2.1. West Central German

Since the 19th century, a number of scholars have made proposals for the division of the German dialect space (see, for example, Behaghel 1891, Bernhardi 1844, Bremer 1892, Martin 1959, Mitzka 1943, and Wiesinger 1983). Niebaum & Macha (2005:217) consider Wiesinger's (1983) division, which is based on a set of phonological and morphological features, the closest to linguistic reality since it clearly indicates the exis-tence of transitional zones. As the authors point out, however, the majority of recent overviews of German dialects differentiate one dialect region from another on the basis of sound shift boundaries, resulting in a conception of dialect space in which one dialect area begins where another ends.Footnote ⁷The Institut für Deutsche Sprache in Mannheim, which maintains the material from which the data treated in this paper are drawn, organizes the sound recordings in its archives according to this type of division. Since the question of dialect differentiation is not central here, I therefore assume divisions of the WCG dialect area according to sound shift isoglosses, adopting those in Beckers 1980 as presented in figure 2 below.

WCG belongs to the central region of the High German dialect area bordering on Low German to the north, East Central German to the east, and Upper German to the south. The WCG region can be subdivided into two larger dialect areas along isogloss 6 in figure 2, the dat/das line, namely Central Franconian and Rhine Franconian. The former comprises the smaller dialect areas Ripuarian and Moselle Franconian, the latter Pfälzisch and Hessian.Footnote ⁸ The dialects spoken in Lorraine (Lothringisch) were not included in this study.

Figure 2.

West Central German dialect area (Beckers 1980:469).

2.2. The Corpus

The corpus for the present study consists of 2,578 clauses from transcripts of 187 dialect recordings in the Datenbank Gesprochenes Deutsch (Spoken German Databank), maintained online by the Institut für Deutsche Sprache. The transcripts, classified as Vollmundart (base dialect), were taken from the Zwirner Corpus compiled between 1955 and 1970 under the direction of Eberhard Zwirner.Footnote ⁹ These include all 73 available transcripts from the Central Franconian dialect area and 114 transcripts from the Rhine Franconian dialect area.Footnote ¹⁰ The number of transcripts from the Rhine Franconian dialect area totals roughly 175, yet only a subset of these was selected in order to strike a better balance between the number of transcripts from both major dialect regions. To this end, only one transcript from the oldest male and female speaker per Planquadrat (quadrant) was included for analysis.Footnote ¹¹

Figure 3 above indicates the quadrants from which at least one transcript was taken. The WCG dialect area as a whole is demarcated by the heavy black line. The smaller dialect areas, further distinguished by the medium-heavy lines, are labeled “a” (Ripuarian), “b” (Moselle Fran-conian), “c” (Hessian), and “d” (Pfälzisch).

Figure 3.

Quadrants represented in the corpus study (map from Haas & Wagener 1992:1611).

The 2,578 total tokens include 1,450 subordinate and 1,128 main clauses that fall into four complex categories: subordinate clause two- and three-verb complexes and main clause two- and three-verb com-plexes.Footnote ¹² Only those tokens whose clausal status could be unambiguously determined were included in the corpus. Consider the examples in 6 and 7.

1. (6)

2. (7)

Since no constituent appears between the subject die Leute ‘the people’ and the verb sollen ‘should’ in 6 or the pronominal subject wir ‘we’ and the verb sollten ‘should’ in 7, it is not possible to determine whether these are examples of main clauses with verb-second word order or subordinate clauses with the 1-X-2-(X) word order. Both clauses are introduced by a subordinating conjunction, dass and wenn, respectively; it is well known, however, that the subordinating conjunction weil may be followed by main clause word order in spoken German (see, for example, Gaumann 1983, Günther 1993, Louden 2000, Schwitalla 2003: 144–146, Wegener 1993). Thus, the type of conjunction alone is not sufficient to determine clausal status. For this reason, only clauses in which an element appears between the subject and first verb were included, as in 8 and 9.

1. (8)

2. (9)

After excluding eight clauses that did not meet this criterion, all those remaining were checked against the audio recordings, entered into a database, and then coded and classified according to verb form and verb type for further analysis.

3.1. Subordinate Clause Two-Verb Complexes

The corpus contains 1,326 subordinate clause two-verb complexes that fall into the syntagms illustrated in 10–16. The distribution of these syntagms is presented in table 1.Footnote ¹⁴

1. (10)

   1. a.

   2. b.

   3. c.

   4. d.

   5. e.

2. (11)

3. (12)

4. (13)

5. (14)

6. (15)

7. (16)

Table 1.

Distribution of subordinate clause two-verb complexes by syntagm (group).

As shown in table 1, the overall preferred order is 2-1, which occurs far more frequently than the 1-2 order, at a rate of 88% to 12%.Footnote ²¹ The group of syntagms involving a past participle and an auxiliary verb, namely PTC + AUX (perfect, subjunctive, passive), and INF + MOD occur most frequently in the corpus, accounting for 1,263 out of 1,326 total tokens (95.2%), and display both the 2-1 and 1-2 orders.Footnote ²² INF + tun and INF + LEX, which occur far less frequently, also display both the 2-1 and 1-2 orders, though the 1-2 order is rare with the former syntagm.Footnote ²³ Finally, the syntagms with the lowest frequencies, the PTC + kriegen, INF + lassen, and the modal perfect, display only the 2-1 order. The last syntagm is very rare in the corpus, occurring in a single token. Table 2 shows the breakdown of word order rates for PTC + AUX (perfect, subjunctive, passive).

Table 2.

Breakdown of word order rates for PTC + AUX.

With the perfect with haben, subjunctive with sein, and werden passive, the rate of the 1-2 order is very low, occurring in only 7 out of 521 tokens (1.3%), 1 out of 18 tokens (5.6%), and 1 out of 111 tokens (0.9%), respectively. The 1-2 order is far more frequent with the perfect with sein, occurring in 64 out of 288 (22.2%) tokens. Finally, the 1-2 order is unattested with the subjunctive with haben.

3.2. Subordinate Clause Three-Verb Complexes

The corpus contains 124 subordinate clause three-verb complexes that fall into the syntagms illustrated in 17–30. The distribution of these syntagms is presented in table 3.

1. (17)

   1. a.

   2. b.

2. (18)

   1. a.

   2. b.

3. (19)

   1. a.

   2. b.

4. (20)

5. (21)

6. (22)

7. (23)

8. (24)

9. (25)

10. (26)

11. (27)

12. (28)

13. (29)

14. (30)

As shown in table 3, the 3-1-2 order occurs most frequently overall, in 58 out of 124 total tokens (46.8%). However, the 3-2-1 order occurs nearly as frequently, in 50 out of 124 total tokens (40.3%). Thus, there is no robust predominance of a single word order as with the subordinate clause two-verb complexes. The most frequently occurring syntagms are the perfect of passive, werden passive + MOD, and modal IPP, representing 91 out of 124 total tokens (73.4%). Instances of the perfect of passive are nearly equally divided between the 3-2-1 and 3-1-2 orders, though the 3-2-1 order is slightly more frequent. With the werde n passive + MOD and modal IPP, the 3-1-2 order occurs most frequently by a wide margin. The remaining syntagms are represented by 33 out of 124 total tokens (26.6%).

Table 3.

Distribution of subordinate clause three-verb complexes by syntagm.

Note that the orders 2-3-1 and 2-1-3 are unattested in the corpus. This fact is unsurprising: The 2-3-1 order is relatively uncommon, limited to modal constructions in Afrikaans, Swabian, and West Flemish, while the 2-1-3 order is conspicuously absent from the inventories of three-verb complexes cross-linguistically and has thus been treated extensively in the generative literature.Footnote ²⁷ (See Wurmbrand 2006:239–242 and Dubenion-Smith 2008:43–62 for discussions of these word orders.)

3.3. Main Clause Two-Verb Complexes

The corpus contains 1,114 main clause two-verb complexes that fall into the syntagms illustrated in 31–48. The distribution of these syntagms is presented in table 4.Footnote ²⁸

1. (31)

2. (32)

3. (33)

   1. a.

   2. b.

4. (34)

5. (35)

6. (36)

7. (37)

8. (38)

9. (39)

10. (40)

11. (41)

12. (42)

13. (43)

14. (44)

15. (45)

16. (46)

17. (47)

18. (48)

Table 4.

Distribution of main clause two-verb complexes by syntagm.

As table 4 shows, the overall preferred order is 3-2, which occurs far more frequently than the 2-3 order, at a rate of 87% to 13%. Note that this distribution is very similar to that for the subordinate clause two-verb complexes (88% 2-1 to 12% 1-2). The syntagms perfect of passive [S], passive [MOD], and modal IPP [H] occur most frequently in the corpus, accounting for 878 out of 1,114 total tokens (78.8%). Strikingly, the 2-3 order does not occur at all with the perfect of passive [S] and passive [MOD], both involving a past participle and a form of werden, while this is by far the preferred order with the modal IPP [H], at a rate of 57.9%. In section 4.1, we see that this distribution can be attributed in part to the combination of verb forms and types that comprise these syntagms. Among the remaining syntagms, both the 3-2 and 2-3 orders occur with seven, but the 2-3 order is preferred with only three: the subjunctive modal IPP [H], ACI IPP [H], and INF + modal perfect [H].

3.4. Main Clause Three-Verb Complexes

The corpus contains 14 main clause three-verb complexes that fall into the syntagms illustrated in 49–55. The distribution of these syntagms is presented in table 5.

1. (49)

2. (50)

3. (51)

4. (52)

5. (53)

6. (54)

7. (55)

Compared to the other complex categories, the main clause three-verb complexes are relatively scarce, as table 5 shows. Nonetheless, 4 of the 6 possible word orders are attested in the corpus: 4-3-2, 4-2-3, 3-2-4, and 2-4-3. The 4-3-2 and 2-4-3 orders occur most frequently, each in 6 out of 14 total tokens (42.9%). Syntagms involving an IPP construction account for the majority of tokens, 11 out of 14 (78.6%). As with the subordinate clause three-verb complexes, there is no robust predominance of a single word order.

Table 5.

Distribution of main clause three-verb complexes by syntagm.

3.5. Verb Projection Raising and Two-Verb Complexes

This section deals with verb projection raising in subordinate clause two-verb complexes and main clause two-verb complexes. Recall from the previous section that 159 out of 1,326 subordinate clause two-verb complexes display the 1-2 order. Of these, 20 (12.6%) involve VPR with a constituent appearing between the members of the verb cluster. The intervening constituents fall into four types, illustrated in 56–59. Their distribution is presented in table 6.

1. (56)

2. (57)

3. (58)

4. (59)

Table 6.

Distribution of VPR intervening constituents, subordinate clause two-verb complexes.

Table 6 shows that the majority of intervening constituents are separable prefixes. Note that unlike standard German (but like languages such as standard Dutch), WCG readily allows separable prefixes to be stranded. This is exemplified in the following example in which the particle an ‘on’ does not appear raised with schneiden ‘cut’.Footnote ³²

1. (60)

Let us now examine verb projection raising in main clause two-verb complexes. Compared to the subordinate clause two-verb complexes, the main clause two-verb complexes display VPR at a much higher rate. Of the 145 clauses with the 2-3 order (13%), 75 (51.7%) involve an inter-vening constituent. The attested VPR constructions fall into five types with a single intervening constituent and an additional 13 types with two or more intervening constituents. Table 7 shows the distribution of con-structions with one intervening constituent, illustrated in 61–65.Footnote ³³

1. (61)

2. (62)

3. (63)

4. (64)

5. (65)

Again, we see that the majority of intervening constituents are separable prefixes. A comparison of tables 6 and 7 reveals that for both the subordinate and main clause complexes, the three most common types of intervening constituent are the same (separable prefix, PP, adjective/ adverb), followed by at least one nominal category (indefinite NP, bare N, definite NP).

Table 7.

Distribution of VPR intervening constituents, main clause two-verb complexes.

3.6. Extraposition and Two-Verb Complexes

In this section, I discuss extraposition in subordinate clause two-verb complexes and main clause two-verb complexes. As noted in section 1, verbal complexes may occur not only with elements appearing inside the cluster, but also with extraposed elements to the right of the cluster. The distribution of extraposition in the subordinate clause two-verb com-plexes is presented in table 8.Footnote ³⁴

Table 8.

Distribution of extraposition, subordinate clause two-verb complexes.

As table 8 shows, extraposition occurs in 55 out of 1,326 total tokens (4.1%). 50 out of 55 instances (90.9%) occur with the 2-1 order, while only 5 out of 55 (9.1%) occur with the 1-2 order. Thus the 1-2 order occurs less frequently than the expected rate (12%) in constructions that display extraposition.

Table 9 shows the distribution of extraposition by constituent type, illustrated in 66–72.Footnote ³⁵

1. (66)

2. (67)

3. (68)

4. (69)

5. (70)

6. (71)

7. (72)

Table 9.

Distribution of extraposed constituents, subordinate clause two-verb complexes.

The majority of extraposed constituents are adjunct PPs, accounting for 30 out of 55 tokens (54.5%). The adjunct PPs, in combination with the other adverbials and da-compound, comprise 47 out of 55 total tokens (85.5%), while argument PPs make up the remaining 8 out of 55 tokens (14.5%). The occurrence of only one instance of an extraposed object NP falls in line with the increased unacceptability of object NP extraposition in the ENHG period to its general ungrammaticality in modern German (see Bies 1996).

Like the subordinate clause two-verb complexes, the main clause two-verb complexes display extraposition. The distribution of extra-position is presented in table 10.

Table 10.

Distribution of extraposition, main clause two-verb complexes.

As table 10 shows, extraposition occurs in 108 out of 1,114 total tokens (9.7%) at a rate significantly higher than in the subordinate clause two-verb complexes (4.1%). Ninety-six out of 108 instances (88.9%) occur with the 3-2 order, while only 12 out of 108 (11.1%) occur with the 2-3 order. Thus the 2-3 order occurs less frequently than the expected rate (13%) in constructions that display extraposition, as we saw above for the 1-2 order in the subordinate clause two-verb complexes.

Table 11 below shows the distribution of extraposition by constituent type, illustrated in 73–80.

1. (73)

2. (74)

3. (75)

4. (76)

5. (77)

6. (78)

7. (79)

8. (80)

Table 11.

Distribution of extraposed constituents, main clause two-verb complexes.

As we saw above with the subordinate clause two-verb complexes, adjunct PPs account for the majority of extraposed constituents in the main clause two-verb complexes as well, comprising 61 out of 108 total tokens (56.5%). These plus the adverbials, da-compound, appositives, and preposition make up 86 out of 108 total tokens (79.6%). Argument PPs account for the remaining 22 out of 108 tokens (20.4%). Extraposed object NPs do not occur in the main clause two-verb complexes.

3.7. Summary of Main Findings

The syntagms and word orders attested in the Zwirner Corpus data were introduced in section 3.1–3.4. For the subordinate clause two- and three-verb complexes and the main clause two-verb complexes, these are presented in the summary tables 12–14 below.Footnote ³⁸

Table 12.

Word orders in WCG, subordinate clause two-verb complexes.

Table 13.

Word orders in WCG, subordinate clause three-verb complexes.

Table 14.

Word orders in WCG, main clause two-verb complexes.

In the subordinate clause two-verb complexes, both the 2-1 and 1-2 orders occur, at a rate of 88% and 12%, respectively. The syntagms INF + LEX, INF + MOD, and perfect with sein display the non-canonical 1-2 order at rates higher than the expected rate. Four word orders are attested in the subordinate clause three-verb complexes, 3-2-1, 3-1-2, 1-3-2, and 1-2-3, occurring at respective rates of 40.3%, 46.8%, 8.1%, and 4.8%. The modal IPP displays the non-canonical word orders 1-2-3 and 1-3-2, the werden passive + MOD the order 1-3-2, and the perfect of passive the order 3-2-1 more frequently than the expected rates. The main clause two-verb complexes display both the 3-2 and 2-3 orders, at rates nearly identical to those in the subordinate clause two-verb complexes: 87% and 13%. The syntagm modal IPP [H] displays the non-canonical order 2-3 at a rate far higher than the expected rate. Finally, the four word orders 4-3-2, 4-2-3, 3-2-4, and 2-4-3 are attested in the main clause three-verb complexes, at respective rates of 42.9%, 7.1%, 7.1%, and 42.9%.

Sections 3.5 and 3.6 presented the distribution of elements in VPR and extraposition constructions, limited to the two-verb complexes. The analysis of VPR showed that the rate of occurrence is far higher in the main clause than in the subordinate clause complexes (51.7% vs. 12.6%) and that separable prefixes comprise the largest group of intervening constituents. Similarly, the rate of extraposition is higher in the main clause than in the subordinate clause complexes (9.7% to 4.1%). For both complex categories, the rate of the non-canonical word order is lower than the expected rate when an element is extraposed, and adjunct PPs comprise the largest group of extraposed constituents.

4.1. Syntagm and Two-Verb Complexes

In this section, I consider the effect of syntagm on word order in subordinate and main clause verb complexes. I begin with subordinate clause two-verb complexes. In the corpus, syntagms with an infinitive display the 1-2 order at a far higher rate than those with a past participle. Table 15 shows that the former favor, and the latter disfavor, the 1-2 order. Table 16 presents the factor weights for these syntagms broken down into specific types.Footnote ⁴²

Table 15.

Effect of general syntagm on word order, subordinate clause complexes.

Table 16.

Effect of specific syntagm on word order, subordinate clause complexes.

As the table illustrates, the syntagms INF + LEX, INF + MOD, and perfect with sein favor the 1-2 order, while this order is disfavored by INF + other (tun, lassen), the subjunctive (haben, sein), the perfect with haben, and the werden passive. The syntagms can be placed on a hierarchy according to factor weight: INF + LEX > INF + MOD > perfect with sein > perfect with haben > werden passive. This hierarchy is similar to those found in previous studies of ENHG. Compare the hierarchies in 81a–d.Footnote ⁴³

1. (81)

   1. a.

   2. b.

   3. c.

   4. d.

The four hierarchies are nearly identical, with the exception of the relative ordering of the constructions perfect with sein and INF + MOD on Ebert's hierarchy. Despite this deviation, the order perfect with sein > perfect with haben > werden passive holds in each case. Thus, it may be concluded that the relationship between syntagm and word order in the subordinate clause two-verb complexes has not changed from Early New High German to modern German, as far as the WCG dialects are concerned.

Let us now consider main clause two-verb complexes. Since 11 of the main clause two-verb syntagms, including two of the three principal syntagms, are unattested with either the 3-2 or 2-3 order and are therefore *knockout* factors, it was not possible to conduct a statistical analysis with each separate syntagm. However, it was possible to analyze the syntagms in three larger groupings, namely syntagms involving an IPP, an infinitive, or a past participle.Footnote ⁴⁴ Recall that sub-ordinate clause two-verb complexes with an infinitive favor, and those with a past participle disfavor, the 1-2 order. This generalization holds for the main clause two-verb complexes as well. As table 17 shows, syntagms with two infinitives favor the 2-3 order, as do the syntagms with an IPP. The syntagms with a past participle, on the other hand, disfavor the 2-3 order.

Table 17.

Effect of verb form on word order, main clause complexes.

The syntagms may also be categorized according to whether the lexical verb occurs with another lexical verb, a modal, or an auxiliary.Footnote ⁴⁵Table 18 shows that when a lexical verb is combined with a modal or another lexical verb, the 2-3 order is favored. In contrast, combinations of a lexical and an auxiliary verb disfavor the 2-3 order.

Table 18.

Effect of verb types on word order, main clause complexes.

In conjunction, the hierarchies IPP > infinitive > past participle and LEX + MOD > LEX + LEX > LEX + AUX correspond roughly to the hierarchy for the subordinate clause two-verb complexes, namely INF + LEX > INF + MOD > perfect with sein > perfect with haben > werden passive, whereby complexes involving a lexical or modal verb and an infinitive more greatly favor the 1-2 order than those involving an auxiliary verb and a past participle. These hierarchies are also reflected in the three most frequently occurring main clause two-verb syntagms, the perfect of passive [S], passive [MOD], and modal IPP [H], presented in table 19.

Table 19.

Distribution of the three most frequently occurring syntagms, main clause complexes.

Here the additive effect of verb form and verb type is particularly striking. With the perfect of passive [S] and passive [MOD], which involve a past participle and the auxiliary werden, both at one end of their respective hierarchies, 0 out of a total 700 tokens (0%) occur with the 2-3 order. The modal IPP [H], on the other hand, which involves verbal elements at the other end of the hierarchies, an IPP and a modal, displays the 2-3 order at a rate of 58.4%, far higher than the expected rate of 13%.

4.2. Verbal Prefix Type and Two-Verb Complexes

This section deals with the effect of verbal prefix type on word order in subordinate and main clause complexes. I begin with the subordinate clause two-verb complexes. The results in table 20 reveal that the 1-2 order is favored by the presence of a stressed separable prefix on the non-finite verb in the complex and disfavored by the presence of an unstressed prefix or the absence of any prefix.

Table 20.

Effect of verbal prefix type on word order, subordinate clause complexes.

Sapp (2006:61–62) points out that the correlation between stressed separable prefix and the 1-2 order might be attributed to an interaction between verbal prefix type and syntagm. Past participles typically begin with the unstressed prefix ge-, and as we saw in section 4.1, syntagms involving a past participle disfavor the 1-2 order. Sapp tests for this interaction using GoldVarb's cross-tabulation function. The results of the cross-tabulation for the Zwirner Corpus data are presented in table 21 below.

Table 21.

Cross-tabulation of verbal prefix type with syntagm, subordinate clause complexes.

These results show that the factor groups syntagm and verbal prefix type are, in fact, independent of each other. For both syntagm groupings, that is, syntagms involving either a participle or an infinitive, the rate of 1-2 is above the expected rates (8% and 24%) when the prefix on the non-finite verb is stressed and below the expected rates for the other two prefix types. Note, however, that for all three prefix types the rate of 1-2 is higher for the infinitive grouping than for the participle grouping. It may be concluded, then, that while syntagm and verbal prefix type are independent factor groups, their effect on word order is additive.

Like the subordinate clause two-verb complexes, the main clause two-verb complexes exhibit an effect of verbal prefix type on word order, illustrated in table 22.

Table 22.

Effect of verbal prefix type on word order, main clause complexes.

The results show that a stressed separable prefix or no prefix on the main lexical verb in the complex favors the 2-3 order. Yet unlike the subor-dinate clause two-verb complexes in which the presence of an unstressed prefix or the absence of any prefix has a disfavoring effect on the 1-2 order, only unstressed prefixes have a disfavoring effect in the main clause two-verb complexes overall.

As with the subordinate clause two-verb complexes, an interaction between verbal prefix type and syntagm can be tested for using GoldVarb's cross-tabulation function, the results of which are presented in table 23.

Table 23.

Cross-tabulation of verbal prefix type with syntagm, main clause complexes.

The results of the cross-tabulation suggest that the effect of verbal prefix type on word order is not independent of syntagm. For syntagms involving two infinitives or an IPP, the rate of 2-3 is above the expected rates (19% and 56%) when the prefix on the main lexical verb is stressed and at or around the expected rates when there is no prefix. For syntagms involving a participle, on the other hand, the rate of 2-3 is at the expected rate (1%) when the prefix is stressed but far above the expected rate when there is no prefix. The fact that the verbal prefix types stressed and no prefix favor the 2-3 order with different syntagms indicates an interaction between the factor groups verbal prefix type and syntagm. Table 24 shows the effect of verbal prefix type on the groupings infinitive/IPP and participle separately.

Table 24.

Effect of verbal prefix type on word order, main clause complexes, infinitive/IPP, and participle.

The favoring effect of a stressed separable prefix on the 2–3 order in the syntagms with two infinitives or an IPP appears less robust than in table 22, while the favoring effect of the absence of a verbal prefix on syntagms with a participle is even clearer.Footnote ⁴⁶

4.3. Focus and Two-Verb Complexes

As discussed in section 4, Sapp (2006, 2007) investigates the relationship between word order in the verbal complex and object focus, in particular contrastive and new information focus, and, by extension, old infor-mation focus.Footnote ⁴⁷ Sapp (2006:58–59, 2007:302–303) finds that contrastive focus strongly favors the 1-2 order in his ENHG corpus. New information focus also favors the 1-2 order, though less so than contrastive focus. Old information focus, on the other hand, somewhat disfavors the 1-2 order. Sapp (2006:81, 2007:304) also determines that new information and contrastive focus favor every attested order except 3-2-1 (1-3-2/1-2-3/3-1-2 in combination) in the subordinate clause three-verb complexes.

Sapp (2006:37, 2007:303) concedes the difficulties inherent in ascertaining the information status of clauses in a written language. Without the intonation cues present in spoken language it is not always possible to correctly determine focus. For this reason, Sapp (2006:58–61) seeks to corroborate his findings based on information structure with focus-related linguistic correlates.

One such correlate is the status of the object NP preceding the verb cluster. Definite NPs, including pronouns, tend to represent old information, while indefinite NPs tend to represent new information and, by extension, new information focus. Sapp also examines the intervening NPs (including pronouns and QPs) in VPR constructions. He assumes that an NP that remains inside the VP and is raised, regardless of its definiteness, is focused.Footnote ⁴⁸

A second piece of evidence that Sapp considers is scrambling, which has been shown to correlate with focus in German (see Haider & Rosengren 2003). If an object appears to the left of either an adverbial immediately preceding the verb cluster or a negator, it is scrambled and represents old information, and if it appears to the right, it is not scrambled and represents new or contrastive information.

In the present study, I chose to preliminarily limit the examination of focus to grammatical correlates. Since the occurrence of objects with adverbials or negators is too rare in the Zwirner Corpus data to determine whether objects have been scrambled, I analyzed only the information status of preceding/intervening object NPs.Footnote ⁴⁹

The preceding NPs in the subordinate clause two-verb complexes are assignable to four groups: definite, indefinite, non-NP, and zero. The definite NPs include NPs preceded by a definite article, possessive pronoun, or NumP, as well as proper names and pronouns.Footnote ⁵⁰ The indefinite NPs include bare Ns, NPs preceded by an indefinite article, a QP (jeder ‘every’, kein ‘no’, alle ‘all’), or a quantificational adverb (viel ‘much’, ein bisschen ‘a bit’), as well as etwas ‘something’, etwas + adjective, alles ‘everything’, and nichts ‘nothing’. Finally, adjectives, adverbs, nicht ‘not’, modal particles, and da-compounds comprise the group of non-NPs. Since the syntagms werden passive, perfect with sein, and subjunctive with sein may not occur with object NPs, tokens with these constructions were excluded from the analysis.

The effect of the information status of the preceding object NP on word order is inconclusive. Table 25 shows factor weights for definites and no constituent above and below 0.5, respectively, and for non-NPs and indefinites around 0.5. However, with p = 0.168, the results for the factor group are statistically insignificant.

Table 25.

Effect of the information status of the preceding constituent on word order, subordinate clauses.

The addition of intervening NPs in the VPR constructions to the indefinite NPs results in a factor weight for this factor above 0.5, but this has little effect on the factor weights overall. As can be seen in table 26, the values are even closer to 0.5 than in table 25, and the p-value is higher.

Table 26.

Effect of the information status of the preceding/intervening constituent on word order, subordinate clauses.

Thus, there is a possible effect of object focus on word order in the subordinate clause two-verb complexes as indicated by the occurrence of precluster indefinite NPs and intervening NPs in the VPR constructions. Because of statistical insignificance, however, no firm conclusions may be drawn.

To test the effect of focus in the main clause two-verb complexes, I once again examined the occurrence of precluster NPs. These NPs are assignable to the same four groups as for the subordinate clause complexes: definite, indefinite, non-NP, and zero. The definite NPs include NPs preceded by a definite article, possessive pronoun, or NumP, as well as proper names and pronouns. Included in the group of indefinite NPs are bare Ns, NPs preceded by an indefinite article, a QP (jeder ‘every’, kein ‘no’, alle ‘all’), or a quantificational adverb (viel ‘much’, ein bisschen ‘a bit’), as well as etwas ‘something’, etwas + adjective, manch ‘many a’ + adjective, alles ‘everything’, and nichts ‘nothing’. Finally, adjectives, adverbs, nicht ‘not’, modal particles, and da-compounds comprise the group of non-NPs. Since the majority of tokens involving a participle are passive constructions (perfect of passive [S], passive [MOD]), which may not occur with an object NP, only the groupings infinitive and IPP were included in the analysis.

As table 27 shows, only when no constituent precedes the verb cluster is the 2-3 order favored. In contrast, the occurrence of any constituent, whether a definite or indefinite NP or non-NP, disfavors the 2-3 order.

Table 27.

Effect of the information status of the preceding constituent on word order, main clauses (infinitive and IPP).

When the intervening NPs in the VPR constructions are added to the indefinite NPs, however, the 2-3 order is favored by the absence of a constituent and by indefinites plus intervening NPs, and disfavored by definites and non-NPs. These results are shown in table 28.

Table 28.

Effect of the information status of the preceding/intervening constituent on word order, main clauses (infinitive and IPP).

Thus, it may be concluded for the main clause two-verb complexes that object focus, as indicated by the occurrence of precluster indefinite NPs and intervening NPs in the VPR constructions, has an effect on word order in the Zwirner Corpus data.

4.4. Extraposition and Two-Verb Complexes

The final factor group tested in the statistical analysis is extraposition. Sapp (2006:57–58) demonstrates that the rate of 1-2 in his corpus is far higher than the expected rate when some constituent is extraposed, but very close to the expected rate when no extraposed constituent appears to the right of the verb cluster. However, Sapp's (2007:303–304) analysis reveals that when extraposed adjunct PPs and argument NPs/PPs are tested separately, the latter but not the former favor the 1-2 order. Since Bies (1996) shows that extraposed arguments in her ENHG corpus tend to be heavy or focused, Sapp (2007:304) concludes that the correlation in his data between the 1-2 order and argument extraposition can be attributed to focus, not to extraposition in general. He takes these findings as a further piece of indirect evidence for the effect of focus on word order.

The effect of extraposition on the 1-2 order in the subordinate clause complexes is shown in table 29.

Table 29.

Effect of extraposition on word order, subordinate clause complexes.

Factor weights for extraposed arguments and adjuncts PPs indicate that these have a disfavoring effect on the occurrence of the 1-2 order, while the factor weight for other extraposed constituents points to a slight favoring effect. No firm conclusions may be drawn, however, since the results are statistically insignificant with p = 0.908.

The results of the analysis of extraposition in the main clause complexes are presented in table 30. Recall from section 3.6 that no argument NPs are attested in these data.

Table 30.

Effect of extraposition on word order, main clause complexes.

In contrast to extraposed adjunct PPs, extraposed argument PPs and other extraposed constituents have factor weights that indicate a favoring effect on the 2-3 order. However, no firm conclusions may be drawn since the results are statistically insignificant with p = 0.806.

4.5. Summary of Main Findings

Four factor groups (syntagm, verbal prefix type, focus, and extraposition) were tested with the statistical analysis program GoldVarb to determine their effect on word order in the subordinate and main clause two-verb complexes. The factor groups syntagm and verbal prefix were shown to have statistically significant effects on word order in both the subordinate and main clause complexes. Focus, as indicated by the occurrence of pre-cluster indefinite NPs and intervening NPs in the VPR constructions, has a statistically significant effect on word order only in the main clause complexes with the groupings infinitive and IPP. Finally, extraposition has no statistically significant effect on word order in either the sub-ordinate or main clause complexes.

A glance at the corpus and the factor groups that have a statistically significant effect on word order in the subordinate and main clause two-verb complexes reveals that 156 out of 159 complexes with the 1-2 order and all 145 out of 145 complexes with the 2-3 order can be linked to one or more of these factor groups. One-hundred forty-nine out of 159 subordinate clause two-verb complexes and 136 out of 145 main clause two-verb complexes involve a syntagm or verb form that favors the 1-2 or 2-3 order, respectively. As for the factor group verbal prefix type, the rates are 75 out of 159 subordinate clause two-verb complexes and 49 out of 145 main clause two-verb complexes. Finally, the factor group focus can be linked to the 2-3 order in 45 out of 135 main clause two-verb complexes in the groupings infinitive and IPP.

Recall from section 1 that Lötscher (1978) proposed the interplay of grammatical, functional, and performance rules to account for word order variation in the verbal complex. In sections 4.1–4.3, we saw how word order in the verbal complex is influenced by the linguistic (grammatical) factor groups syntagm and verbal prefix type as well as the functional factor group focus. While performance has not yet come to the fore, one important finding from section 3.5 was the high rate of VPR in the main clause two-verb complexes. In section 6, I argue that the occurrence of VPR, alone and in conjunction with extraposition, which in this section was not conclusively shown to affect word order, can be attributed in part to the third set of rules in Lötscher's account, namely performance rules. First, however, let us turn to a discussion of the multi-causal account of word order in the verbal complex proposed in Lötscher 1978.

5.1. Grammar

Lötscher (1978:12) formulates a grammatical rule in terms of the position of the verbal elements in a complex in relation to one another. In effect, the instantiation of this rule arranges the verbal elements in their order of subcategorization from right to left, as in standard German for the majority of syntagms, to yield a left-branching structure.Footnote ⁵⁵ The basic orderings derived through this rule may then be altered through inversion in interaction with a performance rule and are subject to functional rules (Lötscher 1978:12–13, 17–18).

I propose here that Lötscher's (1978) grammatical rule be modified and expanded to encompass a rule and factors that interact and feed verbal complex formation. First, a grammatical rule defines, for a particular grammar, the range of word order variation in the verbal complex.Footnote ⁵⁶ This moves us already beyond the scope of a purely core-syntactic investigation concerned with exactly how the orders permitted and not permitted may be predicted, motivated, and constrained. In contrast to such an investigation, I take for granted the range of variation for a given grammar.Footnote ⁵⁷ A set of grammatical factors then constrains the orders defined by the grammatical rule. In sections 4.1 and 4.2, we saw that in the Zwirner Corpus data, the grammatical factor groups syntagm and verbal prefix type have statistically significant effects on word order in subordinate and main clause two-verb complexes. Given a set of all possible word orders in a particular grammar, there will be a number of subsets of these possible word orders, linked to different syntagms.Footnote ⁵⁸ The factor group verbal prefix type may then influence the occurrence of a particular word order within these subsets of word orders.

In the Zwirner Corpus data, for example, two word orders are possible in the subordinate clause two-verb complexes: 2-1 and 1-2. Not every syntagm allows both these orders, however; while both are attested with INF + MOD, for example, only the 2-1 order is attested with PTC + kriegen.Footnote ⁵⁹ Which of the two word orders, 2-1 or 1-2, is realized in complexes with the syntagm INF + MOD is further determined in part by the type of verbal prefix on the lexical verb.

5.2. Function

Lötscher suggests that functional rules also determine word order in the verbal complex. Lötscher (1978:4–6, 17–18) discusses a correspondence between certain word orders, specifically in main clause verbal com-plexes involving a modal and lexical verb plus a non-verbal element, and the position of both sentential accent and rhematic/thematic material.

Sapp (2007) more closely examines the relationship between word order and function—focus in particular. As discussed in section 4.3, Sapp found that in his ENHG corpus, new information and contrastive object focus favor the 1-2 order in subordinate clause two-verb complexes, while in subordinate clause three-verb complexes, focus favors all attested orders but 3-2-1 when taken together. Sapp supports this finding with evidence from scrambling, extraposition, and experi-ments testing the effect of focus on word order in modern German (see Sapp 2007:305–315). As for the Zwirner Corpus data, it was shown in section 4.3 that there is a statistically significant link between focus, as indicated by the occurrence of precluster indefinite NPs and intervening NPs in the VPR constructions, and the 2-3 order in 45 out of 135 main clause two-verb complexes in the groupings infinitive and IPP.

5.3. Performance

The last of Lötscher's (1978) rules deals with the connection between word order and performance. Recall that this rule allows structures that in terms of performance are too complex to be restructured to facilitate processing. On the basis of psycholinguistic evidence indicating that left-branching structures are more difficult to parse than right-branching ones, Lötscher (1978:12–17) proposes a rule to invert the elements in the strictly left-branching verbal complexes generated by the grammatical rule described in section 5.1 above. This inversion rule applies not only to verbal elements, but non-verbal elements as well, allowing for the derivation of VPR constructions. Similarly, Kaufmann (2007:157–158, 198), following Haider 2003:91, 119–123, associates V(P)R structures with lower costs in parsing. He concludes that “the raising of VPs may be the speaker's attempt to reduce the hearer's processing load by producing less complex, right-branching structures” (p. 198).

The assumption that right branching structures are easier to parse than left-branching ones would seem to rest on the notion that the structural and semantic/thematic analysis of arguments is postponed in left-branching phrases, where the occurrence of the head is delayed. This delay places a load on working memory that can be alleviated through mechanisms that enable earlier analysis, for example, through V(P)R. These assumptions are compatible with head-driven approaches to parsing, as advocated by Pritchett (1991). However, it has been argued, on the basis of more recent psycholinguistic evidence from German (Bader & Lasser 1994) and Korean (Koh 1997) and confirmed, for example, in later studies of Japanese (Aoshima et al. 2003, Kamide & Mitchell 1999 among others), that the structural analysis of arguments occurs before the appearance of the verbal head, which belies a head-driven approach. Furthermore, the semantic/thematic interpretation of arguments has been shown to proceed incrementally (see, for example, Bornkessel et al. 2003 and Schlesewsky et al. 2003). If an incremental pre-head attachment account is assumed to be correct, then arguments for the advantage of right-branching V(P)R constructions for the hearer are weakened. In addition, an account that assumes the production of V(P)R structures for the benefit of the hearer entails speaker altruism, the notion that the speaker “packages” utterances in ways easiest for the hearer to parse, which has been shown in the literature to be problematic (see Kirby 1999:60–62, 135–136).Footnote ⁶⁰

Alternatively, the occurrence of V(P)R constructions may be associ-ated with lower derivational costs of some sort for the speaker, as Kaufmann (2007:157–158) also suggests. In section 6, I first outline a speaker-oriented account of VPR based on principles of language processing proposed in Hawkins 1994 and 2004, linking this process to a productive syntactic operation in spoken German (including the Zwirner Corpus data), namely extraposition.Footnote ⁶¹ I then place VPR in the broader context of speaker-hearer dichotomies, specifically those linked to bracketing, a salient characteristic of German.

6.1. Minimize Domains: VPR and Extraposition in the Zwirner Corpus

With the tools of the IC-to-word ratio calculation in hand, let us now examine how Hawkins' principles of MiD and EIC can account, in part, for constructions involving VPR formation and extraposition attested in the Zwirner Corpus data. For the derivation of the sample sentences below, I assume Haegeman's (1992) analysis, which treats VPR as rightward adjunction of the VP with its arguments to a maximal pro-jection, preceded by optional scrambling. Furthermore, I adopt Frey & Pittner's (1998) analysis of adverbials, which, as they argue, are gener-ated in different base positions according to semantic class and may undergo scrambling. Full underlying and surface representations for the sample sentences are provided in the footnotes; however, to facilitate EIC calculations, simplified structures appear in the main text.Footnote ⁷⁰

I begin with a main clause involving both extraposition and VPR:Footnote ⁷¹

1. (89)

Based on this derivation, I posit the clausal structure in 90 as the “basic” order associated with 89.Footnote ⁷²

1. (90)

This sentence, comprising the projections CP, VP₂, and VP₃, consists of the following ICs:

1. 1. CP = C (habe), VP₂

2. 2. VP₂ = V₂ (sollen), VP₃

3. 3. VP₃ = V₃ (gehen), PP₁, PP₂

I assume that C constructs CP, V₂ VP₂, V₃ VP₃, P₁ (mit) PP₁, and P₂ (in) PP₂. Given these ICs and their constructors, the IC-to-word ratios and aggregate score for the basic sentence in 90 are as follows:

1. (91)

I now calculate the ratios and aggregate score for a theoretical sentence with extraposed PP₁ that has been attached to VP₂. Following Hawkins 1994 and 2004, I assume discontinuous constituency.Footnote ⁷³

1. (92)

If we compare the aggregate scores in 91 and 92, we see that the extraposition of PP₁ results in an improvement of the aggregate score over the score for the competing basic clause, like CP extraposition in 88 above.

The aggregate score calculations for a second theoretical sentence with VPR (raising of VP₃) alone and the attested sentence with both VPR and extraposition are as follows:Footnote ⁷⁴

1. (93)

2. (94)

We see that in 93, VPR alone results in an aggregate score only slightly higher than the score for the basic sentence, but in 94, VPR and extra-position together show an additive effect with a greater improvement in aggregate score than with either operation alone.

Next, I calculate the aggregate scores for a pair of attested main clauses with similar S-structures that display VPR alone.Footnote ⁷⁵

1. (95)

2. (96)

I assume that 95 and 96 are related to the basic sentences in 97 and 98, respectively.

1. (97)

1. (98)

I begin with 97, which consists of three projections: CP, VP₂, and VP₃. For present purposes, we need be concerned with only CP and VP₂. The ICs of these phrases are as follows:

1. 1. CP = C (hatte), VP₂

2. 2. VP₂ = V₂ (müssen), VP₃

I assume that C constructs CP, V₂ VP₂, and V₃ (greifen) VP₃.

Below are the calculations of the aggregate scores for the basic 99 and attested 100 sentence with VPR.

1. (99)

2. (100)

As in 93, in 100 the occurrence of VPR alone, involving a VP with an embedded PP, improves the aggregate score only slightly.

For 98, I assume the same projections and ICs as for 97.

1. 1. CP = C (hat), VP₂

2. 2. VP₂ = V₂ (müssen), VP₃

However, there is a difference between these sentences in the constituent structure of VP₃ and the way in which this phrase is constructed. While in 97 the ICs of VP₃ are V₃ and PP, and V₃ constructs VP₃, in 98 the ICs of VP₃ are V₃, DP₁, and DP₂. Furthermore, as Hawkins (1994:393–397) argues, an accusative- or dative-marked determiner can construct VP.Footnote ⁷⁶ Thus, in 98 above, VP₃ is constructed upon the occurrence of the dative-marked D₁ (dem) of DP₁. Based on these assumptions, then, the fol-lowing are the aggregate score calculations for the basic sentence 101 and the attested sentence with VPR 102.

1. (101)

2. (102)

We see that VPR with left-edge construction of VP₃ results in a far improved aggregate score.

I now calculate aggregate scores for the following subordinate clause, which displays both extraposition and VPR:Footnote ⁷⁷

1. (103)

Based on this derivation, I posit the clausal structure in 104, in which AgrP and TP have been conflated as IP for expository reasons, as the basic order associated with 103.

1. (104)

As we can see, the sentence in 104 consists of four projections, CP, IP, VP₂, and VP₃, with the following ICs:

1. 1. CP = C (wo), IP

2. 2. IP = I (soll), DP₁, VP₂

3. 3. VP₂ = V₂ (kriegen), VP₃

4. 4. VP₃ = V₃ (abgemacht), Adv₁, Adv₂, PP, DP₂

I assume that C constructs CP, V₂ VP₂, V₃ VP₃, P (mit) PP, and D₂ (ein) DP₂. The construction of IP, on the other hand, is less straightforward. As the phrasal head, I can construct IP. Additionally, Hawkins (1994: 384–385, 398) proposes that in languages such as German, S (IP) can be constructed by the complementizer C or by a nominative-marked NP (DP) at the left edge of S.Footnote ⁷⁸ Therefore, both wo and ich can also construct IP in 104.

Below, I present the aggregate score calculations for four sentences: the basic sentence 105, theoretical versions with extraposition 106 and VPR 107 alone, and the attested sentence 108 with both extraposition and VPR. To simplify the calculations, I do not include the ratios for CP and VP₂ since these remain constant, and I assume that DP₁ constructs IP.Footnote ⁷⁹

1. (105)

2. (106)

3. (107)

4. (108)

As we can see, extraposition alone results in an aggregate score higher than that of the basic sentence, but VPR alone results in a lower score. Additionally, we see that the score for the sentence with extraposition and VPR together is identical to that for the sentence with extraposition alone.

For cases in which aggregate scores for competing structures are non-optimal and need to be further distinguished since they are very close or identical, Hawkins (1994:82–83) proposes an on-line EIC metric that takes into account not only the size of the PCD but also the position of the ICs within the domain. The idea behind this metric is the preference for a leftward skewing of ICs within a domain.Footnote ⁸⁰ On-line calculations for the four sentences are as follows, again including ratios only for those projections for which they do not remain constant, namely IP and VP₃:

1. (109)

2. (110)

3. (111)

4. (112)

The aggregate scores reveal that if the position of ICs within each PCD is taken into account, then both extraposition and VPR alone result in a score higher than that of the basic sentence, and, in combination, the highest score.Footnote ⁸¹

Finally, I calculate aggregate scores for a subordinate clause that displays only VPR.Footnote ⁸²

1. (113)

I posit that 113 is related to the basic sentence 114 with conflated IP.

1. (114)

114 consists of three projections, CP, IP, and VP, with the following ICs:

1. 1. CP = C (dass), IP

2. 2. IP = I (müssen), DP_1,VP

3. 3. VP = V (kochen), Adv, DP₂

I assume that C constructs CP, DP₁ IP, accusative-marked DP₂ VP, and D₂ (einen) DP₂.

In the following, I present the aggregate scores for the basic sentence 115 and attested sentence 116, based on ratios for only the relevant phrases, IP and VP.

1. (115)

2. (116)

A comparison of the scores for 115 and 116 reveals an improved score for the attested VPR construction over that for the competing basic order.

In conclusion, the example calculations presented above suggest a potential processing-related motivation for VPR, namely more efficient language production through domain minimization and/or leftward skewing of ICs within a domain. In this respect, VPR, an operation involving rightward movement, is related to the more general syntactic operation of extraposition, which also involves rightward movement and can result in more efficient structures. Extraposition was shown in section 3.6 to be productive in WCG verbal complexes. In addition, a sample of 1,082 subordinate and main clauses without a verbal complex from 80 Zwirner Corpus transcripts display extraposition in 366 (33.8%) of these. To be sure, I am not arguing here that the occurrence of VPR constructions in production—let alone the grammaticalization of VPR—can be predicted solely on the basis of MiD.Footnote ⁸³ Rather, efficiency in language production is just one of the factors that can account for the occurrence of VPR.