How theoretical is your (historical) syntax? Towards a typology of Verb-Third in Early Old High German

Verb-Third (V3) constructions, i.e. root clauses displaying more than one element in their prefield, have been attested throughout the history of German. In this paper, I discuss some methodological issues related to the investigation of this structure in Early Old High German (EOHG, eighth to ninth century). In this language stage, the linear syntax of matrix clauses is very unstable—mainly due to the not-yet complete systematization of the V2/verb-final distinction and to the considerable amount of extraposition in main clauses—which makes the detection of this construction in the available texts particularly challenging. Moreover, little substantial and reliable prose data are extant for this period. In order to get a realistic picture of the distribution of the phenomenon at stake in EOHG, it is therefore necessary to evaluate the data by means of diagnostic tests for verb movement and only consider those cases in which the verb is unambiguously located in C and the preverbal area hosts multiple (head or non-head) elements. I will show that: (1) some of the patterns that are usually assumed to be good candidates for the category of V3 in the relevant literature are never attested in the diagnostic dataset; (2) a typology consisting of five non-correlative and three correlative diagnostic V3 constructions can be identified on the basis of an approach based on replicable methods; (3) some of these patterns display historical continuity; (4) non-prose texts are inadequate for syntactic investigations.


Introduction
The syntactic analysis of (older stages of) languages attested only in written form is a notably difficult task. In many cases, the arrangements documented in the existing texts are not transparent with respect to the underlying operations leading to the surface form. Old High German, the earliest stage of the German language, is no exception. Even though in recent years a number of theoretically-informed studies have emerged that provide deeper insights into the structure of this historiolect from different perspectives (cf., among many others, Tomaselli 1995;Axel 2007;Petrova 2009;Hinterhölzl and Petrova 2010;Schlachter 2012;Walkden 2014), some aspects of its syntax are still partly elusive. One of the most interesting puzzles concerning the structure of Old High German is the structure of its left periphery, especially in matrix clauses. While it seems to be generally accepted that in this language stage, some of the relevant features characterizing Present-Day German are already to a certain extent part of the system-e.g., Verb-Second (V2), the presence of a prefield, etc.-the problem of establishing whether and, if so, under what conditions it is possible to have 'exceptions to the rule' such as so-called Verb-Third (V3) or Verb Late(r) arrangements, i.e. deviations from V2, has still not been entirely solved in the literature. The present paper is an attempt to contribute to the growing discussion on this issue in Old High German main structures focusing on the oldest period of this stage, Early Old High German (EOHG, eighth to ninth century). To address the question of what left-peripheral sequences attested in the EOHG texts can be taken to belong to the repertoire of native syntactic constructions of this period, a number of methodological issues need to be taken into account and thoroughly examined.
The paper is structured as follows: in Sect. 2, the relevant premises concerning the syntax of EOHG are discussed. In Sect. 3, complex left-peripheral patterns diverging from prototypical V2, which will be labeled as Verb-Third (V3) here, are discussed with respect to the data attested for EOHG. In Sect. 4, I present a corpus study of EOHG prose texts from which such constructions have been extracted and categorized according to specific theoretical assumptions. In particular, it will be contended that in light of the structural complications deriving from the issues described so far, the only way to establish which of the V3 configurations witnessed in the texts are to be ascribed to the core grammar of (E)OHG is adopting a 'zerotolerance' approach by considering those patterns that: (1) meet the diagnostics for verb movement; (2) are more or less systematically attested independently of the Latin source. In Sect. 5, a comprehensive typology of 'diagnostic' V3 in EOHG is proposed and discussed on the basis of the analysis carried out of the corpus data. Throughout the paper, I follow Rizzi (1997) in assuming that the CP domain is split into a number of different projections and adopt the corresponding terminology. Section 6 shows that not only should the data be analyzed by applying a strictscrutiny approach relying on diagnostic criteria to determine whether V-to-C movement has occurred in these clauses; a further fundamental aspect to take into account is that only prose texts are suitable for the investigation of purely wordorder-related issues. It will be shown that texts exhibiting a metrical structure of any type are characterized by a 'mixed' syntax that includes the structures discussed in Sects. 3-5 and a (virtually unpredictable) number of additional 'V3' patterns that are evidently not part of the native syntax of EOHG. For this reason, such texts should not be used in investigations of clausal syntax. Section 7 summarizes the methodological considerations and the findings discussed in the paper.
2 Clause structure in EOHG EOHG is the oldest attested stage of the German language and encompasses the dialects spoken between the eighth and the ninth centuries south of the Benrath line (Bavarian and Alemannic in the Upper German group and East Franconian, Middle Franconian and (South-) Rhine Franconian in the Central German group), which all underwent the High German consonant shift (German Zweite Lautverschiebung), although to slightly different degrees.
The syntax of this variety allows for a certain amount of verb-order variation in matrix clauses due to the fact that the grammaticalization (in the sense of integration into the speakers' I-language) of the formal differentiation between main and embedded clauses in terms of asymmetric Verb-Second (henceforth: V2) 1 is still not complete and becomes systemic, i.e. obligatory, only from the 12 th century onwards (cf., among many others, Hilkert 2007, 69). Most notably, EOHG displays a conspicuous record of main clauses in which the verb has not left the lower portion of the structure (cf. (1)): (1) enti ubil man fona ubilemo horte ubil fram bringit and evil man from evil.DAT.SG treasure.DAT.SG evil V.PRT bring.IND.PRES.3SG 'and an evil man brings evil things out of the evil stored up in him' et malus homo de malo thesauro profert mala (MF. 9,VI,(19)(20) In this matrix clause, whose syntax differs from that of the Latin (henceforth: Lat.) Vorlage in that the finite verb follows the direct object, the inflected part of the verb (bringit 'brings') is linearized after the verb particle fram ('from'), with which it forms an independent lexical entry (frambringan 'bring (forth)'). This suggests that the verb does not sit in C°, since in configurations in which the finite verb has moved to the left, it obligatorily precedes the particle. Indeed, verb particles are immovable (that is, neither subject to pied-piping nor extraposable) in EOHG, just like in Present-Day German (henceforth: PDG).
If we are to assume-in a non-Kaynian approach-that the TP in EOHG is headfinal like in PDG, 3 there are two possible ways to account for the arrangement in (1): (i) the bare verb (without the particle fram) moves from its base-generation position in V°into the TP-head to the right of V°to acquire the relevant inflectional features with which it surfaces at PF (2a); or (ii) the verb remains in situ and receives its inflection through leftward affix hopping from T to V (2b). This second option corresponds, mutatis mutandis, to what is generally stipulated for English finite clauses, in which the verb never leaves the extended verb projection, but rather "I lowers onto V" (Haegeman 2006, 169): (2) a. b.
Even though verb movement still does not play an unambiguously crucial role in distinguishing main and embedded clauses, it has been established in a number of studies that V-to-C movement appears to be the dominant pattern both in declarative (3) and interrogative (4) clauses in this phase (cf., inter alia, Behaghel 1932;Lippert 1974;Admoni 1990;Dittmer 1992;Dittmer and Dittmer 1998;Greule 2000;Hinterhölzl and Petrova 2005;Petrova 2006;Axel(-Tober) 2007Petrova and Solf 2009;Schlachter 2012;Szczepaniak 2013). In the attested V2 declarative main clauses of EOHG, virtually any element can be fronted to a dedicated specifier in the left periphery, e.g. a full subject DP (3a), a frame-setting adverb (3b) or a past participle (3c). In all three examples below, the linear order of the clause differs from that of the Lat. Vorlage, which indicates that the translator probably felt that the original text could be best rendered into German by adding a missing verb (3a), displacing the subject into the middle field (3b) or paraphrasing the predicate in the Lat. text (3c): (3) a. Dhiin sedhal, got, ist fona euuin in euuin your throne God be.IND.PRES.3SG from eternity in eternity 'Your throne, God, is unto the ages of ages' Sedis tua deus in seculum seculi (Is. 5, 14-15) b. duo santta iħs Zuene iun girono then send.IND.PRET.3SG Jesus two.ACC.PL apostle.GEN.PL 'Then, Jesus sent two apostles' tunc iesus misit duos discipulos (MF. 23,XIV,29) c. Araugit ist in dhes aldin uuizssodes reveal.PST.PART be.IND.PRES.3SG in the.GEN.SG old.GEN.SG law.GEN.SG boohhum dhazs … book.DAT.PL that 'In the books of the Old Testament it is written that …' Pateat ueteris testamenti apicibus … (Is. 14, 7-8) In interrogative clauses, V-to-C is the dominant pattern both in wh-questions (4a), in which the finite verb in C°follows the clause-initial pronoun in [Spec,CP], and in yes/no questions (4b), which exhibit a V1 arrangement. Also in this case, the linear order of the examples substantially differs from that of the Lat. text, which suggests that verb movement to the C-head is already indicative of syntactic autonomy in EOHG, irrespective of whether the first clause position in occupied by an overt constituent like in the examples in (3) or in (4a) or by a silent operator like in (4b): (4) a. Huuer ist dhanne dhese who.NOM.SG be.IND.PRES.3SG however this.NOM.SG man … ? man.NOM.SG 'But who is that man …?' Quis iste uir? (Is. 25, 12-13) b. Muoz man in uirra ta gum be-allowed.IND.PRES.3SG one in Sabbath-day.DAT.PL heilan heal.INF 'Is it allowed to heal on Sabbath days?' Si licet sabbatis curare? (MF. 5,IV,21) Note that this does not imply that the EOHG system allows for any word order in independent clauses (pace Müller and Frings 1959). In general, verb movement occurs in main and not in embedded clauses in EOHG, although it is still not obligatory in the former. 4 The target of this movement can be identified as the CP head (5), and there seems to be no evidence for any intermediate (say, middle-field) position in which the finite verb may surface in verb-movement configurations (Axel 2007, 98;Walkden 2014, 82;cf. Schlachter 2012, 137 for a different analysis). 5 Thus, EOHG can be seen as a phase of transition in which V-to-C is already present in the system to signal syntactic autonomy, 6 but this movement need not take place overtly: To complicate the issue, a number of extraposition phenomena seem to be licit in EOHG that often exacerbate the analysis of main clauses, thereby making it difficult to identify the actual position of the verb (and, as a consequence, of the other elements) at PF. At this stage, virtually all 'heavy' (i.e., non-pronominal and nonfocused) constituents, including arguments, may surface after the right sentence bracket. In (6a), the subject gotes engila ('God's angels') must have been shifted to the postfield, since it occurs after a verb particle, zuo. The fact that the finite verb surfaces to the left of the particle can in turn be taken as unambiguous evidence that it has undergone movement to C°, since: (i) it appears in V1 position in front of an originally adverbial element, thô, which must occupy the middle field, since it is not extraposable; (ii) when thô surfaces postverbally in a main clause exhibiting verb movement, it occupies a very high position in the TP (in the great majority of cases to the immediate right of the verb), suggesting that the verb must have moved into C and not into a middle-field head in this clause (for an exhaustive view of this element, cf. Fuß 2008Fuß , 2018Donhauser and Petrova 2009;Light 2015); (iii) as will become apparent below, verb movement to a medial head is ruled out for independent reasons; (iv) in the corresponding Lat. structure, the subject angeli is linearized to the left of the verb, which further supports the idea of a native status of 5 An anonymous reviewer legitimately points out that this point is a controversial one that cannot simply be dismissed without a thorough discussion, since the possibility for (modern) V2 languages to have a sentence-medial projection for the verb to move in has already been proposed in the literature (e.g., in Travis 1984;Zwart 1997;Haegeman 2001Haegeman , 2002. I agree with that. However, any proposal in support of the hypothesis of a TP-internal landing site for the finite verb in EOHG would be, to the best of my knowledge, highly speculative for the time being. I will therefore stick to Axel's (2007) proposal in this paper and leave the question open as to whether (E)OHG (as well as later stages of the language including PDG) also has a lower (i.e., a TP) V2 position. 6 As Axel (2007, Ch. 2.4) extensively argues on the basis of a large-scale corpus study of (E)OHG prose, genuine (viz. structural, in the sense of 'not only linear') verb movement in this language stage is generally found in independent declarative, imperative, and interrogative clauses, i.e. in non-embedded constructions. This, in turn, is compatible with Lohnstein's (2000) criteria-developed for PDG-that derive clause types compositionally attributing to V-to-C movement a bipartition of possible worlds limited to syntactically autonomous portions of syntactic structures. At least in a general sense, this seems to be the case in EOHG, as well as in PDG (for a more differentiated approach, cf. Struckmeier and Kaiser 2019). this construct. The corresponding operation can represented as in (6b) (also cf. Axel 2009, 32) (7), in which the PP adjuncts appear in clause-final position against the Lat. Vorlage, it must be assumed that these constituents have been moved into the postfield, since both the gerund sprehhendi in (7a), which forms an analytic progressive construction in combination with the copula be in V2 position, and the past participle gascriban in (7b), the non-finite component of a static passive, evidently occupy the right sentence bracket (at least in Höhle's 1986 topological model). Extraposition is still possible in PDG, but it appears to be strongly constrained as compared to (E)OHG (see, e.g., Haider 1996  Although the underlying syntactic operations leading to the surface word order of EOHG are often not transparent, in some cases V-to-C movement can be disambiguated by means of diagnostic tests (mainly developed by Axel 2007) based on the non-extraposability of some categories. In particular, elements like unstressed pronouns (8a), clausal adverbs (8b) and, as mentioned above, verb particles (8c) cannot be right-dislocated. This entails that whenever the finite verb appears to the left of one of such categories in a root clause, it must be assumed to have moved to C°. Note that there is no evidence for a sentence-medial landing site 7 Subjects are very frequently dislocated into the postfield if the lexical verb is unaccusative or displays passive voice. As observed, for example, by Axel (2007Axel ( , 133-134, 2009, this could be indicative of the fact that they behave like underlying objects, as would be expectable in such cases. Although this hypothesis is compelling, under the premise adopted here that EOHG has a basic OV word order, the constituent must have undergone movement to the position in which it appears, since its first-Merge site is VP-internal. for the finite verb, i.e. for a functional head different from V°or C°in which the verb may surface. This means that the post-finite 8 linearization of the abovementioned categories is particularly valuable for determining the exact position of the verb even in seemingly ambiguous clause structures. In (8a), the personal pronoun imo is unstressed, since its referent is given and unfocused. As such, this element cannot have been exbraciated into the postfield over the VP. This pronoun is positioned in the middle field, and the subject (truhtin '(the) Lord') has arguably not moved from its VP-internal base-generation position. Note that in the original text, the corresponding sentence has a verb-final word order. The adverb auar ('however') in (8b) is immovable and cannot be assumed to have moved to the right of the finite verb together with (or driven by the same syntactic operation as) the PP za gote. Also in this case, the structure can be taken to be representative of the translator's native competence, since a corresponding clause in Lat. is missing. In (8c), the occurrence of the particle fora (with which the base verb faran forms the lexical unit forafaran 'precede, go over') in clause-final position testifies the movement of the lexical part of the verb to C°. Here, the original sentence contains a simple lexical verb, ire ('go')-in Lat., particle verbs in the Germanic sense do not exist-and the subject's going before the internal argument is expressed by means of a prepositional relation to te ('you'), which displays the accusative case selected by ante ('before'): A further aspect relevant for determining the exact position of the finite verb in EOHG concerns the structure of the two sentence brackets. In fact, eighth-and ninth-century High German already displays the basic topological structure generally assumed for PDG (cf. Höhle 1986), consisting of a left sentence bracket, corresponding to C°, in which the finite verb surfaces in a V2 clause, and a right sentence bracket, which has its counterpart in V°or T°, depending on what kind of (movement or non-movement) derivation is taken to underlie verb(-particle)-final word orders. Cases in which a synthetic, e.g. passive, verb form in the Lat. clause is rendered in German by an analytic structure filling the left and the right bracket (cf. (9a)) or in which a corresponding clause in the original text is missing (cf. (9b)) are particularly indicative of this. Such arrangements can be argued to instantiate a native construction in EOHG:  Axel (2007) proposes that EOHG is a partial pro-drop language in which (referential) null subjects are licensed by V-to-C movement in matrix clauses (also cf. Weiß 2010, 2011;Volodina and Weiß 2016). 9 This means that, in principle, whenever a root clause displays a null pronominal subject, its finite verb can be claimed to have moved to the left periphery, even when this operation is not visible in the surface syntax. In (10), the subject is a referential pro licensed by the movement of the inflected verb chilaubemes into C°. In this case, the Lat. clause also lacks an overt subject. Note, however, that Lat. is a generalized-pro-drop language in which a root-clause pronominal subject is spelled out only under specific conditions (focus or contrastive interpretation, disambiguation, etc.). For this reason, in many cases subjectless (matrix) clauses in Lat. are translated into German by adding a personal pronoun. 10 Indeed, the EOHG sentence and the Lat. 9 Axel's and Axel and Weiß's proposal of a basically syntactic trigger for null subjects in EOHG resulting from licensing of pro in the middle field via c-command relation to a higher FP responsible for agreement is groundbreaking in that it disposes of the classical idea that this phenomenon is allowed by discourse-semantic factors, notably by the presence of an identifiable antecedent in the text (see, e.g., Eggenberger 1961;Schrodt 2004). Although in other recent studies (e.g., Schlachter 2012, 161-163;Walkden 2013;Walkden and Rusten 2017) the universalizing force of this approach has been relativized, I find that the theoretical core of Axel's (2007) proposal accounts for at least the great majority of the root-clause structures exhibiting a referential null subject. For this reason, I will proceed from this assumption in the present paper. 10 Cf. the following example from Isidor, in which the German and the Lat. clause show a very similar linearization of the constituents, but the German translator added the first-person pronoun ih ('I'): (i) mit imu uuas ih dhanne al dhiz frummendi with him.DAT.SG be.IND.PRET.1SG I then all this.ACC.SG create.GER 'I was creating all of this with him' cum eo eram cuncta componens (Is. 2,(7)(8)(9) text differ in that the former displays the typical syntax of a V2 structure (topicalized adjunct, finite verb in second position, pro licensed in the middle field, direct object in some lower position to the right of the verb), whereas the latter looks like a verb-final construction in which the object filium occupies a pre-finite position.
(10) in dhemu uuorde chilaubemes pro sunu in the.DAT.SG word.DAT.SG believe.IND.PRES.1PL son.ACC.SG 'with this word, we believe in the son' in uerbo filium credimus (Is. 16, 6-7) It may therefore be claimed that the above-mentioned generalizations formulated by Axel (2007) can be used as diagnostic tests to discern main clauses with verb movement to C (i.e., V2 clauses). 11 As the data show, however, such clauses do not systematically correspond to the 'descriptive' notion of V2 generally assumed for some modern Germanic languages like PDG and Dutch.

Veffects in EOHG?
As mentioned above, in the majority of EOHG (declarative) V-to-C clauses, the finite verb occupies the second position predicted by the descriptive notion of V2, that is, it follows one XP that occupies the specifier position of the CP. In spite of this, some cases are attested in EOHG in which more than one element precedes the verb in the left periphery, thereby giving rise to so-called 'V3 effects'. Note that in this paper, 'V3' is used as an umbrella term for all configurations implying verb movement and a multiply-filled prefield, irrespective of whether there are two or more elements occupying the preverbal area and whether these elements are specifiers or heads. Moreover, the entire preverbal area in V-to-C-movement main clauses will be referred to as the prefield; thus, Höhle's (1986) notion of pre-prefield generally adopted for PDG will be dispensed with. In Sect. 5.2, however, the results will also be discussed with respect to the question as to whether the single prefield categories attested are to be considered clause-internal or clause-external in terms of their structural position below or above ForceP.

Linear non-V2 word orders
In (11a), a frame-setting PP appears in first clause position, followed by a focused adverb (approximately meaning 'what follows' and cataphorically referring to the following text passage) to the immediate left of the verb. So arguably carries a newinformation focus stress in this sentence. In (11b), the left periphery contains a pronominal subject and an originally deictic-adverbial element. The CP area in (11c) is occupied by a full-DP direct object preceding an adversative adverb/particle (for the analysis of constructions like (11b) and (11c), cf. Sect. 5.2.1.3 below). Two frame-setters appear in the prefield of the root clause in (11d). These cannot be assumed to form a "conceptual" unit merged at some point of the derivation and spelled out directly into the left periphery or raised there from the middle field because they instantiate two different types of reference (local and modal); that is, they convey two independent pieces of information:  Is. 19,[14][15][16][17] This fact is, per se, not at all new, since it is mentioned in a number of recent works on the syntax of EOHG (cf. particularly Axel(-Tober) 2007Schlachter 2012;Walkden 2014Walkden , 2017. In this paper, however, the data illustrating this phenomenon will be scrutinized and categorized from a different perspective, namely by adopting a diagnostic approach in the attempt to provide a typology of V3 constructions which can be assumed to be part of the native grammar of EOHG beyond any doubt.

Multiply-filled prefield or verb-final construction?
In consideration of the seemingly 'unstable' surface syntax of EOHG and with a focus on the structures illustrated in (11), the following questions will be addressed in what follows: (i) Which diagnostic V3 word orders are attested in EOHG? (ii) Given that most prose texts of OHG are translations, how can it be ascertained whether the single attested V3 clauses are to be ascribed to the translator's native grammar or results from a word-by-word rendering of the Lat. text?
The findings of this study clearly will bear on the analysis of the syntax of V2 and may be particularly relevant for approaches that try to identify the functional structure of the left periphery. Where relevant, I will make some suggestions about the implications and speculate on some of the accounts that might be available. A full analysis in cartographic terms would first require the formulation of the precise assumptions of the cartography of the left periphery. Such an endeavor would be outside the scope of the present paper and will await further work. As will become apparent in the next sections, in the present paper the term 'diagnostic' is used in relation to V3 to identify clause constructions that display a non-ambiguous multiply-filled left periphery, i.e. in which the finite verb can be shown to have moved to C°and which do not replicate the syntactic arrangement of a Lat. Vorlage.
The issues addressed in (i) and (ii) above are of particular interest because a great number of root clauses in which two or more elements appear before the finite verb in the EOHG texts are ambiguous between a syntactic interpretation in which the verb has moved to C°and the prefield is multiply filled and one in which the verb has remained in situ and the constituents occurring to its left are therefore positioned in the middle field. The latter option applies for any matrix structure with verb-final word order in a language stage with optional verb movement due to the fact that the relevant parameters are still not fully grammaticalized.
This, of course, substantially exacerbates the analysis of the available data with respect to the very structure of the left periphery in EOHG. In what follows, the consequences of this issue are elucidated with some illustrative examples from the corpus used for the present study. The main clause in (12) is very often cited in the literature as a good candidate to be categorized as a V3 construction in which firchnussu occupies the canonical finite-verb position (cf., among many others, Tomaselli 1995;Schrodt 2004, 201;Poletto and Tomaselli 2004, 257;Schallert 2007, 39;Speyer 2010;Wiese 2012, 91;Meibauer et al. 2015, 318). This example is generally taken to illustrate a rarely attested native pattern of EOHG in which a direct object is fronted to the first clause position and a weak subject pronoun has been added by the translator in a lower left-peripheral site to the left of the verb (note that in the original Lat. text, no pronoun appears in that position), so that this configuration would imply two arguments occurring in the pre-C°area. However, there does not seem to be any compelling evidence that this clause has an underlying V3 arrangement resulting from V-to-C movement and multiple fronting. 12 In fact, none of Axel's tests for verb movement are applicable here. This arrangement could as plausibly be analyzed as a verb-final construction: (i) in which the object has been scrambled to some intermediate (i.e., middle-field) position over the VP-internal (i.e., in situ) subject; or (ii) which replicates the syntax of the Lat. sentence (whose linear word order also does not help to determine the exact position of the verb) with the sole insertion of a pronoun. Lenerz (1983, 6), indeed, mentions this sentence as an example for a verb-final main clause: (12) Erino portun ih firchnussu bronze.ACC.PL portal.ACC.PL I destroy.IND.PRES.1SG 'I will destroy bronze portals' Portas aereas conteram (Is. 7, 1-2) If we were to assume that the poorly attested configuration in (12) instantiates a native V3 construction of EOHG on the basis of the insertion of a pronoun that is not present in the Lat. Vorlage, then we should-for the sake of consistency-make the same stipulation with respect to structures like (13a). Here, the verb and the adjunct PP in the German sentence both occupy the same surface position as in Lat., but a personal pronoun (ir) occurs in the translated text between umbi dhen in first position and two adverbs, aer and chiuuisso, occurring in pre-finite position. Apparently, the former has been introduced by the translator, while the latter may be taken to translate Lat. enim, which surfaces immediately before the verb in the original text. Also in this case, there seems to be no theoretical reason to give for granted that the verb must be positioned in C°in the German clause. After all, examples like (1) above, reported in (13b) for the reader's convenience, show that (diagnostic) verb-final main clauses are still part of the EOHG system despite the growing incidence of V-to-C in syntactically independent constructions: In-situ finite verbs are attested in the surface syntax of root clauses even when the translator has modified other parts of the clause. In (14), for instance, two framesetting adverbs have been added after the (arguably clause external) particle see translating Lat. ecce, and the subject is not split between the right sentence bracket and the postfield. Considering what has been said about examples like (12) and (13), the question arises as to whether (14) should be analyzed as a structure with underlying V-to-C movement in which the prefield is particularly complex. In this case, excluding clause-initial see, it would contain two deictic adverbs (or one, if these are interpreted as a 'frame-setting unit'), followed by the direct object, a PP adjunct, the subject, and a modal adverb. If this were the case, the assumption should be made that the left periphery of EOHG has hierarchically ordered dedicated positions for all these types of constituents and these may simultaneously occupy the preverbal area. However, no decisive argument can be given in favor of this hypothesis: In cases like (15), the Lat. clause is pro-drop and linearizes the unfocused/unstressed pronominal direct object after the finite verb. It may well be argued that the verb has moved out of the VP in this sentence. 13 In the corresponding EOHG structure, instead, both an overt pronominal subject and the direct object are linearized to the left of the inflected verb. In principle, the idea that the verb occupies C°here would be as speculative as it would be with respect to examples (12)-(14). It is not implausible that these two arguments are located in the middle field and that the adjuncts following the finite verb are extraposed constituents. In (16), where the same pre-(subject[object[Vfin) and postverbal (Vfin[ adjunct) order is found, similar conditions apply. The finite verb could occupy the right sentence bracket in this clause, implying that the two arguments are positioned in the middle field and the adjunct obar erdu ('over the earth') in the postfield. A further methodological problem must be dealt with here: the German clause replicates the exact word order of the Lat. original clause (in this case, the overt realization of the pronoun ego in Lat. reflects a contrastive or focused interpretation of the subject). Given that, as mentioned above, the most relevant prose texts attested for EOHG are translations that are not as free as those emerging from the late Old High German period onwards, it cannot be excluded that the German word order might result from a literal translation of the Lat. construction. This could be due to the fact that the sense of the utterance was understandable even with an arrangement that did not reflect the syntax of native EOHG.
(16) Ih thih giberehtota obar erdu I you.ACC.SG glorify.IND.PRET.1SG upon earth.DAT.SG 'I have glorified thee over the earth' ego te clarificaui super terram (T. 290, 24) Following from the issues raised above, it must be concluded that in a considerable proportion of cases, it is not possible to ascertain whether the position of the inflected verb in a root clause results from V-to-C movement or not in EOHG -and hence, whether the elements preceding it in the overt syntax are positioned in the prefield or in the middle field. This means that any attempts at categorizing a clause as V2 or verb-final must be based on speculation in the absence of more or less explicit evidence obtainable by means of the tests addressed in the previous section. If one, instead, accepts the hypothesis that all the clauses above display V-to-C movement, one might get the impression that 'anything goes' in the left periphery of EOHG, i.e. that it may contain virtually any type of constituent (even with variable serializations, as in for example, the relative order of subject and object in (12) vs. (15)). This clearly cannot be the case.
In the next two sections, a corpus study will be presented in which the V3 configurations attested in the main prose texts of EOHG are considered with respect to their (non-)diagnostic V-to-C status.

Methods
In order to draft a typology of the V3 construct in EOHG, the three major prose texts of this period have been primarily considered, namely: the Old High German Isidor (Is., late eighth century, a Rhine-Franconian translation of Isidore of Seville's De fide catholica contra Iudaeos); the Monsee Fragments (MF., early ninth century, a Bavarian translation of a collection of Christian writings including e.g. Matthew's gospel, a sermon by Augustin of Hippo, and the homily De vocatione omnium gentium); the Old High German Tatian (T., c. 830, an East-Franconian translation of Tatian's Diatessaron).
Other texts of the same period, like the Murbach Hymns (MH., early ninth century, Alemannic) and Otfrid's gospel harmony (O., ninth century, South Rhine Franconian) have also been analyzed, but will not be included in the results to be discussed in Sect. 5. Indeed, they do not represent a good source of syntactic information: the former document basically contains (morphological) glosses that help the reader decipher the Lat. text and display a word order entirely replicating the original one; the latter has a metrical structure consisting of tail-rhyming couplets organized in four-line stanzas. Its surface syntax almost entirely depends on the end-rhyme pattern, which is why this text cannot be entirely representative of the author's unmarked native grammar sensu stricto. As we will see in Sect. 5, however, the investigation of the V3 construct in texts like the Murbach Hymns and Otfrid is useful in order to make general considerations about their (non-)comparability to the sources mentioned above for the goals of this paper. 14 For Isidor, the Monsee fragments and Tatian, the data were extracted from the Referenzkorpus Altdeutsch, 15 the largest repository of morphologically and syntactically annotated texts from the Old High German period. While the V3 data in the former two texts were collected manually by analyzing the single main clauses annotated in the Referenzkorpus, the following search string was used to extract the data from Tatian: 'X' standing for the variable 'introduced clause' versus 'non-introduced clause', 16 'Z' standing for the clause type (e.g., declarative, interrogative, imperative, etc.), and 'Y' standing for one of the parts of speech annotated in the corpus. For the present study, all possible pre-finite-verb combinations between (common) nouns, personal pronouns, interrogative pronouns, adjectives, (discourse) particles and adverbs in Tatian were tested. Second and subsequent 'genuine' declarative conjunct clauses have been excluded from the analysis, since the licensing of a referential null subject in such contexts is not necessarily due to V-to-C movement, but to the high degree of activation of the referent (see, e.g., Walkden 2014, 175-176). 17 The collected main clauses were then categorized into the following four categories: (i) Diagnostic V3 clauses deviating from the syntax of the original text. The term 'diagnostic' refers to the applicability of Axel's tests for verb movement, i.e. 'diagnostic V3 clauses' are structures implying-at least in principle -V-to-C movement and more than one preverbal element in the EOHG clause. For 'diagnostic' structures, I further differentiated between clauses in which the 16 In the Referenzkorpus Altdeutsch, declarative main clauses are annotated differently according to whether they follow a coordinating conjunction in the linear syntax (i.e., they are a non-first conjunct in a syndetic sentential coordination) or not. In particular, declarative non-first conjuncts following a paratactic connector are retrievable via a search string including the category 'CF_I_M' ('I' standing for 'introduced'), while first conjuncts or main clauses not occurring in a coordinative structure are annotated as 'CF_U_M' ('U' standing for 'unintroduced'). 17 See, e.g., coordinative complexes like (i), in which the second and the third conjunct both exhibit a null subject co-indexed with that of the first conjunct (ih 'I'). Such structures occur with a certain frequency in the EOHG prose texts consulted for the present study and cannot be assumed to necessarily instantiate finite-verb movement into C, since the second and third coordinated clause appear to replicate the syntactic arrangement of the first conjunct, in which the linear position of the verb is ambiguous between a V-to-C and an in-situ analysis.
(i) Erino portun ih firchnussu, iisnine grindila firbrihhu endi dhiu chiborgonun hort dhir ghibu (lit.) 'Bronze portals I destroy, iron locks smash and the hidden treasure you.DAT give ' (Is. 7,[1][2][3][4] translator has intervened to modify the syntax (i) and clauses that display exactly the same syntax as in the Lat. Vorlage (ii). 18 Structures in which no verb movement is diagnostically detectable were categorized as 'ambiguous', thereby distinguishingfor the sake of consistency-those with a Lat. syntax from those with a non-Lat. syntax.
Evidently, the two poles of this taxonomy, namely (i) and (iv), correspond to the most and the least conclusive categories, respectively: diagnostic clauses with a non-Lat. syntax can be assumed to be more 'informative' with respect to the native syntax of the EOHG prefield than the other classes; on the other hand, sentences in which there is no evidence for V-to-C movement and all constituents occupy the same linear positions as in the original text could be word-by-word translations that do not reflect the translator's native grammar. The clauses categorized as (ii) and (iii) cannot be taken to be reliable testimonies of the translator's native competence beyond any doubt, either. In (17)-(20), the four categories used to classify the results are exemplified. In (17), two frame-setting(-like) elements 19 occupy the prefield of the German clause, and the verb is followed by a weak (nonextraposable) personal pronoun, while the corresponding Lat. structure is pro-drop and has the verb in final position. It is quite clear that the verb is located in C°in the translated construction; as to example (18), one might, in principle, assume that a referential null subject is licensed in the middle field in the EOHG clause. However, this clause strikingly replicates the word order of the original text. There is no substantial evidence that this construction is native; in (19), we have the opposite scenario: the syntax has been 'adapted' by the German translator mainly insofar as the Lat. synthetic verb form of the clause-final passive is realized analytically, and the subject (daer baum) appears after the finite verb in EOHG, while Lat. arbor precedes cognoscitur in the original text. However, no clue is given that the position of the inflected verb uuirdit results from V-to-C movement in German. As a 18 The main criterion employed to determine the overall (non-)latinizing word order of the attestations was the external syntax of the constituents. Structures in which the linear syntax of the phrases is exactly the same as in Lat. but the internal syntax of (some of the) constituents slightly diverges from the EOHG one were categorized as replicating the syntax of the Lat. sentence. For instance, the relative order of DPinternal elements (e.g., the relative arrangement of possessive pronouns and nominal heads) is sometimes modified by the translator even when the rest of the clause displays the same syntax as in the original text. Cf. example (i), in which thine iungoron ('your disciples') differs from discipuli tui, although the external syntax of the constituents is the same in EOHG and Lat.: As expected, only a minority of all 'linear' V3 clauses display at the same time diagnostic V-to-C movement and a word order diverging from that of the Lat. text. In the next section, the idea will be put forward that, given the intricacies of the EOHG syntax presented above, these 'doubly diagnostic' data are the sole absolutely reliable source of information concerning the possible combinations of left-peripheral elements and their relative word order in EOHG. Of course, this does not at all exclude that further arrangements may have been possible in this language stage; however, it will reveal that some are certainly part of the EOHG system. This methodological approach, based on a strict selection of the data in order to provide an as accurate as possible typology of possible prefield serializations in EOHG, is reminiscent of that adopted, for example, by Danckaert (2017, 90) with respect to the internal structure of the Latin VP and represents, to this author's mind, an ideal prerequisite for the syntactic treatment of any language (stage) attested only in written form.

Frequency
Clause structures with unambiguously complex prefields are attested in all three texts considered for the present study. The preliminary results are summarized in Table 1. In the second column, the overall number of main clauses (including, inter alia, declarative, interrogative, and imperative structures) contained in each of the texts as displayed in the Referenzkorpus Altdeutsch (version 1.0) is given. 20 The third column indicates the total number of main-clause structures showing surface V3 and the corresponding percentage value in relation to the number of main clauses present in the corpus. The fourth column contains the data relative to diagnostic V3 clauses with a non-Lat. syntax; the share refers to the rate of the 'doubly diagnostic' V3-clauses in proportion to all the matrix clauses attested in the corpus. The value in the last column relates to the frequency of the diagnostic multiply-filled structures with respect to the overall number of structures displaying a surface V3 word order (i.e., the values indicated in the third column).
The percentage values in the fourth column of Table 1 show that-unsurprisingly -the (diagnostic) multiply-filled prefield is a marginal option in EOHG in relation to other word orders: its absolute frequency amounts to a maximum of 7%-8%. However, this is not particularly telling, since these values refer to all types of V3 configuration (see below for details). What is more interesting here is that the frequency of diagnostic V3 in relation to all surface V3 clauses in the corpus ranges from about one-fifth (19.32% in the Monseer Fragments) to over one-third (34.84% 20 Note that these numbers, which are based on the output of the digital corpus query, have been slightly modified with respect to the original ones displayed in the Referenzkorpus in light of occasional errors in the tagging (e.g., subordinate clauses annotated as matrix structures and other discrepancies between the categorization and the interpretation of the clause, ambiguous classification of some clause-initial connectors, etc.) and are, to a certain extent, approximate. Also consider that the occurrences in Tatian were filtered with a search string. These values, however, are useful to estimate the general ratio between the overall number of main clauses present in each of the texts and the number of structures displaying a (diagnostic) V3 arrangement.
in Tatian), i.e. the proportion of clauses with surface verb-late word order in which V-to-C movement can be double-diagnosed is quite variable (and not predictable in any obvious way). This entails that only for about 20-34% of the clause structures in which the finite verb does not appear in (linear) second position can it be ascertained that its position results from V-to-C movement. In all remaining cases (73.73% in Isidor, 80.68% in the Monseer Fragments, 65.16% in Tatian), the surface verb-late clauses are basically ambiguous; that is, they cannot be analyzed by means of Axel's tests, or they replicate the Lat. syntax, or both.

Typology
Employing strict empirical methods like the double-diagnosis adopted here may certainly exclude some (rather infrequent) native constructions of EOHG from the analysis which happen to be attested only in ambiguous sentences. I reckon however that the expected result of compiling a taxonomy of V3 constructions that can be ascertained to belong to the native competence of EOHG speakers with a high degree of certainty is worth both the difficulty of the attempt to establish them and the risk of not being completely exhaustive. For the time being, there does not seem to be any other way to reduce the data attested for EOHG prose to the essential.
As for the V3 taxonomy obtainable by means of the methods outlined in Sect. 4, a further interesting result is that, while the surface syntax of non-diagnostic verblate structures seems to allow for a very wide spectrum of possible pre-C°word orders (including virtually all possible argument combinations), the range of patterns attested in diagnostic clauses is quite limited. Two macro-categories can be identified to start with: (i) Prefields hosting non-correlative constructions (i.e., constituents that are not coreferential); (ii) Prefields hosting correlative constructions (i.e., constituents of different nature that are resumed by a co-referential pronominal or adverbial element).
In what follows, the structures belonging to each of these two categories are addressed extensively in order to have a full picture of the V3 configurations attested in EOHG. I will eschew a detailed quantitative analysis of each preverbal sequence, since some of the phenomena described below occur simultaneously. The taxonomies presented in the next sections facilitate the formalization of the phenomenon by means of information-structural categories.

Non-correlative constructions
This macro-category includes multiple combinations of non-correlative constituents (XPs) and/or particles (heads) occupying the preverbal area in V-to-C-movement matrix clauses with a syntactic arrangement diverging from that of the Lat. Vorlage.
For the non-co-referential prefield configurations attested in the data considered for the present study, five different subcategories can be identified: In the next sections, these subcategories are illustrated, compared and embedded into the more general discussion of the makeup of the left periphery in EOHG. Subclass (a) includes preposed adverbial clauses that specify the deictic-referential (temporal, conditional, situational, etc.) frame into which the content of the matrix clause is embedded and are followed by unstressed/non-contrastive pronouns (21a); contrastive-focused arguments, in many cases interrogative pronouns (21b); modal adverb(ial)s (21c); the root-clause predicate (21d), or; illocutive particles (21e) The examples in (21) are reminiscent of data already discussed in the literature, e.g., for PDG (Auer 2000), West Flemish (Haegeman and Greco 2018a, b), and Scandinavian urban vernaculars (e.g., Ganuza 2008; Walkden 2017 for Swedish). 21 This shows not only that a construction attested in EOHG exhibits diachronic continuity until the present and is similar to structures productive in other (presentday) varieties; it also demonstrates that such sequences are possible in systems in which the verb systematically moves to C°in main clauses as is the case in all Germanic varieties with the exception of Present-Day English.

Clausal frame-setter[XP[Vfin
The status and syntactic position of adverbial clauses in Early Germanic is a much debated topic in the literature. Axel (2004Axel ( , 2007 proposes that such structures are syntactically non-integrated and base-generated above the leftmost edge of the root clause in this language stage, arguing that data like (21), in which the adverbial construction appears to the immediate left of a clause displaying 'standard' V-to-C 21 See, e.g., the following examples, in which a preposed adverbial clause is immediately followed by an unstressed/unfocused personal pronoun (i)-(ii) or by a full-DP subject (iii) in the left periphery of the clause.   Ganuza 2008, 132) movement, corroborate this hypothesis. One of her arguments is that particles like eno (normalized form: inu) (cf. (21e)), which are illocutionary in nature (they are in most cases used to introduce a rhetorical question and/or to express the speaker's attitude) 22 and are therefore assumed to be first-merged in ForceP, are systematically linearized below adverbial clauses. ForceP is generally assumed to be the highest clause-internal projection in Rizzi's (1997) model. Hence, if it were the case that inu occupied [Spec, ForceP], then it would be reasonable to assume that preposed adverbial clauses are hosted in a clause-external specifier in V3 configurations. In fact, this is in line with what is often assumed for the modern Germanic V2 languages (see, e.g., Broekhuis andCorver 2016, 1711;Haegeman and Greco 2018b, 236).
However, with respect to special questions like (21e), the following should be noted: in rhetorical interrogatives containing a fronted adverbial clause introduced by ibu ('if'), the latter is not a 'regular' protasis introducing a condition on whose validity the circumstances verbalized in the apodosis depend. The truth value of the matrix ni tuont thaz heidane man is utterly independent of the ibu-clause. In other words, the function performed by the ibu-clause consists in introducing content which is referred to in the main clause, but does not define its validity. In this light, it can be argued that the fronted adverbial clause in (21e) occupies a position above ForceP similar to a Hanging Topic. However, Petrova (2017, 320-321) shows that the particle inu, while being very high in the structure, can occasionally be found above fronted adverbial clauses. In (22a), the preposed ibu-clause is a canonical conditional clause (i.e., it introduces a hypothetical situation of the type 'if it were the case that…'). Its interpretation is crucially different from that in (21e). This might imply that ibu-clauses like (21e) and run-of-the-mill conditionals like (22a) occupy different positions because they contribute to the semantic interpretation of the utterance in different ways. If the particle inu is positioned in ForceP as proposed by Axel (2007), then the adverbial clause in (22a) could be argued to surface below ForceP, i.e. clause-internally.
Moreover, there is evidence that, for example, temporal clauses may surface below subordinating conjunctions, which lexicalize the head of ForceP (Rizzi 1997) or SubP (Haegeman 2004, 77) (22b), and that conditional clauses can occur below epistemic and particle-like adverbs in clause-first position (22c). Note that the clause in (22c) also instantiates the subcategory illustrated in (21c), in which a modal adverbial (here buuzsan einigan zuuiuun 'without any doubt') is spelt out to the right of the adverbial clause in the position immediately preceding the finite verb: Therefore, it appears sensible to assume that adverbial clauses do not systematically occupy an extra-sentential position above ForceP in EOHG, but may do so in cases in which the subordinate structure is semantically unintegrated. 23 A further observation is essential at this point. If we consider sentences like (22c), in which the finite verb can be shown to have moved into C against the Lat., there does not seem to be any obvious reason to claim that three of the four elements occupying the preverbal area (the epistemic adverb chiuuisso, the adverb/particle nu, the adverbial clause and the epistemic PP) must be located clause-externally, i.e. above ForceP. Such an assumption would be highly speculative and empirically unmotivated. At the same time, it is undeniable that these four elements do not form one big constituent, regardless of how they reach their surface position or whether they are moved or base-generated. Following from these considerations-and assuming, with Poletto (2013) and Wolfe (2015aWolfe ( , b, 2016, that languages with a V2 grammar may have the verb either in the highest or in lowest left-peripheral headit suggests itself that the locus of V2 effects is not ForceP, but FinP in EOHG. It can therefore be contended on the basis of these data that EOHG-differently from what 23 Another possible way of accounting for the varying left-peripheral positions of adverbial clauses in EOHG would be to assume that these constructions may surface in (at least) two different clause-internal (i.e., CP) specifiers depending on their specific relation to the content of the matrix clause: resting on the (preliminary) assumption that a particle like inu is a head occupying some high CP position (say, Force°), it could be argued that the clause is located in [Spec,ForceP] when it precedes the particle (to be in a spechead configuration with the particle) and in some lower specifier hosting frame-setting elements when it follows the particle. Since there does not seem to be any direct evidence for an extra-sentential versus high left-peripheral positioning of adverbial clauses in these cases, this question will be left open here. has been assumed for most old Romance varieties-is (still) a Fin-V2 language. This is in line with Axel-Tober's (2018) analysis, in which the finite verb is located in a position lower than what the author calls "the precursor of the modern prefield". Subclass (b) includes non-clausal framesetting constituents, e.g. deictic-referential adverbs or full-PP adverbials, followed by other constituents of the same class-the projection hosting such elements can be taken to be recursive in light of the data attested-as exemplified in (23), where hear and nu are combined in the prefield; 24 or by weak elements like unstressed pronouns (24a) or modal adverbs (24b), or by focused/contrastive constituents (25) With respect to (clausal and non-clausal) frame-setters and irrespective of whether fronted adverbial clauses are located above or below the leftmost edge of the root structure, these data show that deictic constituents generally occupy a site that is higher than focused/contrastive elements and pronouns/low-CP modal adverbs. [Vfin Subclass (c) includes prefield sequences in which a (contrastive or aboutness) topic is followed by a topic marker and optionally by another constituent. By 'topic marker' I mean an originally adverbial element that signals the topic status of the constituent immediately preceding it. The term 'topic particle' or 'topic marker' (German Topikpartikel) has been used in the literature at least since Saebø (2003) to capture the functions of, for example, PDG aber ('however'), nämlich ('indeed'), freilich and allerdings ('certainly'), etc. when they surface in the left periphery of the clause systematically in combination with a non-familiar topic (in the sense of Frascarelli and Hinterhölzl 2007; also see Breindl 2008Breindl , 2011, as in (26a) (adapted from Breindl 2011, 2), where these items can be assumed to lexicalize the contrastive/aboutness status of the preceding topic. They may, therefore, be assumed to be base-generated in the head of a higher (i.e., nonfamiliarity) TopP, while the topicalized XP is dislocated into the specifier of the same projection via movement, as shown in (26b)  Such elements are already documented in EOHG, where they apparently perform exactly the same function as in PDG (cf. Speyer and Weiß 2018, 80-81). In the diagnostic data attested in the three texts considered for the present study, for example, the following items are attested: afur, ouh, danne, all translatable as 'however' or 'namely' depending on the context and signaling the salience of the preceding given or newly introduced topic, which is extracted and (adversatively or consecutively) emphasized from a set of alternatives (cf. (27a)); 26 a further topic 25 Zwart (2005, 28) discusses Dutch items similar to German aber in (26) in terms of 'extra-dependent' constituents to be treated as syntactic interpolations, i.e. as elements that are not part of the syntactic computation of the clause and surface in an apparently clause-internal position. Haegeman and Trotzke (2020), however, argue in a recent paper on non-temporal dan (lit. 'then') in Dutch that at least some of the items of Dutch addressed by Zwart (2005) can be taken to be integrated in the structure. Irrespective of the technical details of the analysis, I agree with Haegeman and Trotzke that some of these elements are clause-internal. 26 If the combination topic[topic marker is followed by another XP in the preverbal area, then this is quite systematically an adverb(ial) element. For instance, in the following example the adverb sus, which is arguably focused/contrastive here since it realizes the head of a comparative phrase, precedes the verb in the prefield:

Topic[(topic marker) ([XP)
(i) Christus auur sus quham fona fater ziuuaare so selp so … Christ indeed so come.IND.PRET.3SG from father.DAT.SG certainly as 'Indeed, Christ certainly came from the father like …' Christus enim ex patre ita emicuit ut … (Is. 4,(19)(20)(21)(22)(23) Given that the spell-out of the topic marker is not obligatory, this observation basically amounts to recognizing that in general, the particleless sequence topic[adverbial[Vfin is native-at least as a marginal option-in EOHG, which is confirmed by the fact that this pattern is attested independently in indicator, which is particularly often attested in Tatian, is tho (lit. 'then', 'there'). 27 This element has never been extensively discussed in terms of topic marking (although this possibility is mentioned en passant in Axel (2007, 225)). In principle, one might assume that tho retains its original deictic (temporal and/or local) meaning even when it appears in this V3 configuration. A closer look at the contexts in which it occurs in Tatian, however, reveals that in the great majority of cases in which it performs this specific function in diagnostic clause structures (30 out of 31), tho follows a shifting topic in a dialogic/interactional context implying alternating turn-taking of at least two (groups of) speakers whose reference is contextually given (cf. (27b)) (cf. Catasso et al. (to appear)). Thus, for this element the same categorization will be employed as for the above-mentioned particles: (27) ([ XP) [ Vfin In subclass (d), we find three types of very high left-peripheral elements that do not seem to interact directly with the linear syntax of the main clause in which they appear and are arguably first-merged in their surface position: (i) adverbial connectives like so in (28a), which are often absent in the Lat. Vorlage and must therefore have been particularly productive in spoken EOHG. Since these items are non-referential, as well as informationstructurally and illocutively vacuous, and systematically occur in clause-first position, they can presumably be assumed to be base-generated in some position above ForceP; (ii) the sentence-initial interjection senu in (28b) (also cf. see in (23)) generally translates Lat. ecce ('behold!', 'look!') and may virtually precede any type Footnote 26 continued the diagnostic dataset, as shown in (ii). Indeed, Speyer (2008, 456-457, his ex. (2d)) shows that this configuration is even still attested in written PDG as a marked structure: (ii) taufi · armhercin · enti gnada ubar · unsih siin simples baptism mercy and grace over us.DAT.SG be.SBJV.3PL always 'Let baptism, mercy and grace always be over us' (no Lat. counterpart) (MF. 68,XLI,(8)(9) of left-peripheral element. Along the lines of what is generally postulated with respect to PDG interjectional expressions in first position like weißt du ('you know'), which also have no impact on the syntax of the root clause, and just like adverbial connectives, the first-merge position of this element is presumably extrasentential; (iii) it has been established that the illocutive particles mentioned above in relation to category (a) are clause-internal (Axel 2007;Petrova 2017) and may occur in combination with other items of the same type. In (28c) pre-finite eno, nu and ia may all be taken to be particles (although the status of nu, lit. 'now', is admittedly ambiguous between an illocutive item and a weakly referential adverb): (28) a. So auh in andreru stedi … uuard dhera so also in other.DAT.SG passage become.IND.PRET.3SG the.GEN.SG dhrinissa bauhnunc sus araughit trinity.GEN.SG concept so reveal.PST.PART 'The concept of the Trinity was revealed … as follows also in another passage' Item alibi … trinitatis sic demonstratur significantia (Is. 19, 14-17) b. Seenu · bidiu · huuer uemes · uuir · za **** behold therefore go-back.IND.PRES.1PL we to 'Behold, therefore we will go back to ****' ecce convertimur ad gentes (MF. 49,XXXI,17) c. enonu ia sint zuelif citi thes tages inu-nu ja be.IND.PRES.3PL twelve hour.NOM.PL the.GEN.SG day.GEN.SG 'Are there not twelve hours in a day?' nonne XII hore sunt diei (T. 229, 27) Thus, items like so in (28a) and see/senu in (28b) (but not the particles in (28c)) can be assumed to occur above ForceP and to select a CP. Their merging site is arguably the head of one of the clause-external Speech-Act Projections proposed by Hill (2013, 2014) for West Flemish and Romanian (see also Cardinaletti 2015 for Italian).

Focus operator[XP[Vfin Subclass (e) is somewhat controversial in that it
presupposes the application to the EOHG data of a theoretical model originally advocated by Jacobs (1983Jacobs ( , 1986 for PDG (cf. Büring and Hartmann 2001; Kleemann-Krämer 2010 for a slightly different implementation), namely the assumption that focus particles (FPs)-i.e., additive, exclusive, and scalar operators such as auch 'also', nur 'only', sogar 'even', etc.-and their domain do not form a constituent, but the focus quantifier is rather (post-syntactically) left-adjoined to a 'complete' V2 clause with a standard left periphery in which the focused XP surfaces, as shown in (29). The reader is referred to the above-mentioned works for the technical details (and, e.g., to Bayer 1996;Reis and Rosengren 1997;Nederstigt 2003;Reis 2005 for a different analysis):

(29) [ CP FP [ CP NP F C°… ]]
Jacobs' and Büring and Hartmann's general hypotheses, despite differing in some points, both entail that all clauses with a prefield constituent modified by a focus marker instantiate a genuine V3 construction. If we explore this possibility for the data attested in EOHG and take sentences like (30), in which the particle ioh ('also') binds an F-marked subject demonstrative, to be representative of the category of multiply-filled-prefield structures, we find that this scenario basically resembles that of PDG: Of course, focus particles do not only bind 'simple' DPs, but can also modify other types of constituents, such as PPs (like in example (28a) above, in which the local adverbial in andreru stedi, the first XP occurring prefield-internally if we assume the connective adverb so to be merged above ForceP, is modified by exactly the same particle).
In this section, V3 configurations of EOHG were considered in which the elements occupying the pre-finite left periphery are not co-indexed (in fact, some of these, e.g. particle-like items, are not referential at all). In what follows, we look at multiply-filled-prefield patterns that consist of referential XPs referring to the same concrete or abstract entity.

Correlative constructions
This macro-category includes correlative prefield combinations that are still attested, mutatis mutandis, in PDG. In particular, three subclasses can be distinguished, namely: These three classes are illustrated and discussed in the next three sub-sections.

Clausal frame-setter[resumptive[Vfin Subclass (a) comprises diagnostic V3
constructions in which a preposed adverbial clause with frame-setting function is resumed by a correlative pronoun or adverb located in a lower preverbal sitearguably a [Spec,FinP]-like position in modern terms, since the resumptive element appears to systematically surface to the immediate left of the finite verb. In EOHG, fronted adverbial clauses very often occur in correlative patterns of the type illustrated in (31) (also see Axel 2004, 29). In (31a), the consecutive adverb thane ('then') resumes a fronted conditional clause; in (31b), tho ('there, then') and the temporal clause introduced by mit diu ('when') are co-referential: Obviously, the theoretical premises addressed in Sect. 5.2.1.1 with respect to the (non-)integrated status of adverbial clauses in this period also apply for the same construction in correlative patterns. It is usually assumed that such clauses are extrasentential in (E)OHG and gradually come to be integrated into the root structure throughout the history of German, so that PDG adverbial clauses are predominantly not followed by any resumptive and generally occur in 'regular' V2 configurations. However, I argue that both integrated and less-integrated/non-integrated adverbial clauses are already part of the language system in EOHG, and that correlative constructions involving a preposed adverbial clause resumed by a pronominal/ adverbial element in pre-finite position are not to be viewed as syntactically unintegrated only in the light of their form. Differently from (standard) PDG, in EOHG constructions involving a fronted adverbial clause show a strong tendency to lexicalize the movement of the clause from its middle-field base-generation site to the prefield in terms of resumptive-pronoun/adverb trace spell-out in a [Spec,FinP]like specifier, along the lines of what has been proposed by Grewendorf (2002a, b) for left-dislocated constituents in PDG. As far as the structural position of clauseinternal adverbial clauses (vs. extra-sentential structures like (21e)) is concerned, it can be assumed that they sit in a relatively high specifier of a projection that we may label 'FrameP', adopting the modern terminology. This specifier must in any case be higher than the one hosting focused constituents (cf. (23b)) and lower than the head in which the complementizer is spelt out (cf. (22b)): In EOHG, just as in PDG, the trace is not obligatorily spelt out in the position immediately preceding the finite verb (i.e. as in (31a)-(31b)); it can also be realized in the lower TP/VP first-merge site of the adverbial clause, like in (33), in which case it cannot be spelt out in other positions. Relying on the diagnostic data attested, weak elements like resumptives (31a)-(31b), argument pronouns (21a) and modal adverbs (21c) occupying a left-peripheral position in a V3 root structure with a fronted adverbial clause possibly compete for the same (or an immediately adjacent) position, since whenever one of these three items is present, the others are excluded (or at least are not attested): 29  [Vfin Subclass (b) includes non-clausal (i.e., adverbial or PP) frame-setters that do not seem to distinguish themselves significantly from clause-internal adverbial clauses in that they are also often resumed by correlative elements that appear in pre-finite position. After all, clausal and nonclausal frame-setting topics do exactly the same job from an information-structural point of view, albeit they have different syntactic labels: they "limit the applicability of the main predication to a certain restricted domain" (Chafe 1976, 50). In (34a), the (here temporally interpreted) adverb thanan ('afterwards', normalized form: danan) is resumed by tho in a lower position. Note that in EOHG, tho does not only serve as a resumptive element, but may itself be a deictic element when it functions as a temporal anaphor referring to an interval of time given in the previous discourse (see, e.g., Donhauser and Petrova 2009). 30 Indeed, diagnostic data are attested in which tho and similar frame-setters are followed, for example, by modal adverbs in the prefield, in turn implying that the full adverb tho is positioned in a higher specifier than the one hosting modal adverbs. As pointed out above, non-contrastive/focused modal adverbs (i) uuanan uns sint in uuostinnu so manigu brot whence us.DAT.PL be.IND.PRES.3PL in desert.DAT.SG so many.NOM.PL bread.NOM.PL 'How may we get enough bread in the desert?' unde ergo nobis in deserto panes tantos? (T. 140, 23-25) sit in a very low projection (presumably [Spec,FinP]) which is either the same as resumptives or an immediately adjacent one. It is compelling to assume that full deictic adverbs like thanan in (34a) and tho in (34b) are located in the very same projection in which (clause-internal) adverbial clauses are hosted, namely FrameP. Just like in the case of fronted adverbial clauses in correlative constructions, the sequence nonclausal frame-setter[modal adverb[resumptive is not attested in the diagnostic dataset. Resumptive elements like tho in (34a) can be argued to lexicalize the movement of the non-clausal adverbial into [Spec,FrameP] via the corresponding specifier (labeled as ' [Spec,FinP]' for the sake of convenience) (35): To sum up: for configurations in which a clausal or non-clausal frame setter is fronted and resumed by a preverbal adverbial element, I assume a movement-based analysis reminiscent of that proposed by Grewendorf (2002a, b) for Left Dislocation in PDG. Of course, this is not the only analysis put forth in the literature for similar constructions in modern Germanic. For instance, scholars like Hoekstra (1999), Zwart (2005), Broekhuis and Corver (2016) propose for Dutch that the resumptive is the element that satisfies V2 and that the initial constituent is main-clause external. In this analysis, the preposed constituent has a status which is similar to that of a Hanging Topic. This is in fact one plausible option to account for the word order in these sentences, and it would not be sensible to adopt one of the two alternatives excluding a priori the other one. However, I conceive such structures as resulting from the operations described above because the construction with the resumptive and the construction without the resumptive have exactly the same meaning. The same goes for the status of the fronted frame-setter: it does not modify the interpretation of the clause any differently whether it is taken up by an adverbial resumptive or not. This seems to suggest that it is all about two possible realizations of one and the same sentence. The application of an analysis à la Zwart (2005) to these constructs, instead, would rather suggest that the resumptive and the resumptiveless versions of a sentence like PDG Wenn zwei und zwei vier ergeben, (dann) ergeben drei und drei sechs result from two different derivations. It is assumed here that a uniform analysis for these two syntactic options (one in which the trace spell-out is pronounced and one in which it is 'silent') represents a more elegant solution.

'Hanging
Topic' Subclass (c), labeled as 'Hanging Topic' here for the sake of convenience, subsumes all structures in which a topic surfacing in the left periphery is resumed by an overt or covert pronominal resumptive linearized either in the lower portion of the prefield or in the middle field. Axel (2007, 204) correctly points out that there is no conclusive evidence that German Left Dislocation (cf. Frey 2004) has already evolved in EOHG, i.e. that there exist no suitable criteria for differentiating Left Dislocation and Hanging Topics as we know them today (for an overview of the structural differences, see Shaer and Frey 2004). At the same time, some of the constructs attested very much look like the PDG Hanging Topic, at least in structural terms. Thus, the category at stake here includes, in principle, clause structures like those illustrated in (36) (cf. Axel 2007, 204-209). In (36a), the pronoun ther resumes the higher DP. This construction is reminiscent of PDG Left Dislocation, since the fronted constituent and the pre-finite correlative (d-)pronoun display the same φ-features. Just like in PDG, however, this serialization is ambiguous between a Left-Dislocation and a Hanging-Topic interpretation, given that the nominal expression in clause-first position exhibits nominative morphology. Moreover, no prosodic clues are available to discern the two constructions. In (36b), a left-peripheral non-d-pronoun resumes a DP surfacing in a higher position. Note that the weak pronoun is linearized to the right of the interrogative pronoun huuemu, exemplifying the rarely attested (but native) pattern in which a focused/contrastive element and a weak pronoun both appear in preverbal position (see fn. 29, where the interrogative pronoun is an adjunct; for a more detailed analysis of (36b), see 5.3). For cases like (36c), Axel (2007, 206-207) proposes two possible analyses: one in which the position of the DP results from XP movement from the middle field (in which got leaves a trace t i ) into the left periphery, and one in which the topic linearized in the CP is resumed by a null resumptive (pro i ) in the middle field. I will remain agnostic as to the status of such sentences. Note, however, that the position of the topic in the latter structure must be clause-internal, since it appears in a lower position than the illocutive particle inu. For such patterns, Axel-Tober (2018, 43) proposes that the topic is hosted in a relatively high position between the projection in which particles like inu are generated and the projection in which, e.g., focused/contrastive material is accommodated (in the author's words, the precursor of the prefield in modern terms). As (36d) shows, there are also constructions in which the constituent in clause-first position occurs to the left of such particles, which possibly characterizes it as a genuine Hanging Topic. In this case, the topic must be in a clause-external projection above ForceP (say, a Disc(ourse)P in Benincà's (2001) and Benincà and Poletto's (2004) spirit): 31 31 Of course, the assumption made above with respect to (36d) presupposes that thú is not part of the vocative phrase that hosts the following constituent capharnaum, but rather an independent Hanging Topic occurring in an even higher projection. This hypothesis cannot be tested. However, if it were the case that thú is a genuine Hanging Topic here, this construction would show that in EOHG, just like in PDG, Hanging Topics may appear to the left of vocatives. The problem of which serialization between Hanging Topic[Vocative and Vocative[Hanging Topic-which is also possible in PDG-is the underlying one (i.e., which of the two arrangements results from movement of an XP to a higher specifier) is left to future research. On the basis of these data, it can be concluded that EOHG has at least one clauseinternal position for Hanging-Topic-like constructions such as (36c)-under the assumption that the DP is resumed by a null pro in the middle field-and one clause-external position for Hanging Topics that are more similar to the same pattern in PDG such as (36d). 32 32 It goes without saying that a good number of clauses containing Hanging Topics (sensu lato) are ambiguous like (36a) and (36c). This is not different from what was observed for adverbial clauses. However, irrespective of what one assumes with respect to the (possible) structural positions of adverbial clauses in the left periphery, as well as of whether, for example, topics of the type in (38c) are resumed by a null pro in the middle field and/or are clause-internal or clause-external, it must be said that there is independent evidence that topics of this type precede adverbial clauses in the visible syntax, as in (i). Note that in this example, the adverbial clause is resumed by a correlative element in the lower part of the prefield (dhuo) against the Lat. Vorlage. It seems that whatever the position of the topic might be (DiscP above ForceP or some TopP in the higher portion of the left periphery), the temporal clause is positioned CP-internally:  (Is. 29,(17)(18)(19)(20)(21) The results of the study presented in this section only concern Verb-Late(r) data that are doubly diagnostic (in the sense elucidated in Sect. 3) and may be taken to reflect the translator's native grammar. Of course, the structures discussed above also occur in the data that are not analyzable by means of syntactic probes; in the non-diagnostic dataset, however, a lot of other serializations are attested which-in the absence of additional evidence-can only be categorized as 'ambiguous' (for instance, the patterns discussed above in examples (12)-(16) in Sect. 3.2).

The argument-plus-argument pattern is non-diagnostic
The results relative to the resumptive and non-resumptive constructions addressed in the previous section are exhaustive (i.e., include all possible serializations attested in the corpus) with respect to data whose validity and representativity in unquestionable. Indeed, the material considered for the present study, besides displaying diagnostic verb movement, consists of matrix clauses that show a syntax diverging from that of the Lat. Vorlage, and there is no reason to think that the translator would have modified the syntactic ordering of the original text to generate ungrammatical structures.
One pattern which is not attested in the diagnostic dataset-but relatively often occurs in the texts-is the combination illustrated in (37). No doubly diagnostic clause in the corpus displays two (non-correlative) arguments (irrespective of case/ syntactic function and specific serialization, e.g., Subject[direct Object, or direct Object[Subject) in its prefield: (37) full-DP argument[full-DP argument ([[…])[Vfin In most clauses that exhibit a word order like (37), no verb movement can be diagnosed, or the EOHG text replicates the syntax of the Lat. text, or both. For instance, a sentence like (38) could in principle be a native construction that has not been modified because it happens to be grammatical although it displays the very same arrangement as in Lat., but there is no evidence for V-to-C movement. This means that, irrespective of whether this example reflects a native or a non-native construct, it could be the case that the finite verb ázun ('ate') is in situ: The non-availability of serializations of this type in the diagnostic dataset is interesting in particular because it suggests a possible division between southern and northern West-Germanic varieties. Walkden (2014, 69) shows that the fronting of two (pronominal and/or full-DP arguments) is possible in Old English (ca. midseventh century-1100). He also discusses examples like (39) for Old Saxon (2014, 97), in which two personal pronouns, the subject and the direct object respectively, precede the finite verb. Assuming for this variety the same diagnostics for verb movement adopted for the High German, it could be argued that in this sentence, scal cannot be in situ, since the sentential adverb diurlîco is not extraposable: (39) Ic eu an uuatara scal gidôpean diurlîco I you.ACC.PL in water.DAT.SG shall.IND.PRES.1SG baptize.INF tenderly 'I shall baptize you tenderly in water' (Heliand 882-883) The data relative to Old Saxon could admittedly be problematic, since the largest text available for this variety, Heliand (first half of the ninth century) is an epic poem that uses the alliterative verse and therefore not ideally suitable for syntactic investigations.
However, Petrova (2012) shows that the very same pattern is attested in Middle Low German prose (ca. 1250-1600), i.e. in the language stage naturally evolving from Old Saxon. This makes it most plausible that the preceding language stage could have had this option. In the example in (40) (Petrova 2012, 175), we observe the opposite pattern to that in (39): here, the direct object immediately precedes the subject in preverbal position. The finite verb is immediately followed by a non-extraposable light pronoun and forms a cluster with the infinite part of the verb in clause-final position. This means that wille ('want') must necessarily be positioned in C°: Whether the patterns illustrated in (39) and (40) were really part of the speakers' grammatical competence or they represent a marginal, stylistically marked construct is not easy to determine. In any case, none of these two options can be discarded.
In Middle High German (ca. 1050-1350), the direct successor of Old High German, no V3 data are attested in which two arguments surface in the prefield (cf. Catasso 2019). I assume the non-availability of this pattern in the diagnostic dataset to be indicative of its absence in the syntactic repertoire if EOHG. Nonetheless, if one does not wish to exclude the possibility that prefield-internal argument-plusargument patterns only 'happen' to appear exclusively in ambiguous clauses (i.e., that they are part of the translator's native grammar, but are accidentally absent from the diagnostic data), one will have to admit that this option must have been very marginal in the system. After all, negative evidence should generally be taken (if not to prove the non-existence of a phenomenon) at least to suggest a possible scenario, and this is particularly true of data of this kind, which cannot be checked with native speakers and do not provide any additional information e.g. about prosody, speaker's intention, etc.
At this point, the question arises as to why in EOHG, a sequence of the type 'frame-setter[(familiar) topic' or 'frame-setter[(modal/cataphoric) adverb' (cf. Sect. 5.2.1) should be licit, whereas no serializations seem to be attested in which two (non-co-indexed) arguments surface in the left periphery. On the basis of the data, this could be accounted for in two ways.
being the unmarked case) in (41b) would disqualify this construction as a 'Hanging Topic'. It would rather lead one to favor an analysis in which the pronoun sia does not function as a 'full' pronominal resumptive, but rather lexicalizes the movement of the topic christes chiburt, which has been case-marked in the middle field, to its landing site in the left periphery via the [Spec,FinP]-like position that has been assumed for adjunct and left-dislocated constituents in Sect. 5.2.2. If this were the case, then: (i) the two examples (36b)/(41a) and (41b) could be taken to instantiate not two arguments (one DP and one pronoun) surfacing in the prefield, but rather only one DP whose cyclical movement through the lowest CP specifier is spelt out; (ii) the two clauses would exemplify sort of an 'imperfect' Left Dislocation type in which the resumptive element is not a d-, but a personal pronoun. This would also explain why this pattern is so poorly attested. 33

Patterns attested in other texts
As mentioned in Sect. 3, two further EOHG texts have been analyzed according to similar criteria to those applied to Isidor, Monsee Fragments and Tatian, but have not been considered in the results of the present investigation.
The Murbach Hymns are a word-by-word translation (interlinear version) of a Lat. text. This makes this text unsuitable for syntactic investigations in general. Virtually every single matrix clause contained in this text replicates the syntactic arrangement of the Vorlage. As a consequence, even diagnostic tests for verb movement produce inaccurate (in the sense of 'only seemingly diagnostic') results. This, in turn, implies that the 'diagnostic' V3 configurations available in this text cannot be taken to be representative of native spoken OHG. Indeed, patterns are attested in the dataset that do not match the findings presented in the previous sections. For instance, the assumption that (42a) might have a native word order (due to the licensing of a null referential pro in the middle field) would lead to the conclusion that the EOHG prefield may host two arguments-in this case, one in the dative (bound by a postnominal focus operator, which could only follow from DPinternal movement if we assume that additive particles are left-adjoined to V2 clauses, cf. Sect. 5.2.1.5, or by an additive coordinating conjunction) and one in the accusative. This arrangement faithfully replicates that of the Lat. text, the postnominal element ioh being the German rendering for the enclitic conjunction -que. The same goes for (42b), in which V-to-C movement could in principle be assumed to have taken place given that the utterance realizes a directive speech act and the verb is in the imperative mood. If that were the case, the preverbal area of this clause would contain an adjectival predicative of the object and the 33 An anonymous reviewer points out that another possibility to analyze the data in (41) would be that neither of the two examples result from V-to-C movement to the left periphery and that their strikingly un-Germanic pattern involving a topic preceding a wh-phrase is to be attributed to the Lat. word order. In principle, it could of course be the case that they are 'exceptions to the rule' in which the arrangement of the original text overrides the translated syntax. If we instead take the (wh-)interrogative nature of these two clauses to diagnostically show that the verb has moved to C°, as proposed by Axel (2007), they are to be categorized as cases of Hanging Topics involving verb raising. corresponding direct object in the accusative, which is otherwise not attested in any other EOHG text. Otfrid's gospel harmony, a non-translational text, is an epic poem: it is one of the first German texts to use the Romance end rhyme, a verse type that gradually replaces the traditional Germanic alliterative pattern. As, for example, Speyer (2016, 116) notes, Otfrid's Evangelienbuch has a more loosely knit metrical pattern than other poetic texts, which means that it may be used (with caution) to investigate specific phenomena that are not directly or exclusively dependent on word order, but it is ultimately not appropriate for purely syntactic analyses, especially if a strict methodological approach is adopted. This poem certainly also contains constructs that are part of the syntactic repertoire of EOHG, but this is not enough in order to compile a typology of possible multiply-filled-prefield constructions in this language stage. This is basically the same problem that was discussed with respect to the alliterative meter of Old Saxon Heliand in Sect. 5.3. A closer look at the 'diagnostic' V3 patterns present in this text shows that even here, arrangements are attested that may be metrically motivated. In (43a), at least two clues could be indicative of verb movement, namely: (i) the post-finite position of the verb particle úz (forming a unit with the lexical verb werfan, uzwerfan meaning 'throw out'), and; (ii) the post-finite position of the non-extraposable adverb sar. If this arrangement were to be imputed to the author's native grammar, then we should assume that complex prefield structures are possible in EOHG that contain the serialization subject pronoun [direct-object pronoun[modal adverb[deictic adverb[modal adverb. This relative ordering is not attested in the diagnostic dataset presented in the previous sections. Indeed, it is quite clear that it depends on metrical principles, given that wár rhymes with utterance-final sar. In order to establish this pattern, the word wúrfun (i.e., the finite verb, whose position opens up a prefield to its left) must be preceded by an appropriate number of syllables. This is most probably the reason why the prefield is so complex in this clause. In (43b), in which the adverb hárto ('hard') would have to surface in the middle field in a diagnostic structure, three pronominally realized arguments occupy the prefield: a (cataphoric) genitival object, a subject and a direct object. Also in this case, the word order is legitimated by the metrical structure of the text: hárto rhymes with worto; indeed, this serialization is not part of the patterns observed above.  ' (O. 5.25, 9) It may be concluded that even a text like Otfrid's gospel harmony, which is often used as a source for syntactic investigations of OHG, is not adequate for investigating the structure of the left periphery of this language stage. If we were to include Otfrid in our corpus and consider all 'diagnostic' V3 configurations attested in this text in combination with the results from the other texts in order to provide a comprehensive typology of Verb-Late(r) constructs, our taxonomy would be contaminated with word orders that are not part of the EOHG native grammar.

Concluding remarks
In this paper, I have discussed some aspects related to the investigation of rootclause V3 in EOHG (eighth to ninth century) by focusing on the possible serializations attested in the pre-finite area of the clause. The main claim was that, in light of the syntactic peculiarities of this language stage, which is characterized by a considerable amount of verb-position variation, extraposition and licensing of null referential subjects, only a very strict methodology can unveil which multiply-filledprefield combinations may be assumed to be part of the native grammar of EOHG with a certain degree of plausibility.
In particular, to determine the relative complexity of the EOHG prefield, three methodological measures were taken: (i) all non-prose texts were a priori excluded from the analysis, reducing the corpus to Isidor, the Monsee Fragments and Tatian; (ii) only clauses with a syntax diverging from the Lat. were considered; (iii) the investigation included only those clauses in which verb movement to C°could be diagnostically ascertained.
The study produced the following results: (i) in approximately 65-80% of surface V3 clauses attested in the corpus, the German structure perfectly replicates the arrangement of the Lat. text and/or V-to-C movement cannot be diagnosed. This means that only a minority of the data available are a reliable source; (ii) five non-correlative and three correlative V3 constructions (and some combinations of these patterns) are frequently attested in the diagnostic dataset, while some of the left-peripheral sequences occurring in the non-diagnostic data (e.g., the fronting of two or more full-DP arguments) are never found in the diagnostic group; (iii) these results lead to the consideration that texts with a metrical structure are not as adequate as prose texts to be included in a corpus for the investigation of issues related to purely syntactic principles.
The data extracted following these methods have the advantage that they are, if not incontrovertible, at least very reliable with respect to their native status. Of course, it cannot be excluded that other possible left-peripheral combinations accidentally only occur in non-diagnostic clauses and can therefore not be counted among the constructs discussed here; nonetheless, the result of having a circumscribed typology of V3 structures that does not bear on conjectures is-at least for the time being-worth the risk of not being completely exhaustive.
The question as to whether EOHG can be assumed to show bottleneck effects (Haegeman 1996;Roberts 2004;Cardinaletti 2010) lends itself to considerable adhoc speculation: it has been shown in Sect. 5.2.2.1 that whenever, for example, a weak subject pronoun, a low-CP modal adverb or a focused/contrastive constituent and a fronted adverbial clause surface together in the left periphery, the latter cannot be resumed by a correlative element. This seems to suggest, at least in very abstract terms, that the movement of such constituents into the CP must occur via [Spec, FinP] (or a similar projection located in the lower portion of the prefield), which blocks movement into the left periphery. Under these premises, it would be irrelevant whether the moved constituent remains in the specifier position of FinP (as is presumably the case for weak pronouns and modal adverbs) or further moves to a higher position (this has been assumed to happen, e.g., with interrogative pronouns), since the assumption could be made that in such cases, the constituents surfacing to the left of the XPs assumed to have moved to the left periphery (in this case: personal and interrogative pronouns, modal adverbs, etc.) are base-generated in that position. This solution, however, could be problematic if we wanted to explain why adverbial clauses with the very same interpretation (say, genuine hypothetical conditional clauses) are sometimes base-generated in the extended prefield and sometimes in a much lower position. The same goes for serializations of the type 'focused constituent (e.g. wh-pronoun)[weak pronoun[Vfin'. If we adopted the assumption that only one element can be dislocated into the CP from the middle field and that this element must move via [Spec,FinP], then we should explain why interrogative pronouns are sometimes moved to the CP and sometimes (i.e., in cases like christes chiburt huuer sia chirahhoda in example (41b)) firstmerged in their surface position. The only convincing evidence for movement through a low-CP specifier comes from frame-setting adverbial clauses resumed by correlative elements. A more thorough investigation of these issues, as well as of the factors licensing multiply-filled prefields in EOHG, is left to future research.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creativecommons.org/licenses/by/4.0/.