE, as Schallert (2006: 139, 172) Haider (2020b), ironically accused himself of "anticipatory plagiarism" because of having anticipated FOFC already in Haider (2000a), as a subcase of the Branching Condition (BBC); see Haider (2013: 132-135). The bibliography of Biberauer et al. (2014) lists this paper (for other purposes), which verbatim states what they claim as their own original idea in their paper. "If a functional projection is a functional extension of the projection of a lexical category, the BC rules out functional heads to the right in general." (Haider 2000a: 48). Flabbergastingly, they did not bother crediting it.
conceived of, as has been laid bare in Haider (2013: 132-135). It is a principle-based explanation and it is stronger than FOFC 31 because it is more restrictive.
], but it may be mistaken for it In their comment, Šimík & Jasinskaja (Š&J) put a most welcome focus on Czech. They subscribe to the widely entertained view according to which Czech is a language with a basic clause structure like English, as they explicate in Jasinskaja and Šimík (in press). We are grateful for their thought-provoking attempts of challenging our theory thoroughly by confronting it with properties of a Slavic language we have considered only in passing in our paper. After all, the essential quality test for a theory is resolute and thorough falsification trials. Gladly we grasp the opportunity of demonstrating that our theory stands the test successfully. For ease of reference, we juxtapose their summary and the version we shall defend in Table 1. Š&J find fault with four properties we attribute to Slavic languages, when applied to Czech, namely (ii), (vi), and (viii), plus property (ix), which they add although it is a subcase of (vii.). Let us start with (viii) and an 'operating instruction' for the list of syntactic properties in the tables below. It is important to language but it is possible for them to share one or the other property, for independent reasons. All we claim is that a language is not [S [VO]] if the relative order of auxiliaries or quasi-auxiliaries is variable. On the other hand, languages with a uniform order of auxiliaries are consistent with any type. Therefore, we gratefully accept the information that in Czech, there are invariant orders of auxiliaries. 32 Corpus data confirm however, that at least the order of quasi-auxiliaries is variable. 33 Let us continue with line (ix), which Š&J added to our table. In fact, free OV/VO order is a sub-instance of (vii.). Here, "free" is to be interpreted as syntactically free, that is, not narrowed down by syntactical conditions, or, as Siewierska and Uhlířová (2010: 109) put it: "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." In OV or VO languages, nominal objects are distributionally restricted by the directionality requirement of the head they depend on. Nominal objects obligatorily precede the base position of the verb in OV. In VO, they obligatorily follow the verb. The order restriction is independent of information structure properties, intonation effects, or other pragmatic preferences. In Czech, and in fact in all Slavic languages, even in Sorbian to a minor extent, 34 objects as well as the subject may precede or follow the verb. This fact cannot be seriously contested. 32 Š&J's Czech example, however, does not match the Polish counterpart. As the authors concede, it involves a semi-lexical verb dát 'give' and a modal chtít 'want', which systematically allows for finite embeddings with the subjunctive marker by, introduced by a complementizer aby 'that'. In contrast, the Polish modal musieć 'must' does not show this behaviour, and B/C/S morati 'must' only selects finite complements in varieties which have lost the infinitive or, at least, pushed it back. So, syntactically, Polish musieć (and Czech muset) differ from chcieć (and chtít). 33 In brackets the number of hits of the respective search restricted to Google books: (1130) lie let let lie 'let lie' ' let lie' 34 Even Sorbianalthough predominantly verb-finalis not strictly SOV, as (i) illustrates (Scholze 2015: 206): (i) Ćeta dari mi rjaneho žurka. aunt gave me beautiful hamster However, and crucially, "free" must not be interpreted as "any order is acceptable anytime." The existence of a pragmatically unmarked or neutral order is fully consistent with the syntactical variation potential. The fact that alternatively available orders are not always freely exchangeable in utterances is an independent issue. Information structure preferences partition the pool of syntactic variants. We do not want to repeat the explication of the interface effects presented in Haider (2020). Let us merely summarize it: "In general, when syntax admits structural variation, this potential is captured and utilized by other subsystems of grammar" (Haider 2020: 375). All variants are syntactically well-formed, but, of course, they are neither equivalent with respect to information structuring, since they may be associated with particular focus, topic, or givenness properties, nor with respect to scoping or binding since word order is an interacting structural variable.
Importantly, the conditions of information structuring do not constrain syntax; they merely exploit the syntactically available options. On the other hand, if syntax does not admit variation, information structuring cannot coerce syntax, otherwise all languages would closely resemble Czech or Russian. It is a fact of Czech that there are acceptable utterances in which a direct object precedes a main verb and that there are acceptable utterances in which a direct object follows a main verb, finite or not, in main as well as in embedded clauses.
The relevant data are familiar and have been reconfirmed on independent grounds, for instance by computational methods, measuring the word order freedom in parsed corpora (tree banks). Kuboň et al. (2016) present the following percentages for Czech, on the basis of a tree-bank corpus of 16,862 main clauses and 11.849 embedded clauses out of 87.913 sentences in total, in comparison with 22 other languages. (18) and (19)  The percentages of Czech and Slovenian OV orders contrast clearly with undisputed SVO languages such as English or Portuguese, for instance. Kuboň et al. do not differentiate between pronominal and non-pronominal objects. This accounts for the "OV" orders in a VO language such as Portuguesea language with pronominals cliticized to finite (auxiliary) verbs 35and English, a language without pronoun cliticization. Kuboň et al. (2016: 15) compare and rank their set of 23 languages by computing four measures of variation (viz. max-min, Euclidian distance, cosine similarity, and entropy). On each of the four measures, Czech ends up in the top group of five languages in the ranking of word order freedom (20). SVO languages, such as English and Portuguese, are in the opposite region, namely in the top group of languages with highly restricted word order. The entropy-ranking for languages with the greatest word order freedom is shown in (20) Let us turn now to another discriminating trait, namely the obligatory preverbal subject position of SVO languages. Actually, this is a type-defining property. The [S [VO]] clause structure implicates an obligatory structural position for the syntactic subject of the clause. This position is outside of, and preceding, the VP. An [S[VO]] clause is ungrammatical when this position is radically empty. This is the case when there is no subject argument available and the position is not filled with an expletive, as in the unacceptable example (21a), contrasting with (21b):

35
(i) O comprador não o teria encontrado. the buyer not it would-have found (21) a. *Dimanche 24 mai, aété procédé à l'installation du conseil municipal. Sunday 24 may, has been proceeded to the-installation of-the municipal council b. Dimanche 24 mai, il aété procédé à l'installation du conseil municipal. 36 Sunday 24 may, it has been proceeded to the-installation of-the municipal council The limiting case of a subjectless construction is the passive of intransitives. If an intransitive verb is passivized, there is no argument left. So, the subject position must be filled with an expletive (21b) in SVO. In French, the expletive is the third person sg. masc. pronoun. Such pronouns falls prey to pro-drop in Romance nullsubject languages. Consequently, Romance pro-drop languages cannot and do not passivize intransitive verbs since the expletive must not be a null pronoun 37 (Haider 2019 ] language, the subject position must not remain empty. If it is empty, the result is ungrammatical, as in English, unless an expletive is adduced for filling the obligatory subject position, as in French. In SOV and T3 languages however, there is no obligatory structural subject position, hence no room for a subject expletive. Czech behaves as expected and predicted for a T3 language, and so do other Slavic languages. The passive of intransitives is not the only source of evidence, of course, but it is the syntactically most straightforward one. There are numerous papers on Czech impersonal constructions (cf. Guiraud-Weber and Kor Chahine 2013) with data such as (23), which have to be analyzed properly (see Szucsich 2006), however. As discussed in detail in Haider (2019: 20), unlike expletives, semantically empty subject arguments are licit null-subjects in pro-drop languages. The presence of an accusative object in (23) is an indication of the presence of a null subject. So, technically, (23a, b) from Guiraud-Weber and Kor Chahine (2013: 12), and (23c) are not subjectless. In the German counterpart (23c), the semantically empty subject es ('it') is audible. The verbs in (23) are agentive verbs used in a variant in which the subject argument is not an agent but unspecified, that is, semantically empty. Let us keep in mind that preverbal objects are ungrammatical in prototypical [S[VO]] languages, unless they have been wh-moved to a clause initial position.
Suddenly has it him without noticeable cause knocked-over 'He was suddenly knocked over, without noticeable reason.' Let us finish the data review with a clear-cut case of a subjectless clause, viz. (24a), taken from Guiraud-Weber and Kor Chahine (2013: 9). Czech and German share the very same construction, namely a copula construction with a dative plus a nominalized verb in a PP. (24a) translates word by word into German (24b), with the exception of the cliticized negation. Such a construction is inaccessible in an [S [VO]] language since it does not contain a subject and, as German (24b) confirms, there is no (hidden) semantically empty subject involved.
(24) a. Petrovi (ne)bylo do smíchu / řeči / zpěvu. Peter Dat (NEG)was Neut PREP laugh Gen / talk Gen / sing Gen 'Peter felt/did not feel like laughing/talking/singing' b. Dem Peter war (*es) nicht nach Lachen / Reden / Singen. the Dat Peter was (it) not PREP laughing / talking / singing Let us summarize the discussion of property (ii). In the Czech sentence structure, a structural subject position is neither obligatorily present nor obligatorily filled. The contrast between Czech and undisputed [S [VO]] languages is clearly demonstrable. Czech behaves as expected and predicted for a T3 language. Let us turn now to property (v.), viz. the absence of the LLC effect for immediately preverbal adjuncts in T3 languages. We predict that the LLC effect is absent in Czech because the VP of a T3 language such as Czech or Russian does not count as a syntactically head-initial VP. Š&J's objection is easy to dismiss. What they present is two sentence pairs, namely their examples numbered (16a, b) and (17a, b), from Czech and from Russian, respectively. They rate one sentence of each pair as ungrammatical. In each case this is a sentence that consists of a noun at the beginning and a verb at the end, with a single, overlong, center-embedded but extraposeable phrase as the 'meat' of this syntactic 'hamburger'. We aren't surprised at all that such utterances would be rated less 'palatable' or even unacceptable, in comparison with their extraposed versions. But, we are surprised that this is presented as counterevidence. Š&J have only shown that utterances are avoided that contain 'very heavy' center-embedded phrases that could be extraposed. This is not the point however. The point is that, independently of their size, pre-VP adjuncts must be head-adjacent to the VP in languages with strictly head-initial VPs. This is easy to test. The test is not restricted to APs. It is sufficient to adduce head-initial PPs.
A search in three big English corpora 41 confirms the effect of LLC in English (25b, c). The expression "should more carefully" is well attested in preverbal position. However, as predicted for a PP, the sequences "should with care," "should with great care," or "should with more care" are absent in these three corpora in the pre-VP position (see Haider 2018). They are frequently found in clause-final or clause-initial positions, however. The same is true for French (très) soigneusement in comparison with avec (grand) soin.
(25) a. She (has) much more carefully examined the case b. *She (has) with great care examined the case c. *She (has) after a few minutes stopped the examination It is easy to locate Czech corpus data showing adverbial PPs in immediately preverbal positions (26). Note that for English, the BNC for instance does not contain a single token of "before midnight," "with great care," or "with pleasure" immediately before the verb although more than one thousand tokens of each PP can be retrieved from this corpus. In sum, Czech is evidently not subject to the LLC constraint. This is what is predicted if the Czech VP is directionally unconstrained, which is a defining T3 property. Eventually, Š&J object to line (vi.) in the table, that is left-branch extractions from noun phrases in preverbal positions. 45 The term 'preverbal' refers to the relevant positions in the clause structure of SVO languages. The preverbal subject and any word order property. 50 Slavonic free word order is a word order type, 51 namely T3, and this type embraces all Slavic languages (with some more specific preferences in Sorbian). The Slavic word order type characterizes a type of languages with a clause structure of mobile verbal heads without directional licensing restrictions, and the concomitant potential for scrambling, or in other words, it is the T3 type. In sum, thanks to the issues raised by Š&J, we are content to find ourselves in a comfortable position for concluding on solid empirical grounds that the evidence for classifying Czech as a Type 3 languages is good, and in fact as good as the evidence for other Slavic languages.