Do orders cross borders?
Acquiring German verb clusters
Dorothee Bliem
*Thesis Research Master’s in Linguistics
Supervisor: prof. dr. Fred Weerman
Second reader: dr. Jan Don
July 3, 2015
Abstract
This paper presents the results from a sentence repetition task (SRT) conducted with 27 L1 (M=7;6) and 26 early L2 (M=7;3) speakers of Austrian German in their acquisition of verb clusters. Previous studies (Meyer & Weerman 2014 for Dutch, De Haan et al. 2014 for Dutch-‐Frisian bilingualism) suggest that ascending orders such as wil lezen (‘want to read’; lezen wil, respectively for the descending alternative) are crucial for the acquisition of this syntactic phenomenon, as they lead to the application of a general verb raising rule. In contrast to Dutch, however, German exhibits quite a rigid descending system and allows for (partly) ascending orders only in complex tripartite constructions, which are furthermore acquired later than bipartite constructions (Meyer & Weerman 2014). In this paper I show that in German the acquisition of order differences among tripartite clusters does not affect the acquisition of bipartite clusters, but that it may influence the acquisition of rigid tripartite clusters. This suggests that in contrast to Dutch, at most the latter are acquired via a verb raising rule. From a theoretical point of view this means that whereas German tripartite clusters can be explained by a verb raising analysis (cf. Evers 1975), the formation of bipartite clusters must be explained in a different way.
1 Introduction
In recent decades, verb clusters have been subject to on-‐going theoretical debates within both generative (e.g. Evers, 1975; Haegeman & Van Riemsdijk) and functionalist frameworks (e.g. De Sutter, 2005; Arfs, 2007). The inseparable sequences of verbs at the end of (West-‐) Germanic
* I would like to express my gratitude to my supervisor prof. dr. Fred Weerman for his guidance and support throughout the
entire course of this project. Furthermore I would like to thank dr. Jan Don for agreeing to serve as the second reader for this thesis. I am moreover indebted to Caitlin Meyer, who gave me valuable suggestions and very useful feedback. I would also like to thank Suzana Ilic for her help with the experiments, Lena Krämer for lending me her voice for the audio recordings, and, last but not least, all of the children from the Volksschule Wörgl who participated in my experiment.
subordinate clauses mainly received attention for the huge variation across, and also within these languages (cf. Wurmbrand 2005). Research on the acquisition of these clusters, however, is still in its infancy. So far, claims have only been put forward based on data from Dutch L1 acquisition (Zuckerman 2001, Meyer & Weerman 2014) and Dutch-‐Frisian bilingualism (De Haan, Faber, Meyer & Weerman 2014). All of these studies suggest that so-‐called ascending cluster orders (e.g. wil lezen ‘want to read’ and wil hebben gelezen ‘want to have read’) are crucial in the acquisition of verb clusters. Whereas these orders were overgeneralised by Dutch Kindergardners, descending cluster orders (e.g. lezen wil ‘want to read’ and gelezen hebben wil ‘want to have read’) were found to be vulnerable. Based on these findings, Meyer & Weerman (2014) propose that Dutch verb clusters are acquired according to one general rule, rather than by the application of several construction-‐specific ones.
Assuming that ascending orders indeed provide such an important trigger for the acquisition of verb clusters, a question arises of how verb clusters are acquired by children, whose L1 provides them with less evidence for ascending orders. Whereas Dutch allows for both ascending and descending orders in most cluster types, German only allows for (partly) ascending orders in complex types of tripartite clusters (IPP constructions and future tense forms of modal verbs). Among all other cluster types, only descending orders (e.g. lesen will ‘to want to read’ and gelesen haben will ‘gelesen haben will’) are grammatical. As tripartite constructions are acquired relatively late, however, this means that acquirers of German not only encounter less evidence for ascending orders, but also start to analyse them at later ages than their Dutch peers.
It shall be the aim of this paper to shed more light on the acquisition of verb clusters in Austrian German (henceforth: AG). By the means of an SRT with both L1 and early L2 acquirers of AG, I also want to investigate whether German verb clusters are acquired via the application of a general rule and, if so, whether this affects both bi-‐ and tripartite clusters. Or, whether the abovementioned differences lead to a construction-‐specific acquisition of different cluster types.
The children’s performance clearly shows that the acquisition of bipartite clusters is not dependent on order differences among tripartite clusters. Though the overall results are not as straightforward with respect to tripartite clusters, an individual analysis reveals a pattern, which suggests that tripartite clusters may be acquired via such a general rule. These findings also yield consequences for a more theoretical analysis, since they imply that at least the formation bipartite clusters cannot be ascribed to verb raising.
The present paper is structured as follows: I will first give an overview of German verb clusters and introduce some theoretical notions that will be recurrent in the subsequent sections. I will then present two types of arguments for the vulnerability of descending clusters orders. My research question is formulated in Section 3. Section 4 provides information about the sentence
repetition task and the participants. The results will be presented in Section 5 and linked to my research question in Section 6. I will summarize my findings in a conclusion.
2 Theoretical Background
Verb clusters are usually defined as inseparable sequences of two or more verbs in clause-‐final position (cf. Evers 1975). They are among others attested in the West-‐Germanic O-‐V languages, mainly German, Dutch, Frisian, West-‐Flemish and Afrikaans (Wurmbrand 2004). A striking property about these clusters is the variation in word order, which is found not only between, but also within the verb cluster languages. Depending on the construction, i.e. the number and types of verbs involved (e.g. modals, auxiliaries, participles), different orders may be required within one language. One construction may furthermore allow for two or more orders as semantically identical alternatives, which adds the notion of optionality to the already puzzling variation within and across verb-‐cluster languages. It is beyond the scope of this paper to give a full account of the variation and optionality in all of the abovementioned languages. For an extensive overview, I refer to Wurmbrand (2004). As the acquisition of (Austrian-‐) German (henceforth AG) constitutes the centre of the present study, I will limit my elaborations in section 2.1 to this variety, sometimes drawing comparisons to verb clusters in Dutch. Besides an overview about verb clusters in AG, the following part will introduce the theoretical notion of verb raising, which will be recurrent in the subsequent sections. In Section 2.2, I will discuss the verb cluster variation in terms of vulnerability. I will present arguments from language contact and first language acquisition, which both suggest that the order, which is found to be most frequent in German, should be vulnerable.
2.1 German verb clusters & verb cluster formation
With only one grammatical word order in 2-‐verb (or bipartite) constructions, German is a language that displays a very rigid verb cluster pattern. Bipartite clusters can be formed by both combinations of infinitives and modals (henceforth: INF-‐MOD), as well as participles and auxiliaries (henceforth: PART-‐AUX). The following examples are taken from Bader & Schmid (2009).
(1) INF-‐MOD cluster
…dass Peter ein Buch lesen muss. …that Peter a book read.INF must ‘...that Peter must read a book’
(2) PART-‐AUX cluster
…dass er dir geholfen hat. …that he you help.PART has ‘...that he has helped you’
In these constructions, the finite verb strictly has to appear in clause-‐final position. Note that this is in line with the underlying O-‐V word order, according to which complements strictly have to precede the finite verb in subordinate clauses:
(3) ...dass er einen Apfel isst. ...that he an apple eats. ‘...that he is eating an apple’
Also Dutch is considered an underlying O-‐V language (cf. Koster 1976). The order among Dutch verb clusters, however, is less strict. Besides the order attested in German, also the reverse is possible in Dutch bipartite clusters:
(4) INF-‐MOD cluster
a. ...dat Peter een boek lezen moet. ...that Peter a book read.INF must
b. ...dat Peter een boek moet lezen. ...that Peter a book must read. INF
both: ‘...that Peter must read a book’
(5) PART-‐AUX cluster
a. …dat hij jou geholpen heeft. …that he you help.PART has
b. …dat hij jou heeft geholpen. ...that he you has help.PART both: ‘...that he has helped you’
These examples only give a brief impression of different types of bipartite clusters and the variation, which is found with respect to the verb order. Whereas (1) and (2) illustrated the rigidity of German
clusters, (4) and (5) showed examples for a less rigid system, as in Dutch bipartite clusters two orders can be used interchangeably.
As far as the formation of verb clusters is concerned, different theories have been put forward. Probably the first is Evers’ (1975) verb raising approach, which implies a process of movement. 1
The Dutch examples in (4) and (5), for instance, already suggest that movement must have taken place, since the surface order seems to contradict the O-‐V word order. For German, however, verb raising would imply a process of vacuous movement. But why would we assume such a process, if the same order may easily be explained by a mere O-‐V analysis?
Evers (1975) defends this analysis by referring to certain characteristics of the verbal sequences. As already mentioned in the first part of this section, inseparability is a defining criterion for verb clusters. Sentences like (6), for example, where the verbs are interrupted by an adverb, are not grammatical in German: 2
(6) ...*dass Peter ein Buch lesen schnell muss. … that Peter a book read fast must
In order to account for this observation, Evers (1975) suggests that also in German, the impenetrable sequences result from a verb raising rule. According to this rule, the head of the complement of the finite-‐verb has to move up in the structure and attach to the matrix-‐verb. Whereas in German bipartite constructions the verbal complement only attaches to the left of the finite verb (also called the descending or 2-‐1 order), examples (4) and (5) have illustrated that both left-‐ and rightward attachment are possible in Dutch, i.e. both 2-‐1 and 1-‐2 orders are grammatical. The numbers are used in order to refer to the verbs’ level of embedment – the higher the number, the deeper the verb is embedded in the clause. The following figure illustrates the process of verb raising with the German example already given in (1), repeated here for convenience:
(1) V-‐MOD cluster
…dass Peter ein Buch lesen muss. …that Peter a book read.INF must
‘...that Peter must read a book’
1 Verb cluster formation is not the only syntactic phenomenon that Evers (1975) ascribes to verb raising. He proposes that
also gapping and quantifier hopping, among others, can be attributed to this process. I refer to his dissertation (1975) for further elaborations on the topic.
2 Note that among the verb cluster languages, certain dialects allow for a restricted set of elements to occur within the
(1a) depicts the underlying order before verb raising has taken place. In the process of verb raising, the verb moves out of its original position and attaches to the matrix verb. As in German, the attachment is leftward, this yields a structure as in (1b).
So far, I have limited my elaborations to clusters containing two verbal elements. However, verb clusters can theoretically be derived with an infinite number of verbs. Then, verb raising is repeated until no more verbs can be raised. (7) illustrates this with a 3-‐verb (or tripartite) cluster:
(7) V-‐MOD-‐MOD cluster
…, dass sie das Buch lesen können muss. …, that she the book read can must ‘...that she must be able to read the book’
(7) illustrates the formation of a V-‐AUX-‐MOD cluster. The finite modal selects another modal verb, which in turn selects the infinitival form. The elements again appear in a strictly descending order. The same holds for AUX-‐AUX-‐V (8) and PART-‐MOD-‐AUX constructions (9). (8a) and (8b) illustrate that also in tripartite clusters, intervening elements are ungrammatical in German:
VP1
VP2
das Buch lesen muss
VP1
VP2
das Buch ti
leseni muss a. before raising b. after raising
a. before raising b. after raising
VP2 muss
VP3 können
das Buch lesen
VP1
VP2 leseni könnenj muss
VP3 titj
das Buch ti
(8) AUX-‐AUX-‐V cluster
a. …dass das Rätsel schnell gelöst worden ist. …that the puzzle fast solve.PART PASSIVE is ’...that the puzzle has been solved’
b. ...*dass das Rätsel gelöst schnell worden ist/gelöst worden schnell ist. ... that the puzzle solved. PART fast PASSIVE is/solved.PART PASSIVE fast is
(9) PART-‐MOD-‐AUX cluster
a. …dass er das Buch schnell gelesen haben muss. …that he the book fast read. PART have must ‘…that he must have read the book’
b. ...*dass er das Buch gelesen schnell haben muss/gelesen haben schnell muss. ... that he the book read. PART fast have must/read. PART have fast must
(8) and (9) show that also German tripartite clusters come in strict descending orders. There are two constructions, however, that allow for deviances from the O-‐V word order. This is the case, when complex tense forms of modal verbs are involved. Then, the surface order may differ from the underlying order, which, following Evers (1975), indicates a process of non-‐vacuous movement.
In AG, deviances are restricted to IPP constructions (infinitivus-‐pro-‐participio) and future tense forms of modal verbs. Wurmbrand (2005) shows that possible orders vary across German dialects. The following is based on a judgment task by Wurmbrand (2005), who shows that also in AG, orders other than 3-‐2-‐1 are accepted:3
(10) Future tense form of modal verbs
a. …dass Peter ein Buch lesen können wird. (3-‐2-‐1) …that Peter a book read.INF can will
b. ...?dass Peter ein Buch wird lesen können. (1-‐3-‐2)
3 A question mark indicates that 75% of Wurmbrand’s informants accepted or were uncertain about the order. The %
indicates that this was the case for at least 50% of her participants.
… that Peter a book will read.INF can
c. ...%dass Peter ein Buch lesen wird können. (3-‐1-‐2) … that Peter a book read.INF will can
all: ‘...that Peter will be able to read a book’
(10a) shows that also with this construction, the standard 3-‐2-‐1 order is perfectly grammatical in AG. Besides this order, Wurmbrand’s informants also accepted other orders, albeit to a more limited extent: (10b) shows that the 1-‐3-‐2 order was judged to be grammatical by at least 75% of her participants. About half of them also accepted the 3-‐1-‐2 order.
Optionality is furthermore attested in so-‐called IPP constructions (infinitivus-‐pro-‐participio). The clusters do not only stand out for their deviant word order, but also because we encounter an unexpected form in these constructions: As the name indicates, an infinitival form has to be used in order to express the prefect tense of modal verbs (see Schmid 2005 for an extensive discussion of the IPP effect). Given the perfect-‐tense meaning, one may expect sentences like (11)
(11) …*dass Peter das Zimmer aufräumen gemusst hat. … that Peter the room tidy.INF have.to.PART has ‘...that Peter had to tidy the room’
Instead of the participle gemusst, however the infinitival form müssen has to be used. This construction allows for the following orders in AG (judgements again based on Wurmbrand 2005):
(12) Orders attested among IPP constructions
a. ...?dass Peter das Zimmer aufräumen müssen hat (?3-‐2-‐1) ... that Peter the room tidy.INF must.INF has
b. ...dass Peter das Zimmer aufräumen hat müssen (3-‐1-‐2) ...that Peter the room tidy.INF has must.INF
c. ...dass Peter das Zimmer hat aufräumen müssen (1-‐3-‐2) ...that Peter the room has tidy.INF must.INF
all: ‘…that Peter had to tidy the room’
The examples in (12) show that the descending order is considered marginal, whereas the deviant orders 3-‐1-‐2 and 1-‐3-‐2 both received better judgments from Wurmbrand’s (2005) informants.
Linking this to Evers’ (1975) verb raising analysis, the examples given in this section show that German distinguishes between two main types of tripartite clusters: One type where movement is vacuous and the surface order equals underlying order; and one type where movement can also result in (partly) ascending orders. Whereas most cluster types belong to type one, there are two kinds of constructions allowing for deviances from the standard 3-‐2-‐1 order. The following table will summarize the different types of verb clusters with their respective orders order(s) in AG:
Construction Verb types Example Attested orders
V-‐MOD INF-‐FIN dass Peter singen kann 2-‐1
V-‐AUX PART-‐FIN dass Peter gesungen hat 2-‐1
MOD-‐MOD-‐V FIN-‐INF-‐INF dass Peter singen können muss 3-‐2-‐1 MOD-‐AUX-‐V FIN-‐INF-‐PART dass Peter gesungen haben muss 3-‐2-‐1 AUX-‐AUX-‐V PART-‐INF-‐FIN dass Peter gewählt worden ist 3-‐2-‐1 AUX-‐MOD-‐V FIN-‐INF-‐INF dass Peter ein Lied singen können wird
dass Peter ein Lied singen wird können dass Peter ein Lied wird singen können
3-‐2-‐1 %3-‐1-‐2 ?1-‐3-‐2 AUX-‐MOD-‐V FIN-‐IPP-‐INF dass Peter ein Lied hat singen müssen
dass Peter ein Lied singen hat müssen ?dass Peter ein Lied singen müssen hat
1-‐3-‐2 3-‐1-‐2 ?3-‐2-‐1
Table 1. Possible orders among verb clusters in Austrian German
2.2 Vulnerability of 3-‐2-‐1
The fact that verb-‐clusters are subject to a lot of language and dialect variation has led to many debates about the very nature of these clusters. In the previous section I have presented one theoretical viewpoint by introducing Evers’ (1975) verb raising analysis. Besides the question about how verb-‐clusters are formed, however, also issues about the complexity of different verb orders have been addressed. Different implications follow, for example, depending on whether one takes a head-‐final or a head-‐initial approach (see Wurmbrand 2005 and Barbiers 2008 for an overview of different analyses). Regardless of the theoretical background assumptions, however, there are two types of evidence suggesting that descending (3)-‐2-‐1 orders represent vulnerable orders. Having demonstrated that German shows a very rigid descending pattern, this may at first sight seem
counter-‐intuitive. Nevertheless, observations among the neighbouring languages English, Dutch and German, as well as empirical investigations on L1 acquisition point to the fact that what is found to be so stable in German should indeed be vulnerable.
Weerman (2006) regards the variation among the West-‐Germanic languages from a perspective of language contact and links this to the theory of the so-‐called “Germanic Sandwich”. This “Germanic Sandwich” is formed by the languages German, Dutch and English, which have experienced different degrees of L2-‐contact throughout their history (see Van Haeringen 1956 for a comparative overview of the three languages). Whereas German has developed in a rather isolated way, English has had a long history of L2-‐contact. Dutch is located between the two languages in terms of contact and therefore constitutes the middle-‐part of the sandwich. The literature suggests (Trudgill 2001, Weerman 2006, among others) that L2-‐contact has a strong influence on a language’s complexity features: Whereas a high degree of L2-‐contact, i.e. a high number of second-‐language learners, may drive grammatical changes into the direction of a grammatical system where complex features for L2 acquisition are avoided, this will not be the case in languages with a low degree of L2-‐ contact. Comparing the grammatical properties of English, Dutch and German indeed shows that these languages do not only form a sandwich in terms of L2-‐contact, but also in some of these L2 features: German, for example, has retained three grammatical genders, followed by Dutch with two grammatical genders and finally by English, which has lost all gender distinctions. Similar developments were observed with respect to the expression of plurality: German exhibits a broad variety of plural morphemes as opposed to English with its uniform –s. Though Dutch displays more variability than English, German still stands out in terms of the plural allomorphy. Comparing the verb order in a sequence of more verbs also shows that Dutch is in an in-‐between position between German and English, which both only allow for one order (1-‐2 in English and 2-‐1 in German). As already illustrated in the previous section, both the 1-‐2 and the 2-‐1 order are grammatical in Dutch:
Degree of L2
contact Language Verb order in a sequence Example high low
English 1-‐2 ...that he has read the book
Dutch 1-‐2 2-‐1 ...dat hij het boek heeft gelezen ...dat hij het boek gelezen heeft
German 2-‐1 ...dass er das Buch gelesen hat
Table 2. Sequential verb orders in English, Dutch and German.
Based on these observations, Weerman (2006) suggests that Dutch is in a transitional state from 2-‐1 to 1-‐2. As 2-‐1 is the only grammatical option in the language with the lowest degree of L2-‐contact,
this descending order could therefore be more complex in acquisition than the reverse. The same applies to tripartite clusters.
A way to test this hypothesis is by taking a look at how children acquire these clusters and especially how they deal with optionality in the input. Assuming that children should start out with a default order and that more complex orders should come in relatively late, data from acquisition can help gain valuable insights in this matter. Up to the present day, the number of studies on the acquisition of verb clusters is limited (Zuckerman 2001; Meyer & Weerman 2014 for Dutch L1 acquisition, De Haan et al. 2014 for Dutch-‐Frisian bilingualism).
Zuckerman (2001) and Meyer & Weerman (2014) approach the optionality in Dutch verb-‐ clusters from a viewpoint of learnability. Both studies show that younger children (around age three) prefer 2-‐1 orders, whereas older children (around age five) show a huge preference for 1-‐2 orders. As the younger children prefer the 2-‐1 order, Zuckerman (2001) takes this as evidence for the fact that descending cluster orders are acquired earlier and therefore represent the default option. Though the children participating in Meyer & Weerman’s (2014) sentence repetition task exhibited the same patterns as found in Zuckerman (2001), Meyer & Weerman do not agree with his reasoning. First, because it is unclear why the older children would show preferences for the more complex option, though this was not found to be the most frequent option in adult speech (cf. Arfs, Coussé & De Sutter 2008, De Sutter 2005, Stroop 2009). Second, because Zuckerman’s (2001) analysis does not provide an explanation for what may finally cause the development towards adult-‐like behaviour. Meyer & Weerman (2014) therefore propose multiple stages in the acquisition of verb clusters. They argue that early instances of 2-‐1 are no real clusters yet, but a mere result of the O-‐V word order. At one point, however, children have to analyse the input, which – on a surface level – provides them with exceptions to the O-‐V order. In order to keep the O-‐V rule intact, children resort to the process of verb raising, or cluster formation. Meyer & Weerman (2014) found that once children start to analyse the deviant input, they show a strong preference for the ascending order, which provides them with obvious evidence for verb raising. Starting out with purely verbal constructions (MOD-‐INF), they soon overgeneralize the ascending order to all types of bipartite clusters. This shows that also from a viewpoint of learnability, descending cluster orders are vulnerable. Only after the acquisition of extra-‐grammatical factors, such as prestige or (in-‐)formality (see Coussé, Arfs & De Sutter 2008), children would turn to adult-‐like behaviour and use more 2-‐1 orders again, which, following Evers (1975) are then also analysed to be the result of verb raising.
Meyer & Weerman (2014) also tested their participants for tripartite constructions. Though the responses were too varied for statistical analyses, the same tendencies were observed as for the bipartite constructions, i.e. there was a clear preference for ascending orders.
In a follow-‐up study, De Haan et al. (2014) investigated the acquisition of bipartite clusters by Dutch-‐Frisian bilingual children. They conducted an adapted version of Meyer & Weerman’s sentence repetition task and tested their participants for both Frisian and Dutch bipartite clusters. In contrast to Dutch, Frisian only allows for the descending 2-‐1 order in bipartite constructions. As for the Dutch version of the test, De Haan et al.’s (2014) participants performed like their monolingual peers. In Frisian, however, they found a considerable amount of 1-‐2 orders, which is not grammatical according to the Frisian prescriptive grammar. Given that the interference was unidirectional, i.e. ‘Dutch’ orders were overgeneralized to Frisian but not the other way around, provides further evidence for the theory that descending orders are vulnerable.
3 Research Questions
The previous section has shown that research on both language contact and language acquisition suggests that descending verb cluster orders are vulnerable. I have presented two types of arguments in favour of this hypothesis: The first one stems from Weerman (2006), who observes that the variation in verb cluster orders is in line with other differences between the Germanic Sandwich languages. He ascribes this variation to different degrees of language contact, which entails that the German descending order would represent the most complex option. The second argument addresses this question from a viewpoint of learnability. Meyer & Weerman (2014) found that monolingual Dutch children at kindergarden age overgeneralize ascending clusters, which leads them to conclude that 1) ascending orders are crucial for the acquisition of verb clusters, since they provide the trigger for verb raising and 2) as a result descending clusters are vulnerable, that is they are more difficult to acquire.
This analysis raises several problems for languages such as German, where we find a much more rigid verb cluster system. In Section 2.1 I have shown that German bipartite clusters only allow for the descending 2-‐1 order, which Evers (1975) considers to be the result of vacuous movement. Following Meyer & Weerman (2014), this entails that the German bipartite input lacks the relevant cue to provide a child with evidence for verb raising. In German, this cue can only be found among clusters, which involve more than two verbal elements, even though also there, descending orders are much more frequent: As already discussed in the theoretical background, partly ascending orders are restricted to future tense forms of modal verbs and IPP constructions. Children would therefore have to rely on these constructions in order to infer a general verb raising rule. Besides the fact that the German input provides children with less evidence for verb raising, this also raises an issue related to the age of acquisition: Meyer & Weerman’s (2014) study suggests that tripartite clusters are acquired relatively late, which entails that German children start to analyse (partly) ascending
clusters much later then their Dutch peers. This, in turn, may have an impact on the way in which these clusters are acquired: Whereas in early stages of the acquisition process (until ~age 3), children mainly infer general rules about their L1, late acquisition is more likely to be construction-‐specific (cf. Meisel, 2009).
Given these differences, Meyer & Weerman’s (2014) analysis raises the question of how verb clusters are acquired in German – whether, despite the fact that there is 1) less evidence for ascending orders and 2) the child will analyse them relatively late, verb clusters are acquired via the application of a general rule. Or, whether late acquisition rather leads to a construction-‐specific acquisition of verb clusters. Based on the preceding discussion, I have sketched three main scenarios, which, as I will show, also yield certain consequences for Evers’ (1975) theoretical analysis:
A first scenario may be that, despite the late acquisition, ascending tripartite clusters lead to the application of a general verb raising rule, which would affect both bi-‐ and tripartite clusters. Once children start to analyse ascending orders, they would take these as evidence for verb raising, which they in turn apply on all other types of verb clusters. Following Meyer & Weerman (2014), this yields the prediction that ascending orders would be overgeneralised to both bi-‐ and tripartite constructions. This scenario would support Evers’ (1975) formal analysis, as it would provide evidence for the fact that verb raising is applied all across the board.
Given the fact that ascending orders are restricted to tripartite constructions, however, it may just as well be the case that a general rule is at play, but that this rule is restricted to tripartite constructions only. This is the second possible scenario, which would suggest that ascending orders are important in the acquisition of tripartite clusters, but that bipartite clusters are acquired in a different way. Overgeneralisations of ascending orders are then only expected among tripartite clusters. In contrast to the first scenario, this would call for an adaptation of Evers’ (1975) analysis: Recall that Evers proposes that all types of verb cluster are the result of verb raising. As Section 2.1 has shown, he uses this reasoning to account for several properties of a cluster, such as the fact that no intervening elements are allowed between its constituents. If, however, the acquisition of ascending tripartite clusters does not lead to a general verb raising rule among bipartite clusters, such properties must be explained in a different way. The fact that adverbs, for example, are not allowed to interrupt the cluster, may then be ascribed to other mechanisms.
The third scenario even yields bigger consequences for Evers’ (1975) analysis. Due to the late acquisition, children may acquire different cluster types in a construction specific way, i.e. there would be no such thing as a general rule. This would therefore predict no overgeneralisation of ascending orders whatsoever. If this is the case, both descending bi-‐ and tripartite clusters, in German, cannot be the result of verb raising and must be explained in an alternative way.
In order to shed more light on these issues, I conducted an SRT with L1 and early L2 acquirers of AG. Studies (see Meisel 2009 for an overview of the relevant literature) have shown that early L2 acquirers go through the same developmental stages as monolingual children. Having the same memory capacity as the L1 group, child L2 speakers enable us to investigate earlier stages of the acquisition process due to the later age of onset. This, in turn, may provide us with a broader picture of how verb clusters are acquired in German and deliver insight into which of the three scenarios may offer the best explanation.
4 Method
In this section I will present the method applied for the present project. Section 4.1 will contain some general information about the sentence repetition task that has been conducted with child L1 and child L2 acquirers of AG. In section 4.2 I am going to present the test design and the two conditions that have been included in the experiment. The pilot experiments will be discussed in section 4.3. In Section 4.4 I will give information about the participants. I will elaborate on the recruitment procedure and the administering of the test in Section 4.5.
4.1 Sentence repetition task
As it is the goal of the present paper to shed more light on the acquisition of verb clusters in German, data is needed that informs us about what children can or tend to produce with respect to those clusters. Eliciting such clusters in a traditional production task, however, would be very challenging as, especially when being tested for tripartite clusters, children may be inclined to produce alternative constructions that are less complex. Instead, I decided to conduct a sentence repetition task, which already provides the children with the grammatical construction in question, but still requires them to use their own grammar. Sentence repetition tasks (SRT; also known as elicited imitation tasks) have already been used successfully in language acquisition research as an alternative to traditional production tasks. In an SRT, the children get to hear sentences and have to repeat what they have heard. In order to make sure the children do not only use their memory alone, the sentences have to be constructed in such a way that they are too long or too complex for their memory capacity. When trying to repeat the sentence, the children should therefore produce what is grammatical according to the current state of their grammars (Chomsky 1964, Eisenbeiss
2010, among others). As far as the design of the test is concerned, especially the sentence length has to be considered carefully: On the one hand, the sentences have to be long enough to exceed the child’s memory capacity, but on the other hand, the children should still be able to remember the gist of the sentence. The literature makes different suggestions with respect to the number of words
or syllables (e.g. Montgomery et al. 1978) for different age groups. Based on Zuckerman (2001) and Meyer & Weerman (2014), who also used an SRT in their study on verb clusters, I opted for a sentence length of 10 words. This is considered to be of a medium difficulty for 4-‐6 year old children. Even though the children in the present experiment are somewhat older (6-‐8 years), I decided to keep this sentence length as I expected the tripartite clusters themselves to add more complexity to the sentence. The pilot experiments showed that the sentences seemed to be appropriate for the age group in mind.
In order to make the task less monotonous and more natural, each sentence was accompanied by a picture that matched its semantic content. The pictures moreover had the advantage that the children were not too focused on merely repeating phonological sequences, as their attention was partly drawn to what the pictures depicted. In case they couldn’t remember relevant parts, they were encouraged to use information from the picture to complete the sentence.
4.2 Design
Section 2 has shown that there are several different types of tripartite constructions in German. So far, hardly anything is known about their acquisition. Testing all of them, however, would result in a costly and time-‐consuming experiment with lots of different variables. In order to make the task feasible, I decided to include one cluster-‐type with a rigid 3-‐2-‐1 order (descending V-‐AUX-‐MOD constructions) and IPP constructions, which involve the same kinds of verbs, but differ with respect to the possible orders. The IPP constructions are moreover a special case, as children need to learn that a perfect tense meaning of a modal verb has to be expressed with an infinitival form.
Factors like prosody or kind of modal verb have been kept constant throughout the task. With respect to the latter, only the modal verb müssen (‘to have to’) was used with the rigid clusters. In the IPP condition, I alternated between müssen (‘to have to’) and wollen (‘to want to’), as there were more stimuli in the IPP condition (15 as compared to only 9 in the rigid condition). In Meyer & Weerman’s (2014) study, the type of modal did not influence the children’s performance, so that I did not expect this to be a confounding variable. As far as prosody is concerned, the sentences were presented with the same intonation patterns. Bader & Schmid (2009), however, report that also this factor did not influence their participants’ judgments on the grammaticality of tripartite clusters, suggesting that they are indeed semantically identical alternatives.
With respect to word frequency, I chose verbs and contexts that should be familiar to the children. Most of the sentences are related to school, family and leisure activities. Moreover I have only chosen inseparable verbs, i.e. no verbs that with separable prefixes, as in Dutch, for example,
prefixes are allowed to intervene between clusters (cf. Barbiers et al. 2008). Though this is not attested for German, I decided to exclude this as a possible confounding variable.
Our variables therefore were cluster type (rigid constructions and IPP constructions) and cluster order. In every condition, the child was presented with a number of grammatical and ungrammatical clusters. The stimuli were pre-‐recorded by a native speaker of AG. The reactions to these stimuli (correct repetitions/conversions to other orders) should indicate what preferences the children have with respect to (tripartite) clusters, i.e. what they can produce or what is most economical for them. Altogether, each child had to repeat 36 sentences, among which 24 experimental items and 12 fillers (to which I return below). Moreover, three trial sentences were included at the beginning. I expected that the children would be able to complete the task within 15-‐ 20 minutes, which was confirmed in the pilot experiments.
In the first condition, it was tested how children handle V-‐AUX-‐MOD clusters where only one order is available in the input. Recall that in this condition, the verbs can only appear in a descending order:
(13) Ich glaube, dass er den Film gesehen haben muss. (3-‐2-‐1) I think that he the movie see.PART have must.
‘I think that me must have seen the movie’
I designed 3 sets of stimuli for this condition. Each semantic content was presented once with the grammatical 3-‐2-‐1 order, once with the ungrammatical 3-‐1-‐2 order (which is a possible option in IPP-‐ constructions) and once with the ungrammatical 1-‐2-‐3 order, which is not grammatical in any cluster types. As already mentioned, only the modal verb müssen (‘to have to’) was used in the rigid condition. Table 3 shows a sample set of stimuli:
Order Sentence Cluster
3-‐2-‐1 Hans denkt, dass Fritz den ganzen Tag geschlafen haben muss *3-‐1-‐2 Hans denkt, dass Fritz den ganzen Tag geschlafen muss haben *1-‐2-‐3 Hans denkt, dass Fritz den ganzen Tag muss haben geschlafen
Table 3. Example set of stimuli for the rigid condition.
Altogether, the children were presented with 9 stimuli from this condition, consisting of three different semantic contents and three different cluster-‐orders (see Appendix A1 for a complete