THE TESTABILITY OF PROMINENT HYPOTHESES ON ACCESS TO UNIVERSAL GRAMMAR IN ADULT SLA

There is a growing awareness among linguists that second language acquisition (SLA) is a promising area of inquiry, due to the fact that an understanding of how SLA occurs may provide a useful perspective on the mental processes involved in language learning and language use in general, as well as on the nature of the language faculty (Epstein, Flynn and Martohardjono 1996:677-678; Flynn 1988:76). A central question in SLA research today concerns the extent to which adult second language (L2) learners have access to the innate linguistic principles that guide first language (L1) acquisition, i.e. to Universal Grammar (UG). As we shall see, there is no shortage of proposed answers to this question.


INTRODUCTION
There is a growing awareness among linguists that second language acquisition (SLA) is a promising area of inquiry, due to the fact that an understanding of how SLA occurs may provide a useful perspective on the mental processes involved in language learning and language use in general, as well as on the nature of the language faculty (Epstein, Flynn and Martohardjono 1996:677-678;Flynn 1988:76). A central question in SLA research today concerns the extent to which adult second language (L2) learners have access to the innate linguistic principles that guide first language (L1) acquisition, i.e. to Universal Grammar (UG). As we shall see, there is no shortage of proposed answers to this question. 1 Among the proposed answers to the question of the accessibility of UG in adult SLA, hypotheses 2 may be divided into six groups. The first group of hypotheses claims that L2 learners have full access to UG, as L1 learners have. The second group also claims that the L2 learner has full access to UG, but these hypotheses further claim that the L2 learner's L1 knowledge is the starting point for SLA. The third group claims that, although L2 learners have full access to UG in SLA, the transfer of L1 parameter settings to the L2 learners' interlanguage (IL) grammar (for the target language) is partial. According to the researchers in the fourth group, L2 learners have partial but direct access to UG. The fifth group of hypotheses claims that the L2 learner has partial and indirect access to UG and also that the L2 learner's L1 knowledge is the starting point for SLA. Researchers in the sixth group claim that L2 learners have no access to UG in SLA and that their L1 knowledge in its entirety is transferred to the L2 learners' IL grammar.
In this article, each of ten prominent hypotheses on SLA will be placed into one of these six groups. The testability 3 of each hypothesis will be evaluated. This entails identifying the test implications derived from each hypothesis. It will also be stated what experimental evidence would falsify each hypothesis. The article will conclude with comments on the state of linguistic theories in the field of SLA.
Before commencing the discussion of the ten hypotheses, the meanings of the terms "testability" and "test implications" need to be clarified. A distinction is made between testability in theory and testability in practice. According to the philosopher Hemple (1966:30), a hypothesis is testable in theory if (i) precise test implications can be derived from the hypothesis; and (ii) it is possible to specify what experimental findings, should they be available, would indicate whether the test implications of the hypothesis are true or false.
According to Popper (1969:36), however, testability is refutability. By this he means that refutability is a necessary condition for empirical status. A hypothesis that is confirmable but not refutable, is, according to Popper, not an empirical hypothesis. To accommodate Popper's view, Hemple's definition of testability must be amended to the following: A hypothesis is testable in theory if (i) precise test implications can be derived from the hypothesis; and (ii) it is possible to specify what experimental findings, should they be available, would indicate whether the test implications of the hypothesis are false.
According to Hemple (1966:30), a hypothesis is testable in practice if (i) precise test implications can be derived from the hypothesis; and (ii) experimental findings which would indicate whether the test implications of the hypothesis are true or false, are in fact available.
In the present article, the actual availability of experimental findings will not be discussed.
One factor which, according to Popper, can adversely affect the testability of a hypothesis, is the non-explicit nature (or obscurity) of the hypothesis. If the content of the hypothesis is not stated clearly and precisely, then it is not possible to derive test implications from it. Where such obscurity is present, it will be mentioned in this article.
Let us now turn to the ten hypotheses on access to UG in SLA.

Epstein, Flynn and Martohardjono's (1996) Full Access Hypothesis
On Epstein et al.'s (1996:707) view, adult L2 learners have full and direct 4 access to all properties of UG. These properties include all parameter settings, functional categories and feature values (Dube 2000:234;White 2000:135), even those that are not instantiated in the learner's L1. The initial state of L1 and L2 acquisition are, therefore, the same. during subsequent stages of SLA. Therefore, one would have to examine the L2 learners' IL grammars immediately after the onset of SLA. One can, however, never be sure that the first data obtained are really representative of the initial stage of SLA, as there might be a silent period before first production, and grammatical acquisition might take place during this silent period (White 2000:140 Where parameter values of the L1 and the L2 differ, parameter resetting 6 needs to take place during SLA. This resetting will take place if the input to which the L2 learner is exposed contains the relevant positive evidence, 7 i.e. evidence that the target L2 instantiates a parameter value that differs from the value instantiated in the learner's L1. Where positive evidence is not sufficient 8 to bring about the desired change from the L1 parameter value to the correct L2 parameter value, this change has to be brought about through exposure to negative evidence, viz. exposure to direct evidence in the form of corrective feedback or explicit instruction 9 .

Testability in theory of White's hypothesis
The following precise test implications can be derived from White's hypothesis.  Schwartz and Sprouse (1994:319), this revision (and perhaps re-revision), or this change in IL values, is possible in theory and occurs gradually as primary linguistic data (PLD) 12 are accommodated by the developing IL grammar.
Possibly, not all parameters are revisable on the basis of PLD alone (Schwartz and Sprouse 1994:319). Therefore, many L2 learners never reach native-like competence in their L2.
Herein lies the difference between the proposal of White (1998) and that of Schwartz and Sprouse. Contrary to White (1998), Schwartz and Sprouse do not believe it possible for parameters to be reset on the basis of exposure to negative evidence. According to Schwartz (1993:156-158), parameters can only be reset during SLA on the basis of exposure to positive evidence. Exposure to negative evidence in the form of explicit instruction or corrective feedback leads to learning about language. The L2 learner may, thus, learn the correct surface form that a structure would have, had the parameter been reset, without actual parameter resetting having taken place.
Where revision and resetting of parameters do take place, such revision and resetting conform at all times to UG principles, making the overall process of L1 acquisition and SLA very similar. To summarise their view, Schwartz and Sprouse (1994:316) state that each successive stage in the L2 learner's IL development emerges on the basis of interaction between the L1 grammar, positive evidence (from the PLD), UG principles and aspects of a (language) learning procedure.

Testability in theory of Schwartz and Sprouse's hypothesis
Precise test implications can be derived from Schwartz and Sprouse's hypothesis, and it is possible to specify what experimental findings would indicate whether the test implications of the hypothesis are false. Therefore, the hypothesis is testable in theory.
The first test implication is the same as the first for White's hypothesis, but the second implication differs from the second one for White's hypothesis. While White (1998) has to prove that parameter resetting (and not mere memorisation of surface structures) occurs after exposure to negative evidence, Schwartz and Sprouse need to prove the opposite, namely that mere learning of correct forms takes place and not changes in the l2 learner's IL grammar.
Using Schwartz's (1993)  If functional projections are found to be present in the L2 learner's IL grammar, then it could be said that only lexical projections were initially transferred and that functional projections were acquired later, i.e. that the learner is not really an initial-state L2 learner but more advanced. As was said when the testability of Epstein et al.'s theory was discussed, one cannot be certain that the first production data obtained from the L2 learner are really representative of the initial stage of SLA, as there might be a silent period before first production during which grammatical acquisition might take place (White 2000:140). A more reliable way of testing for the presence or absence of functional categories in the initial stage might thus be to use grammaticality judgement tasks and comprehension tasks, rather than relying on production data. Eubank (1993/4:183-184) proposes the weak transfer view, in which lexical and functional projections, as well as the headedness of these projections, may transfer from the L1, but the

Krashen's hypotheses
Krashen's theory of SLA consists of five hypotheses (Krashen 1985:1). The content of each of these will be given briefly below.
(i) The Acquisition-Learning Hypothesis claims that there are two ways of developing ability in an L2, namely acquisition (which is a subconscious process of obtaining knowledge of language, identical to the process employed in L1 acquisition) and learning (which is a conscious process of obtaining knowledge about language, by focusing on form). These two types of knowledge are claimed to remain separate.
(ii) The Natural Order Hypothesis claims that the L2 learner acquires language rules in a predictable order, irrespective of the order in which the L2 learner is introduced to the rules in a language class.
(iii) The Monitor Hypothesis claims that learned knowledge acts as a monitor, editing the utterances produced by the acquired knowledge before (or sometimes after) the L2 learner speaks or writes.
(iv) The Input Hypothesis claims that an L2 learner needs comprehensible input, or i+1, in order to acquire his/her L2, where i is the L2 learner's current level of competence and i+1 is the next level along the natural order.
(v) The Affective Filter Hypothesis claims that the L2 learner's affective filter, which is a mental block, can be up (such as when the learner is anxious or unmotivated), preventing input from reaching the L2 learner's Language Acquisition Device (LAD), 16 or down (such as when the learner is not concerned with the possibility or consequences of failure), allowing input to reach the LAD and so allowing SLA to take place.

Testability in theory of Krashen's hypotheses
As was said in the introduction to this article, the non-explicit nature, or obscurity, of a hypothesis adversely affects the testability of that hypothesis. This is due to the fact that, if the content of the hypothesis is not clear and precise, then it is not possible to derive (precise) test implications from it.
Krashen's first hypothesis, viz. the Acquisition-Learning Hypothesis, is a hypothesis for which it is not possible derive test implications, because Krashen does not define "acquisition" and "learning" well enough for these concepts to be evaluated empirically. acts as a monitor for the utterances produced by the acquired knowledge. This lack of clarity stems, among other things, from the fact that "learning" and "acquisition", as well as the knowledge stemming from these two processes, are not well-defined. It is also not clear how one would prove that it is learned knowledge, and not some other mechanism, that does the monitoring.
Because "comprehensible input" has not been defined clearly by Krashen, it is not possible to derive precise test implications from the Comprehensible Input Hypothesis. Krashen does not state exactly what comprehensible input is and it is, therefore, unclear whether L2 learners in fact need it, as Krashen says they do. Furthermore, comprehensible input is supposed to lead the L2 learner to the next developmental level, but these levels have also not been defined. Therefore, it is not possible to determine whether comprehensible input does in fact lead to this next level.
Precise test implications can also not be derived from the Affective Filter Hypothesis, because it has not been clearly stated what this filter consists of. The factors or circumstances causing the filter to be up or down are also not described clearly. Although Krashen gives general descriptions of these factors or circumstances (e.g. Krashen (1985:3) states that the filter will be up when the learner "is unmotivated, lacking in self-confidence, or anxious ..., when he considers the language class to be a place where his weaknesses are revealed"), these factors and circumstances are merely examples and do not constitute an exhaustive list. If it had been possible to derive precise test implications, it would still not be clear what experimental finding(s) would refute this hypothesis. One reason for this is that it would not be possible to prove that the affective filter, and not some other mechanism, blocks the input from the LAD.
A second reason is that it would not be possible to prove that a lack of acquisition can be blamed on input not reaching the LAD. The development and exact functioning of the affective filter will have to be described precisely before empirical research concerning the accuracy of Krashen's Affective Filter Hypothesis can be carried out.
Because precise test implications cannot be derived from four of Krashen's five hypotheses, these four hypotheses are not testable in theory. 17

Felix's Competition Model
According to Felix (1985:51), children younger than 13 years have full and direct access to UG. They make use of a language-specific cognitive (LSC) system 18 to acquire language (both L1 and L2); they are not yet able to perform formal cognitive operations, such as general problem-solving, in order to acquire knowledge. At the age of about 13 years, Piaget's Stage IV (Formal Operations) has its onset. At this stage, it is possible for the L2 learner to make use of general problem-solving skills in SLA, using the general problem-solving cognitive (PSC) system. Both the LSC system and the PSC system can process language input, and, according to Felix, this is why competition arises between the LSC system (which is very effective for language acquisition) and the "stronger" or dominating, insuppressible PSC system (which is far less effective for language acquisition, as it is not languagespecific). UG is available to both the under-13 and over-13 L2 learner, but general problemsolving skills interfere with the over-13 L2 learner's access to UG, because of the competition between the LSC and the PSC.

Testability in theory of Felix's hypothesis
Felix's hypothesis is testable in theory, because precise test implications can be derived from it, and because it is possible to specify what experimental findings would indicate that the test implications of the hypothesis are false. The precise test implications which can be derived from Felix's hypothesis are as follows: (i) If the LSC system and PSC system are in competition with one another, then L2 learners older than 13 years who have good problem-solving skills should be poor at learning their L2, and L2 learners who are successful at learning the L2 should have poor problem-solving skills.
(ii) If the PSC system develops at the age of 13, then children younger than 13 years of age should not be capable of performing any general problem-solving procedures.
(iii) If low motivation and bad attitude can suppress the PSC system (as Bley-Vroman 1989:60 states they do), then L2 learners with poor attitude and low motivation should be the most successful L2 learners (Bley-Vroman 1989:61), because their PSC system is suppressed, leading to little competition with the LSC system.
(iv) If an L2 learner has access to two systems for processing language data, then the two systems should enter into direct competition instead of conspiring to make adults better L2 learners than children (White 1989:179 (iii) that L2 learners with poor attitude and low motivation are the least successful L2 learners. Note, however, that it is difficult to test whether L2 learners with poor motivation and a bad attitude toward SLA are good at learning their L2 and vice versa.
The reason for this difficulty lies in the definition of "motivation" and "attitude": these constructs have not yet been defined clearly enough to allow them to be subject to unambiguous empirical research; and (iv) that, instead of entering into direct competition, the PSC system and the LSC system conspire to make adults better L2 learners than children.
Felix states that the difference between adult and child L2 learners lies in the different learning strategies utilised by adults and children during SLA. He needs to prove that it is in fact a difference in learning strategy and not a difference in the availability or accessibility of UG. If the adult L2 learner no longer has (full) access to UG, then adults necessarily have to employ different language learning strategies than children do (children will use UG while adults cannot). What Felix needs to explain is why, if both adult L2 learners and child L2 learners have access to UG, these two age groups employ different learning strategies rather than both using UG that leads to successful SLA (Schachter 1991:111).
Felix's predictions are based on the assumption that children are more successful L2 learners than adults are. If this assumption is found to be true, then one cannot necessarily conclude that the reason for this is that UG is no longer available or no longer fully accessible to the adult L2 learner. It could also be, among other explanations, that proper triggering data are not available to the adult (Hilles 1991:307). It would also not be possible to determine whether children's success is due to different learning strategies in adults and children because of the unavailability of UG, or simply due to different learning strategies in adults and children. It is clear how these two instances can be distinguished from one another empirically.

Clahsen and Muysken's hypothesis
According to Clahsen and Muysken (1996:722), language learning involves a loss of information specified in UG. Initially, the L1 learner has access to all principles and parameters of UG, but once a certain parameter option has been chosen, the other option(s) are no longer accessible. This entails that once the learner's L1 grammar is mature, this speaker no longer has access to some aspects of UG. Clahsen and Muysken (1989:23)  It is also possible to test for so-called "wild" grammars, i.e. grammars that are not constrained by UG. If one finds such a grammar, then Clahsen and Muysken's hypothesis will, at least in part, be false. However, should one find no evidence of wild grammars, it would not necessarily prove Clahsen and Muysken's hypothesis correct.

Schachter's Window of Opportunity Hypothesis
According to Schachter (1996:185), there is a period before which and a period after which a UG principle is not available to the language learner. In other words, Schachter proposes a sensitive period 19 for parameter resetting: "for certain principles and parameters that depend on interaction with the environment, the principle or parameter will mature; … [after the end of the sensitive period -FS] that parameter will no longer be available for fixing" (Schachter 1996: 185).
For SLA, Schachter (1989:75) proposes that all that remains of UG, is a language-specific instantiation, namely those principles and parameter settings that are instantiated in the L2 learner's L1 grammar. UG in its entirety will thus no longer be available as a knowledge source for the acquisition of an L2.
On this point, Schachter's proposal does not differ from Clahsen and Muysken's. Also, both Schachter and Clahsen and Muysken go on to say that L2 learners make use of a set of cognitive systems or cognitive mechanisms to learn their L2, although Clahsen and Muysken are non-specific about the nature of these mechanisms. Schachter states that these cognitive systems are not specifically designed for SLA. Schachter (1988:231) claims that the L2 learner controls these cognitive systems or mechanisms in that (s)he can decide when to discontinue their operation (the latter would happen when the L2 learner is satisfied with his/her communicative ability). These systems are not the only sources of information on the rules of the target L2 that the L2 learner can make use of during SLA; the L1 grammar and the L2 input data are also available as sources of information.

Testability in theory of Schachter's hypothesis
The specific test implications derived from Clahsen and Muysken's hypothesis can also be derived from Schachter's, and the same experimental evidence would falsify these implications. Two further implications, that could not be derived for Clahsen and Muysken's hypothesis, can be derived for Schachter's hypothesis, viz. the following.
(i) If there exists a sensitive period for parameter resetting, then certain parameters should only be resettable at certain times during SLA, and these "certain times" should be clearly preceded and clearly followed by a period in which no such resetting can take place.
(ii) If learners can in fact wilfully decide to terminate their own SLA, no further L2 learning should take place after an L2 learner has decided to cease learning.
According to Schachter, it is possible to specify which experimental evidence would falsify the first of these two test implications. In explaining how one can test for the existence of a sensitive period for parameter resetting, Schachter (1996:185) reviews a 1992 study by Lee.
Lee studied Korean-speaking learners of American English who had the same length of exposure to the target L2 (3 years), but varying ages of first exposure to the target L2 (from 3 years to more than 22 years). The subjects were tested on their knowledge of the operation of Principle A of Binding theory in English, which includes knowledge of c-command and the Governing Category Parameter. It was found that adolescence (i.e. 11 to almost 14 years old) was the period in which most of the learners reset the Governing Category Parameter. The next most successful groups were the older children (8 to almost 11 years old) and the teenagers (14 to almost 18 years old). The least successful groups were the young children (6 to almost 8 years old) and the adults (19 to almost 26 years old). The results of Lee's study could be regarded as confirming the existence of a sensitive period for parameter resetting, i.e. a period before which and a period after which parameter resetting is less successful than in the sensitive period itself. This is one way in which Schachter's claim of a sensitive period for parameter resetting can be tested and possibly refuted.
It is, however, not clear what experimental evidence would prove the second test implication false. The question is: How would one test whether L2 learners can willfully terminate their own SLA? One would presumably have to make use of self-examination questionnaires. It is, however, unlikely that an L2 learner would be able to determine consciously whether or not he/she has chosen to stop improving his/her L2 knowledge. It is, therefore, doubtful whether this test implication is falsifiable. Therefore, Schachter's hypothesis is not in its entirety testable in theory.

Bley-Vroman's (1990) Fundamental Difference Hypothesis
In brief, Bley-Vroman (1990)  (ii) linguistic, meaning that the difference is caused by a change in the language faculty and not by some general change in learning ability; and (iii) qualitative (and not just quantitative), meaning that the "domain-specific acquisition system" 21 (Bley-Vroman's term) is unavailable, not simply attenuated in SLA.
All children have full access to UG and use a domain-specific learning system to acquire their L1, which accounts for the fact that success is inevitable in child L1 acquisition. Adults, in contrast, no longer have access to UG, but only to a mental representation of their L1 knowledge, and to a general abstract problem-solving system.
According to Bley-Vroman (1996:718), the FDH "permits UG-like effects via the L1". In other words, the L2 learner does not have access to UG, but observes aspects of UG instantiated in his/her L1 (aspects such as structure dependency). The L2 learner expects certain things from the L2 because of his/her L1 knowledge; these expected things include a syntax, a lexicon, a phonological system with syllables, feet and phonological phrases, as well as the potential for an infinite number of sentences to be generated in the L2 (Bley-Vroman 1989:51). The L2 learner makes use of these "expectations", as well as various general cognitive mechanisms (not specifically linguistic, according to Bley-Vroman (1996:718)), in the construction of the L2. In this latter respect, Bley-Vroman agrees with Schachter.
Although Felix and Bley-Vroman both propose a general cognitive or general problemsolving system that adults (but not children) use for SLA, their proposals differ in at least one respect. Felix proposes that the problem-solving system is in competition with UG in adults, whereas Bley-Vroman proposes that the problem-solving system replaces UG in adults. Bley-Vroman's proposal also differs from that of Clahsen and Muysken and that of Schachter.
Where the latter group of authors propose that L2 learners have access to those aspects of UG that are instantiated in their L1, Bley-Vroman proposes that they have no access to UG, but that they observe aspects of their L1 and then construct a surrogate UG. Bley-Vroman has to explain why, if the L2 learner knows all these "expected" things (such as that the L2 has a syntax, a lexicon, and a phonological system with certain structures, as well as that an infinite number of sentences could be generated in the L2) about the target L2, the L2 learner is still not successful at SLA.

Testability in theory of Bley-Vroman's hypothesis
The question arises as to how one distinguishes UG-like effects via the L1 (proposed by Bley-Vroman) from UG effects via the L1 (proposed by an indirect access hypothesis). The "expectations" mentioned by Bley-Vroman seem almost indistinguishable from the "expectations" that are provided by (indirect) access to UG. If UG-via-L1 and UG then deliver the same surface data, then Bley-Vroman will have to explain why his account (the account resorting to a combination of strategies) should receive preference above the theoretically more simple UG account.

CONCLUSION
In this article, a brief overview was given of prominent proponents of the various views on the accessibility of UG in adult SLA. One of the characteristics of a good theory is that it is testable (Mitchell and Myles 1998:5). Therefore, the testability in theory of each hypothesis was considered. It was found that some proposals were testable in theory whereas others were not, or at least not in their entirety. Whether testable (or falsifiable) in theory or not, all these hypotheses on the accessibility of UG in adult SLA continue to exist, together with a host of others not mentioned in this article. White (1996:115) states that perhaps the time has come to stop asking the broad question of whether or not UG is available to L2 learners. According to her, this question has stimulated a great deal of fruitful research, but it is now the time for more detailed focus on the precise nature of the linguistic competence of language learners, a focus that will continue to draw on current linguistic theory. Felix (1991:92) is of the opinion that the history of linguistic theory has taught us that fundamental issues are frequently not resolved by more and new data, but rather by careful examination of the rationale and conceptual perspective underlying opposing views.
According to Felix, this holds true for the UG accessibility debate: discussions on methodological issues (on data collection and analysis, amongst others) often obscure rather than clarify the actual issues at stake.
I agree with neither White nor Felix. Although much research has been conducted on the availability of UG in SLA, one cannot necessarily call this research "fruitful", as it has provided no conclusive answers. We are still left with many hypotheses, some of which are unfalsifiable. According to Rosenberg (1986:343), a field can only be narrowed down to a smaller number of theories once a great deal of data have been collected. The evaluation and elimination of theories on the availability of UG in adult SLA can, therefore, only take place relatively late in the scientific process, once there are enough data.
Therefore, contrary to what Felix claims, more data on the availability of UG in adult SLA are needed (especially as not all the existing data were obtained in empirically sound ways).
Studies need to be duplicated, because not all studies examine that which they set out to examine, and existing research results need to be confirmed with similar findings. Beretta (1991:497) states that, contrary to the already-successful sciences, where single theories tend to dominate, an (apparent) problem of multiple rival theories exists in SLA.
SLA, as a young science, lacks an adequate, agreed-upon theory of language learning (Beretta 1991:493). Seeing that theory evaluation and subsequent theory elimination only take place after much data have been collected, I would propose that it is not time to move away from the question of the availability of UG in adult SLA. We only have rival theories that in many cases cannot be examined because no appropriate data exist. What is needed is more data collected in empirically sound ways, followed by the elimination of falsified theories. This will move us toward a single theory, or at least toward multiple complementary theories, rather than the current multiple rival theories, on the way in which adults learn an L2.

NOTES
1 According to Beretta (1991:495), Long stated in 1985 that there were 15 to 20 theories on SLA. Eight years later, Long (1993:225) stated that there were 40 to 60. According to Gregg (1996: 73), it would be more appropriate to call these "theories in SLA" because they are theories of production, or of variation, or of interaction, etc. Agreed, not all of these theories are/were on the accessibility of UG in SLA, but the figures are quoted here to provide an idea of the quantitative state of theory formation in SLA. 2 "Theories" is the term most frequently used in the literature. Long (1993: 225) states that "theories" (in inverted commas) may be preferred by some, as not all of these so-called "theories" are in fact theories: hypotheses, models, metaphors, frameworks and perspectives are all included in the list of things referred to as "theories". In this article, the term "hypotheses" will be used, because it is unusual to talk of the test implications of theories.
3 Note that theories will not be critiqued on points other than testability and falsifiability, such as on the methodological soundness of the research offered as support for the theory. 4 "Direct" here means that the L2 learner does not access UG through those aspects of UG which are instantiated in the L1. 5 For all these test implications, one assumes that all else is equal, i.e. that non-UG-related factors are excluded. 6 When referring to a change in parameter setting, "resetting" is a problematic term. As Epstein et al. (1996:712) explain, the term could be seen as inaccurate because it implies that a parameter is reset from the L1 value to the L2 value, thereby losing the L1 value and no longer having that aspect of the L1 grammar represented in the language faculty. A more accurate term would be "new parameter setting" (Epstein et al. 1996:712) or "parameter setting", which would be the assignment of a new or different additional value to a parameter where the L1 parameter value differs from that of the L2. Despite its inaccuracy, "parameter resetting" is the popular and most widely used term in SLA writings, and for that reason it is used here. 7 Positive evidence is evidence as to what is grammatical or possible in a language, whereas negative evidence is evidence as to what is ungrammatical of impossible in a language.
8 One instance of when positive evidence does not seem to be sufficient for bringing parameter resetting, involves that SLA of adverb placement in English by L1 speakers of French: The Adjacency Condition on Case Assignment requires a case-receiving NP to be next to its case assigner. For this reason adverbs in English sentences may not occur between the (case-assigning) verb and the (case-receiving) NP, as illustrated in (1a): 1a. *Mary does slowly her homework Apart from this restriction, adverb placement in English is "very free" (White 1989:151), as illustrated in sentences (1b)  In French, the equivalent of (1a), together with the other "English" adverb placement options, is acceptable, as illustrated in (2):

Mary does slowly her homework
Therefore, it can be said that adjacency is not strictly observed in French (i.e. French is a [-strict adjacency] language), whereas adjacency is strictly observed in English (i.e. English is a [+strict adjacency] language).
For the L1 speaker of English learning French as an L2, sentences like (2) serve as positive evidence, i.e. as evidence that something that is not possible in their L1 is in fact possible in their L2.
The L1 speaker of French learning English as an L2 will not only need evidence of what is possible in English (i.e. positive evidence), but will also need negative evidence regarding adverb placement, i.e. evidence that sentences such as (1a) are impossible. Because of the very nature of impossible sentences, they do not occur in the input. The French L2 learner of English will only hear sentences such as (1b) to (1e), but because learners do not listen for non-occurring sentences or "gaps", this L2 learner will conclude that adverb placement in English is as free as it is in French and will, therefore, produce sentences such as (1a). This is one example of positive evidence being insufficient to bring about parameter resetting. 9 Althouogh this was the position of White in 1998White in , see now (2002 claims that L2 learners cannot make use of negative evidence to reset parameters. On White's current view, if the relevant positive evidence needed for parameter resetting is not available, then the parameter cannot be reset, even if negative evidence is provided. 10 See the discussion on the testability of Epstein et al.'s hypothesis for a very brief discussion on the difficulty of determining the "real" initial stage of SLA. 11 As White (2000:137) states, Schwartz and Sprouse claim that more than only the parameter values transfer. They believe that the entire grammar, apart from the lexical and morphological items and the phonetic matrices, constitutes the initial state of SLA. 12 Schwartz (1993:148) states that PLD consist of contextualised utterances in the language environment of the learner. PLD provides the learner with positive evidence. Without exposure to PLD, language development will not take place. 13 The Minimal Trees Hypothesis is in line with the so-called "Weak Continuity Approach" of L1 acquisition, according to which components of UG are available and accessible to children (i.e. L1 learners), but the functional categories need to be acquired on the basis of the input the child receives (Vainnika and Young-Scholten 1994:267). 14 Note that under the Minimalist Programme (Chomsky 1995) most if not all parameters are linked to features and feature strengths of functional categories. Under this approach, Vainikka and Young-Scholten's hypothesis would actually predict that parameter resetting should never be necessary in SLA. The reason why it should not be necessary is that L2 learners do not start out with functional projections that are transferred from their L1; rather, they acquire these functional projections in the same way that an L1 learner would, namely on the basis of the input and with the aid of UG. Therefore, these L2 learners should actually be "spot-on" with parameter values for the L2 from the moment they have acquired functional projections. 15 See Dube (2000: 237-238) for a brief summary of researchers who have argued for null functional elements.