Early German = Slavic?

2.1 The idea is not entirely new

H & S’s approach in terms of variable head positioning (VHP) seems to be very similar in spirit to the double base hypothesis (DBH) proposed by Susan Pintzuk (1999) and others to account for the variable word order of Old English (OE) and a number of other languages (cf. e.g. Santorini 1992 on Yiddish and Taylor 1994 on Ancient Greek). The only difference seems to be that H & S assume that the directionality parameter is unspecified, while Pintzuk assumes that both a head-initial and a head-final setting co-exist in a single grammar. As far as I can see, however, the two approaches do not differ with regard to their empirical coverage. Moreover, the approach by Pintzuk is embedded under a general theory of language change in terms of grammar competition (Kroch 1989), which captures quantitative and variationist aspects of word order change. It is unclear how the latter can be accounted for under H & S’s approach.

2.2 The absence of VO-Aux patterns

Due to the basic similarity with the DBH, the VHP faces the same challenges as the DBH. For instance, it is not clear how the order VO-Aux, which seems to be unattested in the world’s languages (the Final-over-Final Condition [FOFC]; Biberauer et al. 2014; Sheehan et al. 2017) can be systematically ruled out if the heads of two projections XP and YP (with YP directly embedded under XP) may precede or follow their complement. In previous work (cf. e.g. Haider 2010), Haider addresses this problem by assuming that VO-Aux orders present a parsing problem (center embedding) which must be repaired by other operations (e.g. verb raising/verb cluster formation). However, this approach does not seem to capture a basic difference of these two phenomena: Despite the parsing problems they create, center embedding structures do appear in the world’s languages; in contrast, FOFC violations seem to be ruled out cross-linguistically. This suggests that center embedding structures can be created by the grammar, while VO-Aux orders should be ruled out by general principles.^[1]

2.3 Overall similarity of T3 and SOV languages

While I basically agree with H & S’s critical assessment of traditional typological approaches that classes Slavic with regular SVO languages, the alternative analysis in terms of T3 languages is also not without problems. While the authors highlight the idea that the typological classification of languages should be based on structural properties, the difference between SOV and T3 languages turns out to be rather marginal. In fact, it seems to be confined to a single surface property, namely linear order (in contrast to SOV languages, T3 languages allow for surface SVO/VSO orders). With respect to all other criteria discussed in the paper (cf. Table 1 on p. 3 for an overview), T3 and SOV languages are expected to behave on a par. This raises the question of whether it is really justified to posit a major typological difference or whether we should attribute the surface difference between SOV and “mixed” languages to another well-established parameter, namely the absence/presence of information-structurally driven movement operations (see below for another potential difference between T3 and SOV languages, which H & S only briefly mention, namely the availability of verb raising/verb cluster formation).

2.4 Pathways of word order change

There is general agreement that Proto-Indo-European (PIE), the common ancestor of the Germanic, Romance, Celtic, Slavic languages (and many other families) was an SOV-language (cf. e.g. Lehmann 1993 and the work cited there). If we accept this view, an analysis that treats Slavic and Early Germanic uniformly as T3 languages must assume a number of changes in basic word order type:

1. – For prehistoric and historical stages of Slavic, a change from SOV (PIE) to T3.

2. – For prehistoric and historical stages of the Germanic VO-languages, a change from SOV (PIE) to T3 to SVO.

3. – For prehistoric and historical stages of German, H & S must assume a trajectory for word order change from SOV (PIE) to T3 (Early Germanic) back to SOV (Early New High German (ENHG)/present-day German).

The traditional view that present-day German still reflects the basic OV character of earlier stages does not require a double typological shift from SOV to T3 and back to SOV and thus seems to be the more parsimonious option. Moreover, the idea that SOV languages may shift to T3 languages raises a number of additional questions, since these types are very similar (see 2.3). In particular, we might wonder which kind of evidence/independent change might lead learners to reanalyze syntactic patterns generated by an SOV grammar in terms of a T3 grammar. Furthermore, it is quite mysterious why T3 quite rapidly gives way to strict VO and OV in Germanic, while it remains stable in Slavic.

3.1 OV/VO and the placement of elements that resist extraposition

Despite their close genetic affiliation and their surface similarity, the early Germanic languages do not exhibit identical syntactic properties. Focusing on Old English (OE) and OHG, I will argue that the relevant (by now well-known) differences do not warrant a uniform treatment of these languages in terms of a T3/VHP grammar. Rather, it will become clear that while OE might fit the bill, OHG is typologically closer to the present-day Germanic OV languages than to Slavic.

In the literature on the OV-VO alternation in early Germanic, it is commonly assumed that the relative ordering of verb and auxiliary, and the position of verbal particles represent indicators of basic word order properties. Cases where the finite auxiliary follows the non-finite verb as in (1a), or where verbal particles precede the lexical verb as in (1b) are commonly taken to signal an OV base (cf. e.g. Pintzuk 1999; but see Elenbaas and van Kemenade 2014 for critical discussion):^[2]

Relevant orders are robustly attested in both OE and OHG, as illustrated in (2) and (3). However, as is well known, OE and OHG also exhibit orders where arguments and adjuncts follow the verb (cf. e.g. Pintzuk 1999 for OE, and Axel 2007 for OHG). Some OHG examples are given in (4).

Traditionally, surface VO orders as in (4) are analyzed as the result of extraposition from an underlyingly preverbal position (cf. e.g. van Kemenade 1987 on OE; Axel 2007 and Sapp 2016 on OHG; but see e.g. Biberauer and Roberts 2005 for an alternative analysis). However, as pointed out by Pintzuk (1999), there are VO orders in OE where an analysis in terms of extraposition runs into difficulties. Relevant cases involve postverbal placement of prosodically light elements such as pronouns, (verbal) particles, and monosyllabic adverbs, which cannot undergo extraposition in the Germanic OV languages:^[3]

According to Pintzuk (1999), orders as in (5) support the hypothesis that basic VO was a structural option in OE in addition to an OV base (see also Haider 2012, 2014). She analyzes these facts in terms of the DBH, i.e., conflicting settings of the head parameter for both V and Infl, which, as noted above, is very similar to VHP/T3. So it seems that, in principle, OE might be amenable to an analysis along the lines proposed by H & S. Next, I would like to address the question of whether this approach is also viable for OHG.

At first sight, it appears that examples with postverbal pronominal elements can also be found in early OHG texts:

However, upon closer inspection it turns out that the word order of the OHG clauses in (6a–b) is identical to the Latin source; in particular, it appears that the postverbal position of the pronoun in (6a–b) simply mimics the ordering found in the corresponding Latin clause and thus does not constitute clear evidence in favor of the existence of a VO base order option in OHG.^[4] Still, there are a few cases where the postverbal placement of a pronominal element (in most cases a reflexive pronoun) cannot be attributed to properties of the source text (cf. e.g. Dittmer and Dittmer 1998; Schallert 2010; among others):

In (7), the postposed reflexive pronoun sih translates a preverbal Latin pronoun; in (8), the Latin text lacks a relevant pronominal element. However, these cases are exceedingly rare in OHG and subsequent historical stages of German (cf. e.g. Axel 2007; Sapp 2014, 2016; Fuß 2018; and Table 1 below). Thus it seems that postverbal placement of object pronouns was not freely available in OHG, which casts doubt upon the claim that OHG was characterized by a mixed OV/VO grammar similar to OE (pace Haider 2010; Schallert 2010; and H & S). Basically the same goes for the placement properties of verbal particles and light adverbs such as thô, dhar, or nû. With respect to the former, Axel (2007: 109) notes that “in contrast to Old English, in OHG there are hardly any cases with post-verbal particles attested in subordinate clauses with particle verbs”.^[5] A search conducted in the OHG Reference Corpus (ReA) shows that the bulk of cases where a light adverb follows the verb in an embedded clause comes from Otfrid’s gospel harmony.^[6] However, it is a well-known fact that evidence from Otfrid does not always provide reliable cues on OHG word order, since the syntactic patterns are often shaped by requirements of rhyme and meter. This holds also true of deviating word orders involving postverbal light adverbs, as illustrated with (9), where the final position of thar ‘there’ seems to be determined by the needs of rhyme:

Thus, it appears that OHG differs from OE in that it exhibits a strong tendency to avoid postverbal placement of light elements that resist extraposition such as pronouns, verbal particles, and light adverbs. This suggests that in contrast to OE, OHG does not support multiple base orders in the VP domain.^[7]

Crucially, OHG differs in this respect also from the Slavic languages, where postverbal placement of pronouns and light adverbs seems to be easily possible. In Polish, for example, relevant orderings are somewhat marked, but perfectly grammatical. This is shown in (10) and (11) for object pronouns and light adverbs, respectively.^[8]

To conclude, it turns out that evidence from the placement properties of phonologically light elements that generally resist extraposition suggests that OHG systematically differs from alleged T3 languages such as Old English and Polish. Instead, the relevant facts seem to support the view that German has not been affected by a major change in basic word order throughout its recorded history, revealing a diachronic continuity ranging from OHG over middle high German/early new high German to present-day German.

3.2 Quantitative evidence

The impression that OHG is more “OV-ish” than OE is further supported by quantitative evidence. For example, Dittmer and Dittmer (1998) point out that word orders that deviate from the Latin source text in the Tatian translation exhibit a clear preference for preverbal placement of arguments and adjuncts. Table 1 shows that cases of VO order that have no model in the Latin source text are very rare, while there is a large number of OV orders that are independent of or deviate from the Latin source.

In a similar vein, based on a survey of more than 2,500 embedded clauses from OHG, MHG, and ENHG, Sapp (2014, 2016 concludes “that there has been no parametric change in the headedness of the VP from PIE to OHG to modern standard German“ (Sapp 2016: 404) and that present-day German extraposition “aside from decreased frequency, is largely similar to the medieval construction” (Sapp 2014: 154), in that it affects essentially the same range of constituents (mostly clauses, PPs, and heavy or focused NPs). This historical continuity would come as a surprise if VO orders were generated by different mechanisms in the various historical stages of German (T3 in OHG/MHG, OV + extraposition in present-day German). Of course, H & S could still say that the few apparently clear-cut cases of VO order suffice to support their view that OHG was characterized by a T3 grammar. But this raises the general question of whether we are willing to accept the notion that the word order options made available by a language’s type of grammar are only rarely exploited, or whether we prefer to attribute the relevant few cases to the workings of other factors (such as marked stylistical operations).^[9]

3.3 Verb clusters

As noted above, V-Aux order is often taken to be a reflex of a language’s basic OV character (cf. already Greenberg 1963). However, it is a well-known fact that in the Germanic OV-languages, alternative serializations of selecting and selected verb are also possible (for different historical stages and varieties of German cf. e.g. Maurer 1926; Behaghel 1932; Robinson 1997; Schmid and Vogel 2004; Schmid 2005; Axel 2007; Sapp 2011a, 2011b). Importantly, as pointed out by H & S, these reordering possibilities are only found in the Germanic OV-languages, while VO-languages exhibit a fixed ordering of finite and non-finite verb forms (cf. e.g. Haider 1993; Vikner 2001). In particular, the formation of verb clusters (in combination with restructuring effects) is a typical feature of many continental West Germanic OV-languages. This is also true of early German (cf. e.g. Robinson 1997). Example (12) shows that in addition to V-Aux order (cf. (2b) above), Aux-V is also a possible serialization in OHG. Moreover, the fact that the order of the finite and the non-finite verb deviates from the order found in the Latin source suggests that verb raising/verb cluster formation was a well-established trait of the early OHG grammar.

This conclusion is supported by the observation that early OHG also exhibits variable ordering in three-verb clusters:

As pointed out by H & S, this phenomenon seems to be absent in the Slavic languages, which exhibit variable linearization of verb forms (to the extent that auxiliary and modal verbs are available), but crucially “the verbs do not form a cluster of the kind known from OV languages” (H & S, p. 24). The data in (12) and (13) clearly show that OHG is not subject to such a restriction. Instead, it seems to pattern with the present-day Germanic OV-languages.^[10]