L1 and L2 acquisition between sentence and discourse: Comparing production and comprehension in child Dutch

Lingua 117 (2007) 1930–1958 www.elsevier.com/locate/lingua

L1 and L2 acquisition between sentence and discourse: Comparing production and comprehension in child Dutch Sharon Unsworth * Utrecht University, Trans 10, 3512JK Utrecht, The Netherlands Received 18 May 2006; received in revised form 17 September 2006; accepted 13 November 2006 Available online 16 May 2007

Abstract This paper compares production and comprehension data from children acquiring Dutch as a second language (L2) with children acquiring Dutch as a first language (L1). Using the same experimental methodology for all subjects, it investigates how these two groups of learners acquire direct object scrambling, a property of Dutch which requires syntactic, semantic and discourse knowledge. A comparison of production and comprehension data from the same individuals suggests that in child L2 acquisition – as observed in child L1 acquisition – there is a discrepancy between production and comprehension in this domain. Contrary to what is generally expected to be the case, however, it is production which is more targetlike than comprehension. Several explanations for this finding are explored, including the lack of discourse integration and processing problems. # 2007 Elsevier B.V. All rights reserved. Keywords: Child second language acquisition; Scrambling; Production vs. comprehension; Discourse integration

1. Introduction: child L2 acquisition In recent literature, there has been renewed interest in child L2 acquisition (e.g. Blom and Polisˇenska`, 2006; Haznedar, 2003; Hulk and Cornips, 2006; Ionin and Wexler, 2002; McDonald, 2000; Pre´vost, 2003; Schwartz, 2003, 2004; Weerman et al., 2003). Specifically, it has been claimed that child L2 acquisition can be informative with respect to both child L1 acquisition and adult L2 acquisition (Lakshmanan, 1995; Schwartz, 1992, 2003, 2004; Unsworth, 2005a).

* Tel.: +31 30 2536014; fax: +31 30 2536000. E-mail address: [email protected]. 0024-3841/$ – see front matter # 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.lingua.2006.11.009

S. Unsworth / Lingua 117 (2007) 1930–1958

1931

The term child L2 acquisition is generally used to distinguish successive child bilingualism from simultaneous child bilingualism, on the one hand, and successive child bilingualism from successive adult bilingualism, on the other. Here, the following definition is adopted: (1)

Definition of L2 child A non-native acquirer whose initial exposure to the target language is between the ages of four and seven years.

The lower bound in this definition is set at age four because at this age, we can assume most (purely) grammatical principles (and, for example, the phonology) of the first language to be in place (e.g. Goodluck, 1986; Guasti, 2002). The upper bound is set at age seven because this is the age which has been put forward as the start of a decline in the ability to reach nativelike levels of L2 proficiency (DeKeyser, 2000; Johnson and Newport, 1989, 1991). (See Unsworth, 2005a:5–7 for relevant discussion.) The focus here is on the comparison between child L2 and child L1 acquisition. Such a comparison is interesting because there are both similarities and differences between these two groups of learners. These similarities and differences relate to age (at time of testing) and to L1 transfer. In the case presented here, there is complete overlap between L2 and L1 children in terms of age (at time of testing). The two groups of course differ, however, because the children in the former group already (by definition) have knowledge of another language, which allows for the potential of L1 transfer. Comparing L2 children with L1 children will shed light on the role of these two factors in child L2 acquisition and the role of age in child L1 acquisition. The fact that L2 children already have knowledge of another language is probably the most obvious difference between L2 and L1 children. Given that most aspects of morphosyntax and phonology are in place by age four and that, as stated in (1), child L2 acquisition is defined as L2 acquisition where first exposure is between the ages of four and seven, it will generally be the case that L2 children differ in a non-trivial way from L1 children, because they will already have linguistic knowledge which they could transfer to their interlanguage grammar. Although this might, to a certain extent, seem like stating the obvious, it is only relatively recently that the issue of L1 transfer in child (as opposed to adult) L2 acquisition has been systematically investigated. There is an increasing body of evidence which demonstrates that as in adult L2 acquisition, the initial stages of child L2 development are characterised by L1 transfer (Haznedar, 1997; Whong-Barr and Schwartz, 2002 amongst others). The present paper contributes to this body of research. The extent to which L1 and L2 children differ (or are similar) in terms of their age (at time of testing) partly depends on the property of language in question. Put simply, L2 children may be either younger or older than the age at which the relevant property of language is acquired by the L1 child. The relevance of age (at time of testing) depends on the importance attributed to it in a given acquisitional theory. There exist theories which attribute particular developmental stages in L1 acquisition to maturation, that is, to biological changes occurring in the brain at a given chronological age. For example, Rizzi’s (1993/1994) Truncation Hypothesis and Wexler’s (1998) Very Early Parameter Setting, the details of which need not concern us here, are both acquisitional accounts which are couched in terms of maturation. Such accounts entail very specific predictions about whether L2 children (and adults) will pass through similar stages as L1 children, namely they predict that if a particular developmental stage in L1 acquisition, occurring at age X, is maturationally determined, then any L2 child who is older than age X should not pass through that developmental stage and any L2 child who is age X or younger should pass through the relevant stage. Following this logic, then, and given the definition of child L2 acquisition

1932

S. Unsworth / Lingua 117 (2007) 1930–1958

adopted here, L2 children should not pass through any maturationally determined developmental stages which are claimed to occur before the age of four (as in the accounts put forward by Rizzi (1993/1994) and Wexler (1998)). Studies which have examined child L2 acquisition in a maturational light include Henry and Tangney (1996), Ionin and Wexler (2002), Park (2000) and Schwartz (2004), amongst others.1 There are of course other factors which may interact with age but which may not be a direct result of maturation. These include, for example, limited input, pragmatic and lexical development. Direct object scrambling in Dutch, the focus of the present paper, has been related to pragmatic development (Kra¨mer, 2000; see discussion below). Kra¨mer demonstrates that L1 Dutch children fail to assign scrambled indefinite objects a targetlike, specific interpretation as late as age seven. The observation that a non-targetlike developmental stage may persist in L1 acquisition well beyond the age of four rather complicates any child L1child L2 comparison. If a particular property of the target language is only acquired by L1 children at say, age eight, L2 children will not necessarily be expected to acquire that property before age eight. Once beyond the relevant age, however, targetlike acquisition should in principle be possible for the L2 children, too.2 As noted above, the target language property in question in the present paper is direct object scrambling in Dutch. For L1 children, it has been found that after a period of no or optional scrambling, scrambled objects are produced fairly consistently at an early age, around 3–4 years (Schaeffer, 2000), whereas targetlike interpretation of scrambled indefinite objects is not acquired until much later (Kra¨mer, 2000; Philip, 2003). There is thus a discrepancy between production and comprehension. This paper investigates the acquisition of scrambling by L2 children, asking the following three questions: (i) (ii) (iii)

In production, do L2 children pass through the same stages as L1 children? In comprehension, do L2 children show the same age-related effect? Do L2 children also show a discrepancy between production and comprehension?

The paper is organised as follows: section 2 outlines the relevant facts about scrambling in Dutch. Sections 3 and 4 present the production and comprehension data, respectively, and section 5 compares the production and comprehension data from the same sub-group of L2 children. Section 6 outlines the significance of the observed results and evaluates various explanations for the findings. 2. Direct object scrambling in Dutch Scrambling is a cover term used for different permutations of verbal arguments and other sentential constituents, such as adverbials, in languages such as Dutch, German and Japanese. 1 Of course, if maturation, i.e. the particular age X, is not considered a crucial factor in the developmental stage in question, it does not matter whether L2 children pass through the same developmental stage, albeit somewhat later in terms of chronological age, as L1 children. 2 An anonymous reviewer asks whether there are any target language properties which L2 children may be reasonably expected to acquire before L1 children. Imagine there is a target language property, X, which is acquired by L1 children at the age of, say, five. It is possible that, if L2 children have knowledge of the same (or similar) property in their L1, they will be able to use this knowledge and potentially acquire X before L1 children. Let us imagine another target language property, Y. Here, it is possible that, again, as a result of existing L1 knowledge, L2 children require less exposure to the TL in order to acquire Y, and in this sense, they might acquire Y before L1 children.

S. Unsworth / Lingua 117 (2007) 1930–1958

1933

In (2)-a, the direct object een bal ‘the ball’ appears to the right of the sentential adverbial, whereas in (2)-b, it is to the left of the adverbial. This latter position is known as the scrambled position. (2)

a.

De jongen heeft twee keer [een bal] gegooid the boy has two times a ball thrown ‘The boy threw a(ny) ball twice.’

[non-specific]

b.

De jongen heeft [een bal] twee keer gegooid the boy has a ball two times thrown ‘The boy threw a (certain) ball twice.’

[specific]

Scrambled and non-scrambled orders are assumed to be generated by the grammar by movement of the object to an adjoined position (e.g. de Hoop, 1992).3 Scrambling of indefinite NP objects is exploited at the syntax-external interfaces to mark interpretive differences. Thus, in examples (2) above and (3) below, there is a clear difference in interpretation between the nonscrambled (a) and the scrambled (b) orders: in the former, the direct object has a non-specific (or narrow scope) interpretation, whereas in the latter, it has a specific (or wide scope) interpretation, as indicated by the English glosses. (3)

a.

De jongen heeft [geen (niet + een) bal] gegooid the boy has no not + a ball thrown ‘The boy didn’t throw a(ny) ball.’

b.

De jongen heeft [een bal] niet gegooid the boy has a ball not thrown ‘The boy didn’t throw a (certain) ball.’

[non-specific]

[specific]

The L2 children in the present study are native-speakers of English. English does not have scrambling, that is, it does not have an overt movement operation which induces that same type of interpretive effects as scrambling in Dutch. The difference between a specific and non-specific indefinite object NP is not overtly realised in the syntax. The indefinite object NPs in (4) are thus ambiguous between the two readings, i.e. they can either refer to any ball or to a certain ball: (4)

a. b.

The boy threw a ball twice The boy didn’t throw a ball

Following traditional analyses (Koster, 1975), Dutch is assumed to be underlyingly head-final (OV), whereas English is head-initial (VO). Adverbs are assumed to be VP-adjoined in both English and Dutch, and following Pollock (1989), negation is assumed to head its own phrase, NegP, also VP-adjoined. The English equivalents of the Dutch scrambled (and non-scrambled) sentences in (2) and (3), given in (4), generally involve a sentence-final adverb. The exact position of post-verbal adverbials is very unclear (Haider, 2004). Here, it will be assumed that they are right-adjoined to the VP. Following Van Geenhoven (1998), indefinite NPs are assumed to derive their interpretation from the context in which they appear. The fundamental idea underlying Van Geenhoven’s 3

What this position is exactly is not relevant for present purposes.

1934

S. Unsworth / Lingua 117 (2007) 1930–1958

approach is that unless there is some indication to the contrary, indefinite NP objects are always interpreted as being part of the verb (that is, as predicative indefinites). Scrambling is one such way of indicating that this is not the case, that is, that the indefinite NP object has a different interpretation, namely that of a free variable. Following Kamp (1981) and Heim (1982), Van Geenhoven assumes that indefinites are variables, but unlike these authors, she assumes that they do not introduce variables themselves. Working in a Discourse Representation Theory framework (Kamp and Reyle, 1993), Van Geenhoven assumes that indefinite NPs get their interpretation from the context in which they appear: the indefinite’s variable is introduced either by the semantically incorporating verb or by ‘accommodation’. On Van Geenhoven’s account, each transitive verb has two lexical entries: one where it is semantically incorporating and one where it is not, that is, one which is truly transitive. When an indefinite direct object is incorporated by the verb, the property expressed by the indefinite becomes a condition on the argument slot of the verb, that is, it restricts the values of the argument which the verb takes. For example, if the indefinite a ball incorporates into the verb throw, given formally in (5), the result is the predicate ball-throw, and consequently, the indefinite can no longer be regarded as a true argument. (5)

lPlx. 9y [P(y) ^ throw (y)(x)]

Scrambling in Dutch is viewed as a language-specific means as indicating that the indefinite object should not be incorporated. Rather, scrambled objects are true arguments of the verb whose variable is bound at some higher level. More specifically, the indefinite’s variable is introduced by accommodation (following van der Sandt, 1992). This is given for sentence (2)-b in (6). (6)

[Er is een bal x]accommodation De jongen heeft x twee keer gegooid there is a ball x the boy has x two times thrown ‘There is a ball x. The boy has thrown x twice.’

Accommodation operates at the level of discourse. Variables introduced at the top-most level of discourse representation are assumed to be interpreted existentially (cf. Kamp and Reyle, 1993). Indefinites whose variables are bound in this way thus automatically escape the scope of any sentential adverbials or negation, whereas indefinites which are incorporated into the verb will remain in the scope of such elements. As the English glosses to the scrambled sentences in (2)-b and (3)-b indicate, one of the few ways of indicating the specific (or ‘free variable’) reading in English is to use an adjective such as certain. Using the cardinal one, as in (7), leads to a preference for a specific interpretation, but a non-specific interpretation is not ruled out. (7)

The boy threw one ball twice

To sum up, then, scrambling of indefinite NP objects can be said to disambiguate meaning. Acquiring scrambling requires the learner to come to know that: (i) Dutch has the syntactic option of scrambling, (ii) scrambled indefinite object NPs have a specific interpretation, and crucially, (iii) these NPs cannot have a non-specific interpretation. 3. The production of scrambled indefinite objects This section investigates whether children produce the targetlike form (scrambled vs. nonscrambled) in the context consistent with the relevant interpretation (specific vs. non-specific).

S. Unsworth / Lingua 117 (2007) 1930–1958

1935

3.1. L1 children L1 scrambling studies have found that before children scramble in a targetlike fashion, they pass through stages where they either fail to scramble at all and/or they scramble optionally (e.g. Eisenbeiss, 1994; Schaeffer, 2000). In order to determine whether L1 children produce the scrambled and non-scrambled forms in the targetlike contexts, a modified version of Schaeffer (2000) was carried out with monolingual Dutch-speaking 5-year-olds (n = 13) and adult native-speakers (n = 11).4 Subjects were presented with stories consistent with either a specific or non-specific interpretation of the object. After each story, a puppet asked a question designed to elicit a scrambled object in the specific indefinite condition and a non-scrambled object in the non-specific condition. All test items involved negation. Example scenarios are provided in (8) and (9). (Small capitals indicate sentential stress where relevant.) A scrambled response is expected in scenario (8) and a non-scrambled response in (9). (8) Context:

Example scenario for specific indefinite condition Mickey staat in de tuin Mickey stands in the garden ‘Mickey is standing in the garden.’

Mickey:

Kijk, wat een mooie bloemen, zeg! Gele bloemen, rode bloemen. look what a beautiful flowers say yellow flowers red flowers ‘Look, what beautiful flowers!’ ‘Yellow flowers, red flowers.’ Ik hou van bloemen. Ik ga ze plukken. I love from flowers I go them pick ‘I love flowers. I’m going to pick them.’ Maar dan is de tuin leeg but then is the garden empty ‘But then the garden will be empty.’ Dus een rode bloem ga ik NIET plukken so a red flower go I not pick ‘So I’m NOT going to pick a red flower.’

Puppet:

WAT gaat Mickey niet doen? What goes Mickey not do ‘WHAT’S Mickey not going to do?’

a.

Mickey gaat een (rode) bloem niet plukken Mickey goes a (red) flower not pick ‘Mickey’s not going to pick a (certain) (red) flower.’

b.

4

Mickey gaat niet een/geen (rode) bloem plukken Mickey goes not a no (red) flower pick ‘Mickey’s not going to pick a(ny) (red) flower.’

The difference between specific and non-specific indefinites was unclear in the original experiment (see Unsworth, 2005a:111–112 for relevant discussion).

1936

(9) Ernie:

S. Unsworth / Lingua 117 (2007) 1930–1958

Example scenario for non-specific indefinite condition Mmm, ik heb zin om iets te vangen met mijn I have desire to something to catch with my hengel fishing rod ‘Mmm, I fancy catching something with my fishing rod.’ Kijk, een vijver! Er zitten kikkers in. Look a pond there sit frogs in ‘Look, a pond! There are frogs in it.’ Maar een kikker is moeilijk om te vangen but a frog is difficult around to catch ‘But a frog is difficult to catch.’ Ik denk niet dat ik dat kan I think not that I that can ‘I don’t think I can do that.’ Dus dat ga ik NIET doen so that go I not do ‘So I’m NOT going to do that.’

Puppet:

WAT gaat Ernie niet doen? What goes Ernie not do ‘WHAT’S Ernie not going to do?’

a.

Ernie gaat een kikker niet vangen Ernie goes a frog not catch ‘Ernie’s not going to catch a (certain)frog.’

b.

Ernie gaat niet een/geen kikker vangen Ernie goes not a no frog catch ‘Ernie’s not going to catch a(ny) frog.’

Each condition contained six tokens. However, both adult and child subjects sometimes produced responses which, although acceptable in terms of their content, did not contain an indefinite object NP. Examples of such alternative responses in the specific indefinite condition are provided in (10). When subjects responded in this way, the experimenter playing the puppet pretended not to understand so that they would reformulate their answer, in the hope that this would lead to a useable response. This was a relatively successful tactic. (10) a.

Alternative responses in the specific indefinite condition Mickey gaat dat ene bloemetje niet plukken Mickey goes that one flower not pick ‘Mickey isn’t going to pick that one flower.’

b.

Mickey gaat niet alle bloemetjes plukken Mickey goes not all flowers pick ‘Mickey isn’t going to pick all flowers.’

S. Unsworth / Lingua 117 (2007) 1930–1958

c.

Mickey gaat er e´e´ntje laten liggen Mickey goes one leave lie ‘Mickey is going to leave one of them.’

d.

Mickey gaat het laatste bloemetje niet plukken Mickey goes the last flower not pick ‘Mickey isn’t going to pick the last flower.’

1937

As a result of the use of these acceptable, alternative responses, the number of tokens in each condition differs per subject. The number of tokens per child ranged from two to six, with each child producing on average 4.3. 3.1.1. Group results The group results for the L1 children are presented in Table 1.5 Both the adults and children make a clear distinction between the specific and the non-specific indefinite conditions, scrambling in the former and but (generally) not in the latter (Wilcoxon signed ranks test: Z = 3.064, p = .002, for the adults; Z = 2.677, p = .007, for the children).6 The two groups do not differ significantly from each other in the non-specific indefinite condition (Mann–Whitney test (two-tailed): Z = 1.734, p = .083), but they do in the specific indefinite condition (Mann– Whitney test (two-tailed): Z = 2.417, p = .016).7 Despite being significantly different from the adults in these two conditions, the children are clearly on their way to becoming adults, scrambling at an average rate of 61.3%. 3.1.2. Individual results A comparison of the individual results in the two indefinite conditions shows that virtually all (10/12) children who produce scrambled objects in the specific indefinite condition and who also produce tokens in the non-specific condition make the targetlike distinction: they scramble the Table 1 L1 children and L1 adults (production): group results Group

n

Condition Specific indefinite (target: scrambled)

5-year-olds Adults

13 11

Non-specific indefinite (target: non-scrambled)

%

S.D.

%

S.D.

61.3 92.9

35.0 12.6

15.2 0

31.3 0

Average percentage scrambled objects. 5

The figures are presented as the average percentage of scrambled responses per group because, as a result of the number of tokens differing per subject, it is this figure which is used in the statistical tests. 6 When adults scramble, they usually (88.1% (37/42)) realise the object as the stressed indefinite/cardinal e´e´n as opposed to the unstressed indefinite een. The status of e´e´n, that is, whether it is a stressed indefinite or a cardinal, is unclear. There is, however, evidence to suggest that it does not behave like other cardinals (Barbiers, 2005) and for this reason, e´e´n is included with the unstressed indefinite een here. The L1 children realise 47.1% (30/57) of scrambled objects as e´e´n. 7 The fact that the p-value in the non-specific indefinite condition is approaching significance is undoubtedly because the adults never scramble, whereas there are three scrambled tokens in the child data (out of 24 for the group as a whole). See section 3.2.2 for further discussion of these tokens.

1938

S. Unsworth / Lingua 117 (2007) 1930–1958

object (sometimes (n = 6) or always (n = 4)) in the specific indefinite condition and they leave it unscrambled in the non-specific indefinite condition. For most children (9/12), this distinction is categorical and for two, it is relative. The exception is one child who scrambles all the time in both conditions. 3.1.3. Developmental sequence and summary Combining these data with previous findings (Schaeffer, 2000), the following developmental sequence for the L1 acquisition of scrambled (specific) indefinite objects can be inferred: (11)

Stage I II-a II-b

Description No scrambling Optional scrambling Scrambling

Word order patterns produced Negation-Object-Verb only Object-Negation-Verb and Negation-Object-Verb Object-Negation-Verb only

Summarising, by the age of five, the option of scrambling indefinite object NPs is used consistently (by some children at least), and when it is used, it is in a targetlike fashion. 3.2. L2 children This section examines whether L2 children produce the targetlike form in the relevant context, and whether they pass through the same stages as L1 children. The L2 children in question are all native-speakers of English. Given that L2 children have been found to show L1 transfer in the initial stages of development, it is expected that these children will at first fail to scramble and that their utterances will reflect an underlying VO word order (as opposed to the target OVorder). Using exactly the same methodology as with the L1 children, data were collected from 25 English-speaking children acquiring Dutch. Their age at first exposure ranged from 4;0 to 7;1 (mean = 5;6, S.D. = 1;0), their age at time of testing from 5;3 to 17;4 (mean = 9;3, S.D. = 2;4), and their length of residence from 0;2 to 13;0 (mean = 3;8, S.D. = 2;6).8 All had had some Dutch language instruction at the international schools which they attended. Note, however, that scrambling is not explicitly taught in the L2 Dutch classroom. This was confirmed by Dutch language teachers as well as consultation of grammar books. Instruction is often given on (general) word order and the use of geen ‘no/not any’ (cf. (9)-b), but the interpretive facts or constraints under investigation here are not discussed. In fact, most of the (native-speaker) teachers consulted were not consciously aware of the relevant distinction.9 A picture description task was also administered to collect spontaneous data, which, following Whong-Barr and Schwartz (2002), were used to provide a global indication of L2 proficiency. This proficiency measure also allows approximate developmental stages to be inferred from the cross-sectional scrambling data in the following way. On the basis of this measure, learners are divided into low, mid and high proficiency groups. It is assumed that in their acquisition of scrambling, the low proficiency learners will be the least targetlike, the high proficiency learners will be the most targetlike, and the mid proficiency learners will be somewhere in-between these 8

Thirteen of these subjects are reported on elsewhere (Unsworth, 2004); twelve are new. The following books, which include most of the standard textbooks, were consulted: Dutch: An Essential Grammar (Schetter and van der Cruysse, 2002), Teach Yourself Dutch Grammar (Quist and Stuik, 2000), Essential Dutch Grammar (Stern, 1984), Grammatica in Gebruik. Nederlands voor Anderstaligen (Bakx et al., 1999), Help! Een Cursus voor Buitenlanders (Ham et al., 1998), Nederlands voor Buitenlanders: De Delftse Methode (Montens and Sciarone, 1991) and Code Nederlands (van Kalsbeek et al., 1998). 9

S. Unsworth / Lingua 117 (2007) 1930–1958

1939

Table 2 L2 children (production): average percentage targetlike responses (organised by proficiency group)

Low Mid High

Specific indefinite (target: scrambled)

Non-specific indefinite (target: non-scrambled)

n

%

S.D.

n

%

S.D.

10 6 7

23.3 71.7 91.7

35 40 14

8 5 6

47.9 70 83.3

52 45 41

two groups. The scrambling behaviour of the low proficiency learners is therefore considered to represent an earlier developmental stage than the mid proficiency learners, and likewise for the mid proficiency learners with respect to the high proficiency learners. Combining the experimental tasks on scrambling with an independent measure of proficiency will thus allow us to derive a developmental sequence for the acquisition of this particular property of the target language on the basis of cross-sectional data.10 Each child is assigned a proficiency score which is based on three measures, two measures of complexity and one of accuracy. Verbal density (number of finite and non-finite verbs divided by total number of T-units11) is used as a measure of morphosyntactic complexity, and lexical richness or diversity (Guiraud’s index: V/HN (where V refers to type and N to token)) is employed for lexical complexity. Morphosyntactic and lexical accuracy is calculated on the basis of the rate of errorfree T-units. These three sub-scores are combined into a single score using principal components analysis with the result that subjects each have a single standard normal (z) score as their final proficiency score. These z-scores ranged from 2.06 to 2.12 for subjects who completed the production task and from 1.95 to 1.70 for subjects who completed the comprehension task (see section 4.2). Subjects were divided into three proficiency groups (low, mid, high) on the basis of these scores. (For a detailed motivation of the proficiency measure and an outline of how each sub-score was calculated, see Unsworth, 2005a, Chapter 4.) 3.2.1. Group results Some subjects failed to produce any utterances containing an indefinite NP: two in the specific condition and six in the non-specific condition. For this reason, the number of children per proficiency group differs per condition. The group results for the remaining subjects are given in Table 2. In the specific indefinite condition, the percentage of scrambled objects increases with proficiency, the high proficiency group scrambling consistently.12 Games Howell post-hoc tests reveal that the mid and high proficiency groups do not differ significantly from the native adults (reported on in the previous section), but the low proficiency group does (mean difference (MD): 70.1%, p = .001 for low proficiency group; 21.2%, p = .619 for the mid proficiency group; 1.9%, p = .998 for the high proficiency group). The low proficiency group is also significantly different from the high proficiency group (MD: 68.8%; p = .001). When subjects fail to scramble, they produce one of the following two types of utterance: 10 Ideally, such a developmental sequence should be confirmed within one and the same learner by means of a longitudinal study on scrambling. 11 A T-unit is defined as ‘one main clause plus whatever subordinate clause and nonclausal expressions are attached to or embedded within it’ (Hunt, 1970:14). 12 L2 children realise 52.6% (30/57) of scrambled objects as e´e´n rather than een.

1940

(12)

S. Unsworth / Lingua 117 (2007) 1930–1958

a.

Mickey gaat niet een bloem plukken Mickey goes not a flower pick

b.

Mickey gaat niet plukken een bloem Mickey goes not pick a flower

With a Negation-Object-Verb order, the utterance in (12)-a is similar to the non-scrambled utterances produced by L1 children. The utterance in (12)-b, with a Negation-Verb-Object order, is never produced by L1 children, however. The L2 children’s responses in this condition correlate with age at time of testing (r = .905, p = .000) and proficiency (r = .721, p = .000), but not with length of exposure (r = .155, p = .481). Proficiency is the best predictor of targetlike responses (r2 = .519). In the non-specific indefinite condition, the percentage of targetlike (non-scrambled) responses also increases with proficiency.13 Two types of utterance were counted as nontargetlike (at this stage of the analysis at least): those containing a scrambled indefinite, as in (13)-a, and those with Negation-Verb-Object order, as in (13)-b. (13)

a.

Ernie gaat een kikker niet vangen Ernie goes a frog not catch

b.

Ernie gaat niet vangen een kikker Ernie goes not catch a frog

In the non-specific condition, there is no statistically significant difference between the different proficiency groups (Kruskal–Wallis: d.f. = 2, x2 = 2.588, p = .274), but when the native adults are included in the analysis, the results are significant (Kruskal–Wallis: d.f. = 3, x2 = 9.440, p = .024). Responses in this condition did not correlate with length of exposure (r = .316, p = .187), age at time of testing (r = .111, p = .652) or proficiency (r = .220, p = .380). 3.2.2. Individual results When individual subjects’ responses on the two conditions are compared, the following pattern emerges. There are six children who never scramble at all, in either condition. A further eight sometimes or always scramble in the specific indefinite condition, but never in the nonscrambled condition, and five sometimes or always scramble in both conditions.14 Note that producing the scrambled form in the non-specific condition is unexpected rather than ungrammatical. A scrambled form with a targetlike specific reading is not inconsistent with the context. Take the example scenario given in (9). The scrambled form, given in (9)-a, states that there is a frog such that that frog has not been caught. This is true in the given context because 13 Two types of non-scrambled utterance were counted as targetlike, namely those containing the targetlike, suppletive form geen ‘no’ and those with the non-suppletive form niet een ‘not a’. This is because use of the suppletive form arguably reflects the acquisition of lexical knowledge rather than syntactic knowledge, the latter being the focus here, and the utterance containing the niet een form reflects targetlike knowledge with respect to the indefinite’s syntactic position. 14 The data from these five children might suggest that there is something like a ‘scramble everything’ stage. The most important argument against such a stage stems from learnability considerations. If children were to pass through a stage in which they consistently scramble all indefinite objects across negation, they would never encounter any evidence that, when a non-specific interpretation is intended, the scrambled form is non-targetlike. This is because a negative sentence containing a non-scrambled indefinite, such as (9)-b, entails the equivalent sentence with a scrambled indefinite, as in (9)-a: if it is true that Ernie did not catch a frog, then it is also true that – assuming the set of frogs is not empty – there is a frog which Ernie did not catch. See Unsworth (2005a:258) for further arguments against such a stage.

S. Unsworth / Lingua 117 (2007) 1930–1958

1941

Table 3 Developmental sequence for L2 acquisition of scrambling of specific indefinite object NPs Stage

I II III-a III-b

Description

No scrambling No scrambling Optional scrambling Scrambling

Word order patterns produced

Negation-Verb-Object Negation-Object-Verb Object-Negation-Verb Object-Negation-Verb

only only and Negation-Object-Verb only

No. of children in each stage Low

Mid

High

3 3 3 1

1 2 3

2 5

all the frogs have not been caught. Thus, although the experiment was set up such that subjects were discouraged from interpreting the context in this way, such an interpretation cannot be ruled out. Interestingly, it is only subjects who scrambled in the specific condition who also scrambled in the non-specific condition (likewise for the three L1 children who scrambled (one token each) in the non-specific condition). In other words, there are no subjects who scramble in the nonspecific condition, but not in the specific condition. 3.2.3. Developmental sequence and summary Using the proficiency levels as a guide, it is possible to infer the following developmental sequence for the acquisition of scrambled (specific) indefinites by English-speaking L2 children shown in Table 3. Stage I results from the transfer of the left-headed VP in English. Other studies examining the L2 acquisition of a Germanic OV language by native-speakers of a VO language have also observed the same word order pattern in the early stages of L2 development; see Clahsen (1988) on adult L1 Romance/L2 German, Felix (1977) and Molony (1977) on child L1 English/L2 German, and le Roux and Wilsenach (1999) on child L1 English/L2 Afrikaans. That the Negation-Verb-Object order constitutes the initial stage in the developmental sequence for the acquisition of scrambling by English speakers is supported by the observation that this is the only stage with exclusively low proficiency children. The underlying syntactic structure in this stage is given in (14). (14)

[IP Mickey [I will [NegP not [VP [V pick] a flower]]]]

The observation that transfer characterises the initial stages of the child L2 acquisition process is consistent with previous findings. It is this stage which distinguishes child L2 acquisition from child L1 acquisition. The remaining three stages are identical to those proposed for L1 acquisition. Summarising, the option of scrambling the indefinite object is used by most (16/23) subjects. Dividing the subjects into proficiency groups suggests that in their acquisition of scrambled indefinite (specific) objects, a ‘no scrambling’ stage precedes an (optional) scrambling stage. Furthermore, the L2 children generally make targetlike distinctions: the scrambled form is produced in the specific condition and the non-scrambled form in the non-specific condition. 4. The comprehension of scrambled indefinite objects This section determines whether children assign the targetlike interpretation (specific vs. nonspecific) to the relevant form (scrambled vs. non-scrambled indefinite).

1942

S. Unsworth / Lingua 117 (2007) 1930–1958

4.1. L1 children Studies on the L1 acquisition of the comprehension of scrambled indefinite object NPs have shown that (i) in the early stages, no distinction is made between scrambled and non-scrambled objects (Bergsma-Klein, 1996; Kra¨mer, 2000), and (ii) the targetlike specific reading of the scrambled indefinite object is not acquired until very late (Kra¨mer, 2000; Philip, 2003). Adopting Van Geenhoven’s (1998) analysis (see section 2), Kra¨mer (2000) proposes that children’s failure to interpret scrambled objects in a targetlike fashion results from poor discourse integration. She claims that discourse integration, and in particular ‘bridging’ (Clark, 1977), is required in order to identify the accommodation site of the indefinite’s variable. Children experience problems integrating discourse until relatively late (Karmiloff-Smith, 1979; Yuill and Oakhill, 1991). When discourse integration fails, the indefinite cannot be interpreted as a free variable and the default process of incorporation takes place. As a result, the indefinite object is interpreted predicatively, that is, as though it were not scrambled. The oldest child L1 subjects in Kra¨mer’s (2000) study were seven years old and their responses differed almost maximally from those of adult native-speakers. In order to determine just how late acquisition takes place, Kra¨mer’s truth value judgement task on scrambling across negation, as well as a second task on scrambling across the adverbial twee keer ‘twice’, were carried out with 99 monolingual L1 Dutch children aged between seven and 13 years old. In both tasks, each item was presented as a story with four pictures. In each story, a protagonist is seen to manipulate two out of three objects, the third of which is always left unmanipulated. In the negation task, this is highlighted physically by the experimenter. A silly puppet, who is asked to say what happened, then produces a scrambled utterance, such as (15). In the scenario corresponding to this particular sentence, a boy catches two out of three fish, leaving one fish remaining in the pond. The scrambled sentence is thus consistent with the context.15 (15)

De jongen heeft een vis niet gevangen the boy has a fish not caught ‘The boy didn’t catch a (certain) fish.’

[target: acceptance]

The task on scrambling with twee keer ‘twice’ was similar, except that the three objects were always different colours and the one unmanipulated object was not highlighted. For example, in a story about a girl and three monkeys, the girl tickles a (red) monkey and then another (brown) one. The puppet then produces an utterance containing either a scrambled or non-scrambled indefinite object NP, as in (16)-a and (16)-b, respectively. The scrambled utterance is inconsistent with the context because it states that one particular object was manipulated twice, whereas in the non-scrambled utterance, it is irrelevant whether one or two objects were manipulated. The nonscrambled utterance is thus consistent with this context. (16)

15

a.

Het meisje heeft een aap twee keer gekieteld the girl has a monkey two times tickled ‘The girl tickled a (certain) monkey twice.’

[target: rejection]

b.

Het meisje heeft twee keer een aap gekieteld the girl has two times a monkey tickled ‘The girl tickled a(ny) monkey twice.’

[target: acceptance]

The non-scrambled equivalent, which is inconsistent with this context, was not tested because Kra¨mer (2000) shows that this is already interpreted in a targetlike (non-specific) fashion at the age of four.

S. Unsworth / Lingua 117 (2007) 1930–1958

1943

There were five test items in each task; in the negation task there were eight fillers and in the twee keer task 16. These always contained the same word order (viz. scrambled or non-scrambled) as the test items (see Unsworth, 2005a, Chapter 6, for more details). As in Kra¨mer’s original experiment, the intonation used for all items was the most natural for the scrambled order, namely with the pitch rise for the VP starting on the negator niet ‘not’, and the indefinite determiner was always pronounced as een \3n\ rather than the stressed indefinite/cardinal form, e´e´n ‘A/one’. Although the use of the stressed indefinite/cardinal form e´e´n is arguably more natural in scrambled position than the unstressed indefinite een, the latter form was used for two reasons. First, using een rather than e´e´n ensures that it is the relation between the relative position of the object and its interpretation which is being tested (as opposed to the form of the object and its interpretation), and second, as noted in section 2, the cardinal form in English, one, often induces a specific interpretation. If the English-speaking L2 children were to allow the specific interpretation of the e´e´n form in scrambled position, it would be unclear whether this was the result of targetlike knowledge on formmeaning restrictions or the result of L1 transfer. 4.1.1. Group results In the non-specific condition for scrambling across twee keer (cf. (16)-b), none of the child groups differed significantly from the adults (ANOVA: d.f. = 5, F = 1.224, p = .305). Table 4 presents the group results for the negation task and the specific condition in the twee keer task.16,17 There is a clear developmental progression from non-targetlike towards targetlike with increasing age. Children are not (completely) targetlike until very late, however, and with negation, even the oldest children are still not completely targetlike. Post-hoc Games Howell tests were used to determine which groups differed from each other. The results are given in Table 5. Significant differences are highlighted. For both scrambling across negation and scrambling across twee keer, there appears to be a leap towards targetlike behaviour between ages 8 and 9. For scrambling across twee keer, there is a further leap between ages 11 and 12. Wilcoxon signed-ranks tests were carried out to determine whether the percentage of targetlike responses in the negation task differs from the percentage of targetlike responses in the twee keer task. The results, given in the final column in Table 4, show that for most of the child L1 groups and for the adult controls, no distinction is made between the two. For the 12- and 13-year-olds, however, there is a significant difference between responses on the negation task and responses on the twee keer task. This is because these children are targetlike on twee keer, but not on negation, whereas the younger groups are equally non-targetlike on both (and the adults are equally targetlike on both). 4.1.2. Individual results Following Kra¨mer (2000), subjects were classified as having an acceptance, rejection or mixed response pattern depending on whether they accepted all or all but one of the test items, rejected all or all but one of the test items, or did something in-between. The individual results are 16

Children with two or more incorrect fillers were excluded; these were one 8-year-old and one 12-year-old in the negation task and one 12-year-old in the twice task. As a result of these exclusions, the number of children in each of the two tasks differs for the relevant age groups in Table 4. Children who were excluded from one of the two tasks were not included in the Wilcoxon signed-ranks test reported in the final column of Table 4. 17 Unlike in the production experiment, the number of tokens (n = 5) per child was identical. Therefore, the raw numbers are also provided here (cf. fn. 5).

1944

S. Unsworth / Lingua 117 (2007) 1930–1958

Table 4 L1 children (comprehension): group results for negation and twee keer tasks

presented for negation in Fig. 1 and for twee keer in Fig. 2. The targetlike response pattern is in black. In the negation task this is acceptance, whereas in the twee keer task this is rejection. Most (12/14) of the adults accept all or all but one of the test items. The remaining two adult control subjects accept three out of the five items. The majority of the older children also have a pattern of acceptance. Most of the children in the younger two groups, however, have a rejection pattern. A clear developmental pattern is thus found, with a difference between children younger than nine years old, on the one hand, and children aged nine years and older, on the other. This tallies with the differences observed in Tables 4 and 5. Turning now to the response patterns on the twee keer task, let us begin with the adult controls. Most of the adults (12/14) reject the non-specific interpretation for the indefinite object NP scrambled across twee keer as expected.18 The majority of the children in the youngest two groups have a (non-targetlike) pattern of acceptance. At the other end of the spectrum, virtually all the children in the oldest two groups have a (targetlike) pattern of rejection. In the 9- and 11-year-old groups, the majority of subjects have a pattern of rejection or a mixed pattern, but still a not insignificant number of subjects has a pattern of acceptance. Broadly speaking, there appear to be three different groups of children: the 7- and 8-year-olds, who are generally nontargetlike, the 9- and 11-year-olds who are more targetlike than the youngest children but still not at adultlike levels and the 12- and 13-year-olds who are more or less consistently targetlike. Again, this tallies with the differences observed in Tables 4 and 5.

18

Contrary to expectation, however, one adult consistently accepts the scrambled test sentences. Initially, this subject spontaneously rejected the first test item, as expected, but subsequently, during the following filler item, he decided that if the NP was not realised as e´e´n, the scrambled sentence was acceptable after all. The production data from the native adults presented above indicate that e´e´n is indeed the preferred realisation of the scrambled indefinite object (cf. fn. 6). Nevertheless, the observation that the vast majority of adult native speakers reject the non-specific reading for the scrambled indefinite when it is not realised as e´e´n demonstrates that it is the position rather than the form of the NP which is associated with the specific meaning. It seems that this particular subject was applying metalinguistic knowledge in this task. The remaining adult subject rejects three scrambled test items and accepts two. This could be considered as noise.

S. Unsworth / Lingua 117 (2007) 1930–1958

1945

Table 5 L1 children (comprehension): results of post-hoc Games–Howell test for negation task and specific condition in twee keer task

1946

S. Unsworth / Lingua 117 (2007) 1930–1958

Fig. 1. L1 children (comprehension): distribution of individual response patterns on negation task.

There is a significant moderate correlation between children’s responses on the two tasks (r = .558, p = .000). Combined with the group results reported in Table 4, this suggests that for the younger groups at least, subjects’ responses on the two tasks are more or less the same. In order to ascertain whether this really is the case, it is necessary to examine the results on the two tasks for each subject individually. The following broad patterns can be discerned. There are: (a) 50 children who have a mixed or targetlike response pattern on the negation task and on the twee keer task; (b) nine children who have a mixed or targetlike response pattern on the negation task but a non-targetlike response pattern on the twee keer task; (c) 14 children who have a mixed or targetlike response pattern on the twee keer task but a non-targetlike response pattern on the negation task; and (d) 24 children who have a non-targetlike response pattern on both tasks.

Fig. 2. L1 children (comprehension): distribution of individual response patterns on twee keer task (scrambling condition).

S. Unsworth / Lingua 117 (2007) 1930–1958

1947

Amongst the 50 subjects with the first pattern, 43 have a targetlike response pattern on both tasks and as expected, the number of children with this pattern increases with age. There are a number of children (23.7% (23/97)) who are more targetlike on one task than on the other. These are more or less equally split across both tasks, that is, there are around the same number of children who are more targetlike on the negation task than on the twee keer task (pattern (b)) as children who are more targetlike on the twee keer task than on the negation task (pattern (c)). Neither of these two patterns is specific to a particular age group, that is, both patterns are relatively evenly distributed across the different age groups. Note, however, that in the oldest group, there are only subjects who are better on twee keer than on negation. It seems, thus, that for around one fifth of the child L1 subjects, their responses on the two tasks do not necessarily match each other. How this finding should be interpreted will be discussed in section 6.1. 4.1.3. Summary The results confirm previous findings showing that the scrambled form is first assigned a nonspecific interpretation before a targetlike specific interpretation. The targetlike interpretation is not completely acquired until (at least) age 12 years. 4.2. L2 children In terms of age at time of testing, the L2 children in the present study are of a comparable age to the L1 children reported on above. The L2 children who are the same age as those L1 children (claimed to) have limited discourse integration are predicted to behave similarly to the L1 children, that is, they should fail to assign a specific interpretation to a scrambled indefinite NP. Note, however, that these L2 children are also expected to provide non-targetlike responses as a result of transfer. Despite the overlap between the predictions made by L1 transfer and limited discourse integration, it should, nevertheless, be possible to tease these two factors apart, namely in the following way. Consider limited discourse integration as the crucial factor first: on the assumption that discourse integration develops with age, then on this scenario, a correlation between age at time of testing and targetlike scrambling responses would be expected, irrespective of proficiency level. Crucially, if the age-effect in child L1 acquisition is also at play in child L2 acquisition, younger L2 children of even the highest proficiency level will fail to assign the targetlike specific interpretation to scrambled indefinite objects. Next, consider L1 transfer as the crucial factor: in this case, a correlation is expected between L2 proficiency and targetlike scrambling responses, that is, as L2ers become increasingly proficient in the L2, they should become more targetlike. Importantly, this should be possible at any age. If both of these factors, that is, if (age-dependent) limited discourse integration and L1 transfer are at play, it would be expected that only older, high proficiency L2 children would be more likely to have targetlike responses. In order to test these predictions, data were collected from 36 English-speaking children who were learning Dutch as a second language. Their age at first exposure ranged from 4;0 to 7;3 (mean = 5;7; S.D. = 1;0), their age at time of testing from 7;3 to 14;11 (mean = 10;4; S.D. = 2;3) and their length of residency from 0;7 to 10;5 (mean = 4;9; S.D. = 2;11). Again, all had had some Dutch language instruction at international schools which they attended. Note that the necessity of the age-based analysis only became apparent once data collection had commenced. For this reason, the subjects are not evenly distributed across the different age groups. Twelve of these children also completed the production task. Their responses on the two experiments will be compared in section 5. Exactly the same method was used as with the L1 children.

1948

S. Unsworth / Lingua 117 (2007) 1930–1958

Table 6 L2 children (comprehension): group results (organised by age at testing) on negation and twee keer Age group

n

7-year-olds 8-year-olds 9-year-olds 10-year-olds 11-year-olds 13/14-year-olds

3 10 3 5 5 8

Negation (target: yes)

n

Mean %

S.D.

0 (0/15) 38.0 (19/50) 13.3 (2/15) 48.0 (12/25) 48.0 (12/25) 30.0 (12/40)

0 49.4 23.1 50.2 50.2 33.8

5 10 3 5 5 8

Twee keer (target: no)

Wilcoxon signed-ranks test

Mean %

S.D.

Z

Significance

0 (0/15) 0 (0/50) 86.7 (13/15) 52.0 (13/25) 32.0 (8/25) 67.5 (27/40)

0 0 23.1 50.2 46.0 42.7

.000 1.890 1.633 .000 .743 1.876

p = 1.000 p = .059 p = .102 p = 1.000 p = .458 p = .061

4.2.1. Group results In the non-specific condition for scrambling across twee keer (cf. (16)-b), the child groups did not differ significantly from the native adults (Kruskal–Wallis: d.f. = 6, x2 = 7.364, p = .289). Table 6 presents the results for the specific condition in the twee keer task and for the negation task. Unlike the L1 children, there is no clear-cut developmental progression from non-targetlike younger children to more targetlike older children. The only significant differences between groups on the negation task are between the 7-year-olds and native adults (Games Howell post-hoc: MD = 92.9%, p = .000) and between the 13/14-year-olds and the native adults (MD = 62.9%, p = .010). On the twee keer task, both the 7-and the 8-year-olds differ significantly from the 13/14-year-olds (both groups: MD = 67.5%, p = .022) and the native adults (both groups: MD = 90.0%, p = .000). Table 7 presents the results organised by proficiency level. On the negation task, there is no clear pattern of development. Games-Howell post-hoc tests reveal that none of the proficiency groups differ from each other, and that the low and high proficiency groups differ from the native adults (low: MD = 58.8%, p = .005; high: MD = 67.0, p = .000), but the mid proficiency group does not (MD = 43.1, p = .171). On the twee keer task, there is some evidence of development between the low and mid proficiency groups, although the only significant difference is between low and the high groups (MD = 43.2, p = .022). Once again, the low and high proficiency groups differ from the native adults (low: MD = 83.4%, p = .000; high: MD = 40.0, p = .038), but the mid proficiency group does not (MD = 37.4, p = .331). It should be noted, however, that the mid proficiency group has considerably fewer subjects than the other two groups and that this may have contributed to this result. Once again, Wilcoxon signed-ranks tests were carried out to determine whether the percentage of targetlike responses in the negation task differs from the percentage of targetlike responses in the twee keer task. The results, given in the final column in Tables 6 and 7, show that Table 7 L2 children (comprehension): group results (organised by proficiency) on negation and twee keer tasks

Low Mid High

n

Negation (target: yes) %

S.D.

12 6 16

35.0 (21/60) 50.0 (15/30) 26.3 (21/80)

.46 .41 .40

n

14 6 16

Twee keer (target: no)

Wilcoxon signed-ranks test

%

S.D.

Z

Significance

7.1 (5/70) 53.3 (16/30) 50.0 (40/80)

.27 .45 .47

1.294 .105 1.557

p = .196 p = .916 p = .120

S. Unsworth / Lingua 117 (2007) 1930–1958

1949

for all groups, there is no significant difference between responses on the two tasks. For the 8-year-olds and the 13/14-year-olds, the difference is almost significant, however. The 8-year-olds are better on negation than on twee keer and vice versa for the 13/14-year-olds. On the negation task, there is no correlation between targetlike responses and length of exposure (r = .057, p = .748), proficiency score (r = .042, p = .819) or age at time of testing (r = .159, p = .346). On the twee keer task, a significant moderate correlation was found between targetlike responses in the scrambling condition and length of exposure (r = .530, p = .001) and age at time of testing (r = .517, p = .001), and a weak-moderate correlation between targetlike responses and proficiency score (r = .456, p = .007). A regression analysis with proficiency and age at time of testing shows that age is the best predictor of targetlike responses (r2 = .244, p = .003). However, when length of exposure is included in the analysis, this is the best predictor (r2 = .257, p = .002). Note, however, that the amount of variance which these predictor variables explain in the data is minimal at 25.7% for length of exposure and 24.4% for age at time of testing (cf. 51.9% of variation is explained by proficiency in the production data). Because the number of targetlike responses is correlated with proficiency and the oldest L2 children are generally more proficient, there is little variation in terms of scrambling responses in this oldest group. This is unfortunate because this is the only group which contains children who are as old as or older than the child L1 groups which were found to be consistently targetlike, that is, the 12- and 13-year-olds. In other words, to tease apart the factors of age and transfer in the manner outlined in the preceding section, it would be necessary to have a larger group of older L2 children at different proficiency levels. Close examination of the available data reveals some interesting patterns, however. In the 13/14-year-old group, there are two non-targetlike subjects (out of eight); one has the lowest proficiency score in the group and the other the fourth lowest. Amongst the 15 children in the youngest two groups, four have a proficiency score which is higher than the lowest proficiency score for the targetlike older children and, nevertheless, these four provide non-targetlike responses, suggesting that age is the crucial factor. 4.2.2. Individual results Individual L2 children were also categorised according to the response pattern which they produced: acceptance, rejection or mixed. The results for the negation task are presented in Fig. 3 and for the twee keer task in Fig. 4.

Fig. 3. L2 children (comprehension): distribution of individual response patterns (by age at time of testing) on negation task.

1950

S. Unsworth / Lingua 117 (2007) 1930–1958

Fig. 4. L2 children (comprehension): distribution of individual response patterns (by age at time of testing) on twee keer task (scrambling condition).

A non-targetlike pattern of rejection is observed in the majority of children in the youngest three age groups and in half of those in the oldest age group. In the 8-, 10- and 11-year-old groups, just under half of subjects have the targetlike pattern of acceptance. In the oldest group, there is just one subject with this pattern and three with a mixed pattern, that is, three subjects who sometimes correctly accept the specific interpretation for the indefinite object scrambled across negation. When the children are divided according to proficiency, the following pattern is observed: the majority of subjects in the low and high proficiency groups have a (non-targetlike) pattern of rejection. The children in the mid proficiency group are evenly distributed across the three response patterns. Around one third of the low and mid groups and one fifth of the high group have a targetlike pattern of acceptance. On the twee keer task, all the children in the 7- and 8-year-old groups have a non-targetlike pattern of acceptance. In all remaining groups except the 12-year-olds, the majority of children have either a mixed or rejection pattern, that is, they correctly reject the non-specific interpretation for the scrambled form some or all of the time. (With respect to the 9-year-olds, it is perhaps worth bearing in mind that there are only three subjects in this group.) When the children are divided according to proficiency, a clear developmental pattern is observed. The number of children with the targetlike rejection pattern increases with proficiency: one in the low group, two in the mid group and eight in the high group. Two mid proficiency children have a mixed pattern, as does one high proficiency child. Ten children make a relative distinction between the scrambled and non-scrambled conditions, accepting the non-scrambled utterances more frequently than the scrambled utterances. Seven children make an absolute distinction between the two conditions: they consistently accept the nonscrambled utterances and consistently reject the scrambled utterances. There was no significant correlation between subjects’ responses on the two experiments for either group (L2 children: r = .040, p = .821). As with the L1 children, the results were examined for each subject individually. There are: (a) seven children who have a mixed or targetlike response pattern on the negation task and on the twee keer task; (b) seven children who have a mixed or targetlike response pattern on the negation task but a non-targetlike response pattern on the twee keer task; (c) seven children who have a mixed or targetlike response pattern

S. Unsworth / Lingua 117 (2007) 1930–1958

1951

on the twee keer task but a non-targetlike response pattern on the negation task; (d) 13 children who have a non-targetlike response pattern on both tasks. Out of the seven children with pattern (a), two have a targetlike pattern on both tasks. These children are both in the high proficiency group. There are roughly the same number of subjects who are more targetlike on the negation task than on the twee keer task as those with the opposite pattern. These children are split relatively evenly across the three proficiency groups. The children with pattern (c), that is, those who are better on twee keer than on negation, predominate in the high proficiency group and the older age groups. 4.2.3. Summary L2 children are able to acquire the interpretive constraints on scrambling in the sense that there are children who make the relevant distinctions.19 On the negation task, the various age groups did not significantly differ from each other, whereas on the twee keer task, significant differences were observed between younger and older children. These differences resemble those found in the child L1 data. 5. Production and comprehension compared This final section compares the production and comprehension data for a subset of the L2 children (n = 11) and asks whether the L2 children show the same discrepancy between production and comprehension as observed in the child L1 data.20 Subjects were always tested on production before comprehension, within a period of approximately 2 weeks. Comparing production of scrambled objects (across negation) with comprehension of scrambled objects (across negation and twee keer), several responses patterns are observed: (i) always targetlike production with non-targetlike comprehension (two subjects on the negation task and five on the twee keer task); (ii) sometimes targetlike production with non-targetlike comprehension (two for negation, two for twee keer); (iii) always targetlike production and always targetlike comprehension (five for negation, two for twee keer); (iv) sometimes targetlike production but always targetlike comprehension (one for negation, one for twee keer); and (v) sometimes targetlike production and sometimes targetlike comprehension (one for negation). At first sight, it appears that every possible pattern occurs. There is, however, one clear generalisation which can be made over these data. Subjects who (always or sometimes) have targetlike comprehension, (always or sometimes) have targetlike production. This does not hold in the opposite direction, however: subjects who (always or sometimes) have targetlike production do not (always) have targetlike comprehension. These data are thus suggestive of a discrepancy between production and comprehension: there are subjects with targetlike production but non-targetlike comprehension and those subjects who have (partial) targetlike comprehension always have (partial) targetlike production. This pattern replicates the direction of the production-comprehension discrepancy observed in L1 children.21

19 The acquisition of the interpretive constraints on scrambled indefinite objects presents a poverty-of-the-stimulus problem – see Unsworth (2005a, 2005b) for arguments. 20 A twelfth subject also completed both tasks but she only produced one token with an indefinite in the production task. She is hence excluded from the present discussion. 21 Note, however, that there are no subjects who fail to produce scrambled objects. The claim that there is a productioncomprehension discrepancy would be further strengthened if it could be shown that all subjects who fail to produce scrambled objects also fail to interpret them correctly.

1952

S. Unsworth / Lingua 117 (2007) 1930–1958

6. Discussion and conclusion Two experiments on Dutch scrambling were conducted with L1 and L2 children. The first, a production experiment, was used to determine whether children produced the targetlike form (scrambled vs. non-scrambled) in the context consistent with the relevant (specific vs. nonspecific) interpretation of the indefinite object. Most of the L1 and L2 children made the relevant distinction: on the whole, when they scrambled, this was restricted to specific indefinite objects. It was suggested that prior to scrambling consistently, both groups of children pass through a no scrambling stage, followed by an optional scrambling stage. When the low proficiency L2 children failed to scramble, they often produced utterances with Negation-Verb-Object order, transferred from their L1 English. The answer to the first research question posed in section 1, namely whether, in production, children pass through the same stages as L1 children, must therefore by a qualified yes: the developmental stages which the L2 children pass through include those of L1 children, but they also include an initial stage of L1 transfer. The existence of an L1 transfer stage is in accordance with previous findings on child L2 acquisition. The second experiment tested whether children assign the targetlike interpretation (specific vs. non-specific) to the relevant form (scrambled vs. non-scrambled). Consonant with previous findings, the L1 children regularly attributed a non-targetlike, non-specific interpretation to the scrambled form and this was found to persist until quite late, namely until (at least) age 12. A similar age effect was also observed, albeit not quite so clear-cut, for scrambling across twee keer in the child L2 data. The restricted set of subjects and the confound between age at time of testing and proficiency meant that disentangling the different factors involved is rather tricky, however. The answer to the second research question posed in section 1, namely whether, in comprehension, L2 children show the same age-related effect as L1 children, is therefore a tentative yes, at least for scrambling across twee keer. 6.1. Comparing scrambling across negation and scrambling across twee keer In both groups, we find children whose responses on one task do not necessarily correspond with their responses on the other. One approach to accounting for these results would be to take them at face value and to reason that given the observed differences between the two tasks, these tasks must be tapping different kinds of knowledge. Such a conclusion would lead us to ask whether, contrary to the analysis adopted here (and alternative analyses available in the literature), scrambling across negation somehow differs from scrambling across twee keer. To conclude on the basis of the current results that this were the case would, I fear, be a little hasty. Before we can seriously entertain this possibility, a second explanation should be considered. Both the L1 and L2 results suggest that there may be a problem with the negation task. Gualmini (2004) argues that children (and adults) accept the specific interpretation of indefinites in negative sentences more readily when negation is used felicitously. Negative sentences are easier to process in a context where certain felicity conditions are met (De Villiers and TagerFlusberg, 1975; Wason, 1965). For example, the sentence It’s not raining is easier to process if there is some expectation that it should be raining. The negative sentences used here state that there is an object (e.g. a fish) such that that particular object has not been manipulated (e.g. caught). In order for the use of such sentences to be felicitous, it is necessary to build up the expectation that all the objects should be caught (cf. Hulsey, Hacquard, Fox and Gualmini, 2004), but the experimental scenario does not do this. It is possible that this infelicitous use of negation contributed to the observed results. It may account for why the L1 children generally scored

S. Unsworth / Lingua 117 (2007) 1930–1958

1953

worse on negation than on twee keer. With respect to the L2 children, it should be noted that (child and adult) English-speakers vary in their willingness to assign a specific reading to an indefinite embedded under negation (Gualmini, 2003; Miller and Schmitt, 2004; Su, 2001). Together with the infelicitous use of negation, which discourages them from assigning the targetlike specific reading to the scrambled sentence, this might account for the lack of development and generally non-targetlike responses in the L2 data on the negation task.22 If there is a methodological problem with the negation task, this could account for the observation that there are some subjects who are targetlike on the twee keer task but not on the negation task. In both groups of children, there was, however, a similar proportion of subjects who were successful on the negation task but not on the twee keer task. There are two differences between the tasks which might account for this finding. First, in the negation task, the object to which the scrambled indefinite refers is present in the pictures used in the task, whereas this is not the case in the twee keer task. Recall that on the latter task, the main character manipulates two different objects, an action which is consistent with the non-specific reading of the object. The specific reading must be inferred by the subject. This could potentially make the twee keer task more difficult.23 The second difference concerns the targetlike response on each of the two tasks. Due to the different entailment relations between the scrambled and non-scrambled sentences, the experimental scenario used in the negation task verifies the targetlike reading of the test sentence, that is, the targetlike response is ‘yes’. The scenario used in the twee keer task, however, falsifies the targetlike reading of the test sentence and consequently the targetlike response is ‘no’. It is possible that this discrepancy may also be reflected in the differences between responses on the two tasks observed for some subjects. If this were the case, we would expect that on a task which tested the comprehension of scrambling across twee keer using an experimental scenario which verified the targetlike reading, subjects should produce comparable responses on this and the negation task (as long as any problems with the infelicity of negation, as discussed above, were first rectified of course). Such an experiment will be left for future research.24 6.2. Why production before comprehension? Putting such methodological considerations aside for the moment, let us now consider the relation between production and comprehension. Although the two experiments were not conducted with the same group of L1 children, the results are clearly in line with previous findings, which suggested a production-comprehension discrepancy in this domain: targetlike production appears to precede targetlike comprehension. This pattern was also observed in the data from those L2 children who participated in both experiments. Assuming that this discrepancy is real, what might be its cause? Kra¨mer (2000:128) suggests that comprehension may lag behind production because ‘the hearer’s task of reconstructing the speaker’s intentions is more complicated than the speaker’s task of putting her intentions into words’. De Hoop and Kra¨mer (2005/2006) formalise this idea within an Optimality Theory 22

Similar results were observed with English-speaking L2 adults using the same task (Unsworth, 2005a). It might be objected that the twee keer task does not conform to the condition of plausible dissent (Crain and Thornton, 1998), and that this may contribute to the high proportion of non-targetlike responses. This seems unlikely, however, given that in a similar experiment on scrambling across twee keer, conducted by Philip (2003), the condition of plausible dissent was incorporated and the (L1) children tested there were generally less targetlike than those here. 24 It is also possible that negation is conceptually harder, although this is only likely to cause a real problem for the youngest children. 23

1954

S. Unsworth / Lingua 117 (2007) 1930–1958

approach. They claim that a child’s ability to take the speaker’s perspective relates to being able to optimise bidirectionally.25 Bidirectional optimisation is acquired relatively late and for this reason, targetlike comprehension of scrambled objects is also acquired late (see Hendriks and Spenader, 2005/2006 for a similar approach to the acquisition of pronouns). One problem for this account, which the authors acknowledge, is that bidirectional optimisation is also required for targetlike production. On de Hoop and Kra¨mer’s approach, then, targetlike production and comprehension should emerge simultaneously. The data presented here suggest that this is not the case (see also Unsworth, 2005a: 373–375 for relevant discussion). Hurewitz et al. (2000) provide a processing-based account for the production-comprehension discrepancy which they observe in the acquisition of ‘garden-path’ sentences. They claim that comprehension is a less controlled process than production (in the sense that it is controlled by the speaker rather than the hearer) that requires the rapid integration of different probabilistic cues from various sources. Consequently, children may experience problems whilst constructing this ‘comprehension machinery’ because they have not yet learned the relevant cues required for successful interpretation and/or their insufficient processing capacities hinder the integration of these cues. One possible cue to the specific interpretation of scrambled indefinites is the form of the indefinite. When native adults produce scrambled indefinites, they usually realise them as the stressed indefinite/numeral e´e´n ‘one’ (see fn. 6 and Unsworth, 2005a:253–256). It is possible that in the input children hear, the form of the scrambled indefinite is a more salient cue than its position, and that when, in the experimental context, this cue is absent, children are unable to assign the scrambled indefinite the targetlike interpretation.26 Building on Kra¨mer’s (2000) work, Miller and Schmitt (2004) make a very explicit suggestion as to how limited discourse integration is responsible for children’s failure to allow the specific interpretation of indefinite objects (in negative sentences). Adopting an analysis of indefinites put forward by Geurts (2002), they propose that children’s failure to interpret specific indefinites in a targetlike fashion results from their inability to background the relevant set from which the indefinite (on this analysis, a covert partitive) object is taken. Miller and Schmitt use a comparable scenario to the one used here, where one object from a group of objects is left unmanipulated. They modify the scenario, however, such that the link between the unmanipulated object and the set from which it was taken is made clearer by using objects 25 A detailed explanation of bidirectional optimisation is beyond the scope of this article. Briefly, whereas unidirectional optimisation only considers optimal forms for given meanings (production) or optimal meanings for given forms (comprehension), bidirectional optimisation (Blutner, 2000) works with form-meaning pairs (production and comprehension together). See Blutner, de Hoop and Hendriks (2006) for more details. 26 An anonymous reviewer suggests that children’s poor performance on the interpretation task results from the fact that the partitive context employed in the experimental context strongly favours a numeral – rather than a specific – reading of the indefinite ‘een N’. Presenting children with the unstressed indefinite rather than the stressed indefinite/cardinal form, e´e´n, which is the preferred means of expressing this partitive reading, might indicate to them that the partitive reading is not intended; the implication here is that had the children heard e´e´n instead of een, their answers would have been more targetlike. Work on the L1 acquisition of English suggests that such a result is unlikely, however. Lidz and Musolino (2002) found that when 4-year-old L1 English children were presented with a numeral (‘two N’), they still preferred the non-specific, i.e. narrow scope, interpretation. Whether a similar result will obtain for (L1/L2) Dutch children is an empirical matter which will be left for future research. It is, however, worth noting that the native-speaker controls had few problems interpreting the scrambled, unstressed indefinite specifically, which suggests that the stressed indefinite/ cardinal form is not required for targetlike interpretation. Thus, even if children’s responses were to improve upon hearing ‘e´e´n N’ instead of ‘een N’, there would still be a difference between the children’s and adults’ behaviour needing an explanation.

S. Unsworth / Lingua 117 (2007) 1930–1958

1955

which form a more natural group and which are associated with a physical entity connecting them together, such as the candles on a birthday cake, for example. Additionally, in their experimental scenario, the unmanipulated object was also highlighted in a separate picture. They find that the L1 English 4-year-olds they tested correctly accepted the specific reading for sentences such as (17) in 92.3% (48/52) of cases, which is considerably higher than previous studies (e.g. Su, 2001). (17)

Timothy didn’t blow out a candle

Miller and Schmitt claim that, in accordance with Kra¨mer’s (2000) proposal regarding limited discourse integration, this significant proportion of targetlike responses results from their having provided children with explicit discourse referents, which made it easier for them to integrate discourse. Hence, their responses were more targetlike than in experiments where this link is not provided. If limited discourse integration contributes to the observed delay in the child (L1/L2) Dutch data, it is expected that by altering the experimental scenarios accordingly, the number of targetlike responses produced by L1 Dutch children should also increase. Carrying out such a study will be left for future research. Acknowledgements This paper has benefited from input from audiences at the inaugural Generative Approaches to Language Acquisition North America (University of Hawai’i at Manoa¯, 17 December 2004), the Linguistics Colloquium at the University of Essex (24 March, 2004), the Workshop on Language Acquisition between Sentence and Discourse (Radboud University, Nijmegen, 12 May 2005) and the Lisbon Workshop on Production vs. Comprehension in the Acquisition of Syntax (Universidade Nova de Lisboa, 6 June, 2005). In particular, I wish to thank Ce´cile De Cat, Harald Clahsen, Peter Coopmans, Helen de Hoop, Irene Kra¨mer, Karen Miller, Cristina Schmitt and Bonnie D. Schwartz for informative discussion and feedback. Two anonymous reviewers also provided insightful comments which allowed me to clarify several points. All shortcomings remain mine, of course. The data reported on here were collected as part of my Ph.D. at the Utrecht Institute of Linguistics, The Netherlands. References Bakx, J., Jetten, M., Korebits, L., 1999. Grammatica in Gebruik. Nederlands voor Anderstaligen [Grammar in Use. Dutch for Speakers of Other Languages]. Intertaal, Amsterdam. Barbiers, S., 2005. Variation in the morphosyntax of one. Journal of Comparative Germanic Linguistics 8, 159–183. Bergsma-Klein, W., 1996. Specificity in Child Dutch: An Experimental Study. Unpublished MA thesis. Utrecht University. Blom, E., Polisˇenska`, D., 2006. Verbal inflection and verb placement in first and second language acquisition. In: Vliegen, M. (Ed.), Proceedings of the 29th Linguistics Colloquium. Peter Lang, Amsterdam. Blutner, R., 2000. Some aspects of optimality in natural language interpretation. Journal of Semantics 17, 189–216. Blutner, R., de Hoop, H., Hendriks, P., 2006. Optimal Communication. CSLI, Stanford. Clahsen, H., 1988. Critical phases of grammar development. A study of the acquisition of negation in children and adults. In: Jordens, P., Lalleman, J. (Eds.), Language Development. Foris, Dordrecht, pp. 123–148. Clark, H.H., 1977. Bridging. In: Johnson-Laird, P.N., Wason, P.C. (Eds.), Thinking. Readings in Cognitive Science. Cambridge University Press, Cambridge, pp. 411–420. Crain, S., Thornton, R., 1998. Investigations in universal grammar. In: A Guide to Experiments on the Acquisition of Syntax and Semantics, MIT Press, Cambridge, MA.

1956

S. Unsworth / Lingua 117 (2007) 1930–1958

de Hoop, H., 1992. Case Configuration and Noun Phrase Interpretation. Ph.D. dissertation, published (1996). University of Groningen, Garland, New York. de Hoop, H., Kra¨mer, I., 2005/2006. Children’s optimal interpretations of indefinite subjects and objects. Language Acquisition 13, 103–123. DeKeyser, R.M., 2000. The robustness of critical period effects in second language acquisition. Studies in Second Language Acquisition 22, 499–533. De Villiers, J.G., Tager-Flusberg, H.B., 1975. Some facts one simply cannot deny. Journal of Child Language 2, 279–286. ¨ berlegungen und Eisenbeiss, S., 1994. Kasus und Wortstellungsvariation im deutschen Mittelfeld. Theoretische U Untersuchungen zum Erstspracherwerb [Case and Word Order Variation in the German ‘Mittelfeld’. Theoretical Considerations and Investigations in First Language Acquisition] In: Haftka, B. (Ed.), Was Determiniert Wortstellungsvariation? [What Determines Word Order Variation?]. Westdeutscher Verlag, Opladen, pp. 279–298. Felix, S.W., 1977. Interference, interlanguage and related issues. In: Molony, C., Zobl, H., Sto¨lting, W. (Eds.), Deutsch im Kontakt mit Anderen Sprachen [German in Contact with Other Languages]. Scriptor, Kronberg, pp. 237– 258. Geurts, B., 2002. Specific indefinites, presupposition and scope. In: Ba¨uerle, R., Reyle, U., Zimmermann, T.E. (Eds.), Presuppositions and Discourse. Elsevier, Oxford. Goodluck, H., 1986. Language acquisition and linguistic theory. In: Fletcher, P., Garman, M. (Eds.), Language Acquisition. Cambridge University Press, Cambridge, pp. 49–68. Gualmini, A., 2003. Some knowledge children don’t lack. In: Beachley, B., Brown, A., Conlin, F. (Eds.), Proceedings of the 27th Boston University Conference on Language Development, Cascadilla Press, Somerville, MA, pp. 276–287. Gualmini, A., 2004. Some knowledge children don’t lack. Linguistics 42, 957–982. Guasti, M.-T., 2002. The Growth of Grammar. MIT Press, Cambridge, MA. Haider, H., 2004. Pre- and postverbal adverbials in VO and OV. Lingua 114, 779–807. Ham, E., Tersteeg, W.H.T.M., Zijlmans, L., 1998. Help! Een Cursus voor Buitenlanders (Deel 1/2) [Help! A Course for Foreigners. (Part 1/2)]. NCB, Utrecht. Haznedar, B., 1997. Child L2 acquisition of English: a longitudinal case study of a Turkish-speaking child. Unpublished Ph.D. dissertation, University of Durham. Haznedar, B., 2003. The status of functional categories in child second language acquisition: evidence from the acquisition of CP. Second Language Research 19, 1–41. Heim, I.R., 1982. The Semantics of Definite and Indefinite Noun Phrases. Unpublished Ph.D. thesis. University of Massechussetts. Hendriks, P., Spenader, J., 2005/2006. When production precedes comprehension: an optimization approach to the acquisition of pronouns. Language Acquisition 13, 319–348. Henry, A., Tangney, D., 1996. Functional categories in child L2 acquisition of Irish. In: Stringfellow, A., Cahana-Amitay, D., Zukowski, A. (Eds.), Proceedings of the 20th Boston University Conference on Language Development, Cascadilla Press, Somerville, MA, pp. 118–127. Hulk, A., Cornips, L., 2006. The acquisition of definite determiners in child L2 Dutch: problems with neuter gender nouns. In: Unsworth, S., Parodi, T., Sorace, A., Young-Scholten, M. (Eds.), Paths of Development in L1 and L2 Acquisition. John Benjamins, Amsterdam, pp. 107–134. Hulsey, S., Hacquard, V., Fox, D., Gualmini, A., 2004. The question-answer requirement and scope assignment. In: Csirmaz, A., Gualmini, A., Nevins, A. (Eds.), Plato’s Problem: Problems in Language Acquisition. MIT, Cambridge, MA, pp. 71–90. Hunt, K.W., 1970. Syntactic Maturity in Schoolchildren and Adults. University of Chicago Press, Chicago. Hurewitz, F., Brown-Schmidt, S., Thorpe, K., Gleitman, L., Trueswell, J.C., 2000. One frog, two frog, red frog, blue frog: factors affecting children’s syntactic choices in production and comprehension. Journal of Psycholinguistic Research 29, 597–626. Ionin, T., Wexler, K., 2002. Why is ‘is’ easier than ’-s?: acquisition of tense/agreement morphology by child second language learners of English. Second Language Research 18, 95–136. Johnson, J.S., Newport, E.L., 1989. Critical period effects in second language learning: the influence of maturational state on the acquisition of English as a second language. Cognitive Psychology 21, 60–99. Johnson, J.S., Newport, E.L., 1991. Critical period effects on universal properties of language: the status of subjacency in the acquisition of a second language. Cognition 39, 215–258. Kamp, H., 1981. A theory of truth and semantic interpretation. In: Groenendijk, J., Jansen, T., Stokhof, M. (Eds.), Formal Methods in the Study of Language. Mathematisch Centrum, Amsterdam. Kamp, H., Reyle, U., 1993. From Discourse to Logic. Kluwer, Dordrecht.

S. Unsworth / Lingua 117 (2007) 1930–1958

1957

Karmiloff-Smith, A., 1979. A Functional Approach to Child Language: A Study of Determiners and Reference. Cambridge University Press, Cambridge. Koster, J., 1975. Dutch as an SOV language. Linguistic Analysis 1, 111–136. Kra¨mer, I., 2000. Interpreting Indefinites. An Experimental Study of Children’s Language Comprehension. Ph.D. dissertation. Utrecht University. Lakshmanan, U., 1995. Child second language acquisition of syntax. Studies in Second Language Acquisition 17, 301–329. le Roux, C., Wilsenach, C., 1999. Transfer from L1 to L2: a study of parameter values in the Afrikaans of a child L2 learner. In: le Roux, C., Aarons, D., Holtzhausen, L. (Eds.), Fokus op taalverwerving. [Focus on language acquisition]. Stellenbosch Papers in Linguistics, Stellenbosch, pp. 79–109. Lidz, J., Musolino, J., 2002. Children’s command of quantification. Cognition 84, 113–154. McDonald, J.L., 2000. Grammaticality judgments in a second language: influences of age of acquisition and native language. Applied Psycholinguistics 21, 395–423. Miller, K., Schmitt, C., 2004. Wide-scope indefinites in English child language. In: Van Kampen, J., Baauw, S. (Eds.), Proceedings of Generative Approaches to Language Acquisition 2003. LOT Occasional Series, Utrecht, pp. 317– 328. Molony, C., 1977. ‘Ich bin sprechen deutsch aber’: the sequence of verb and word order acquisition of an American child learning German. In: Molony, C., Zobl, H., Sto¨lting, W. (Eds.), Deutsch im Kontakt met Anderen Sprachen [German in Contact with Other Languages]. Scriptor Verlag, Kronberg, pp. 274–295. Montens, F., Sciarone, A.G., 1991. Nederlands voor Buitenlanders: De Delftse Methode. Boom, Amsterdam. Park, H., 2000. When-questions in second language acquisition. Second Language Research 16, 44–76. Philip, W., 2003. Extreme delay in the acquisition of a syntax-semantic interface rule: the case of the obligatorily specific Dutch indefinite. Paper given at Modular Interaction in Acquisition workshop: Free University, Amsterdam [October 23–24]. Pollock, J.-Y., 1989. Verb movement, Universal Grammar and the structure of IP. Linguistic Inquiry 20, 635–670. Pre´vost, P., 2003. Truncation and missing inflection in initial child L2 German. Studies in Second Language Acquisition 25, 65–97. Quist, G., Stuik, D., 2000. Teach Yourself Dutch Grammar. Hodder and Stroughton, London. Rizzi, L., 1993/1994. Some notes on linguistic theory and language development: the case of root infinitives. Language Acquisition 3, 371–393. Schaeffer, J.C., 2000. The Acquisition of Direct Object Scrambling and Clitic Placement: Syntax and Pragmatics. John Benjamins, Amsterdam. Schetter, W.Z., van der Cruysse, I., 2002. Dutch. An Essential Grammar. Routledge, London. Schwartz, B.D., 1992. Testing between UG-based and problem-solving models of L2A: developmental sequence data. Language Acquisition 2, 1–19. Schwartz, B.D., 2003. Child L2 acquisition: paving the way. In: Beachley, B., Brown, A., Conlin, F. (Eds.), Proceedings of the 27th Boston University Conference on Language Development, Cascadilla Press, Somerville, MA, pp. 26–50. Schwartz, B.D., 2004. Why child L2 acquisition? In: Van Kampen, J., Baauw, S. (Eds.), Proceedings of Generative Approaches to Language Acquisition 2003, LOT Occasional Series, Utrecht, The Netherlands, pp. 47–66. Stern, H.R., 1984. Essential Dutch Grammar. Dover Publications, Mineola, NY. Su, Y.-c., 2001. Scope and specificity in child language: a cross-linguistic study on English and Chinese. Unpublished Ph.D. dissertation. University of Maryland. Unsworth, S., 2004. Child L1, child L2 and adult L2 acquisition: differences and similarities. In: Brugos, A., Micciulla, L., Smith, C.E. (Eds.), Proceedings of the 28th Boston University Conference on Language Development, Cascadilla Press, Somerville, MA, pp. 633–644. Unsworth, S., 2005a. Child L2, Adult L2, Child L1: Differences and Similarities. A Study on the Acquisition of Direct Object Scrambling in Dutch. Ph.D. thesis. Utrecht University. Unsworth, S., 2005b. Overcoming the poverty-of-the-stimulus: scrambled indefinites in English-Dutch interlanguage. In: Brugos, A., Clark-Cotton, M.R., Ha, S. (Eds.), Proceedings of the 29th Boston University Conference on Language Development. Cascadilla Press, Somerville, MA, pp. 627–638. van der Sandt, R.A., 1992. Presupposition projection as anaphora resolution. Journal of Semantics 9, 333–377. Van Geenhoven, V., 1998. Semantic Incorporation and Indefinite Descriptions: Semantic and Syntactic Aspects of Noun Incorporation in West Greenlandic. CSLI Publications, Stanford. van Kalsbeek, A., Huizinga, M., Kuiken, F., 1998. Code Nederlands. ThiemeMeulenhoff, Zutphen. Wason, P.C., 1965. The contexts of plausible denial. Journal of Verbal Learning and Verbal Behavior 4, 7–11.

1958

S. Unsworth / Lingua 117 (2007) 1930–1958

Weerman, F., Bisschop, J., Punt, L., 2003. L1 and L2 acquisition of Dutch adjectival inflection. Paper presented at Generative Approaches of Language Acquisition 2003, Utrecht, The Netherlands [4th September]. Wexler, K., 1998. Very early parameter setting and the unique checking constraint: a new explanation for the optional infinitive stage. Lingua 106, 23–79. Whong-Barr, M., Schwartz, B.D., 2002. Morphological and syntactic transfer in child L2 acquisition of the English dative alternation. Studies in Second Language Acquisition 22, 579–616. Yuill, N., Oakhill, J., 1991. Children’s Problems in Text Comprehension: An Experimental Investigation. Cambridge University Press, Cambridge.