Second Language Research http://slr.sagepub.com/
Where are the minimal trees? Evidence from early Zulu L2 subordination Busi Dube Second Language Research 2000 16: 233 DOI: 10.1191/026765800666067410 The online version of this article can be found at: http://slr.sagepub.com/content/16/3/233
Published by: http://www.sagepublications.com
Additional services and information for Second Language Research can be found at: Email Alerts: http://slr.sagepub.com/cgi/alerts Subscriptions: http://slr.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav Citations: http://slr.sagepub.com/content/16/3/233.refs.html
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Second Language Research 16,3 (2000); pp. 233–265
Where are the minimal trees? Evidence from early Zulu L2 subordination Busi Dube
University of Edinburgh
The Minimal Trees Hypothesis (Vainikka and Young-Scholten 1994; 1996a; 1996b) proposes that second language (L2) initial state grammars lack functional categories because functional categories are not subject to transfer effects. The aim of this article is to argue that, to some extent, functional categories instantiated in the learner’s first language (L1) transfer to the initial state of L2 syntactic development. On the basis of Zulu interlanguage (IL) data on the acquisition of the obligatory declarative complementizer ukuthi (‘that’) by English native speakers, it is argued that Comp contains a null complementizer system which has sufficient syntactic content to generate subordination in the learners’ initial Zulu IL grammar. Regarding transfer of functional categories, it is suggested that null Comp evident at the Zulu L2 initial state is transferred from English, the subjects’ L1.
I Introduction The status of functional categories (FCs) in initial developing grammars is debated in both first language (L1) and second (L2) language acquisition research. In both cases the debate has focused on the nature of initial state grammars with respect to FCs. The issue is whether FCs are present at the onset of the acquisition process or whether these develop gradually (for L2 acquisition, see White, 1996; for L1 acquisition, see Atkinson, 1996). In L2 acquisition, one can identify at least two contrasting views regarding the status of functional projections in initial L2 grammars. Firstly, the Minimal Trees Hypothesis (MTH; Vainikka and Young-Scholten 1994; 1996a; 1996b) proposes that all functional projections – i.e., IP, CP and DP – are absent in the very early stages of L2 development. According to the MTH, L1 functional structure does not form part of the L2 initial mental representation. The initial absence of surface morphology is taken as evidence for the absence of FCs. In contrast, the Full Transfer/Full Access (FT/FA) Address for correspondence: Busi Dube, PO Box 1126, Bramley 2018, RSA; email:
[email protected] © Arnold 2000
0267-6583(00)SR172OA
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
234 Minimal Trees Hypothesis and Zulu L2 acquisition hypothesis (Schwartz and Sprouse, 1994; 1995a; 1995b; 1996) proposes that all functional projections are initially present and fully specified in initial L2 grammars, resembling those allowed in the L1 grammar. Arguments in support for the FT/FA are based on the early presence of syntactic phenomena linked to FCs. (Note that Epstein, Flynn and Martohardjono (1996; 1998) argue for the presence of functional categories (without transfer) at the L2 initial state; they suggest that functional categories are accessed directly from Universal Grammar.) In this article, the issue of transfer of FCs in early L2 grammars is addressed. By looking at the syntax of early L2 subordination, evidence is presented that suggests that the initial English–Zulu interlanguage grammar (ILG) has the functional category COMP (contrary to the claims of the MTH), although this is an Englishlike Top-type C. In a split-comp hypothesis that and null-that occupy distinct complementizer-type positions. The claim is that the lexical complementizer that is in the head C which projects into a CP. Hence it is referred to as a CP-type C. Null-that is hypothesized to be in the head Top(ic) which projects into a Topic Phrase (TopP), hence it is referred to as a Top-type C. For a detailed discussion of a split-comp hypothesis, see Rizzi, 1995; Nakajima, 1996; Müller and Sternefeld, 1993. It is argued that the initial preference of a null-complementizer system is an indication that there are no deficits in syntactic representations at the initial state of L2 syntactic development. The conclusion drawn is that initial L2 grammars do not differ from mature state grammars with respect to the availability of syntactic positions (as the MTH proposes), but rather with respect to the availability of phonetically null functional heads which are generally not allowed in mature state grammars that require lexical realization of functional heads. II Minimal trees hypothesis The main thrust of the MTH is that what transfers from the L1 into L2 acquisition is confined to lexical projections and their linear orientation. Thus, an initial L2 state consists of lexical projections only. Subsequent development takes the form of a gradual development of functional structure with lower level functional projections such as IP appearing before higher level projections such as the CP. The MTH makes an across-the-board prediction that there will never be instances of transfer effects associated with functional projections in ILGs. The initial ILG is, therefore, assumed to be devoid of any functional structure, either in the form of the
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 235 L1 functional architecture or the L2. Regarding the nature of initial L2 grammars, the MTH suggests that the difference between the early ILG and mature state grammars is a structural one; i.e., it is strictly a syntactic difference. Initial L2 grammars have missing syntactic positions or an incomplete syntactic tree as opposed to a fully-fledged syntactic tree characteristic of mature state grammars. Essentially, where subjects are presented with acceptability judgements, it seems the MTH would predict that beginner learners will make random or inconsistent judgements on any sentences involving functional structure, whereas when they finally acquire the L2like functional structure, their judgements will be consistent with what is allowed in the target language (TL) grammar. There are two types of evidence that can be used to falsify the MTH. Firstly, any evidence that there are indeed transfer effects associated with functional projections in the earliest stages of development would be damaging to the MTH (we discuss some of the more relevant evidence below). Second, the MTH could be falsified by evidence from mature state grammars that exhibit syntactic phenomena which imply functional structure in the absence of phonetic content associated with the respective functional heads. If mature state grammars evidence phonetically null functional heads then there is no reason for a universal grammar-constrained initial state grammar not to be characterized by functional categories with null content. This claim is captured in the FT/FA where it is assumed that each IL state is compatible with the class of grammars permitted by universal grammar (UG). Thus, if UG permits phonetically null functional heads, then IL grammars can be characterized as containing such heads. III Minimal trees: damaging counter evidence 1 Dasha, Erdem and Patty’s interlanguage grammars Gavruseva and Lardiere’s (1996) study on the L2 acquisition of English by Dasha, a native Russian speaker, provides a challenge to the MTH. The implicational acquisition order predicted in the MTH is not evident in Dasha’s ILG. Dasha’s production of embedded clauses is above the criterial level set by the MTH (i.e., 60% production) suggesting presence of CP, whereas her use of agreement marking is below 40% and her suppliance of modals and auxiliaries remains below 60%, which, in the MTH model, would be taken as evidence that IP is not yet projected. Evidence from Dasha’s grammar is damaging to the structure-
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
236 Minimal Trees Hypothesis and Zulu L2 acquisition building mechanism proposed in the MTH as it suggests that CP is projected before IP. The gradual development predicts that CP cannot be projected before the lower level projections like IP. Yet in Dasha’s grammar there is a criterial overt suppliance of CP elements before the criterial overt suppliance of agreement, tense, auxiliaries and modals, all of which are properties of IP. The problem posed for the MTH is that either the gradual development claim or the type of evidence used – i.e., overt suppliance of lexical material associated with functional elements – must be wrong. From a purely theoretical point of view, it is not possible for any grammar to project a full CP without any lower level projections (for details, see Grimshaw, 1994: 82). It follows, then, that Dasha’s utterances must include an IP. The availability of IP is further confirmed by Dasha’s knowledge of an abstract case-assigning feature in IP. Her initial representation of English includes the knowledge that nominative case is assigned by elements in IP. Gavruseva and Lardiere claim that this knowledge has been transferred from the learner’s L1. They conclude that Dasha has a full functional tree ‘despite the lack of overt lexical elements associated with these categories’ (1996: 235). Dasha’s knowledge of nominative case assignment is problematic for the MTH as it calls to question the reliability of the criterial level of overt suppliance of functional elements as evidence for the projection of FCs. Dasha produces less than 60% of IP-level functional elements, which in an MTH account suggests absence of the FC; and yet Dasha’s syntactic knowledge in the form of nominative case assignment suggests otherwise. Gavruseva and Lardiere’s findings dove-tail with those from Haznedar (1997) and Lardiere (1998). In Haznedar, the claim is that Erdem, a Turkish child L2 learner of English, has a mechanism of case assignment transferred from his L1 before the criterial overt suppliance of some IP level functional elements such as agreement. Similarly, Lardiere shows that Patty, an adult Chinese learner of English, has a CP projection although her suppliance of verbal morphology is almost nonexistent. Patty, like Erdem and Dasha, has a mechanism of nominative case assignment, which implicates the presence of an IP projection. The importance of these studies is that they draw attention to the methodological and theoretical problems of relying on overt production as a criterion for the availability of functional structure in developing L2 grammars. This is not an implicit accusation to the MTH for relying exclusively on overt production data. In fact, it is stated in the MTH that ‘both morphological and syntactic evidence’ (Vainikka and Young-Scholten, 1996a: 16) is used. However, the
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 237 MTH employs ‘the strict criterion of an overt complementizer’ (p. 19) for the projection of a CP. Because the structure under consideration is the CP, it is necessary to examine the reliability of overt production data more closely. If the criterion of overt production were to be used, it would lead to an underestimation of the learners’ L2 competence. More importantly, although these learners do not show a criterial level of overt production, there is syntactic evidence in the form of case assignment, which implicates the projection of functional categories. This suggests that there are no deficits in their knowledge of syntactic representations of the relevant structures. If there is any deficit in their grammars, this deficit is in the phonetic realization of the relevant functional heads. Alternatively, the deficit may be a pragmatic one in that functional elements may be omitted due to performance constraints. (For example, Kaplan and Selinker (1997) give examples of various types of lexical items that are elided by L2 learners and these are not limited to functional elements. Kaplan and Selinker suggest the possibility that the ‘absent’ elements may be due to performance factors.) In this regard, lack of overt suppliance cannot reliably indicate the absence of functional categories. The question this claim raises is whether functional heads with null content are a feature of natural language grammars. 2 Null content in mature state grammars A number of researchers have argued for null functional elements. For example, it has been shown that Chichewa has tenseless ideophonic sentences that may, prima facie, suggest the ‘absence’ of a TP projection in the language (Kulemeka, 1993; 1997). However, Kulemeka uses syntactic evidence in the form of verb movement to T as evidence of the presence of a TP projection in the language. Progovac (1998) notes that while there are no articles in SerboCroatian a pronoun/noun asymmetry suggests the projection of a DP. In Serbo-Croatian pronouns precede intensifying adjectives while nouns follow intensifying adjectives. Baptista (1997) and Luchesi (1993) also argue for the projection of a DP in the absence of overt articles in Cape Verdian Creole. The position of nouns relative to intensifying adjectives is also taken as evidence for the projection of a DP. In addition, Nakajima (1996) provides a wide range of syntactic evidence (from English) to show that that-less clauses are CPs headed by a null complementizer, rather than IPs (a number of theoretical linguists suggest that that-less clauses are not CPs; see Speas, 1993; Radford, 1997; Boskovic, 1996; Doherty,
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
238 Minimal Trees Hypothesis and Zulu L2 acquisition 1997). What is worth noting is that in all these studies evidence for projection of the relevant functional category is in the form of a syntactic correlate. Thus, absence of functional heads alone is not taken as evidence of the absence of the relevant functional categories. 3 Underspecification vs. missing FCs Some researchers attribute absence of a particular functional head in an ILG to underspecification. (There is no uniform interpretation of the term underspecification; see Sorace et al., 1998. Generally, the term is used to account for variability in that sometimes there is lexical realization and sometimes there is none.) For instance, in Wexler’s (1994; 1998) analysis of children’s tenseless clauses, derivations without an overt tense marking still converge because Tense is not required by the computational system. In this account, underspecification of functional heads has morphosyntactic reflexes in the form of the absence of surface morphology (e.g., the absence of determiners in determinerless DPs or C in Comp-less Comps; Dube, 1997). But the correlate syntax of the ‘underspecified’ functional head is present in the grammar (see Demuth, 1995;Dube, 1997; 1998). A number of studies in L2 research have argued for the projection of FCs in the absence of lexical realization of the relevant head; e.g., Prévost, 1997; Lardiere and Schwartz, 1997; Prévost and White, 1999. In contrast, a ‘missing’ functional category has dire consequences on the grammar as it has direct implications on the computational system. Such a grammar would have syntactic reflexes, such as lack of scrambling, if there were no DP or lack of wh-questions, relative clauses or subordination where there is no CP. This is what the MTH claims for initial L2 systems (Vainikka and Young-Scholten, 1996a: 16). A grammar with missing functional categories would imply a deficit in the computational system. In view of the above, a missing functional category implies a representational deficit while, in a grammar with underspecified functional heads, the respective functional projections exist at the level of syntactic computation (for a similar argument in L1 acquisition, see Jakubowicz et al., 1997). The fact that mature state grammars such as Chichewa, Cape Verdian Creole and Serbo-Croatian permit functional heads without lexical content suggests that lexical realization is not required by the computational system. If this be the case, then derivations in ILGs converge with mature state grammars in as far as syntactic representations are concerned.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 239 IV Tensed C in English and Zulu 1 Syntactic differences between English and Zulu CP In line with Culicover (1991; 1993) and Nakajima (1996) overt-that and null-that are independent heads occupying different C positions, namely a CP-type head and a Top-type C head (for similar arguments in German and Dutch, see Müller and Sternefeld (1993; Hoekstra, 1993a; 1993b), for arguments for Gungbe, see Aboh, 1997). This suggests that in English the traditional CP node is split into two independent functional heads, i.e., a CP-type C which generates overt-that and a Top-type C which generates null-that clauses (see Nakajima, 1996), as shown (1). 1)
CP Spec
C' C [–wh] that ukuthi
TopP Spec
Top' Top [null-that] [if ]
IP
Since English instantiates both CP- and Top-type C, sentences (2) and (3) are acceptable in the language. 2) John thinks that Mary is a dunce. 3) John thinks ø Mary is a dunce.
On the other hand, in Zulu, as in Spanish1 and in French2, only CP-type C is found, hence sentence (4) is acceptable while (5) is not. (The following abbreviations are used in the glosses in examples (4) to (17): AgrS = subject agreement; Tns = tense; Asp = aspect; PT = part.) 1
For example, in Spanish the lexical complementizer que (‘that’) is obligatory; hence, sentences like *Marta dijo ø Mario compro un libro (‘Marta said Mario bought a book’) are unacceptable. In contrast, sentences such as Marta dijo que Mario compro un libro (‘Marta said that Mario bought a book’) are acceptable. 2 In French tensed sentential complements must obligatorily have the lexical complementizer que (‘that’).Sentences like *Muriel a dit ø Nicole acheté un livre (‘Muriel said Nicole bought a book’) are unacceptable, whereas sentences like Muriel a dit que Nicole acheté un livre (‘Muriel said that Nicole bought a book’) are acceptable.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
240 Minimal Trees Hypothesis and Zulu L2 acquisition 4)
U-Thoko u-cabanga ukuthi u Busani yi-siduphunga. a/the-Thoko AgrS-think that a/the-Busani is-dunce ‘Thoko thinks that Busani is a dunce.’
5) * U-Thoko u-cabanga ø u-Busani yi-siduphunga. (a/the-Thoko AgrS-think ø a/the-Busani is-dunce) ‘Thoko thinks ø Busani is a dunce.’
The main syntactic difference between English and Zulu in so far as the instantiation of tensed C is concerned is that English allows both C-types – i.e., CP and Top-type C – while Zulu only allows a CP-type C. In this regard, English is a superset grammar. But what needs to be established is, structurally or syntactically, the differences between a grammar that allows CP-type C and one that allows Top-type C. 2 Distinction between CP-type C and top-type C The distinction between CP-type C and Top-type C is based on distributional evidence. For a detailed discussion of the differences between the two C-types, see Nakajima, 1996. Firstly, CP-type Cs can occur as complements to all sorts of lexical heads. CP-type C can occur as a complement of V, A and N as shown in the English examples (6a)–(8a) and Zulu examples (9a)–(11a). 6) a.
John thinks that Mary is a dunce.
(Complement of V)
7) a.
I am sure that Mary is a dunce.
(Complement of A)
8) a.
I must provide evidence that this work is original.
(Complement of N)
6) b.
John thinks ø Mary is a dunce.
(ø, complement of V)
7) b.
I am sure ø Mary is a dunce.
(ø, complement of A)
8) b. * I must provide evidence ø this work is original. 9) a. U-Johane u-cabanga
(ø, complement of N)
ukuthi u Thoko
yi siduphunga. (Complement of V) a/the-John AgrS-think that a/the-Thoko is-dunce ‘John thinks that Thoko is a dunce.’
10) a. U-Thoko u-qinisek-ile ukuthi i-ngane zi-ntshontsh-e amaswidi. (Complement of A) a/the-Thoko AgrS-sure-Tns that a/the-children they-steal-Tns sweets ‘Thoko is sure that the children stole sweets.’
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 241 11) a.
Ngi-mele ngi-veze
ubufakazi ukuthi lo-msebenzi mu-sha. (Complement of N) I-must AgrS-provide evidence that this-work be-new ‘I must provide evidence that this work is original.’
9) b. * U-Johane u cabanga
ø u Thoko
yi siduphunga. (ø, complement of V) a/the-John AgrS-think ø a/the-Thoko is-dunce ‘John thinks ø Thoko is a dunce.’
10) b. * U-Thoko u-qinisek-ile ø i-ngane zi-ntshontsh-e amaswidi. (ø, complement of A) a/the-Thoko AgrS-sure-Tns ø a/the-children they-steal-Tns sweets ‘Thoko is sure the children stole sweets.’ 11) b. * Ngi-mele ngi-veze
ubufakazi ø lo-msebenzi mu-sha. (ø, complement of N) * I-must AgrS-provide evidence ø this-work be-new ‘I must provide evidence that this work is original.’
From the above examples we can infer that CP-type C occurs as a complement of V, A and N (see the English examples (6a)–(8a) as well as the Zulu examples (9a)–(11a). On the other hand, Top-type C cannot occur as a complement of N3 although it can readily occur as a complement of V and A; see examples (6b)–(8b). Second, CPtype C can occur in positions where the clause is not contiguous with its head, e.g., as sentential subjects – as in (12) and (15) – in complement clauses with an intervening indirect object between the V and the complement – as in (14) and (17) – and where the complement has been fronted – see examples (13)–(16). This is the case because, even when separated from their heads, the information that the CP-type clause has been ‘dislocated’ can be inferred from the s-selection properties of their heads. 12) That/*ø she’s awake is certain. 13) That/*ø she’s awake, I don’t know. 14) We must show the magistrate that/*ø this is correct.
(subject position, originally complement of A) (originally, complement of V) (complement of V with intervening indirect object; note that Nakajima (1996) describes these as ‘dislocated positions’
3
Some complements of N allow comp deletion, but these were not included as part of the study.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
242 Minimal Trees Hypothesis and Zulu L2 acquisition 15) Ukuthi/*ø u-se-theng-e
i-moto
ng-amanga.
(Subject position)
that/*ø AgrS-Asp-buy-PT a/the-car be-lie ‘That/*ø she has bought a car is a lie.’ 16) Ukuthi/*ø u sa khangele,
cha a-ng-azi.
(originally, complement of V)
That/*ø AgrS-Asp-awake Neg Neg-I-know ‘That/*ø she is still awake, I don’t know.’ 17) Ku-mele si-khombise i-mantshi
ukuthi/*ø loku ku-qondile (originally complement of V) We-must AgrS-show a/the magistrate that/*ø this be-correct ‘We must show the magistrate that this is correct.’
Top-type C can only occur as a complement of V and A for reasons of selection feature checking. Nakajima (1996) and Svenonious (1994) point out that categorial features which are independently necessary to specify the categorial status of heads serve as features for selection checking. Top(ic) is a verbal head and it will be checked off by the heads which have this categorial feature, i.e., V and A. In addition, Top-type C cannot occur in displaced positions (i.e., in positions where the complement is distant from its head). If a Top-type C moved up to, for instance, a subject position, it will be in a position higher than its matrix head. To undergo selection feature checking, it will have to lower to the head that has the same categorial features – i.e., V or A – and this type of movement is illicit as it is a lowering movement. Displaced positions are ‘islands’ or ‘barriers’ to head movement, hence the lowering movement will yield an illicit head chain (Nakajima, 1996). By and large, the reason why a Top-type C cannot occur in cases like (14) and (16) is that the clause, which is complement of the V, is separated from the verb by an intervening indirect object. Thus sentences like (12) share with sentences like (13) to (17) the property that the complement clause is separated from its head (A or V), either by the presence of intervening material – as in (14) and (17) – or by the fact that the clause has fronted – as in (15) and (16) – hence these require a CP-type C. In conclusion, Zulu instantiates a CP-type C. It hence requires an obligatory lexical complementizer ukuthi as a complement of V, A, N and in positions where the clause is not contiguous with its head, e.g., as sentential subjects or in clauses with an intervening indirect object between the verb and the complement. English, on the other hand, instantiates both CP-type C and Top-type C.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 243 3 Hypotheses and predictions In order to test whether the initial ILG has functional categories and, if so, whether these are in their L1 form or not, it was hypothesized that if initial ILGs have missing functional projections, as the MTH suggests, then the initial English–Zulu ILG will not evidence any syntactic phenomena that imply functional structure. It was further hypothesized that if, on the other hand, functional structure does transfer, then the initial Zulu ILG will evidence syntactic phenomena that imply L1-like functional structure of tensed C. The null hypothesis was that functional structure will be present in initial L2 grammars and that it will be L1-based. Since English (our subjects’ L1) instantiates both CP-type C and Top-type C and Zulu only has a CP-Type C, the following prediction can be made: English-speaking beginner L2 learners of Zulu will discriminate between Top-type C (–comp) and CP-type C (+comp) sentences by: 1) accepting both [+comp] and [–comp] complements of V and A (since English allows such sentences with and without a lexical complementizer). 2) accepting [+comp] and rejecting [–comp] in sentences where the clause is distant from its head, e.g., in sentential subjects and in clauses where there is an intervening indirect object between the complement and the V. (In English a lexical complementizer is required in these sentence types.) 4 Definition of terms Sorace (1996:381) defines indeterminacy as ‘indefiniteness of status in the speaker’s grammatical competence’. In an acceptability judgement task, this indefiniteness might be manifested by ‘an inability to distinguish between acceptable and unacceptable sentences’ (p. 381). An operational measure of indeterminacy would be instances where the difference between the means of the acceptable and the unacceptable sentences is not statistically significant. The learner accepts both alternatives (i.e., the acceptable vs. the unacceptable) to the same degree. In contrast, if a judgement is determinate or definite, then the difference between the means of the acceptable and the unacceptable sentence will be statistically significant. However, Sorace’s definition of indeterminacy raises problems of interpretation in that acceptable and unacceptable are defined in terms of native-speaker judgements. An immediate problem in a study such as the present one would be to interpret
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
244 Minimal Trees Hypothesis and Zulu L2 acquisition [–comp] and [+comp] in V and A complements, both of which are acceptable in English. If the syntactic analysis used is that of English, we would expect no significant difference between the two sentence types (both sentences are acceptable to the same degree in English) and yet these judgements would still be determinate.4 V The experimental study 1 Methods The participants were asked to do an acceptability judgement test that was designed to test, among other things, the acquisition of tensed C in Zulu by native speakers of English. A numerical magnitude estimation technique was used as a measurement scale (see Bard et al., 1996; Cowart, 1997). In the magnitude estimation procedure, subjects are asked to form or construct a continuum of acceptability by assigning any number of their choice to the first sentence (called the modulus) they see on a screen while listening to the same sentence being played on tape. In the present study, both auditory and visual stimuli were provided for the necessity of providing both stimuli, see Cook, 1993: 240. Subjects are then asked to assign successive numbers to the sentences that they both read on screen and hear on tape in proportion to the number assigned to the first sentence presented to them as the modulus. For example, if the second sentence is more acceptable (by a certain degree), then a higher number (reflecting that degree) than that assigned to the first sentence should be assigned to the second; if it is less acceptable (by a certain degree), then a lower number (reflecting that degree) should be assigned. See Appendix 1 for a summary of the instructions given. There was a 10-second interval before the following sentence was heard. The time interval had been determined on the basis of the observations made during the piloting phase of the study when intervals of both five and eight seconds were experimented with. The rationale in using a numerical magnitude estimation procedure in the present study was that, being a timed procedure, it is hypothesized that it elicits immediate judgements and does not give subjects time to access their metalinguistic knowledge of the relevant structures. It is also assumed that numerical magnitude 4 As indicated by an anonymous reviewer, a major problem that arises here is Bley-Vroman’s (1983) comparative fallacy. The results were interpreted in terms of whether one sentence type was preferred over another. A preference indicates varying degrees with which sentence types are accepted. Thus, sentence X is preferred over sentence Y if the mean of X is larger than that of Y. If the difference between the two means is statistically different, this could be an indication of a strong preference for X.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 245 estimation makes it possible to capture degrees of indeterminacy in the learner’s judgements and gives greater certainty (than an ordinal scale) of obtaining the subjects’ subjective impressions of sentence acceptability. In designing the acceptability judgement test, the standard experimental control techniques suggested in Schütze (1996) and Cowart (1997) – such as random sampling of stimulus materials and counterbalancing for order effects (see Derwing, 1973) – were followed. The order of the stimulus sentences was randomized so that two consecutive sentences testing the same syntactic structure did not succeed each other. In constructing the test sentences, the control and experimental sentences were identical in every way except for the syntactic structure under investigation. The sentences were controlled for length; sentence length ranged from 5–10 words. The deciding factor on sentence length was the structure under consideration. The entire study consisted of IP- and CP-level structures. In Zulu CP-level structures are generally longer as they consist of complex sentences. CP-level structures were 8 words or more while IP level sentences were five to six words in length. In order to control for vocabulary difficulty, vocabulary booklets which contained all the vocabulary items that were to be used in the experimental sentences were provided to teachers/lecturers and all potential participants four months prior to the commencement of the research. As a result, even the elementary group, most of whom had been on a Zulu language course for a period of three months, had already had access to the vocabulary items that were used in the test sentences. However, subjects were not allowed to consult vocabulary booklets during the experiment. 2 Subjects The experimental study consisted of 151 native speakers of English learning Zulu at various stages of proficiency, as well as subjects who were working in Zulu-oriented jobs. There were also 38 native speakers who served as controls.5 The 151 learners included 5
Some researchers object to the use of native speaker controls (e.g.,Sorace, 1996). The reason behind this objection is that by including a native speaker control group one is not treating the IL grammar as an independent system but rather as a system parasitic to the TL and thus introducing Bley-Vroman’s (1983) ‘comparative fallacy’. In this study a native speaker control group was included on the basis that ‘being a native speaker does not confer one with papal infallibility on one’s intuitive judgements’ (McRae, cited in Paikeday, 1985), as has been shown by subjacency studies. Subjacency studies have shown that even native speakers reject grammatical sentences 10% to 27% inaccurately. In this regard,the inclusion of a native speaker control group was essential to determine how native speakers who supposedly know the language judge the same sentences presented to the experimental subjects.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
246 Minimal Trees Hypothesis and Zulu L2 acquisition students studying and teachers teaching Zulu at primary and high schools in the Johannesburg and Pietermaritzburg areas. There were also trainee teachers, university and college lecturers and professors in Zulu departments. The experimental group also included those who were native speakers of English working in Zulu-related jobs, such as newsreaders, journalists and senior editors. The 38 native controls were students of law and engineering at the universities of the Witwatersrand and Natal. L2 proficiency was determined on the basis of scores on a cloze test that was administered to all subjects, including the native controls. The reason for using the cloze as an independent test was that there was a lot of variation amongst the subjects in terms of the quality and quantity of input they were exposed to, such that the criterion of years or months of exposure to the TL would not have been useful. Some of the subjects had lived in a Zulu-speaking environment, while others had private tutors; some had been taught by native speakers while others had always been taught by nonnative speakers, thus suggesting that the quality and quantity of input the participants had been exposed to differed greatly. Based on the scores of the cloze test, the subjects were then divided into five proficiency levels, excluding the native control group. The experimental subjects were grouped as follows: • • • • • •
Group Group Group Group Group Group
1: most elementary group, the beginner group (nns1); 2: low intermediate group (nns2); 3: high intermediate group (nns3); 4: advanced group (nns4); 5: most advanced, the near-native group (nns5); 6: the native control group (ns).
In order to determine whether the mean scores of the cloze of the six groups differed significantly, a one way ANOVA with the scores in the cloze test as a dependent variable was conducted. The mean scores of the six groups differed significantly (F = 1682.9196, p < 0.0001).A post-hoc Tukey test was conducted in order to determine which of the group means differed significantly. In the Tukey test, the results of the pairwise comparison of means showed that all the six groups differed significantly, thus suggesting that the six groups were drawn from six different proficiency populations. 3 Test sentence structures In order to test whether subjects used any functional categories in their initial grammar and, if so, if these were in their L1 form, five sentence structures were used: ukuthi as a complement of V, A, N,
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 247 and ukuthi in ‘displaced’ clauses where the complement is distant from its head, i.e., sentential subjects and clauses where there is an intervening indirect object between the verb and the complement (see Appendix 2 for examples of sentences used). For each sentence structure, there were two tokens designed for each sentence-type each appearing with or without a complementizer, making a total of 20. Filler sentences were also used. VI Results The results were analysed as follows: firstly, descriptive statistics were used to calculate the geometric means (see Tables 1 and 2 for mean scores). Lodge (1981) describes the geometric mean as a standard measure of central tendency for magnitude data. In simple terms, a geometric mean (or, as it is better known, an antilog of logs) is a ‘harmonic mean’, i.e., a mean that ‘evens out’ the magnitude scale. In order to calculate the geometric mean, the variables were initially converted into logs, and the arithmetic mean of the logs was then calculated. This was followed by a multivariate analysis of variance (MANOVA) with repeated measures in order to determine which of the effects were significant. Where effects (i.e., main as well as interactional) were found to be statistically significant, post-hoc Tukey tests were conducted to make pairwise comparison of means in order to determine which of the means differed significantly. 1 Ukuthi complement sentences Figure 1 shows the mean scores of the judgements on the ukuthi complement sentences by the six groups (see Table 1 for the mean scores of ukuthi across complement types). Recall that sentences with the lexical complementizer ukuthi are grammatical in Zulu, while sentences without the complementizer are ungrammatical. In contrast, [–comp] sentences are grammatical in English only in the V and A complements. The equivalent [–comp] sentences in N complement sentences are ungrammatical in English and in Zulu. Figure 1 presents the judgements given by the subjects at different levels of proficiency. Beginner learners show a significant preference for [–comp] sentences in the V and A complements only. The intermediate groups do not discriminate between sentence types. The difference between the proficiency levels in the overall mean acceptability of sentences is confirmed in the main effect for level of language development in the ANOVA test (F=(5,181) = 2.93, p < 0.014). Tukey tests (tabled at q = 4.03) show that the difference
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
248 Minimal Trees Hypothesis and Zulu L2 acquisition
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 249 Table 2
+C –C
Overall mean scores of the effect of ukuthi in complement sentences nns1
nns2
nns3
nns4
nns5
ns
2.0170 2.7869
2.6931 2.4830
2.5444 2.4824
2.5532 2.3294
3.0426 1.5489
3.3816 1.1584
Note: To calculate the overall mean of the effect of ukuthi the mean logs of all [+comp] sentences are added together and then divided by the number of complement types.
V+c
4
V–C 3.5
A+C A–C
3
N+C 2.5
N–C
2 1.5 1 0.5 0 nns1
nns2
nns3
nns4
nns5
ns
Proficiency groups Key: V – verb; A – adjective, N – noun; +C – [+comp]; –C – [–comp] Figure 1
Mean acceptability scores for Ukuthi complement sentences
is due to the judgements of all the non-native groups compared to the native speaker control group (p < 0.05). The pattern of results in the judgements of V complement sentences resembles that in the A complement sentences. In the A complement sentences, it is the beginner group that discriminates between [–comp] and [+comp] sentences. The beginner group has a preference for [–comp] sentences. The low intermediate to the advanced groups do not distinguish between the two sentence types. The near-native speakers have a preference for the grammatical [+comp] sentence across complements. Their judgements are similar to those of the native control group. On the whole, grammatical sentences are accepted more than ungrammatical ones (see Table 4). This is confirmed by a statistically
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
250 Minimal Trees Hypothesis and Zulu L2 acquisition
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 251 Table 4
Overall mean scores of the effect of ukuthi in displaced clauses
+C –C
nns1
nns2
nns3
nns4
nns5
ns
2.4097 2.4056
2.4765 2.2296
2.5758 2.3418
2.9872 2.4164
3.2393 1.3671
3.3949 1.3378
Note: The overall mean score of the effect of ukuthi is obtained by adding the means of the [+comp] sentences and then dividing the outcome by the number of clause types.
highly significant main effect of ukuthi (F(1,181) = 20.78, p < 0.0001). There is a developmental trend in the discrimination of the two sentence types as confirmed by a statistically significant (F(5,181) = 21.14, p < 0.0001) interaction of the main effects of ukuthi and level of language development. As Figure 2 shows, the beginner group discriminates between the two sentence types and shows a preference for the ungrammatical [–comp] sentence while the nearnative speakers and the native controls have a preference for the grammatical [+comp] sentence. The native speakers seem to like the unacceptable [–comp] sentence. A possible explanations for this could be that the native group consisted of subjects who were also near-native speakers of English, and this could have had an effect on their native intuitions, i.e.,they were not a representative sample. However, the fact that there is a significant difference between the two means indicates that the two sentences were perceived differently. In the within-group comparisons, post-hoc Tukey tests (tabled at q = 4.62) show that there is a statistically significant difference in 4 3 2 1 0 nns1
nns2
nns3
nns4
nns5
ns
Proficiency groups [+C]: sentence with lexical complementizer [–C]: sentence without lexical complementizer Figure 2
Overall mean acceptability of ukuthi effect by level
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
252
Minimal Trees Hypothesis and Zulu L2 acquisition
the acceptability of the acceptable and the unacceptable sentence. Beginner learners have a preference for the unacceptable [–comp] sentence over the [+comp] sentence. In the pairwise comparisons of the mean acceptability of the [+comp] and the [–comp] sentences by the near-native group, Tukey tests show a significant difference in the judgements of the grammatical vs. ungrammatical sentence, with a strong preference for the grammatical sentence. A similar significant difference is found in the judgements of the native control group. The beginner group and the most advanced nonnative group have different preferences. While the beginner group prefers the ungrammatical [–comp] sentence, the most advanced group prefers the grammatical [+comp] sentence in the same way as the native control group. No significant difference is found in the within-group comparisons of the judgements of the two sentence types by the intermediate groups (i.e., from the low intermediate to the advanced groups). These groups do not distinguish between acceptable and unacceptable sentences, suggesting that their judgements are indeterminate or, perhaps, that they accept both sentences as they would in English. It is important to note, however, that the beginner and near-native groups appear to accept both sentence types as well, just to differing relative degrees. Figure 1 also shows that the beginner group’s judgements of the N complement are different from those of A and V complements. (Recall that ukuthi complement of N sentences is a property of CPtype C while [–comp] in V and A complements is a property of Toptype C. By treating complement of N sentences differently from complement of A and V sentences may be an indication that the beginner group differentiates between CP- and Top-type C.) The beginner group does not distinguish between [–comp] and [+comp] in their judgements of the N complement sentences. This difference in the acceptability of [+comp] and [–comp] sentences between complements is confirmed by a statistically highly significant (F(2,366) = 7.30, p < 0.001) interaction between the main effect of ukuthi and that of complement type. The pattern of results in the judgements of N complement sentences by the beginner group resembles that in their judgement of the ‘displaced’ clauses. 2 Ukuthi in ‘displaced’ clauses Figure 3 reports the mean scores of the judgements on displaced clauses, i.e., in clauses in which ukuthi as a sentential subject and in clauses with an intervening indirect object between the head V and the complement by the six groups (see Table 2 for mean
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 253 S+C
4
S–C 2
Int+C Int–C
0 nns1
nns2
nns3
nns4
nns5
ns
Proficiency levels Notes: S = sentential subject; C = complementizer; Int = intervening indirect object between complement and head Figure 3
Mean acceptability scores of ukuthi in displaced clauses
scores). Just like the ukuthi complement sentences, the [+comp] cases have a lexical complementizer introducing the sentential subject. The [+comp] sentences are grammatical in both English and Zulu. The [–comp] cases do not have a lexical complementizer and they are ungrammatical in both English and Zulu. The most elementary non-native group – i.e., the beginner group – does not distinguish between the ungrammatical (i.e., [–comp] ) and the grammatical (i.e., [+comp]) in its judgements of sentential subject sentences. Similarly, the beginner group does not distinguish between the [–comp] and [+comp] sentences in complement clauses where there is an indirect object between the complement and its head. The beginner group’s judgements in both displaced clauses resemble those of complement of N sentences. It seems that beginner learners do not distinguish between grammatical [+comp] and ungrammatical [–comp] in all the sentences that are diagnostic of properties of CP-type C. The effect of level of language proficiency is significant (F(5,183) = 3.60, p < 0.004).In post-hoc Tukey tests, this significance is shown to be between the comparisons between the means of the first four non-native proficiency groups in comparison to the near-natives and native controls (p < 0.05). The effect of ukuthi is also statistically significant (F(1,183) = 5.47, p < 0.020). There is a developmental trend in the acceptability of these two sentence types, as confirmed by a statistically highly significant interaction between the effect of level of language proficiency and the effect of ukuthi (F(5,183) = 13.17, p < 0.001). In the post-hoc Tukey test,the results of the pairwise comparisons of the group judgements of the [+comp] and the [–comp] sentences show that the elementary groups (beginner and high intermediates) do not discriminate between the two sentence types. This replicates the indecisiveness shown by the same groups’ judgements of ukuthi in complement of N sentences. However, on the extreme end of the
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
254 Minimal Trees Hypothesis and Zulu L2 acquisition proficiency scale, the advanced, near-native and native groups differentiate between [+comp] and [–comp] sentences. These groups judge the [+comp] sentences statistically significantly different from the [–comp] sentences (p < 0.05), with a preference for the [+comp] sentence. It is also important to note that while near-natives have a greater difference than the advanced group, this difference is not statistically significant. In summing up the results, the following observations can be drawn: 1) Beginner learners prefer [–comp] to [+comp] in the case of V and A complements and accept the two equally in the other sentence types which are diagnostic properties of CP-type C (i.e., in displaced clauses and complement of N sentences). 2) The low and high intermediate and the advanced groups fail to distinguish [–comp] from [+comp] in all sentence types. 3) The near-natives show the same preferences as the native speakers, that is, they prefer [+comp] to [–comp] in all sentence types. 4) The results of both the judgements of ukuthi complement and ukuthi in ‘displaced’ clauses of the low and high intermediate learners suggest a prolonged period of indeterminate judgements. VII Discussion The present experiment investigated whether L2 learners have the functional category comp in its L1 form in the very early stages of L2 acquisition or whether comp is absent from their grammars. The MTH predicts that complementation would only be evidenced in the ILG in its L2 form and that an initial L2 grammar would be characterized by inconsistent or random judgements as learners at this stage do not have any mental representation of functional structure. So what do our results show? What empirical and theoretical generalizations emerge? The beginner group’s preference for the [–comp] sentences in the complement of V and A sentences and their indeterminate judgements in the other cases might seem inconsistent with our prediction that the L1 final state constitutes the L2 initial state. We would have expected English L2 learners of Zulu not to distinguish between [+comp] and [–comp] sentences in the V and A complements as both are diagnostic of the properties of a CP projection in their L1. We would also have expected the same learners to accept only [+comp] sentences in ‘displaced’ clauses. But
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 255 they did not. Does this suggest that the L1 final state does not constitute the L2 initial state? Could this be evidence that there is no complementation at the Zulu L2 initial state? Are their judgements random and inconsistent? Given these results, the FT/FA is not fully supported in so far as CP-type C was not shown to be active at the initial stage of Zulu L2 syntactic development, as represented by the beginner group. While the early preference for [–comp] could be evidence of L1 representation in the IL, the fact that the L2 learners do not appear to know when null complementizers are not possible (e.g., in displaced clauses) is a problem. However, if we examine these learners’ judgements closely, they are consistent with an initial L2 state that reflects some properties of the English complementation system. In a split-CP analysis, English has both a Top-type C and a CP-type C. Recall that in Section IV it was established that a Toptype C grammar permits [–comp] as a complement of V and A while a CP-type C permits [+comp] as a complement of V, A, N and in cases where a clause is distant from its head. The judgements of the beginner group suggest that learners in the very early stages of L2 development are using a Top-type C grammar as an initial Zulu L2 grammar. While in the V and A complement sentences, beginner learners have determinate judgements, this is not the case in their judgements of the other clauses where the lexical complementizer is discontiguous with its head. In these cases, they do not distinguish in acceptability between [–comp] and [+comp]. Their judgements in these two structures are, to a large extent, compatible with a Top-type C grammar. A Top-type C grammar permits [–comp] as a complement of V and A but disallows it in clauses where the complement is distant from its head (as in sentential subjects and in clauses where there is an intervening indirect object between the verb and its complement). The systematic and significant preference for [–comp] in V and A complements indicates transfer of Top-type C which is instantiated in the L1. Because Top-type C is instantiated in the learner’s L1 and not in the L2, the L1 is a potential knowledge source for this C-type. Thus, the findings of the present study suggest ‘partial’ transfer of the English complementizer system. Does this suggest that CP-type C is absent in the Zulu IL grammar? While beginner learners have a preference for [–comp], they also accept [+comp] sentences, although to a lesser extent. One possible explanation for this outcome is that the functional head for CPtype C is initially underspecified. Thus, beginner learners prefer [–comp] sentences in V and A complements because this is compatible with a specified Top-type C (whose head does not need
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
256 Minimal Trees Hypothesis and Zulu L2 acquisition to undergo lexical realization). In fact, in the FT/FA of Schwartz and Sprouse (1994; 1996), full transfer will still be motivated on the basis of conceptual considerations. Schwartz and Sprouse (1996: 66) suggest that it is ‘difficult to imagine what sort of mechanisms would be involved in extracting a proper subsystem as the basis for a new cognitive state’. Following Schwartz and Sprouse (1994; 1996) it is possible that CP-type C is present in the Zulu IL grammar, as it would not be plausible that learners extract only Top-type C in order to construct their L2 complementation system. By and large, the results in the present study do not provide conclusive evidence on transfer of the totality of the English complementation system (i.e., both Top-type C and CP-type C) to the L2 initial state. They also do not support an initial stage of L2 syntactic development without functional categories (as proposed in the MTH). Although there is no proper explanation as to why CP-type C was not as active as Top-type C, for the beginner group consistently to show properties which are diagnostic of Top-type C (i.e., the significant preference of [–comp] in V and A complements) suggests the presence of complementation, hence a CP projection at the initial stage of Zulu L2 syntactic development. The results confirm the predictions that beginner learners will discriminate between CP-type C and Top-type C by accepting [–comp] in complement of V and A sentences. It also confirms the hypotheses that (1) the L1 functional geometry transfers and (2) initial state systems have functional projections. Thus, there is no representational deficit at the initial state. Beginner learners have a C-system in their IL representational system. In this regard, the Zulu IL data challenges the ‘no-functional-categories in initial grammars’ claim made in the MTH. 1 Subsequent IL development The absence of a significant preference for either sentence type by the ‘middle’ groups can be explained in terms of grammar competition. It could be assumed that the judgements are indeterminate as a result of competition between alternate stages. While at the initial state the L2 input is processed on the basis of those hypotheses generated through the assumption that the L2 is like the L1, by the intermediate stage learners have noticed that the L2 input is not totally analysable on the basis of their L1 system. But whatever knowledge of Zulu L2 complementation they have acquired, it cannot, as yet, be the sole system generating complementation in the ILG. Thus, ‘middle’ groups use both alternative forms drawn from the L1 and the L2 knowledge sources.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 257 It is also noticeable that the middle groups evidence a protracted period of restructuring. The prolonged restructuring phase may be due to the misleading Zulu evidence in the L2 input data. Since English is a superset grammar, allowing both [+comp] and [–comp], the Zulu primary linguistic data (PLD) is obscure in that nothing informs the learner that Top-type C is disallowed in mature Zulu. On encounter with Zulu PLD where the complementizer ukuthi is overt, native speakers may conclude that perhaps native speakers of Zulu have a preference for [+comp], rather than that [–comp] is disallowed. The judgements of the near-natives suggest that their intuitions converge with those of native speakers. The near-native speakers have a significant preference for [+comp] sentences across complement types, and this could be an indication that Zulu complementation has been acquired.Because of the superset nature of English complementation, native speakers of English learning Zulu are not expected to have positive evidence that [–comp] is not permitted in the target language. It is therefore possible that most advanced learners are able to access ‘subtle positive evidence’ or ‘indirect positive evidence’ of an as yet unknown nature to trigger the acquisition of complementation. 2 An attempt at stagiation On the basis of the experimental evidence presented above, it is possible to identify three discrete developmental stages in the ILG. While some researchers in this area extrapolate backwards and posit an initial state grammar to be represented by some hypothetical stage 0 (see Schwartz and Sprouse, 1994; 1996; Robertson and Sorace, 1998), we have taken our initial state grammar to be represented by the beginner group. This is based on the observation that most of the learners in the beginner group had been exposed to Zulu in a formal classroom situation for a period no longer than three months. It is therefore reasonable to assume this is as initial a state as one could expect. a Stage 1: Initial state grammar: From the experimental evidence presented in this study, the initial state grammar or the very early grammar with respect to the acquisition of declarative complementizer selection in Zulu L2 is a null-C system (or in Nakajima’s (1996) account, a Top-type C) which is transferred from mature English.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
258 Minimal Trees Hypothesis and Zulu L2 acquisition b Stage 2:The intermediate grammar (grammar competition phase): This stage can be characterized as the optionality phase. Since our initial state grammar was L1-like, and since acquisition is an incremental process (see Robertson and Sorace, 1998), it is reasonable to assume that optionality evident in the intermediate stage is a result of grammar competition.6 In this regard, the weakening of the L1 knowledge strength in accounting for Zulu input necessitated the observed optionality. While the English-like knowledge system is found inappropriate for the L2 input data, the newly acquired L2 knowledge system is still not yet strong enough to be the sole knowledge system that the L2 learner could rely on. As a result, there is grammar competition at the level of mental representation. Arguably, the observed optionality in the IG is a result of grammar change where one form is replacing another (for similar arguments on language change3 see Kroch 1989; Santorini, 1991). c Stage 3: Mature state grammar (survival from optionality phase): This stage is the mature state grammar where the relevant L2 structures have been acquired and optionality is resolved. At this stage, the most advanced non-native speakers show the same preferences as the native controls. At least with respect to the structures investigated here, this grammar converges with the TL grammar. VIII Conclusions This study set out to investigate: (1) whether L1 functional categories transfer to the L2 initial state and (2) whether initial L2 grammars lack functional categories as proposed in the MTH. Neither claim is supported by the findings of the present study. The results on the acquisition of tensed C in Zulu by native speakers of English challenge the ‘no functional categories in initial L2 grammars’ view expressed in the MTH. While the MTH claims that ‘CP and the complementation information associated with CP is not transferred from the learner’s L1’ (Vainikka and Young-Scholten, 1996a: 32), the Zulu IL data challenges this position. Top-type C, which is instantiated in the subjects’ L1, is active in their initial state 6 Because English, the subject’s L1, permits optionality, it is possible to assume that this is still L1-like. But the problem with this explanation would be that English does not permit [–comp] as a complement of N and in ‘displaced’ clauses. Yet the intermediate groups do not show this distinction in their judgements. Instead, [+comp] and [–comp] sentences are accepted to the same degree in all sentence types and this suggests that this might not be transfer of the L1 grammar.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 259 grammar, thus suggesting that properties of L1 functional categories transfer. On the assumption that Top-type C implicates the projection of CP, there must be a CP projection in the Zulu IL grammar to evidence properties of Top-type C (i.e., the preference of [–comp] in V and A complements). Specifically, the Zulu IL data shows that the CP layer is not missing. On the other hand, the Zulu data does not provide conclusive evidence on whether the totality of the L1 functional geometry transfers. Thus, although the Zulu ILG has complementation, there is no clear-cut indication of full transfer of the English complementation system at the Zulu L2 initial state. CP-type C is shown not to be as active as Top-type C in those sentences which are diagnostic of its properties, i.e., in ‘displaced’ clauses and complement-of-N sentences. However, by virtue of having complementation, the Zulu initial IL grammar – and, perhaps, L2 grammars in general – does not differ with mature state grammars with respect to the availability of syntactic positions as suggested by MTH. IX References Aboh, E. 1997: A split IP and CP approach: evidence from Gungbe. Unpublished paper presented at the ACAL28 Conference, Cornell University, Ithaca, NY. Atkinson, M. 1996: Now, hang on a minute: some reflections on emerging orthodoxies. In Clahsen, 1996, 451–85. Baptista, M. 1997: The referential system of Caperverdian Creole: the role of null determiners. Unpublished paper presented at the 2nd WOCAL conference, University of Leipzig. Bard, E.G., Robertson, D. and Sorace, A. 1996: Magnitude estimation of linguistic acceptability. Language 72(1), 32–68. Bley-Vroman, R. 1983: The comparative fallacy in interlanguage studies: the case of systematicity. Language Learning 33, 1–17. Boskovic, Z. 1996: Selection and the categorial status of infinitival complements. Language Research 30, 387–412. Clahsen, H., editor, 1996: Generative perspectives on language acquisition: empirical findings, theoretical considerations and crosslinguistic comparisons. Amsterdam: John Benjamins. Cook, V. 1993: Linguistics and second language acquisition. London: Macmillan. Cowart, W. 1997: Experimental syntax: applying objective methods to sentence judgements.Thousand Oaks, CA: Sage. Culicover, P. 1991: Topicalization, inversion and complementizers in English. In Delfito, D., Evaraet, M., Eve, A. and Stuurman, F., editors, Fifth symposium on comparative grammar. Utrecht: University of Utrecht, 2–43. Culicover, P. 1993: Evidence against ECP account of that-t Effect.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
260 Minimal Trees Hypothesis and Zulu L2 acquisition Linguistic Inquiry 24, 557–67. Demuth, K. 1995: On the ‘underspecification’ of functional categories in early grammars. In Lust, B., editor, Syntactic Theory and First Language Acquisition. Hillsdale, NJ: Lawrence Erlbaum. Derwing, B.L. 1973: Transformational grammar as a theory of language acquisition: a study in the empirical, conceptual and methodological foundations of contemporary linguistics. Cambridge: Cambridge University Press. Doherty, C. 1997: Clauses without complementizers: finite IPcomplementation in English. The Linguistic Review 14, 197–220. Dube, S. 1997: Complementizer-less comps in the acquisition of declarative complementizer selection by native speakers of English learning Zulu as a second language. Unpublished paper presented at the University of Edinburgh Joint Postgraduate Conference of the Department of Linguistics and Applied Linguistics, University of Edinburgh. ––––– 1998: The initial state of Zulu L2 syntax: a study of the emergence of a CP projection in the Zulu grammar of native English speakers. Unpublished PhD thesis, University of Edinburgh. Epstein, S., Flynn, S. and Martohardjono, G. 1996: Second language acquisition: theoretical and experimental issues in contemporary research. Behavioural and Brain Sciences 19, 677–714. ––––– 1998: The strong continuity hypothesis in adult L2 acquisition of functional categories. In Flynn, S., Martohardjono, G. and O’Neil, W., editors, The generative study of second language acquisition. Hillsdale, NJ: Erlbaum, 171–99. Gavruseva, L. and Lardiere, D. 1996: The emergence of extended phrase structure in child L2 acquisition. In Stringfellow, A., Cahana-Amitay, D., Hughes, E. and Zukowski, A., editors, Proceedings of the 20th Annual Boston University Conference on Language Development, Vol. 1. Sommerville, MA: Cascadilla Press, 225–36. Grimshaw, J. 1994: Minimal projection, heads and optimality. Technical Report 4. New Brunswick. NJ: Rutgers University Centre For Cognitive Science. Haznedar, B. 1997: L2 acquisition by a Turkish speaking child: evidence for L1 influence. In Hughes, E. et al., 1997, 245–56. Hoekstra, E. 1993a: On the parametrization of functional projections in CP. Proceedings of the Northeast Linguistic Society 23, 191–204. ––––– 1993b: Dialectal variation inside CP as parametric variation. In Werner, A. and Bayer, J., editors, Linguistische Berichte Sonderheft 5: Dialeksyntax. Oplagen: Westdeutscher Verlag. Hughes, E., Hughes, M. and Greenhill, A., editors, 1997: Proceedings of the 21st Annual Boston University Conference on Language Development, Vol. 1. Sommerville, MA: Cascadilla Press. Jakubowicz, C., Müller, N., Riemer, B. and Rigaut, C. 1997: The case of subject and object omissions in French and German. In Hughes, E. et al., 341–42. Kaplan, T.I. and Selinker, L. 1997: A Review of Klein, E.C. 1993) Toward
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 261 second language acquisition. Second Language Acquisition Research 13(2), 170–86. Kroch, A.S. 1989: Reflexes of grammar in patterns of language change. Language Variation and Change 5, 199–244. Kulemeka, A.T. 1993: The status of the ideophone in Chichewa. Unpublished Doctoral Dissertation, Indiana University. ––––– 1997: The ideophone predicate: a case of tenseless sentences in Chichewa. Unpublished paper presented at the 2nd WOCAL conference, University of Leipzig. Lardiere, D. 1998: Case and tense in the fossilised steady state. Second Language Research 14(1), 9–26. Lardiere, D. and Schwartz, B. 1997: Feature marking in the L2 development of deverbal compounds. Journal of Linguistics 33, 327–53. Lodge, M. 1981: Magnitude scaling: quantitative measurement of opinions. London: Sage. Luchesi, D. 1993: The article systems of Cape-Verde and Sao-Tomè Creole Portuguese: general principles and specific factors. Journal Of Pidgin And Creole Languages 8, 81–108. Müller, G. and Sternefeld, W. 1993:Improper movement and unambiguous binding. Linguistic Inquiry 24, 461–507. Nakajima, H. 1996: Complementizer selection. The Linguistic Review 13, 143–64. Paikeday, T.M. 1985: The native speaker is dead! Toronto: Paikeday. Prévost, P. 1997: Truncation and root infinitives in second language acquisition of French. In Hughes, 1997, 453–64. Prévost, P. and White, L. 1999:Accounting for morphological variation in second language acquisition: truncation or missing inflection? In Friedman, M.-A. and Rizzi, L., editors, The acquisition of syntax. London: Longman, 202–35. Progovac, L. 1998: Determiner phrases in a language without determiners. Journal of Linguistics 34, 165–79. Radford, A. 1997: Syntactic theory and the structure of English: a minimality approach. Cambridge: Cambridge University Press. Rizzi, L. 1995: The fine structure of the left periphery. Unpublished manuscript, Department of Linguistics, University of Geneva. Robertson, D. and Sorace, A. 1998: Losing the V2 constraint. In Klein, E. and Martohardjono, G., editors, The development of second language grammars: a generative approach. Amsterdam: John Benjamins. Santorini, B. 1991: Scrambling and INFL in German. Unpublished manuscript. University of Pennsylvania, PA. Schütze, C.T. 1996: The empirical base of linguistics: grammaticality judgements and linguistic methodology. Chicago, IL: The University of Chicago Press. Schwartz, B.D. and Sprouse, R. 1994: Word order and nominative case: a longitudinal study of (L1 Turkish) German interlanguage. In Hoekstra, T. and Schwartz, B.D., editors, Language acquisition studies in generative grammar:articles in honour of Kenneth Wexler from 1991
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
262 Minimal Trees Hypothesis and Zulu L2 acquisition GLOW workshops. Philadelphia, PA: John Benjamins, 317–68. ––––– 1995a: L2 Cognitive states and absolute L1 influence. Unpublished manuscript, University of Durham/Indiana University. ––––– 1995b: Why the initial state in L2A is the steady state from L1A: a reply to Eubank (1993/94). Unpublished manuscript, University of Durham/Indiana University. ––––– 1996: L2 Cognitive states and the full transfer/full access model. Second Language Research 12(1), 73–106. Sorace, A. 1996: The use of acceptability judgements in second language acquisition research. In Ritchie, W. and Bhatia, T., editors, Handbook of language acquisition. San Diego, CA: Academic Press, 375–409. Sorace, A., Heycock, C. and Shilcock, R. 1998: Introduction. Lingua 106. Special issue on language acquisition: knowledge representation and processing, 1–22. Speas, M. 1993: Null arguments in a theory of economy of projections. In Benedicto, E. and Runner, J., editors, University of Massachusetts Occasional Papers 17: Functional Projections. Amherst, MA: University of Massachusetts, 179–208. Svenonious, P. 1994: C-selection as feature checking. Studia Linguistica 48, 133–55. Vainikka, A. and Young-Scholten, M. 1994: Direct access to X'-theory: evidence from Korean and Turkish adults learning German. In Hoekstra, T. and Schwartz, B.D., editors, Language acquisition studies in generative grammar:papers in honour of Kenneth Wexler from 1991 GLOW workshops. Philadelphia, PA: John Benjamins, 265–316. ––––– 1996a: Gradual development of L2 phrase structure. Second Language Research 12, 7–39. ––––– 1996b: The early stages in adult L2 syntax: additional evidence from Romance speakers. Second Language Research 12(12), 140–76. Wexler, K. 1994: Optional infinitives, head movement and the economy of derivation.In Lightfoot, D. and Hornstein, N.,editors, Verb movement. Cambridge: Cambridge University Press, 305–50. ––––– 1998: Very early parameter setting and the unique checking constraint: a new explanation of the optional infinitive stage. Lingua 106, 23–79. White, L. 1996: Clitics in child L2 French. In Clahsen, 1996, 335–68.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 263 Appendix 1
Instructions for ME task
1. In this task what you have to do is to assign numbers to sentences in proportion to how acceptable you think they are. That is; give the first sentence you see ANY NUMBER you wish. Then assign the successive sentences numbers depending on how acceptable you think they are in proportion to the first. We will do some examples together to illustrate this point. Example 1: We will practise by estimating line length in proportion to one another. Look at the following 3 lines carefully. 1. __________ 2. _____ 3. _______
1. -----------2. -----------3. ------------
Now think of any number to represent the length of line 1. Any number will be fine. Write it down on the dotted line. How long do you think line 2 is in proportion to line 1? 1/2 as long? Then multiply the number you gave line 1 by 1/2 and write down the answer in the dotted line of number two. For example, if you had estimated line 1 to be 6cm, after multiplying this number by 1/2 you get 3, you then write down three in the dotted line of number 2. Now how long do you think line 3 is in proportion to the first, 3/4 or 2/3? Then assign it a number that is three quarters or two thirds of the number assigned to line 1. Example 2: We can estimate the acceptability of sentences in the same way by assigning numbers on how acceptable you think they are in proportion to the first sentence. Look at sentence 1. 1. Ubaba uhlezi isitulo. Do you think this sentence is acceptable? Assign this sentence any number of your choice. Now look at sentence 2. 2. UThoko ubone abafana. Is this sentence more or less acceptable than the first? Now look at sentence 3. 3. UThoko abafana ibhubesi amangisi ngoba. How acceptable is this sentence in proportion to the first sentence? In proportion to the second sentence? 2. As you can see there are degrees of acceptability. What you have to do is to assign numbers depending on how acceptable you think the sentence is. If you think sentence 2 is half as acceptable as sentence 1 then assign it half the amount you gave sentence 1. If you think sentence 3 is 1/5 or
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
264 Minimal Trees Hypothesis and Zulu L2 acquisition 1/3 as acceptable as sentence 1 then assign one third or one fifth the number given to sentence 1.Remember, the more acceptable the sentence is, the higher the number it gets. 3. In doing this task, you will both see and hear one sentence at a time. Please keep to the speed dictated by the tape. 4. There are 32 sentences here and it should take approximately 6 minutes to complete the task. 5. Write your answers in the DOTTED LINES. Please remember NOT to write your name on the answer sheet. Appendix 2
Test sentences
Ukuthi complement sentences (a) +/– ukuthi as a complement of V 1. * Abafana bacabanga uThabo uthenge imoto. 2. Abafana bacabanga ukuthi uThabo uthenge imoto. 3. * Ngicabanga uThoko uzokudla amabhontshisi. 4. Ngicabanga ukuthi uThoko uzokudla amabhontshisi. (b) +/– ukuthi as a complement of A 5. * Ugogo uqinisekile abantwana bahlala eThekwini. 6. Ugogo uqinisekile ukuthi abantwana bahlala eThekwini. 7. * Ingane ziqinisekile amaswidi aphele izolo. 8. Ingane ziqinisekile ukuthi amaswidi aphele izolo. (c) +/– ukuthi as a complement of N 9. * Ngimele ngiveze ubufakazi lomsebenzi musha. 10. Ngimele ngiveze ubufakazi ukuthi lomsebenzi musha. 11. * Simele siveze ubufakazi lokhu kuqondile. 12. Simele siveze ubufakazi ukuthi lokhu kuqondile. Ukuthi in displaced clauses (d) +/– ukuthi as a tensed sentential subject 13. * useye esikoleni kuyiphutha elikhulu kakhulu. 14. Ukuthi useye esikoleni kuyiphutha elikhulu kakhulu. 15. * usethenge imoto kuyajabulisa kakhulu. 16. Ukuthi usethenge imoto kuyajabulisa kakhulu. (e) +/– ukuthi in complements with an intervening indirect object 17. * Kumele sikhombise imantshi lokhu kuqondile. 18. Kumele sikhombise imantshi ukuthi lokhu kuqondile. 19. * Kumele batshele isela lokhu kubi. 20. Kumele batshele isela ukuthi lokhu kubi.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011
Busi Dube 265 Filler examples (i) +/– pre-expression 21. Umndeni wakubo, ngazi unina yedwa. 22. Kumndeni wakubo, ngazi unina yedwa. 23. Amadolobha amakhulu aseAfrika, ngazi iKapa yodwa. 24. Kumadolobha amakhulu aseAfrika, ngazi iKapa yodwa. (ii) Future tense, +/– zoku (‘will’) 25 * Isalukazi siya edolobheni ngesonto elizayo. 26. Isalukazi siyazokuya edolobheni ngesonto elizayo. 27. * Abafana bathe baya eiskolo ngesonto elizayo. 28. Abafana bathe bazokuya esikolo ngesonto elizayo. (iii) +/– V-raising over adverb 29. * UThoko njalo uzigeza izingubo zabantwana. 30. UThoko uzigeza njalo izingubo zabantwana. 31. * Uthisha kahle ubafundisa abantwana benkosi. 32. Uthisha ubafundisa kahle abantwana benkosi.
Downloaded from slr.sagepub.com at PENNSYLVANIA STATE UNIV on July 14, 2011