Investigating ' Full Transfer ' : Preliminary Data From The Adult L 2 Acquisition Of Afrikaans

In generative linguistics, many researchers agree that (something like) Universal Grammar (UG) must play some role in second language (L2) acquisition, since the logical problem of first language (L1) acquisition1 seems to hold for L2 acquisition, as well: in L2 acquisition, as in L1 acquisition, the complex, abstract system of knowledge that the learner ends up acquiring exceeds to a great extent the primary linguistic data that the learner receives as input (see White 1989 for more detailed arguments). However, these researchers are still debating the following two questions as to the exact role of UG in L2 acquisition: (i) what constitutes the initial state of L2 acquisition? and (ii) is parameter resetting possible in situations in which the value of a certain parameter differs for the learner's L1 and the target L2? A number of hypotheses have been proposed in response to either one or both of these questions (see, for example, Clahsen and Muysken 1986, 1989; Epstein, Flynn and Martohardjono 1996; Eubank 1993/1994, 1994, 1996; Hawkins and Chan 1997; Vainikka and Young-Scholten 1994, 1996; also White 2003 for an overview). One such hypothesis is the Full Transfer Full Access hypothesis (FTFA) (e.g. Schwartz and Sprouse 1994, 1996; White 1989, 2003) which proposes the following answers to the above questions: (i) the learner's L1 grammar (including L1 parameter settings) constitutes the initial state of L2 acquisition (= full transfer), and (ii) the L2 learner has access to UG in its entirety and, hence, parameter resetting is possible in L2 acquisition (= full access).


Introduction
In generative linguistics, many researchers agree that (something like) Universal Grammar (UG) must play some role in second language (L2) acquisition, since the logical problem of first language (L1) acquisition 1 seems to hold for L2 acquisition, as well: in L2 acquisition, as in L1 acquisition, the complex, abstract system of knowledge that the learner ends up acquiring exceeds to a great extent the primary linguistic data that the learner receives as input (see White 1989 for more detailed arguments). However, these researchers are still debating the following two questions as to the exact role of UG in L2 acquisition: (i) what constitutes the initial state of L2 acquisition? and (ii) is parameter resetting possible in situations in which the value of a certain parameter differs for the learner's L1 and the target L2? A number of hypotheses have been proposed in response to either one or both of these questions (see, for example, Muysken 1986, 1989;Epstein, Flynn and Martohardjono 1996;Eubank 1993Eubank , 1996Hawkins and Chan 1997;Young-Scholten 1994, 1996; also White 2003 for an overview). One such hypothesis is the Full Transfer Full Access hypothesis (FTFA) (e.g. Schwartz andSprouse 1994, 1996;White 1989White , 2003 which proposes the following answers to the above questions: (i) the learner's L1 grammar (including L1 parameter settings) constitutes the initial state of L2 acquisition (= full transfer), Simone Conradie 2 and (ii) the L2 learner has access to UG in its entirety and, hence, parameter resetting is possible in L2 acquisition (= full access).
The study reported in this paper was designed to test the predictions made by the FTFA 2 by investigating the L2 acquisition of two syntactic parameters, the Split-IP parameter (SIP) (Thráinsson 1996) and the V2 parameter, in Afrikaans by native speakers of English and German, respectively. German has the same parameter settings as the L2, Afrikaans (namely, [+SIP] and [+V2]), whereas English differs from these two languages with respect to the settings of both parameters ([-SIP] and [-V2]). Hence, the study compares two learner groups to each other (one whose L1 has the same parameter settings as the L2 and one whose L1 has different parameter settings) in addition to comparing each of these groups to native speakers of the L2. This paper reports preliminary data collected from 9 beginner learners of Afrikaans (5 German-speaking and 4 English-speaking) and will thus only be concerned with the 'full transfer' claim of the FTFA. 3 I will first discuss the two parameters (sections 2.1 and 2.2) and their settings in the three languages under investigation (section 2.3) and will then turn to the study itself (section 3). Section 4 provides a brief conclusion.

Two syntactic parameters 4
In sections 2.1 and 2.2, I will first illustrate the properties of the four parameter settings ([±SIP] and [±V2]) by means of English and Icelandic examples and will then turn to the other relevant languages (Afrikaans and German) in section 2.3. The reason for this is twofold. Firstly, although Icelandic is not one of the languages involved in the study reported in this paper, it is the language used to represent [+SIP] languages in the source paper (Bobaljik and Thráinsson 1998). And secondly, headedness and the presence/absence of non-V2 verb movement are established issues for English and Icelandic, but remain controversial in the case of Afrikaans and German.

The Split-IP parameter
Pursuing the idea that languages might differ as to which functional categories they have in their IP-complex 5 (see, for example, Iatridou 1990), Thráinsson (1996: 262)  (1) (a) There appeared a train in the distance.
(b) *There has a cat eaten the mice.
expletive has some cat eaten mice-the 'A cat has eaten mice.' The assumption here is that the subject NP of a transitive verb is not allowed to remain VPinternal (see, for example, Bobaljik and Jonas 1996;Alexiadou and Anagnostopoulou 1997;Chomsky 1995 OSCs are constructions in which the direct-object NP has been moved leftward over some element that is taken to mark the left edge of VP, such as a sentence-medial adverb, floating quantifier, or negative element (for the sake of convenience I will refer to these as "left-edge markers").
In the examples in (3) and (4) the direct object NP three books / þrjár baekur has been moved from its VP-internal position over the negative element not / ekki. As can be seen from these examples, such object shift is allowed in Icelandic (4) but not in English (3).
(3) *I did three books not read. Another property that BandT (1998) link to the setting of the SIP is whether or not the verb raises out of VP in non-V2 environments. This follows from their particular theory of checking (BandT 1998: 39-45), which can be summarized as follows. The crucial assumptions that they make are that (i) "the features of a projection are those of its head", (ii) "movement occurs solely for the purposes of feature checking" and (iii) "features are checked in all and only local relations to a head" (where the "local relations" are specifier-head, headcomplement and head-head) (BandT 1998: 39). This means that in [-SIP] languages verb raising will not be required in non-V2 environments: if V and I have features that need to be checked against each other (as is assumed to be the case), then this checking will simply occur between I and VP, since they are in a head-complement (hence, local) relation to each other (see assumption (iii) above) and the features of VP are those of V (see assumption (i) above). Consequently, to determine the setting of the SIP in a language, one simply has to determine whether or not the verb raises out of VP in non-V2 environments. If the verb has raised out of VP, then it will precede left-edge markers, and this will in turn indicate that the language is [+SIP]. If the verb has not raised out of VP, then it will follow left-edge markers, and this will in turn indicate that the language is [-SIP]. This is illustrated by the examples in (5) and (6) below: in the [-SIP] language English, the verb read follows the adverb often, whereas in the [+SIP] language Icelandic, the verb las precedes the adverb oft.

The V2 parameter
In Table 1, it is stated explicitly that the position of the verb relative to left-edge markers is only indicative of the setting of the SIP in non-V2 environments. This is because it is commonly assumed that in V2 environments the head C (the head of the complementizer phrase) has some feature that requires the verb to raise into C, regardless of whether the language is [+SIP] or [-SIP]. Therefore, the position of the verb in V2 environments is linked to the setting of the V2 parameter, and not the SIP.
The relevant feature in C that is linked to the V2 parameter is said to be present in V2 languages (forcing verb raising into C) and absent in non-V2 languages (prohibiting verb raising into C). The consequences of the setting of the V2 parameter are visible in non-subject initial main clauses (NSIMCs). In V2 languages, a sentence-initial adverbial phrase or topicalized object is in Spec,CP and is immediately followed by the verb (which is in second position, in C). In non-V2 languages, a sentence-initial adverbial phrase or topicalized object is adjoined to IP and is immediately followed by the subject (in Spec,IP), and the verb appears in third position (inside VP). 7 Consequently, V2 languages, such as Icelandic, have V2-NSIMCs (i.e., NSIMCs in which the verb appears in second position -see (7)), whereas non-V2 languages, such as English, have V3-NSIMCs (i.e., NSIMCs in which the verb appears in third position -see (8)).
the-fox (ACC) shot Olaf (NOM) with this shotgun 'The fox Olaf shot with this shotgun.' The fox Olaf shot with this shotgun.

Hypothesis and predictions
As was mentioned in Section 1, the study reported here was designed to test the predictions made by the FTFA with respect to the L2 acquisition of the SIP and the V2 parameter, 10 by comparing English-speaking learners of Afrikaans to German-speaking learners of Afrikaans.
The FTFA predicts that the learners will start out with the L1 settings of the SIP and the V2 parameter, and that the English-speaking learners will be able to reset these parameters so that both groups of learners will end up with the correct settings for Afrikaans (see Table 2).

Tasks
To determine the setting of the SIP and the V2 parameter in the L2 learners' interlanguage (IL) grammars, they were asked to complete three tasks (in this order): a sentence manipulation task, a grammaticality judgment task, and a short truth-value judgment task.
(Examples of the test items on the sentence manipulation task and the grammaticality judgment task are provided in the Appendix.) In the sentence manipulation task (based on that in White 1991), the subject was handed a set of randomly shuffled Afrikaans word cards and asked to form a sentence that (s)he found acceptable, using all of the cards. Once the subject had formed a sentence, this was recorded and the subject was asked whether (s)he could form another sentence using the same cards.
This continued until the subject could not form another sentence. The subject was then presented with the next set of cards and the steps repeated. There are 18 sets of word cards that the subject had to manipulate: 6 SIMC sets (3 x adverb and 3 x negative element), 6 NSIMC sets (3 x sentence-initial adverb and 3 x topicalized object) and 6 OSC sets (3 x adverb and 3 x negative element). 11 An illustrative example is given in (17) In the grammaticality judgment task (based on the written preference task in White 1991), the subject was presented with pairs of Afrikaans sentences and asked to circle one of the options below each pair: "Only (a) is possible", "Only (b) is possible", "Both possible", "Both impossible" or "Don't know". The subject was presented with 55 pairs of sentences to be judged: 5 distracter pairs, 15 SIMC pairs (5 x sentence-medial adverb, 5 x negative element and 5 x floating quantifier), 10 NSIMC pairs (5 x topicalized object and 5 x sentence-initial adverb), 20 OSC pairs (5 x adverb, 5 x negative element, 5 x floating quantifier and 5 x indefinite NP) and 5 TEC pairs. 12 An example of a test pair is given in (18)

Only (a) is possible Only (b) is possible Both possible Both impossible Don't know
Note that this task includes 5 OS pairs that involve shifting an indefinite object (the OSCindef pairs). At issue here are the semantic constraints on OS that determine which objects are allowed to undergo OS in a [+SIP] language. The semantic distinction between VP-external (i.e. shifted) and VP-internal (i.e. unshifted) objects seems to involve the distinction between "old information" and "new information" but, as Thráinsson (2001: 193) notes, "various terms  Thráinsson 2001 for some discussion of this.) The most accurate generalization seems to be that "the weak/existential reading is incompatible with OS and Scrambling but objects having the strong/quantificational/specific reading do not necessarily have to shift or scramble" (Thráinsson 2001: 193). Furthermore, the semantic constraints on OS differ crosslinguistically, as is illustrated by the sentences in (12) and (16)  were included to establish that the German-speaking learners would accept non-OSCs under certain circumstances, namely when the object has the semantic properties of an "unshiftable" object. Under the assumption that the indefinite objects in the OSC-indef pairs on the grammaticality judgment task are "unshiftable", the prediction was that the Afrikaans Controls, the beginner German participants and the beginner English participants would all (correctly) reject these items: the Afrikaans Controls and the German participants because an indefinite object is "unshiftable", and the English participants because their L1 does not allow OS in the first place (regardless of the semantic properties of the object-NP). However, it should be noted that this assumption (that the objects in the OSC-indef pairs are "unshiftable") turned out to be incorrect, as will be discussed in section 3.4 below.
The truth-value judgment task (based on that in Dekydtspotter, Sprouse and Thyre 1999) was designed to determine whether the learners had knowledge of one of the semantic effects of OS in Afrikaans (especially the English participants, since their L1 does not allow OS).
At issue is the following: In Afrikaans, if a direct object NP is modified by a number (e.g. drie boeke 'three books') and this NP remains VP-internal in a negative clause, then the negative element takes scope over the direct object and the number modifying it. However, if this NP is shifted out of the VP and across the negative element, then the direct object and the number modifying it take scope over the negative element. This results in a clear interpretive difference, which is illustrated by the sentences in (19) and (20). Keeping the sentences in (19) and (20) in mind, consider the context in (21): (21) Sven is taking a course in English literature at the university. For this week's class he had to read 4 books but he only read 2. Fortunately, the class got cancelled and now Sven has another weekend to try and get the reading done.
Given the context in (21), the non-OS sentence in (19) meaning "It is not true that Sven has read three books" is true (since Sven has only read two books, not three), while the OS sentence in (20) meaning "There are three books that Sven hasn't read" is false (since there are only two books that Sven hasn't read, not three).
In the truth-value judgment task the participants were presented with contexts such as that in (21) (which were provided in their L1) 13 and these contexts were followed by an Afrikaans OS sentence or non-OS sentence, which the participants had to judge as true or false, given the preceding context. There were 10 sentences to be judged, each with its own context: 2 distracter items, 4 OS items and 4 non-OS items.
Investigating 'Full Transfer': Preliminary Data From The Adult L2 Acquisition of Afrikaans 13

Participants
In the study reported in Conradie (2002) (see also Conradie, to appear), the three tasks described above were used to test 15 advanced English-speaking learners of Afrikaans 14 and I concluded that the results provided evidence for the 'full access' part of the FTFA because these results indicated that the learners had managed to reset the SIP and the V2 parameter from their L1 (English) settings to their L2 (Afrikaans) settings. However, although the participants had grown up in homes where only English was spoken, they had all been born and raised in South Africa, so that they would have been exposed to Afrikaans from birth. With this goal in mind, I am now employing the three tasks mentioned in Section 3.2 to investigate the L2 acquisition of the SIP and the V2 parameter in Afrikaans by native speakers of English and German, respectively, who were not born in South Africa and who only started learning Afrikaans when they came to stay in South Africa (either temporarily or permanently) as adults. As this study is part of ongoing research, Section 3.4 below only offers preliminary results from 20 Afrikaans native speaker controls and 9 beginner adult learners of Afrikaans, which bear on the 'full transfer' part of the FTFA.
The 9 L2 learners are non-South-African students who were taking an Afrikaans Beginner's course through Stellenbosch University. These students had only arrived in South Africa a few months earlier and had never been exposed to Afrikaans before. There were 5 Germanspeaking learners and 4 English-speaking learners. They were tested after 30 hours of instruction, but at this stage both groups still had trouble with the simple Afrikaans vocabulary items used in the tasks, although they had received a vocabulary list with all of the words translated into English and German a few days before the testing and they were allowed to refer to this list during testing. In fact, most of the participants (English as well as German), had to refer to the list constantly during testing and sometimes translated sentences on the grammaticality judgment task word-for-word with the help of the list before deciding whether they found the sentences acceptable or unacceptable in Afrikaans. Consequently, one cannot really conclude anything on the basis of the results obtained during this testing session. The exact same students were then tested again at the end of their Afrikaans course, when they had received 60 hours of instruction. These results are discussed directly below and, although no statistical tests have been performed on these data yet because of the very small learner groups, the tendencies within the two groups are clear enough to justify some discussion.

Results
Recall that, following the FTFA, I predicted that the German participants would perform much better on the tasks than the English participants, because both groups were still beginner learners and, therefore, they were presumably still operating with the parameter settings that they had transferred from their L1s.
The results of the sentence manipulation task are presented in Table 3    groups: they show a strong preference for SIMCs in which the verb follows an adverb or a negative element, and for V3-NSIMCs, and they did not form any OSCs.
The results of the grammaticality judgement task are given in Table 4 as percentage of accurate judgments for each of the sentence types. 16 Afrikaans Controls accepted these OSCs as grammatical. Recall that the participants were asked to judge these isolated sentences without any context (see example 18), so that it is very likely that (as native speakers) they were simply able to imagine a context in which the object does have the semantic properties of an object that can be shifted, and the sentence is, therefore, grammatical. As was mentioned earlier (section 3.2), the semantic distinction between "shiftable" and "unshiftable" objects is not at all clear-cut and can indeed not be reduced to the distinction between definite and indefinite objects.
Furthermore, the semantic constraints on OS differ cross-linguistically in subtle ways. I would like to propose that this accounts for the German participants' low scores on the OS items: these participants are, as beginner learners, still unsure as to how exactly the semantic constraints on Afrikaans OS differ from the semantic constraints on German OS. Regardless of whether this account is valid, the English participants fare even worse on the OS items than the German participants. They do score 70% for the OS items involving an indefinite object, but recall that they were expected to do well on this category. The English participants were correct in rejecting the indefinite OS-items, but they (presumably) rejected them for the wrong reason. They did not reject them because the object had the wrong semantic properties; rather, they rejected them for the same reason that they rejected the other OS-items, namely, because their L1 does not allow OS, regardless of the semantic properties of the object. 17 On all of the other categories in the grammaticality judgment task, the predictions of the FTFA are borne out in that the German participants' scores are much higher than the Eng participants' scores. This is also true of the SIMC items involving a negative element, even though here the German participants' score is not quite as high as their other scores. This is due to the optionality also witnessed in the sentence manipulation task. If one considers the total scores of the three groups, the predictions of the FTFA are again borne out in that the German participants fare much better than the English participants. In fact, overall, the English participants are performing exactly at the level of chance (which is 25% on this task, since, for every sentence pair, there are four possible responses to choose from). Once more data have been collected, statistical tests will be performed in order to determine whether the differences between the two learner groups are statistically significant, but at this stage the tendencies within the two groups on the sentence manipulation task and the grammaticality judgement task fit well with the predictions made by the 'full transfer' claim of the FTFA.
The results of the truth-value judgment task are given in Table 5 as the percentage of accurate judgments on the different categories of this task. that, at this stage, their IL grammars for Afrikaans would be unable to interpret OSCs. When they are then forced to try and interpret these constructions in a task, they resort to a strategy involving a kind of logical calculation in the following way.
The two example sentences from the truth-value judgment task are repeated here as (22)  As a consequence of Afrikaans word order properties, the lexical verb is always surrounded by the two negative elements in OSCs (for example, nie gelees nie in (23)). It might be that the learners have noticed that if the verb is surrounded by the two negative elements, it is the verb that is being negated, and the sentence is about "how many things were not V-ed" (for example, how many books were not read). This calculation happens to lead them to the correct interpretation and gives them a high score on the OS items. In Non-OSCs, on the other hand, as one can see in (22), there is no single element (such as the verb or the object) that is surrounded by the two negative elements. In these cases, the participants don't know whether it is the verb or the number that is being negated and they have to resort to guessing, which would explain why they perform at chance level on the No-OS items.

Conclusion
To summarize, the study reported in this paper was designed to test the predictions made by the FTFA with respect to the L2 acquisition of the SIP and the V2 parameter in Afrikaans. As this study forms part of ongoing research (see note 3), only preliminary results are reported in this paper, namely those obtained from 9 beginner learners of Afrikaans who are native speakers of German and English, respectively. The German participants fared much better on the sentence manipulation task and the grammaticality judgment task than the English participants, and since all of these learners received the same instruction over the same period of time from the same instructor, we can conclude that the differences between these two groups are attributable to the grammars that they started out with. Furthermore, the tendencies observed within the two groups on all three of the tasks can easily be accounted for by the FTFA. In this way, these preliminary results provide some support for the 'full transfer' part of the FTFA. Currently, more beginner adult L2 learners of Afrikaans as well as a number of advanced adult L2 learners of Afrikaans are being tested. As soon as more data have been collected, (i) statistical tests will be performed in order to determine whether the observations made with respect to the beginner learner groups discussed in this paper, hold for a larger beginner learner group as well, and (ii) the data from the advanced learners will be used to test the predictions made by the 'full access' part of the FTFA. Finally, these initial data suggest that a larger database might offer some insight into the semantic constraints on OS and how these differ cross-linguistically, an area that deserves attention given that these constraints are not yet well understood. (See note 3 for a reference to work on the complete data set.) 1.
The assumption that there is a logical problem of L1 acquisition, is not undisputed.
However, see Baker and McCarthy 1981 and Hornstein and Lightfoot 1981 for some convincing arguments for a logical problem of L1 acquisition.

2.
Note that, of course, a number of alternative hypotheses exist. However, it is impossible to design an experiment that will test all of these hypotheses at once.
Consequently, as is stated explicitly here, the study reported in this paper focuses on testing the predictions made by a single hypothesis, the FTFA. See note 3 below for a reference to work that discusses alternative L2 acquisition hypotheses and their success/failure in accounting for the data reported here (as well as additional data).

3.
It should be noted that additional data have subsequently been collected from 24 beginner (4 English-speaking and 20 German-speaking) and 15 advanced (5 Englishspeaking and 10 German-speaking) adult L2 learners of Afrikaans. These data are not included in the present paper, which (as is indicated by the title) is meant to present the preliminary data available at the time of initial submission to SPIL. For an analysis of the complete data set as well as more in-depth discussions of the issues raised in the present paper, the reader is referred to my Ph. investigating the Full Transfer Full Access hypothesis. In this dissertation I argue that the additional data from the beginner learners support the conclusions reached on the basis of the smaller data set in the present paper (i.e., the data are consistent with 'full transfer') and that the data from the advanced learners support the 'full access' claim of the FTFA. The dissertation also includes data from the L2 acquisition of the SIP and the V2 parameter in French by adult native speakers of Afrikaans.

4.
This section provides a brief (and therefore incomplete) discussion of the relevant syntactic parameters. The reader is referred to Conradie (2002) and Conradie (to appear) as well as the source paper (Bobaljik and Thráinsson 1998) for more detailed discussions.

5.
The term 'IP-complex' is taken from Bobaljik and Thráinsson (1998: 38) and refers to "the collection of inflectional heads and their phrases that together make up the articulated IP".

6.
In constructions such as (1a), on the other hand, which contain an unaccusative verb and an expletive, the expletive occurs in Spec,IP and the subject-NP (a train in (1a)) remains in the complement position of VP (where it is base-generated -see, for example, Perlmutter and Postal's (1984) Unaccusative Hypothesis).
7. This is true for non-V2 languages that are [-SIP]. I will not consider non-V2 languages that are [+SIP] (such as French) in this paper.

8.
The assumption here is that SIMCs are non-V2 environments (i.e. IPs/AgrSPs, and not CPs) in V2 languages such as Afrikaans. This assumption is actually untenable under BandT's framework and it is made here purely to simplify the presentation of the results of the study in this paper. See Conradie (to appear) for some discussion about the syntactic status of SIMCs.

9.
In Afrikaans, the second (final-neg) nie has no semantic content and always appears in clause-final position. Its syntactic status need not concern us here as it is the first (clause-medial) nie that serves as a left-edge marker and corresponds to English not and German nicht. See Oosthuizen 1998 for a proposal as to the syntactic status of clause-final nie.

10.
This study is set in the framework proposed by Bobaljik and Thráinsson 1998 The reasons for the small number of distracter pairs on this task are identical to those mentioned in note 11 with respect to the sentence manipulation task: because there are a large number of items on this task (110 sentences to be judged) and the construction types under investigation (n=10) are diverse, it is highly unlikely that participants would be able to determine what was being tested. Furthermore, adding more items to this (already long) task would doubtlessly make it tiring for the participants.
13. Following Dekydtspotter et al., contexts were provided in the participants' L1s to ensure that they would understand the contexts, something which is, of course, necessary for them to accurately judge the test sentences as true or false. 14.
The tasks employed in the study reported in Conradie (2002) differ slightly from those employed in the study reported here and described in Section 3.2, as some revisions were made after the earlier testing sessions (e.g. the inclusion of OS test pairs involving an indefinite object in the grammaticality judgment task).
15. See Schwartz 2003 on potential insights to be gained by comparing child L2 acquisition to adult L2 acquisition. Also, Unsworth 2003Unsworth , 2004 conducted a study on the L2 acquisition of Dutch by native speakers of English (some child L2ers and some adult L2ers), investigating whether these learners could acquire some semantic and syntactic aspects of OS, a phenomenon not allowed in their L1 grammar. She employed a truth-value judgment task and an elicited production task and argues that the results of her study show that both the child and the adult L2ers were able to acquire the relevant properties of the L2.

16.
See note 3 for reference to work that includes an analysis of individual participants' performance as well as an investigation into the FTFA's prediction that performance on related construction types (e.g., performance on OSCs and performance on TECsboth linked to the setting of the SIP) should cluster.

17.
Two anonymous SPIL-reviewers suggested that, contrary to what I claim, the Englishspeaking learners might actually have IL grammars that allow OS but that they simply did not form any OSCs on the sentence manipulation task or accept (most of the) OSCs on the grammaticality judgment task because they could not imagine contexts in which the objects would have the semantic properties that would make them "shiftable" -see again the brief discussion in section 3.2 on the semantic constraints on OS. However, this potentially confounding non-grammatical factor (participants'

24
(in)ability to imagine the required context) should affect all three groups, or at least the two learner groups, equally (precisely because it is a non-grammatical factor). The conclusions reached here are thus supported by a comparison of the two learner groups to each other instead of an analysis of each groups' performance separately.
Furthermore, the analysis of the two learner groups' scores on the OSC pairs in the grammaticality judgement task is supported by an error analysis performed on the larger data set examined in my Ph.D. dissertation (see note 3). At first sight, the German-speaking learners and the English-speaking learners perform similarly on the OSC-categories, both achieving low scores overall. However, an error analysis shows that the two L1-groups make different types of errors on these categories: most of the errors that the German-speaking learners make involve them accepting only the OSC, whereas most of the errors that the English-speaking learners make involve them accepting only the Non-OSC.
18. The fact that the English-speaking learners' group performance on the Non-OSCs in this task is at chance level, is not the result of a bimodal distribution (i.e., some learners performing very well and other learners performing very poorly); individual results show that all of these learners performed at chance level.