To what extent do you think innate linguistic knowledge is involved in second language learning?

2

Principles of Universal Grammar in L2 acquisition

2.1 UG and the logical problem of L2 acquisition

As discussed in chapter 1, UG is motivated on learnability grounds: the subtle and abstract knowledge attained by native speakers goes far beyond the input that they receive as young children. In L2 acquisition, learners are faced with a similar task to that of L1 acquirers, namely the need to arrive at a system accounting for L2 input. In addition, L2 learners are also faced, at least potentially, with a logical problem of language acquisition, in that there are abstract, complex and subtle properties of grammar that are underdetermined by the L2 input (Schwartz and Sprouse 2000a, b; White 1985a, 1989). If it turns out that the L2 learner acquires abstract properties that could not have been induced from the input, this is strongly indicative that principles of UG constrain interlanguage grammars, parallel to the situation in L1 acquisition. This is true even if the linguistic competence of L2 learners differs from the linguistic competence of native speakers. In other words, it is not necessary for L2 learners to acquire the same knowledge as native speakers in order to demonstrate a poverty-of-the-stimulus situation in L2 acquisition; it is sufficient to show that L2 learners acquire complex and subtle properties of language that could not have been induced from the L2 input.

However, L2 learners already have a means of representing language, namely the grammar of the mother tongue. Thus, it might be that there is, in fact, no underde- termination problem: if L2 learners demonstrate the relevant kind of unconscious knowledge, it might be the case that they are drawing on the L1 grammar, rather than on UG itself, as argued, for example, by Bley-Vroman (1990) and Schachter (1990).

Thus, the strongest case for the operation of principles of UG in interlanguage grammars can be made if learners demonstrate knowledge of subtle and abstract linguistic properties which could neither have been learned from L2 input alone nor derived from the grammar of the mother tongue. In other words, there should be underdetermination not only with respect to L2 input but also with respect to the L1 grammar. Furthermore, one must also be able to rule out the possibility of learning on the basis of explicit instruction or by means of general learning

22

C o p y r i g h t 2 0 0 3 . C a m b r i d g e U n i v e r s i t y P r e s s .

A l l r i g h t s r e s e r v e d . M a y n o t b e r e p r o d u c e d i n a n y f o r m w i t h o u t p e r m i s s i o n f r o m t h e p u b l i s h e r , e x c e p t f a i r u s e s p e r m i t t e d u n d e r U . S . o r a p p l i c a b l e c o p y r i g h t l a w .

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX AN: 125056 ; White, Lydia.; Second Language Acquisition and Universal Grammar Account: s9814295.main.ehost

2.1 UG and the logical problem of L2 acquisition 23

principles (not specifically linguistic). For such reasons, L2 researchers try to identify situations involving a poverty of the L2 stimulus, where the available L2 input together with existing grammatical knowledge cannot account for acquisition unless one assumes that interlanguage grammars are constrained by UG (Schwartz and Sprouse 2000a, b; White 1989, 1990).

In summary, to demonstrate convincingly that interlanguage grammars are con- strained by principles of UG, the following conditions should hold:

i. The phenomenon being investigated must be underdetermined by the L2 input. That is, it must not be something that could be acquired by observa- tion of the L2 input, including statistical inferencing based on frequency of occurrence, on the basis of analogy, or on the basis of instruction.

ii. The phenomenon should work differently in the L1 and the L2. That is, it must be underdetermined by the L1 grammar as well. In this way, transfer of surface properties can be ruled out as an explanation of any knowledge that L2 learners attain.

2.1.1 The Overt Pronoun Constraint in L2

Let us reconsider the Overt Pronoun Constraint (Montalbetti 1984) (see chapter 1, section 1.3.1) in the context of L2 acquisition. Recall that, in [+null subject] languages, overt pronoun subjects of embedded clauses cannot receive a bound variable interpretation, hence cannot take quantified expressions or wh- phrases as antecedents, in contrast to null pronoun subjects. The sentences in (1) to (2) (repeated from chapter 1) illustrate the contrast in question. In the Span- ish sentence in (1a), the overt pronoun él cannot have the quantifier nadie as its antecedent, whereas this interpretation is possible with the null pronoun in (1b). Similar facts obtain for the Japanese sentences in (2).

(1) a. *Nadiei cree [que éli es inteligente] Nobodyi believes that hei is intelligent

b. Nadiei cree [que proi es inteligente] Nobodyi believes that (hei) is intelligent

(2) a. *Darei ga [karei ga kuruma o katta to] itta no? Whoi nom hei n om car ac c bought that said Q ‘Who said that he bought a car?’

b. Darei ga [proi kuruma o katta to] itta no? Whoi nom (hei) car ac c bought that said Q ‘Who said that (he) bought a car?’

Is there a logical problem for the L2 learner of a null subject language like Spanish or Japanese with respect to this interpretive restriction on embedded

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

24 Principles of Universal Grammar in L2 acquisition

pronominal subjects? Consider first the L2 input. Discovering the restriction on the basis of input represents as much of a challenge for the L2 learner as it does for the L1 learner. In general, overt and null pronouns appear in similar or over- lapping contexts (see chapter 1, table 1.1), they take similar antecedents, and there appears to be nothing in the L2 input to signal the difference between them as far as bound variable status is concerned. Thus, frequency of occurrence in the input is unlikely to provide any useful clue as to when pronouns may occur and under which interpretations. The L2 learner, like the L1 learner, somehow has to discover that, in a restricted and rather uncommon set of sentence types, an overt pronoun cannot appear with a particular interpretation (as a bound variable), even though it can appear with another interpretation (with a referential antecedent). In addition, classroom input does not appear to be helpful in this respect. Accord- ing to Pérez-Leroux and Glass (1997), this issue is not discussed in L2 Spanish textbooks or taught in L2 classrooms; Kanno (1997) makes the same observations for L2 Japanese. In other words, knowledge of the interpretative restriction on overt pronouns is underdetermined by the L2 input, satisfying the first of the two conditions described in section 2.1.

As for attaining such knowledge on the basis of the mother tongue, if L2 learners are native speakers of a [−null subject] language, such as English, nothing in the L1 grammar would allow them to arrive at the appropriate distinction, since overt pronouns in English are not restricted in the same way. Thus, an investigation of the acquisition of Spanish or Japanese by native speakers of English would also meet the second requirement outlined in section 2.1.

Acquiring the interpretive constraint on overt pronouns in L2, then, constitutes a classic learnability problem. If L2 learners are successful in this domain, it would strongly support the claim that interlanguage grammars are UG-constrained. Recently, researchers have conducted experiments to investigate whether adult learners observe the Overt Pronoun Constraint. Pérez-Leroux and Glass (1997, 1999) have examined L2 Spanish, while Kanno (1997, 1998b) has investigated L2 Japanese. In both cases, the L2 learners are native speakers of English. Thus, both conditions for investigating whether the interlanguage grammar is UG-constrained are met: the L2 input underdetermines the phenomenon being investigated and the L1 is not a potential source of information about restrictions on overt pronouns. The issue, then, is whether L2 learners in fact behave in ways that are consistent with this constraint. If they do, UG is implicated since other potential sources of such behaviour have effectively been eliminated.

In Pérez-Leroux and Glass (1999) (see box 2.1), L2 learners of Spanish at different levels of proficiency were tested by means of a task which involved translating biclausal sentences from the L1 English into the L2 Spanish, following written contexts (in the L1) which strongly favoured either a quantified antecedent

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.1 UG and the logical problem of L2 acquisition 25

Box 2.1 Overt Pronoun Constraint (Perez-Léroux and Glass 1999)

Languages: L1 = English, L2 = Spanish. Task: Translation from English into Spanish. Each sentence preceded by a para- graph (in English) to provide a context. Sample stimuli:

Bound variable context Referential context

The court charged that some journalists had been in contact with the jurors. Several of them were questioned by the judge.

In the O.J. Simpson trial, it is clear that the press has a negative bias against the defendant in their reporting. Some journalist said that he was a wife-beater.

To translate: To translate: No journalist admitted that he had talked to the jurors.

But no journalist said that he is guilty.

Results:

Table 2.1.1 Production of null and overt pronouns (in %)

Bound variable stories Referential stories (# = 4) (# = 4)

Null Overt Null Overt

L2 groups Elementary (n = 39) 57.7 34 21.2 67.9 Intermediate (n = 21) 73.8 26.2 35.7 59.5 Advanced (n = 18) 93.1 0 58.3 31.9

Native speakers (n = 20) 85 13.7 31.3 67.5

n = number of subjects # = number of stimuli

(within the same sentence) or a discourse-based referential antecedent (external to the sentence) for the embedded subject pronoun. If the interlanguage grammar is constrained by the Overt Pronoun Constraint, translations should disfavour overt pronouns and favour null subjects where the antecedent is quantified, since overt pronouns are prohibited in this context. In the case of the referential stories, on the other hand, either kind of pronoun is grammatical. As can be seen in table 2.1.1, production of overt pronouns is significantly lower in bound-variable contexts than in referential contexts and this is true of all groups. Correspondingly, the use of null pronouns is significantly higher following bound variable contexts than referential contexts.1 These results suggest that L2 learners, like native speakers,

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

26 Principles of Universal Grammar in L2 acquisition

distinguish between referential and bound variable interpretations of pronouns, largely disallowing overt pronouns in the latter context, compared with their use in other contexts. Thus, the results are supportive of the claim that interlanguage grammars are constrained by the Overt Pronoun Constraint, even at the elementary level.

However, ideally, it would seem that the incidence of overt pronouns in bound variable contexts ought to be 0%, a result achieved only by the advanced group. Pérez-Leroux and Glass attribute the incidence of overt pronouns in the elementary and intermediate groups to a tendency to overuse overt pronouns in general (based on the L1). In many cases of L2 (and native speaker) performance, there will be such additional factors that come into play. Indeed, what is important here is not the absolute figures but rather the fact that there are significant differences in per- formance across different sentence types. The issue is whether the interlanguage grammar shows evidence of certain distinctions (in this case in the incidence of overt subjects with referential as opposed to quantified antecedents). If learner per- formance on one sentence type differs significantly from performance on another, this suggests that the interlanguage grammar represents the relevant distinction (whatever it may be). If these sentence types were represented in the same way in the interlanguage grammar, such differences would be unexpected; instead, the sentences should be treated the same. (See Grimshaw and Rosen (1990) for related discussion relevant to L1 acquisition and Martohardjono (1993, 1998) for L2.)

Turning now to L2 Japanese, Kanno (1997, 1998b) investigates whether adult learners know the prohibition on quantified and wh antecedents for overt pronouns. (See box 2.2.) Her task was a coreference-judgment task, quite different from the task used by Pérez-Leroux and Glass. Subjects were presented with biclausal sentences with quantified and referential main-clause subjects and overt or null embedded pronoun subjects; they had to indicate whether or not the embedded pronoun could refer to the subject of the main clause.

Kanno found that native speakers and L2 learners alike differentiated in their treatment of overt pronouns depending on the type of antecedent involved (quan- tified or referential) (see table 2.2.1). Native speakers overwhelmingly rejected the interpretation where an overt pronoun took a quantified antecedent, respond- ing instead that the overt pronoun must take a sentence-external referent. This was not due to a general prohibition against quantified antecedents, since these were accepted in the case of null subjects. Nor was it due to any general dislike of sentence-internal antecedents, since referential antecedents within the same sentence were accepted. The L2 learners showed a remarkably similar pattern of results; their performance was not significantly different from the controls. Both native speakers and L2 learners, then, appeared to be following the Overt Pronoun Constraint, disallowing quantified antecedents for overt pronouns. Furthermore, Kanno took the precaution of including a control group of native speakers of

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.1 UG and the logical problem of L2 acquisition 27

Box 2.2 Overt Pronoun Constraint (Kanno 1997)

Languages: L1 = English, L2 = Japanese. Task: Corereference judgments. Japanese biclausal sentences, each followed by a question asking who was performing the action described in the embedded clause. Sample stimuli:

Null pronoun with quantified/wh antecedent

Null pronoun with referential antecedent

Dare ga asita uti ni iru to itteiru n desu ka. Tanaka-san wa raisyuu Kyooto e iku to itteimasita yo.

(Who says that (he) is going to stay home tomorrow?)

(Mr Tanaka was saying that (he) is going to Kyoto next week.)

Q: Who do you suppose is going to stay home tomorrow?

Q: Who do you suppose is going to Kyoto next week?

(a) same as dare (b) another person (a) Tanaka (b) someone other than Tanaka

Overt pronoun with quantified/wh antecedent

Overt pronoun with referential antecedent

Dare ga kyoo kare ga uti ni iru to itteiru n desu ka.

Tanaka-san wa raisyuu kare ga Tokyoo e iku to iimasita yo.

(Who says that he is going to stay home today?)

(Mr. Tanaka said that he would go to Tokyo next week.)

Q: Who do you suppose is going to stay home today?

Q: Who do you suppose will go to Tokyo next week?

(a) same as dare (b) another person (a) Tanaka (b) someone other than Tanaka

Results:

Table 2.2.1 Acceptances of antecedents for null and overt pronouns (in %)

Quantified antecedents Referential antecedents (# = 10) (# = 10)

Null Overt Null Overt

L2 learners (n = 28) 78.5 13 81.5 42 Native speakers (n = 20) 83 2 100 47

English who judged equivalent sentences in English (with overt pronouns in the lower clause). This group allowed quantified antecedents and referential an- tecedents for overt pronouns to an equal (and high) extent, suggesting that the L1 English is not the source of the L2 learners’ behaviour in Japanese.

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

28 Principles of Universal Grammar in L2 acquisition

There are a number of differences between these two sets of experiments, both in the methodology and in the results. In Pérez-Leroux and Glass’s experiments, the task was a production task, involving translation; in Kanno’s studies, learners had to make coreference judgments. In Pérez-Leroux and Glass’s task, the subject of the main clause was always a quantified NP; referential antecedents had to be found elsewhere within the discourse provided by the story. In contrast, Kanno varied the nature of the main-clause subject (quantified/wh expression versus a referential NP), such that the referential antecedent was always within the same sentence as the embedded pronoun. There is one obvious difference in the results of the two sets of studies which may be attributed to these task differences. In the case of most groups, Pérez-Leroux and Glass found that translations of the sentences following a referential context tended to disfavour null subjects; overt pronouns were used instead (see table 2.1.1). Kanno, on the other hand, found that referential antecedents were accepted for null subjects (see table 2.2.1). Montalbetti (1984) suggests that an overt pronoun is preferred when its referential antecedent is not within the same sentence; this would account for the preference observed by Pérez-Leroux and Glass.

Despite this difference in the treatment of null pronouns and their antecedents, the crucial issue concerns treatment of overt pronouns with quantified antecedents. Here, results from the two studies are consistent with each other: L2 learners of Spanish and Japanese alike show significantly lower use or acceptance of overt pronouns with quantified antecedents, as do native speakers. Thus, taken together, the experiments of Pérez-Leroux and Glass (1999) and Kanno (1997) provide evidence for Overt Pronoun Constraint effects in different L2s, with learners at different levels of proficiency and with different methodologies being employed.

While these results are compelling, we should bear in mind that we are interested in the grammars of individual speakers rather than groups. The Overt Pronoun Constraint, as a principle of UG, should constrain the grammar of each individual learner. Group results that suggest UG-consistent behaviour on the part of L2 learners may in fact conceal potentially problematic individual variation. While Pérez-Leroux and Glass do not provide analyses of results from individuals, Kanno does. Kanno found that 100% of the native-speaker controls and 86% of the L2 learners demonstrated consistent behaviour with respect to overt pronouns with quantified antecedents, where consistency is defined as the rejection of quantified antecedents for overt pronouns in 4 or 5 cases out of 5. This level of consistent rejections by individuals suggests that the group results are indeed an accurate reflection of individual linguistic competence.

However, in another study, Kanno (1998a, b) found greater variability in indi- vidual results. A different group of English-speaking intermediate-level learners

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.1 UG and the logical problem of L2 acquisition 29

of Japanese was tested at two different points in time, with a twelve-week interval between the two sessions. The task was the same as the one in the previously de- scribed experiment. At both test sessions, there was a significant difference between coreference judgments involving a quantified antecedent, depending on whether the embedded pronoun subject was null or overt. In the latter case, acceptance of coreference was significantly lower, consistent with the Overt Pronoun Constraint. There were no differences in group performance across the two test sessions. At the individual level, however, with consistency defined as above, Kanno found that only 9 of 29 L2 learners consistently excluded quantified antecedents for overt pro- nouns in both test sessions. A further 15 subjects showed consistent rejection in one session or the other. Thus, at the individual level, operation of the Overt Pronoun Constraint was not consistent across subjects or over time.

Kanno (1998b) suggests that, while the Overt Pronoun Constraint does constrain interlanguage grammars, it cannot always be accessed in performance, perhaps because it has not been activated, due to insufficient exposure to suitable input. However, there is something unsatisfactory about this explanation, in that lack of relevant input is precisely the motivation for principles like the Overt Pronoun Constraint in the first place. Once the L2 learner discovers that Japanese is a null-subject language allowing both null and overt pronouns, the Overt Pronoun Constraint ought to come into effect.

For the sake of the argument, however, let us grant that there may be individual variability of the kind suggested by Kanno. What would it mean for the claim that UG constrains interlanguage grammars if individual learners do not in fact observe a particular constraint consistently? Grimshaw and Rosen (1990) consider apparently problematic results in L1 acquisition where children appear to accept violations of another principle of UG, namely, Principle B of the Binding Theory. Grimshaw and Rosen point out that a number of performance factors may intervene to conceal underlying competence. They suggest that L1 acquirers may ‘know’ UG principles (unconsciously) but nevertheless fail to observe or ‘obey’ them in certain circumstances. In other words, competence and performance will some- times diverge, such that sentences that are ruled out as ungrammatical violations of some principle of UG are nevertheless accepted in certain cases. According to Grimshaw and Rosen, even when this happens, it is nevertheless possible to show that children know the principle in question. If the child’s grammar were not constrained by some principle, one would expect grammatical and ungrammati- cal sentences to be treated alike, since the principle ruling out the ungrammatical sentences would not be available. Therefore, if children treat grammatical and un- grammatical sentences differently (accepting significantly more of the former than the latter), this is sufficient to show that the two sentence types are not the same in the child grammar and that the principle in question must be operating, even

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

30 Principles of Universal Grammar in L2 acquisition

if children do not perform with a high degree of accuracy on the ungrammatical sentences. Linguistic competence accounts for the distinction between grammat- ical and ungrammatical sentences; performance factors account for the failure to observe the distinction absolutely.

Let us apply this logic to the Overt Pronoun Constraint results from individual learners in Kanno’s study. Looking only at the sentences involving overt pronouns and quantified antecedents, Kanno found that some subjects, at both testing ses- sions, failed to consistently exclude the ungrammatical interpretation, leading her to conclude that the Overt Pronoun Constraint (hence, UG in general) is not con- sistently accessible. Following Grimshaw and Rosen’s proposal, however, it would be more appropriate to compare each learner’s acceptance of antecedents for overt pronouns, i.e. referential (grammatical) with quantified (ungrammatical). (In other words, the comparison that Kanno makes at a group level should also be made at the individual level.) If interlanguage grammars are constrained by the Overt Pronoun Constraint, then we expect each subject to show a significant difference in acceptance of grammatical versus ungrammatical interpretations. This should be so even if performance on particular ungrammatical sentences is somewhat vari- able. While Kanno does not provide data to allow one to check this point, she does show that the L2 learners were much more consistent in their coreference judg- ments involving grammatical interpretations, which suggests that, individually, they are in fact distinguishing between licit and illicit interpretations.

To summarize the findings relating to the Overt Pronoun Constraint, Pérez- Leroux and Glass (1997, 1999) and Kanno (1997, 1998a, b) have shown that learners of different L2s, at different proficiency levels, tested by means of differ- ent tasks, show significant differences in their treatment of overt pronouns with quantified antecedents (illicit) versus null pronouns with quantified antecedents or overt or null pronouns with referential antecedents (licit). Group and individual results suggest that L2 learners are making the relevant distinctions, distinctions which could not have been derived from the L1 grammar or the L2 input alone, supporting the claim that interlanguage grammars are subject to the Overt Pronoun Constraint.

2.1.2 Process versus result nominals in L2 French

The previous section discussed L2 knowledge of a principle of UG which restricts the distribution and interpretation of embedded pronominal subjects. We turn now to an examination of another aspect of grammar involving an interplay between syntactic and interpretive factors, namely the distinction between two types of nominals, known as process and result nominals. The research described

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.1 UG and the logical problem of L2 acquisition 31

below addresses the question of whether there is a learnability problem relat- ing to the syntax–semantics interface, a relatively new field of inquiry in the L2 context. Again, the underlying assumption is that knowledge of the property in question must have its origin in UG, the issue being whether L2 learners reveal unconscious knowledge of subtle distinctions between the two types of nominals, distinctions which are unlikely to be derivable from the L2 input alone or from the L1.

Dekydtspotter, Sprouse and Anderson (1997) look at whether L2 learners are sensitive to differences between process and result nominals in L2 French. A process nominal describes an event or something ongoing; a result nominal names the ‘output of a process or an element associated with a process’ (Grimshaw 1990: 49). The English nominal destruction illustrates the distinction. In (3a), destruction refers to the process of destroying the city; the enemy brings about the destruction (i.e. it is the agent), while the city undergoes the destruction (i.e. it is the theme). In (3b), on the other hand, destruction refers to the outcome, the result of the destroying. (Examples from Grimshaw 1990: 52.)

(3) a. The enemy’s destruction of the city was awful to watch. (process) b. The destruction was awful to see. (result)

Dekydtspotter et al. (1997) investigate dyadic process and result nominals in L2 French, that is nominals taking an agent and a theme argument (as in (3a)). There are restrictions on the arguments of such nominals, in particular, restrictions on the form of the agent. Consider the sentences in (4) (a result nominal) and (5) (a process nominal). In both cases, the theme is expressed in a phrase introduced by de (‘of’), while the agent can be introduced by par (‘by’). In (4a), la 9e is the theme and Karajan is the agent; in (5), Tokyo is the theme, while Godzilla is the agent. So far, then, result and process nominals pattern alike.

(4) Result nominal a. la version de la 9e par Karajan

the version of the 9th by Karajan ‘Karajan’s version of the Ninth symphony’

b. la version de la 9e de Karajan the version of the 9th of Karajan

(5) Process nominal a. la destruction de Tokyo par Godzilla

the destruction of Tokyo by Godzilla ‘Godzilla’s destruction of Tokyo’

b. *la destruction de Tokyo de Godzilla the destruction of Tokyo of Godzilla

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

32 Principles of Universal Grammar in L2 acquisition

The two kinds of nominals differ, however, in that result nominals also permit the agent to be introduced by de, as in (4b). In process nominals, on the other hand, this is not possible, as shown by the ungrammaticality of (5b). Thus, result nominals permit multiple phrases introduced by de (namely, both theme and agent), occurring in either order, whereas process nominals do not. (Nevertheless, native speakers have a preference for the agent of a result nominal to be introduced by par rather than de.)

This difference between result and process nominals within French (as well as other languages) stems from independent restrictions on argument structure and event structure originating in UG. In particular, in process nominals with multiple de arguments there are problems in assigning the agent role, violating the principle of Full Interpretation, which requires that arguments be interpretable at LF. (Unfortunately, Dekydspotter et al. give little detail on precisely how this constraint is played out in the case of the different types of nominals that they investigate.2) In consequence of an independent parametric difference between French and English, the distinction between these nominals is partially obscured in English. English disallows multiple of phrases in these contexts, as can be seen by looking at the literal glosses for (4b) and (5b). Instead, the agent appears as a prenominal genitive, whether the nominal is process or result, as can be seen by the translations of (4a), where the agent is expressed as Karajan’s, and (5a) where it appears as Godzilla’s. (Alternatively, English can express the agent in a by-phrase, equivalent to the par-phrases in French: the destruction of Tokyo by Godzilla; the version of the 9th by Karajan – again, there is no distinction between the two types of nominal in this respect.)

Dekydtspotter et al., following Carstens (1991), attribute this crosslinguistic difference (namely, the possibility of multiple de arguments in French and the im- possibility of multiple of arguments in English) to a parametric difference between the two languages. As described in chapter 1, section 1.4.1, nouns in French raise from N to the head of a functional projection, Num (Bernstein 1993; Carstens 1991; Valois 1991). When a nominal raises to Num, it governs both its arguments and case is assigned under government (via de-insertion) to both theme and agent. In languages like English, in contrast, nouns do not raise to Num; in consequence, a nominal will never govern its agent and case assignment to the agent via of insertion is ruled out, case assignment being achieved, instead, by an alternative mechanism (Spec–head agreement).

Turning to the L2 logical problem for English-speaking learners of French, at issue is whether they will acquire the distinction between process and result nominals, with respect to the possibility of multiple de-phrases. It is unlikely that the L2 input is sufficient to signal this difference, since what has to be acquired

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.1 UG and the logical problem of L2 acquisition 33

is knowledge that sentences like (5b) are ungrammatical, whereas sentences like (4b) are possible. Since both types of nominal allow the agent to be introduced by par (as in (4a) and (5a)), the potential for overgeneralization is considerable. Nor is this a topic that is specifically taught in L2 French classrooms (though contrasts between de and par in other contexts may be).

It also seems highly unlikely that properties of the L1 English would be sufficient to allow learners to arrive at the relevant distinction in the L2. Although there are subtle distinctions between process and result nominals in English, these are not manifested in the same way. Because English does not have N-raising, it does not allow multiple of -phrases. There is no distinction in how the agent is realized in process and result nominals; in both cases, the agent is found either as a prenominal genitive (Godzilla’s destruction; Karajan’s version) or in a postnominal by-phrase (the destruction by Godzilla; the version by Karajan). Thus, the L2 acquisition of the distinction between process and result nominals with respect to multiple de- phrases constitutes a learnability problem. If English-speaking learners of French show knowledge of the distinction, this would suggest that UG constrains the interlanguage grammar in this domain.

Dekydtspotter et al. tested for knowledge of the process/result distinction with respect to multiple de-phrases in an experiment involving English-speaking adults learning French. (See box 2.3.) The task was an acceptability judgment task, with written scenarios providing a context (i.e. somewhat similar to the proce- dure adopted by Pérez-Leroux and Glass (1997, 1999), except that Pérez-Leroux and Glass asked for translations not judgments). Since the context is crucial to establish the interpretation of the nominal as either process or result, the scenarios were presented in the L1 (again, like Pérez-Leroux and Glass), in order to ensure that the L2 learners would make their judgments on the basis of the appropriate interpretation. The test sentences to be judged were presented in French.

Native speakers of French distinguished sharply and significantly between the two types of nominals, accepting multiple de-arguments in result nominals to a significantly greater extent than in process nominals, where they were largely rejected. (See table 2.3.1.) The relatively low rate of acceptance even in result nominals reflects the fact that par is the preferred way of realizing the agent. The L2 learners at all levels showed the same distinction between process and result nominals, with the advanced learners not being significantly different from the controls. The beginners and intermediate subjects in general showed a much higher acceptance rate of multiple de phrases with both types of nominals; nevertheless there is a significant difference between the two sentence types, in the expected direction, suggesting that the distinction is represented in the grammar; this is true both at the group level and at the individual level.

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

34 Principles of Universal Grammar in L2 acquisition

Box 2.3 Process and result nominals (Dekydtspotter et al. 1997)

Languages: L1 = English, L2 = French. Task: Acceptability judgments of sentences containing both a theme and an agent introduced by de. Each sentence preceded by a paragraph (in English) to provide a context. Sample stimuli:

Result nominals (grammatical) Process nominals (ungrammatical)

Jean loves nineteenth-century train stations. La Gare du Nord, Paris’s northern train station, still has some of that feel, even though the steam engines are long gone. Whenever Jean looks at a catalogue of Monet’s works (in particular the work entitled La Gare du Nord), he gets a feel for what it would have been like to be there. Not surprisingly,

Shocking and disturbing, yes, but nonetheless true! The executioner’s wife was having an affair, and the only time she could meet with her lover was when her husband the executioner was on the job. She read in the newspaper that a couple of traitors had been sentenced to death on Friday, so she sent a note to her lover, saying:

Jean adore la peinture de la Gare Viens chez moi vendredi pendant du Nord de Monet. l’éxécution de mon mari des traitres

Feels possible in the context? Feels possible in the context? Yes No Cannot decide Yes No Cannot decide

Results:

Table 2.3.1 Acceptances of multiple de arguments (in %)

Result nominals Process nominals (# = 10) (# = 10)

L2 groups Beginners (n = 38) 69.21 53.51 Intermediate (n = 32) 71.56 48.96 Advanced (n = 20) 63.5 24.17

Native speakers French (n = 48) 50.42 15.65 English (n = 24) 22.08 10.53

In summary, L2 learners at all levels of proficiency showed an asymmetry in their acceptance of multiple phrases introduced by de in result and process nominals. The results suggest that the interlanguage grammar is constrained by UG, since the distinction between process and result nominals appears to come neither from the L1 nor from the L2 input.

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.1 UG and the logical problem of L2 acquisition 35

Nevertheless, a caveat is in order. Like Kanno (1997), Dekydtspotter et al. included a group of native speakers of the L1 who judged English translations of the French test items, all of which are ungrammatical in English, in order to see whether learners might simply be judging L2 sentences on the basis of their mother tongue. Overall, multiple of -phrases were rejected in both types of nominal in contrast to the L2 results, suggesting that the L1 grammar is unlikely to be the source of knowledge of the distinction with respect to the L2 sentences. Nevertheless, the difference in their judgments on the two sentence types in the L1 was significant; in other words, even though both sentence types were considered fully ungrammatical, the process nominals with two of -phrases were even less likely to be accepted than the result nominals. Thus, one cannot totally exclude the possibility that the distinction shown by the L2 learners was in some way based on the L1 grammar.

2.1.3 Principles of UG in early interlanguage grammars: the ECP

So far, the experimental research that we have considered has mainly targeted intermediate or advanced learners, a tacit assumption being made that if some UG principle can be shown to constrain later interlanguage grammars, it would also have constrained the grammars of earlier stages. Indeed, if interlan- guage grammars are UG-constrained, this should be so from the earliest stages of L2 acquisition.3 In other words, we expect to find evidence of UG principles functioning in the interlanguage grammars of beginners or low-proficiency L2 learners, all other things being equal. However, a methodological problem arises in this context. UG principles are proposed as an explanation of very subtle and ab- stract linguistic phenomena, as we have seen; in many cases, these principles relate to properties exemplified in complex sentences. In consequence, some of the sen- tence types that are typically investigated by linguists and acquisition researchers are, for reasons independent of the UG issue, too difficult for beginner-level L2 learners to deal with.

Kanno (1996) has shown that Japanese case drop, a phenomenon that occurs in simple sentences, can be used to investigate whether the interlanguage grammars of beginners and low-proficiency learners are constrained by UG. Japanese has case particles, including nominative /-ga/ and accusative /-o/, as shown in (6a) (examples from Kanno). When an object is marked with accusative case, the particle may be dropped in informal spoken language, as in (6b); however, it is ungrammatical to drop the nominative particle marking the subject, as in (6c). Thus, there is an asymmetry here, with respect to the type of particle that can be dropped.

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

36 Principles of Universal Grammar in L2 acquisition

(6) a. John ga sono hon o yonda. John nom that book ac c read-past ‘John read that book’

b. John ga sono hon yonda. John nom that book read-past

c. *John sono hon o yonda. John that book ac c read-past

Japanese case drop meets the criteria discussed above with respect to learnability considerations. Firstly, the relevant generalization (that accusative particles may be dropped while nominative may not) is not derivable on the basis of the L1, assuming an L1 without case particles, such as English. Secondly, it does not appear that this property could be acquired on the basis of L2 input alone, since the input underdetermines the case drop property. If the learner were to try and make the relevant generalization on the basis of ‘noticing’ dropped particles in the input, the potential for overgeneralization is considerable. In addition to permitting accusative case drop, Japanese has a topic marker /-wa/ which can be dropped (Kuno 1973), as shown in (7a), where the topic marker is present, and (7b), where it has been omitted.4

(7) a. Johni wa Hanako ga proi sono hon o yonda to itta. Johni t o p Hanako nom (hei) that book ac c read-pa st that said ‘Speaking of John, Hanako said that he ( = John) read that book.’

b. Johni Hanako ga proi sono hon o yonda to itta. Johni Hanako n om (hei) that book ac c read-past that said ‘Speaking of John, Hanako said that he ( = John) read that book.’

Since subjects can be topicalized and /-wa/ can be omitted from a topicalized subject, it would not be unreasonable to assume that any particle marking a subject can be dropped. In addition, the nominative /-ga/ can be dropped when it occurs on the complement of a stative verb or when it marks the subject of an unaccusative verb.5 Furthermore, case drop is apparently not taught in Japanese L2 classrooms, although it is, presumably, exemplified in classroom input.

Fukuda (1993) attributes the asymmetry in case particle deletion (nomina- tive prohibited, accusative permissible) to the Empty Category Principle (ECP) (Chomsky 1981), a principle which accounts for a variety of subject–object asym- metries across languages. The ECP requires nonpronominal empty categories (i.e. traces or variables) to be properly governed. An empty category in object position is properly governed by the verb, whereas an empty category in subject position is not properly governed. Hence, in a number of different situations, empty cate- gories are permitted in object position but not subject position. Assuming that null case particles are empty categories of the relevant sort, it is permissible to omit an accusative case particle because the empty particle is properly governed by

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.1 UG and the logical problem of L2 acquisition 37

Box 2.4 Case-particle deletion (ECP) (Kanno 1996)

Languages: L1 = English, L2 = Japanese. Task: Grammaticality judgments. The naturalness of sentences is assessed on a scale of 1 (unnatural) to 3 (natural). Sample stimuli:

Missing accusative particle in sentences with 2 overt arguments (grammatical)

Missing accusative particle in sentences with 1 overt argument (grammatical)

Suzuki-san wa dono biiru nomimasita ka? Dono biiru nomimasita ka? (Which beer did Mr(s) Suzuki drink?) (Which beer did (s/he) drink?)

Missing nominative particle in sentences with 2 overt arguments (ungrammatical)

Missing nominative particle in sentences with 1 overt argument (ungrammatical)

Dono gakusee biiru o nomimasita ka? Dono gakusee nomimasita ka? (Which student drank beer?) (Which student drank (it)?)

Results:

Table 2.4.1 Case drop: mean naturalness scores (from 1 to 3)

2 overt arguments (# = 8) 1 overt argument (# = 8) ac c no m ac c no m

L2 learners (n = 26) 2.4 1.76 2.58 1.64 Native speakers (n = 20) 2.6 1.36 2.86 1.31

the verb.6 In contrast, a null nominative particle would not be properly governed, hence nominative particles must be overt.

Kanno (1996) investigated whether beginners learning Japanese are sensitive to this asymmetry in case-particle deletion, that is, whether they distinguish be- tween grammatical sentences with dropped accusative particles like (6b) and ungrammatical sentences with dropped nominative particles like (6c). Subjects were tested on a grammaticality-judgment task in which they had to assess the naturalness of sentences, like those in (6), with and without case particles. (See box 2.4.)

Results revealed significantly greater acceptance of accusative case drop over nominative case drop, and no significant differences between L2 learners and native speakers. This is true for sentences with one or two arguments expressed overtly. (See table 2.4.1.) These results, then, are consistent with the proposal that a UG principle (the ECP) functions in the early interlanguage grammar. Kanno was able

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

38 Principles of Universal Grammar in L2 acquisition

to establish this without having to resort to the kinds of complex sentences that have previously been used to investigate knowledge of the ECP in L2 English (e.g. Bley-Vroman, Felix and Ioup 1988; White 1988) or L2 German (Felix 1988).

If the interlanguage grammars of beginners show evidence of being constrained by the ECP, one expects the same in the case of more advanced learners. In a subsequent study, Kanno (1998a) reports results which largely confirm the orig- inal findings. The second study involved intermediate-level learners of Japanese. Group results again revealed significant differences between acceptance of miss- ing nominative and accusative case particles, consistent with the operation of the ECP. In this follow-up study, Kanno also looked at individual results and found that most individuals exhibited the relevant contrast, although not necessarily as strongly as native speakers.

In contrast, Kellerman, van Ijzendoorn and Takashima (1999), testing Dutch- speaking L2 learners, were unable to replicate Kanno’s finding of a sensitivity to the subject–object asymmetry in case-particle deletion. This attempted replication, however, is somewhat problematic, in that it involved teaching a miniature artificial language to students not previously exposed to Japanese, in a single session. This artificial language had the following characteristics: it was SOV, subjects and objects could be omitted (in other words, it was a prodrop language), and overt subjects and objects were always case-marked. Students were explicitly taught these properties, by means of translation; they had no exposure to input involving missing case particles and they were not taught that case markers could be dropped. Immediately after the teaching phase, they were explicitly told that there was one rule of this language which they had not been taught and which they were to try to discover. Test sentences manipulated dropped nominative or accusative case particles, the idea being that, if interlanguage grammars are constrained by the ECP, dropping accusative case would be permissible, while dropping nominative would not. Learners showed some acceptance of dropped case marking, both nominative and accusative. A second study modified properties of the artificial language (using real Japanese lexical items instead of invented ones) and the test instructions, as well as involving somewhat older subjects; it was otherwise similar to the first study. Again, results showed no subject–object asymmetry in acceptance of case particle deletion.

Kellerman et al. claim that these results cast doubts on Kanno’s conclusion that early L2 grammars are constrained by the ECP. However, there are a number of problems with these two studies which in turn cast doubts on their own conclusions. The ‘L2’ was artificial (the learners knew this – they were told that they were dealing with an extra-terrestrial language); learners were taught just three explicit rules, via translation. Exposure to the ‘L2’ was extremely brief. It seems very likely that learners simply treated the whole thing as a problem-solving exercise, rather

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.2 Alternative accounts 39

than as a genuine language-learning situation, in which case the ECP is simply irrelevant, since it is a constraint on natural language systems, not on systems arrived at by other means.7 Hence, the absence of ECP effects in Kellerman et al.’s study is hardly surprising, leaving Kanno’s results unchallenged. (Indeed, even if Kellerman et al.’s results had supported Kanno’s, one would still have to question their validity.)

Potentially more problematic for Kanno’s claim that L2 learners observe the ECP is an additional experiment by Kellerman and Yoshioka (1999). This experi- ment involved genuine L2 acquisition, testing Dutch-speaking learners of Japanese. The task was a grammaticality-judgment task. Whereas each of Kanno’s test sen- tences contained one dropped case particle, either nominative or accusative (as well as a case-marked NP in those sentences which contained two overt arguments), Kellerman and Yoshioka included additional sentence types, to provide a greater variety of overt and null case-marking combinations. Their reason for doing so was the incorrect and unmotivated assumption that the ECP predicts a hierarchy of acceptability, as follows (where + means overt case marking and −means no case marking):

(8) +nom +ac c ≥ +n o m – ac c > –nom +ac c > – n o m – ac c Kellerman and Yoshioka’s results show no evidence for this hierarchy; hence, they conclude that the ECP does not constrain the interlanguage grammar of these learners. In fact, as pointed out by Kanno (1999), no such hierarchy is predicted by the ECP, which simply says that nominative case particles in subject position should not be deleted. The ECP is neutral as to whether accusative case particle deletion will take place at all (the ECP certainly does not require this) and whether overt accusative marking is preferable to null. As Kanno points out, if one holds the accusative case markers constant (comparing +nom+ac c to –nom +ac c or +nom–ac c to –nom–ac c), Kellerman and Yoshioka’s subjects do show a significant difference in their acceptance of sentences with and without nominative case particles, in favour of the former, as predicted by the ECP account.

2.2 The logical problem of L2 revisited: alternative accounts

So far, we have considered three different linguistic properties assumed to stem from UG: (i) differences between overt and null pronouns (attributed to the Overt Pronoun Constraint); (ii) the distinction between process and result nominals (attributed, indirectly, to restrictions on argument structure); (iii) subject–object asymmetries in case-particle deletion (a consequence of the ECP). In each case, the claim has been that there is a learnability problem, the L2 input alone (whether

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

40 Principles of Universal Grammar in L2 acquisition

naturalistic input or classroom input including instruction) being insufficient to allow the learner to arrive at the relevant distinctions. The assumption has also been that the relevant properties could not have been arrived at on the basis of the L1: in the case of the studies involving the Overt Pronoun Constraint, learners were chosen whose L1s do not permit null subjects, with overt pronouns behaving quite differently in the two languages; in the case of process and result nominals, the L1 and L2 differed in terms of how arguments of these nominals are realized; in the case of case particle deletion, the L1 was a language without case particles at all. In all the studies described above, L2 learners showed behaviour which revealed unconscious knowledge of subtle properties of the L2, consistent with the claim that their grammars are UG-constrained.

In this section, we consider attempts to provide alternative accounts of L2 learners’ successes in acquiring the kinds of subtle distinctions discussed above. Certain researchers deny that there is an underdetermination problem in such cases; hence, they claim, there is no need to assume that interlanguage grammars are UG-constrained. Two different lines of argument have been advanced. The first questions the poverty-of-the-stimulus claim, by trying to show that the L2 input is in fact sufficient to allow the relevant contrasts to be induced without recourse to principles of UG. The second accepts that the L2 input underdetermines the unconscious knowledge that L2 learners attain but maintains that this knowledge derives from the L1 grammar rather than UG.

2.2.1 L2 input

Kellerman and colleagues take the position that properties of the L2 in- put are sufficient to explain L2 learners’ differential treatment of nominative and accusative case drop. While they failed to find a subject–object asymmetry in case- particle deletion in their three studies, they accept the validity of Kanno’s (1996) finding of a robust difference between acceptances of deleted accusative versus nominative particles.

Kellerman and Yoshioka (1999) and Kellerman et al. (1999) account for the differences in their results and Kanno’s in terms of differences in the L2 input. They suggest that case-particle drop would be relatively easy to formulate as a pedagogical rule and would be easy to learn on the basis of instruction. However, as neither their subjects nor Kanno’s had received such instruction, this is a moot point. They also suggest that naturalistic input may show a statistical bias in favour of dropped object particles and that L2 learners in Hawaii (where Kanno’s subjects were tested) are likely to have been exposed to such input. Their performance, then, could be explained in terms of properties of the L2 input and, in their view, in terms of the input alone. But this begs the original question: while input with

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.2 Alternative accounts 41

dropped accusative case particles is certainly necessary to motivate case drop in the first place, this does not mean that it is sufficient to account for the knowledge that L2 learners acquire. The input will exemplify object NPs with and without case markers, subject topics with and without topic markers, as well as nominative markers missing on subjects of a subclass of verbs. The potential to overgeneralize, thus dropping nominative case markers on subjects in general, remains even if there is a preponderance of accusative particle drop in the input. It is important to understand that the UG approach does not deny the importance of input. But the claim is that input alone is not enough.

However, suppose, for the sake of the argument, that one could somehow demon- strate that a particular phenomenon (in this case, dropping of accusative particles) could successfully be acquired on the basis of statistical frequency in the in- put. While this might show that case-particle drop does not constitute a genuine learnability problem, it would not dispose of the learnability problem in general. The fact that there may not be a logical problem of L2 acquisition with respect to one phenomenon does not mean that there is never a logical problem of L2 acquisition.

2.2.2 The L1 grammar as the source of knowledge of UG principles

According to another approach which argues against the need to invoke UG, the complexity of the mental representations achieved by L2 learners is ac- knowledged, as well as the fact that input alone cannot account for such complexity. Instead, the unconscious knowledge attained by L2 learners is claimed to derive from the L1 grammar (Bley-Vroman 1990; Clahsen and Muysken 1989; Schachter 1989, 1990). For example, there are a number of subject–object asymmetries in English which do not involve case particles but which do implicate the ECP. The L2 learner of Japanese, then, may have been able to arrive at the relevant properties of case drop in Japanese on the basis of very different properties of English which are subject to the same constraint. Similarly, in the case of process and result nominals, there are a number of subtle differences between them in English (e.g. Grimshaw 1990), which might somehow explain the results obtained by Dekydtspotter et al. (1997) for L2 French. Thus, it might be argued that knowl- edge of the properties described in this chapter (as well as other principles discussed in the literature) stems from the mother tongue and, hence, fails to provide evi- dence that interlanguage grammars are UG-constrained independently of the L1 grammar.

Indeed, for many principles of UG, it appears that the L1 can never be completely ruled out as a source of the L2 learner’s unconscious knowledge. Since the L1 is a natural language and since many UG principles manifest themselves in the L1 in

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

42 Principles of Universal Grammar in L2 acquisition

some form or other, it will often be hard or impossible to disentangle the two (Hale 1996).8 Nevertheless, the results described so far suggest that L2 learners are able to apply UG principles to totally new domains, including data that do not occur in the L1. As we have seen, L2 learners observe UG constraints even when they apply to very different phenomena in the L2 (restrictions on arguments of process nominals; ECP and case-particle drop), ruling out surface transfer as an explanation. Thus, it appears that there is considerable flexibility in the system; UG constraints can be applied to new data, and to situations that are entirely different from what pertains in the L1. It is not clear that this would be the case if UG principles were only somehow being ‘reconstructed’ on the basis of how they operate in the L1. Furthermore, as will be discussed in chapter 4, where parameters of UG are concerned, it is a relatively straightforward matter to distinguish between L1 parameter settings and other settings, hence to discover whether or not L2 learners are restricted to options realized in the L1.

In any case, an approach which sees the issue as an either/or matter (UG or the L1) is misconceived. This is a false dichotomy. It is inappropriate to contrast UG with the L1 as the source of UG-like knowledge; rather, both appear to be involved. This issue will be considered in greater detail in chapters 3 and 4.

2.3 Problems for the UG claim: wild interlanguage grammars

So far, results from the studies discussed in this chapter are consistent with the claim that UG constrains interlanguage grammars. L2 learners demonstrate unconscious knowledge of subtle contrasts which are by no means transparent in the L2 input and which are not realized in any obvious way in the L1. This unconscious knowledge is unexpected if interlanguage grammars are not UG- constrained. While the source of this knowledge may, at least in some cases, be UG as instantiated in the L1, it is clear that the constraints are applied to completely new data and to phenomena that do not exist in the L1. Learner grammars, then, conform with UG.

There is another way of exploring whether or not UG constrains interlanguage grammars. UG defines what a grammar is, determining what mental represen- tations can and cannot be like. Natural language grammars fall within a range sanctioned by UG. L1 acquirers are limited by the hypothesis space provided by UG, which reduces the number of logical possibilities that have to be entertained in order to arrive at a grammar for the language being acquired. Developing L1 grammars, in other words, are ‘possible’ grammars (White 1982). At least with respect to the properties we have considered so far, the same appears to be true of interlanguage grammars, in the sense that they fall within the range sanctioned

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.3 Problems for the UG claim: wild grammars 43

by UG. Interlanguage grammars, then, exhibit characteristics typical of natural language.

Grammars that do not conform to principles of UG have been variously described as impossible (White 1982, 1988), rogue (Thomas 1991a), illicit (Hamilton 1998) or wild (Goodluck 1991; Klein 1995a); the latter term will be adopted here. If interlanguage grammars are UG-constrained, wild grammars are predicted not to occur in L2 acquisition. In other words, interlanguage grammars should be restricted to properties found in the L1 and/or the L2, and/or natural languages in general. If it can be shown that interlanguage grammars do not conform to properties of natural language, this would suggest that the operation of UG is in some way impaired.

Recently, there have been proposals that interlanguage grammars are in fact sometimes wild. Two phenomena will be considered in this context: reflexives (Christie and Lantolf 1998) and null prepositions (Klein 1993b, 1995a). In both cases, it has been suggested that the interlanguage grammar shows a cluster of properties that is illicit, hence that the grammar is not sanctioned by UG.

2.3.1 Reflexive binding

There has been considerable research which investigates whether reflex- ives in interlanguage grammars are constrained by Principle A of the Binding Theory. Binding Theory places constraints on coreference between various kinds of NPs, Principle A being concerned with properties of anaphors, such as reflexives (himself , herself , etc.) (Chomsky 1981a). According to Principle A, an anaphor must be bound in its governing category; effectively, it must take an antecedent in a local domain (usually the same clause), as shown in (9). In (9a), coreference is possible between the reflexive, herself , and the subject, Mary, because they are within the same clause. In (9b), on the other hand, herself cannot refer to the sub- ject of the higher clause, Mary, but it can refer to the subject of the lower clause, namely Susan.

(9) a. Maryi blamed herselfi b. Maryi thought that Susanj blamed herself*i/j

As is well known, reflexives differ crosslinguistically as to the domain in which they must be bound. While English reflexives require their antecedents to be within the same clause (i.e. local), there are many languages, such as Japanese, which permit the antecedent to be in a different clause. The Japanese sentence in (10) is ambiguous, with zibun (‘self’) able to have either the local subject, Susan, or the main-clause subject, Mary, as its antecedent. When the antecedent is in a different clause from the reflexive, this is referred to as non-local or long-distance binding.

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

44 Principles of Universal Grammar in L2 acquisition

(10) Maryi ga Susanj ga zibuni/j o semeta to omotta. Mary nom Susan nom self ac c blamed that thought ‘Mary thought that Susan blamed herself.’

In addition to domain (locality), orientation is also important, that is, whether the antecedent is a subject or not. Again, languages differ in this respect. An English reflexive can take a subject or a non-subject as its antecedent, whereas Japanese reflexives are restricted to subject antecedents. An English sentence like (11a) is potentially ambiguous, with either the subject or the object available as the antecedent, though context will usually favour one interpretation over the other. The Japanese sentence in (11b), on the other hand, only has one interpretation, where the subject, kanja-ga, is the antecedent.

(11) a. The patienti asked the nursej about herselfi/j b. Kanjai ga kangofuj ni zibuni/∗j no koto nitsuite tazuneta.

patient nom nurse dat self g e n matter about asked ‘The patient asked the nurse about herself.’

Much of the earlier L2 research on reflexives was conducted within the frame- work of Manzini and Wexler (1987) and Wexler and Manzini (1987). They pro- posed two parameters to handle crosslinguistic differences in how Principle A operates: the Governing Category Parameter, which dealt with domain, and the Proper Antecedent Parameter, dealing with orientation. Explicit discussion of the possibility of wild interlanguage grammars first arose in this context (Eckman 1994; Thomas 1991a). However, most of the L2 research conducted on the Gov- erning Category Parameter and the Proper Antecedent Parameter was not directed at the issue of wild grammars but rather to the question of whether L2 learners can reset parameters and whether or not they observe the Subset Principle (Finer 1991; Finer and Broselow 1986; Hirakawa 1990; Thomas 1991b). (See White (1989) for review.)

According to subsequent linguistic analyses, parameterization of this kind is not involved. Instead, two different types of anaphors are assumed, with differing properties. Each anaphor type has a cluster of properties associated with it, as shown in (12).

(12) a. Xmax (phrasal) reflexives. These are morphologically complex, require local antecedents and allow subjects and (in some languages) non-subjects as an- tecedents, e.g. himself , herself in English; taziji (‘himself’) in Chinese; kare- zisin, kanojo-zisin (‘himself’, ‘herself’) in Japanese.

b. X0 (head) reflexives. These are monomorphemic, allow non-local antecedents (as well as local), and require the antecedent to be a subject, e.g. ziji in Chinese; zibun in Japanese.

While accounts differ in their details, they share the central insight that both phrasal and head anaphors undergo LF movement (Cole, Hermon and Sung 1990;

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.3 Problems for the UG claim: wild grammars 45

Katada 1991; Pica 1987; Reinhart and Reuland 1993, amongst others). Differences in domain and orientation fall out from the categorial differences between the reflexives: X0 reflexives raise by head movement to Infl (itself a head). In complex sentences, they can move out of the clause in which they originate, raising from one Infl to another. In consequence, they can be interpreted with a long-distance antecedent in a higher clause; this antecedent must be a subject because only a subject will c-command the reflexive in Infl. Xmax reflexives, on the other hand, are maximal projections which can adjoin only to the nearest maximal projection, namely the VP in which they originate. There they remain in the binding domain of either a local subject or a local object. In general, then, current accounts of reflexives agree on the following properties:

(13) a. Long-distance anaphors must be subject-oriented b. Anaphors which allow non-subject antecedents must be local

It is in this context that there has been extensive consideration of what would con- stitute a wild grammar (Christie and Lantolf 1998; Hamilton 1998; Thomas 1995; White, Hirakawa and Kawasaki 1996; Yuan 1998). Christie and Lantolf (1998) hypothesize that a UG-constrained grammar will show a correlation between do- main and orientation: if the interlanguage grammar allows long-distance reflex- ives, then reflexives will be subject-oriented; if the interlanguage grammar has local reflexives, then non-subject antecedents will be permitted. If interlanguage grammars are not UG-constrained, on the other hand, there will be a breakdown in these relationships.

Christie and Lantolf (1998) investigate the acquisition of reflexives in L2 Chi- nese (which has a long-distance reflexive, ziji) and L2 English.9 (See box 2.5.) In the case of L2 Chinese, the learners are speakers of English, a language which only allows local reflexives. In the case of L2 English, learners are native speakers of Chinese or Korean, both of which have long-distance reflexives. (These languages also have phrasal reflexives which behave like English himself/herself .)

In order to understand what might be going on in the case of reflexives, it is necessary to establish how L2 learners interpret certain sentences, whether, for example, an English sentence such as (9b), repeated here as (14), has only one possible interpretation in the interlanguage grammar (with Susan as the antecedent of herself ), as is the case in English, or whether it is ambiguous (with either Mary or Susan as the antecedent of herself ), as it would be in Japanese or Chinese.

(14) Mary thought that Susan blamed herself.

Christie and Lantolf developed a truth-value-judgment task for this purpose. In a truth-value-judgment task, a particular sentence is paired with a particular context which is provided by a story, a picture, or a scenario acted out in front of the subjects (on video, for example). Subjects have to indicate whether the sentence

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

46 Principles of Universal Grammar in L2 acquisition

Box 2.5 Reflexives (Christie and Lantolf 1998)

Languages: L1 = English, L2 = Chinese and L1 = Chinese/Korean, L2 = English. Task: Truth-value judgments. Subjects indicate whether a sentence provides a true statement about what is going on in an accompanying picture. Sample stimuli:

Local domain (true) Non-local domain (false)

(Picture showing Grover hitting Grover) Bert says that Grover is hitting himself on the head.

(Picture showing Grover hitting Bert) Bert says that Grover is hitting himself on the head.

Subject-oriented (true) Object-oriented (true)

(Picture showing a book about Big Bird) Big Bird is giving Oscar a book about himself.

(Picture showing a book about Oscar) Big Bird is giving Oscar a book about himself.

is true in the context provided. In Christie and Lantolf’s tasks, pictures were used (see box 2.5 for example stimuli). The advantage of such tasks is that subjects are not being asked to make explicit grammaticality judgments as to the form of the sentences. Indeed, all sentences in a truth-value-judgment task are generally grammatical under some interpretation. Rather, subjects are ostensibly being asked something about the meaning of the sentence. Nevertheless, their judgments reveal something about the form of the grammar, in this case the range of interpretations the grammar permits for reflexives.

As Christie and Lantolf recognize, the crucial issue is the grammar of the indi- vidual; in other words, for any particular individual, is the grammar UG-consistent or is it wild? (For related discussion, see also Eckman 1994; Thomas 1995; White et al. 1996.) Results were analysed using cluster analysis, to see whether there is a correlation between domain and orientation in individual grammars. Christie and Lantolf found no evidence for such clustering in the case of either L2. Somewhat surprisingly (given their assumptions), the control groups (Chinese and English native speakers) also failed to show evidence of clustering. It is hard to argue that L2 learners do (or do not) have wild grammars if even native speakers behave in a way that suggests their grammars are wild, a problem that Christie and Lantolf recognize.

Furthermore, Christie and Lantolf’s assumption about clustering of domain and orientation is questionable. As Thomas (1995, 1998) points out, the properties de- scribed in (13) only carry a one-way implication. That is, while long-distance

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.3 Problems for the UG claim: wild grammars 47

anaphors must be subject-oriented, it is not the case that all subject-oriented anaphors must allow long-distance antecedents. There are languages with subject- oriented anaphors which are bound only in a local domain, for example, Japanese zibun-zisin (Katada 1991) and the French reflexive clitic se (Pica 1987). There is nothing in the LF movement account to prohibit this. Similarly, while a non- subject antecedent implies a locally bound reflexive, there is no requirement that all locally bound reflexives permit object antecedents.

As Thomas discusses, Christie and Lantolf incorrectly presuppose a two-way implication between domain and orientation. They assume that a local binding domain necessarily implies the possibility of non-subject antecedents, leading them to the conclusion that L2 learners had arrived at illicit grammars. However, this is not in fact the case. Consider, for example, L2 learners of English who permit only local antecedents for reflexives like himself and herself , thus having the domain right, but also reject object antecedents, thus having orientation wrong. Such a grammar would fail to show the clustering relationship that Christie and Lantolf expected. Nevertheless, the grammar would in fact be perfectly legitimate, although not the appropriate grammar for English – this is precisely how Japanese zibun-zisin behaves, for example.

In addition, as Thomas points out, numerous investigations of the acquisition of English reflexives have found that learners and native speakers have a strong preference for subject antecedents even where object antecedents are possible. For example, given a sentence like John showed Bill a picture of himself , native speakers are more likely to interpret John (the subject) as the antecedent of the reflexive. In certain contexts, L2 learners and native speakers may reject interpre- tations which their grammars in fact permit. This makes the results of the cluster analysis even harder to interpret, since such preferences may conceal the full extent of grammatical knowledge.10

Indeed, Thomas (1995) suggests that languages which do not have long-distance reflexives, such as English, should not be used for investigating whether do- main and orientation are correlated. Instead, one should concentrate on L2s with long-distance reflexives, such as Chinese or Japanese. But here, too, determin- ing whether learners arrive at wild grammars is not as straightforward as Christie and Lantolf imply. If learners accept long-distance antecedents for reflexives, thus getting the domain right, and also accept objects as antecedents, this is not nec- essarily indicative of a wild grammar. It depends crucially on whether the object antecedent is found within a local domain or whether it is a long-distance object. In other words, there is an important distinction between (15a), where the object antecedent is in a local domain, and (15b), a biclausal sentence, where the reflex- ive is construed as having a non-subject antecedent outside the clause in which it appears.

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

48 Principles of Universal Grammar in L2 acquisition

(15) a. The nurse asked the patienti about herselfi b. *Black-san ga Whitei-san ni [Grey-san ga zibuni o

Black-Mr nom White-Mr dat Grey-Mr nom self ac c mita to] iimashita saw that said ‘Mr Black said to Mr White that Mr Grey saw himself.’

Long-distance reflexives in some languages (Icelandic sig or Serbo-Croatian sebe) can take a local subject or object as antecedent; non-local antecedents must always be subjects, however. In other words, equivalents of (15a) are grammatical, while interpretations like (15b) are never possible. Evidence for a wild grammar, then, would be provided if non-local non-subjects were considered acceptable antecedents for a reflexive, in other words if sentences equivalent to (15b) were permitted. In this case, there would be a violation of the requirement that the antecedent of the reflexive must c-command the reflexive at LF.

In their cluster analysis, Christie and Lantolf compare the possibility of local object antecedents (such as Chinese equivalents of (15a)) with long-distance subject antecedents (such as Chinese equivalents of (10)). Finding that both are accepted, they conclude, incorrectly, that the grammar is wild. (Eckman (1994) does the same, and draws the same conclusion.) In fact, Christie and Lantolf only had one test item relevant to the issue of long-distance objects, i.e. only one item similar to (15b). Some learners accepted the long-distance non-subject antecedent in this case, which makes it appear that their interlanguage grammars might in- deed not be UG-constrained. However, given the fact that only one sentence was involved, there is considerable doubt as to the validity and generalizability of this result.11

Thomas (1995) overcomes these shortcomings by investigating the L2 acqui- sition of the Japanese reflexive zibun, by learners at low and high levels of profi- ciency. Again, the task was a truth-value-judgment task (see box 2.6). Crucially, stimuli included four biclausal sentences with long-distance objects as potential antecedents, similar to (15b).

Table 2.6.1 presents group results, where it can be seen that long-distance object antecedents are indeed accepted by low-proficiency learners, to about the same extent that local object antecedents are accepted. The results on local objects, though inappropriate for Japanese, do not provide evidence of a wild grammar but the acceptance of long-distance object antecedents is problematic. Thomas provides a further analysis in terms of individual performance, since it is properties of the individual grammar that are at issue. Using a criterion of three or four responses (out of four) to determine individual consistency,12 Thomas concentrates on the twenty-three learners who consistently accept non-local antecedents for zibun, excluding from consideration those who treat zibun only as a locally bound

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.3 Problems for the UG claim: wild grammars 49

Box 2.6 Reflexives (Thomas 1995)

Languages: L1 = English, L2 = Japanese. Task: Truth-value judgments. Contexts provided by written scenarios (3–5 sen- tences), illustrated by pictures. Each scenario is followed by a statement. Subjects indicate whether the statement is true. Sample stimuli:

Biclausal: LD subject (true) Biclausal: LD object (false)

Scenario: B hits A Scenario: A likes B’s book A wa B ga zibun o butta to iimasita (A said that B hit self)

C wa B ni A ga zibun no hon ga suki da to iimasita (C told B that A likes self’s book)

Monoclausal: local subject (true) Monoclausal: local object (false)

Scenario: A describes A’s problems to B Scenario: A asks B questions about B A wa B ni zibun no mondai ni tuite hanasimasita

A wa B ni zibun no koto ni tuite iroiro na situmon o simasita

(A spoke with B about self’s problems) (A asked B various questions concerning self )

Results:

Table 2.6.1 Responses of true (in %)

LD subject *LD object Local subject *Local object (# = 4) (# = 4) (# = 4) (# = 4)

L2 groups Low (n = 34) 55 54 85 47 Advanced (n = 24) 57 14 96 17

Native speakers (n = 34) 89 18 93 13

reflexive. The question, then, is whether these learners have unconscious knowl- edge of the requirement for non-local antecedents to be subjects. While the majority (70%) showed appropriate orientation, 30% of the learners (six in the low- proficiency group and one advanced) consistently allowed long-distance object an- tecedents (accepting interpretations like (15b)), a property not permitted in natural language.

Other researchers have also reported acceptance of binding to long-distance objects by L2 learners and controls, though to a much lesser extent. White et al. (1996) report for L2 Japanese that one learner (out of thirteen) consistently

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

50 Principles of Universal Grammar in L2 acquisition

accepted long-distance object antecedents, as did one native speaker (out of ten). Thomas (1995) also found that 10% of her Japanese native speaker controls showed the same ‘wild’ behaviour. For L2 Chinese, Yuan (1998) found that three learn- ers out of fifty consistently accepted long-distance objects, while no controls did so.

It is not unreasonable to assume a certain amount of ‘noise’ in the data; perfor- mance at 100% accuracy is unusual in any experimental attempts to get at linguistic competence. Investigation of binding principles is no exception. However, while performance factors may provide a reasonable explanation of the occasional native speaker or L2 learner who deviates from expected behaviour, it is unlikely that noisy data can provide an adequate account of the performance of Thomas’s sub- jects, where 30% of those who have a long-distance reflexive in L2 Japanese also allow its antecedent to be a non-local object. Thomas herself assumes that there is a competence issue here; she suggests that these subjects have misanalysed zibun as a pronoun rather than a reflexive.13 As such, it can take any non-local antecedent. In other words, these L2 learners adopt an alternative analysis, which happens to be inappropriate for the L2, such that their grammars nevertheless fall within the bounds of UG.

This solution in turn raises the issue of falsifiability. By changing the analysis of zibun from anaphor to pronoun, the prospect of a wild interlanguage grammar has been avoided. Does this mean that one can always change the analysis in order to avoid such problems? It is important to understand that the alternative proposal itself makes testable predictions. In order to explore further the pos- sibility that zibun is a pronoun in the interlanguage grammar of at least some learners, one would have to include additional sentence types which investigate how learners treat pronouns in L2 Japanese. Hyams and Sigurjonsdottir (1990) raised a similar possibility for L1 acquisition of Icelandic, namely that children treat the reflexive sig as a pronoun; they rejected this possibility precisely be- cause there turned out to be differences in the way children treated other pronouns and sig.

Another possibility is suggested by Hamilton (1998), who argues that English- speaking learners of Japanese may be treating reflexives as logophoric, i.e. anaphors which are exempt from binding principles and which can be bound non-locally within the discourse (e.g. A picture of myself would be nice on that wall). (See Reinhart and Reuland (1993) for a recent treatment of logophoricity, including constraints on when an anaphor can be logophoric.) Again, this would mean that these interlanguage grammars are in fact licit. And, again, this would have to be demonstrated independently with relevant stimuli, as Hamilton (1998) tries to do for L2 English.

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.3 Problems for the UG claim: wild grammars 51

To sum up, as far as reflexive binding is concerned, it has been suggested by Christie and Lantolf (1998) that L2 learners arrive at grammars where domain and orientation fail to cluster, with a result that their grammars may be wild rather than UG-constrained. Thomas (1995) notes, however, that this claim rests on a mis- conception of what the permissible and impermissible grammars are. Controlling for this by examining only the issue of whether a long-distance reflexive permits a non-local non-subject antecedent (which is illicit), Thomas finds that the major- ity of learners have UG-constrained grammars of anaphora. However, a minority appears to have a wild grammar. Thomas (1995) and Hamilton (1998) offer alter- native analyses of this behaviour which suggest that the interlanguage is, after all, UG-constrained, though not the appropriate grammar for Japanese.

2.3.2 Null prep

Another example of a purported wild interlanguage grammar is provided by Klein (1993b, 1995a). Klein investigates a phenomenon which she calls null prep, whereby L2 learners of English omit prepositions in obligatory contexts. English has a number of prepositional verbs (verbs taking PP complements), where the preposition cannot be deleted, as shown in (16a, b). The preposition also appears in questions, as in (16c, d), as well as relatives (16e, f), whether these involve preposition stranding (16c, e) or the more formal pied-piping (16d, f). It is not possible to omit the preposition in such contexts, as shown in (17). It is omission of the kind illustrated in (17) which Klein terms null prep.

(16) a. The student is worrying about the exam. b. *The student is worrying the exam. c. Which exam is the student worrying about? d. About which exam is the student worrying? e. Here’s the exam that the student is worrying about. f. Here’s the exam about which the student is worrying.

(17) a. *Which exam is the student worrying? b. *Here’s the exam that the student is worrying.

Although null prep is not possible in English, it is found in certain languages, such as Brazilian Portuguese, several dialects of Spanish and French, as well as Haitian Creole. However, it is relatively rare and, according to Klein’s (1993b) review of a number of languages that exhibit null prep, it is found only in relative clauses (equivalents of (17b)), being prohibited in wh-questions (equivalents of (17a)). The examples in (18), from Haitian Creole (Klein 1995a) illustrate this point:

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

52 Principles of Universal Grammar in L2 acquisition

(18) a. Twa zanmi-yo ap pale de sinema sa a. three friend-p are talking about movie this t o p ‘The three friends are talking about this movie.’

b. Men sinema que twa zanmi-yo ap pale a. here movie that three friend-p are talking t o p ‘Here is the movie that the three friends are talking (about)’

c. *Ki sinema twa zanmi-yo ap pale a? what movie three friend-p are talking t o p? What movie are the three friends talking (about)?

Klein observes that relative clauses in languages allowing null prep show char- acteristics which suggest that they are not derived by syntactic movement: (i) in lieu of a null prep, relative clauses can contain an overt resumptive PP, consisting of a pronoun with a preposition cliticized to it, as shown in (19a) (from a dialect of Greek) – resumptives in general are characteristic of lack of movement (Sells 1984); (ii) relative clauses are introduced by complementizers rather than relative pronouns, as shown in (19b, c), from Brazilian Portuguese – relative pronouns are characteristic of movement.

(19) a. to rafio pou douleve (s’afto) ine mikro. the office that he-works (in-it) it-is small ‘The office that he works (in) is small.’

b. a mô que eu falei the girl that I spoke ‘The girl that I spoke (to)’

c. *a mô quem eu falei the girl who I spoke ‘The girl who I spoke (to)’

Klein analyses null-prep relatives as containing a null resumptive PP, which alternates with an overt resumptive PP.14 Null prep is not permitted in wh-questions or relative clauses derived by movement (as in English), on the other hand, because this would constitute an ECP violation: the null preposition would be unable to properly govern the empty category resulting from wh-movement.

Given such restrictions, if null prep were to be found in interlanguage grammars either in relative clauses derived by movement or in wh-questions, this would constitute evidence of a wild grammar, violating the ECP. Previous research has reported sporadic use of null prep in L2 (Bardovi-Harlig 1987; Mazurkewich 1984a). Klein devised a series of experiments to investigate whether null prep is in fact a robust phenomenon in interlanguage grammars, and whether it occurs illicitly, i.e. in movement contexts.

Klein (1993b, 1995a) tested adult learners of English from a variety of language backgrounds and at different levels of proficiency. Learners were asked to judge the grammaticality of sentences lacking prepositions and to correct them if they

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.3 Problems for the UG claim: wild grammars 53

Box 2.7 Null prep (Klein 1995a)

Languages: L1s = various, L2 = English. Task: Grammaticality judgments and corrections. Sample stimuli (ungrammatical): Declarative: The delivery boy applied a new job last week Relative: This is the job which/that/Ø the delivery boy applied last week Question: Which job did the delivery boy apply last week? Results:

Table 2.7.1 Acceptances of null prep in questions and relatives (in %)

Questions (# = 9) Relatives (# = 9) L2 groups Beginners (n = 55) 69 78

Intermediate (n = 66) 52 57 Advanced (n = 75) 30 35

Native speakers (n = 40) 1 2

considered them to be wrong. (See box 2.7.) In other words, they were expected to insert the missing preposition. As Klein points out, it is crucial to establish learners’ knowledge of the subcategorization properties of the verbs in question. For example, if a learner does not know that the verb worry requires a PP comple- ment introduced by about, absence of the preposition does not constitute a case of the null prep phenomenon but simply indicates a lack of knowledge of the sub- categorization requirements of this particular verb. For this reason, the first step in Klein’s analysis of the results was to concentrate on those declaratives which learners identified as ungrammatical and corrected by the insertion of a preposition. Such corrections indicate that learners are aware that the verbs in question require a PP complement. Where the declarative was correctly identified as requiring a preposition, Klein then looked at whether learners know that the preposition is also obligatory in the corresponding questions and relatives. If questions and/or rela- tives with missing prepositions are accepted while the corresponding declaratives are rejected, this is taken to be evidence of illicit null prep.

Results show extensive acceptance of null prep forms in both wh-questions and relative clauses, even at the highest proficiency level (see table 2.7.1); all groups are significantly different from each other and from the controls. There is a decline in acceptance of null prep with increasing level of proficiency. Klein’s other studies (Klein 1993a, 1995b) suggest that null prep is systematic, occurring across different age groups and L1s (including L1s with and without null prep), in both relatives and wh-questions.

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

54 Principles of Universal Grammar in L2 acquisition

While null prep is permissible in relative clauses in certain languages, Klein argues that it is always illicit in wh-questions. Since the learners she studied ac- cepted null prep in wh-questions, and since her methodology rules out the possibil- ity that this is just a general subcategorization error, the results appear to suggest a wild interlanguage grammar, violating the ECP, even in the case of those of advanced proficiency. In addition, if relative clauses are derived by movement, their grammars are wild in this respect as well, since null prep is not permitted in such cases (compare (18b and c)). However, Dekydtspotter, Sprouse and Ander- sen (1998) have challenged Klein’s assumption that null prep is prohibited in the case of structures involving wh-movement. They show that null prep is found in wh-questions in Yoruba, as well as in popular French. Dekydtspotter et al. pro- pose an alternative account of the null prep phenomenon in terms of preposition incorporation, arguing that it obeys general constraints on incorporation, hence suggesting that a null prep interlanguage grammar is not wild after all. (See Klein (2001) for counter-arguments.)

2.4 Methodological issues

The experiments described in this chapter have used a variety of different methodologies to try to determine the nature of the interlanguage grammar. Tasks have included elicited production (translation), grammaticality and acceptabil- ity judgments, coreference judgments, and truth-value judgments. The judgment tasks fall into two distinct types, some concentrating on the form of sentences (for example, Kanno’s (1996) study of case drop – see section 2.1.3) and some concentrating on interpretation (for instance, the studies by Christie and Lantolf (1998) and Thomas (1995) on reflexives – section 2.3.1). In the case of judgments designed to probe learners’ interpretations of particular sentence types, we have seen that it is crucial to provide some kind of context for the interpretation in ques- tion. Even within the judgments directed at interpretation, some tasks are more metalinguistic than others. For example, Kanno (1997, 1998b) requires subjects to choose explicitly between potential antecedents for pronouns; Dekydtspotter et al. (1997) ask learners whether a sentence ‘sounds possible’ in a given context. In contrast, truth-value-judgment tasks merely require the learner to indicate whether or not a particular sentence is true in a particular context. (However, see White, Bruhn-Garavito, Kawasaki, Pater and Prévost (1997) for discussion of problems associated with such tasks: even with this methodology, the effects of preferences for certain interpretations over others cannot be fully eliminated.)

Another welcome trend is the move towards analysing data in terms of the per- formance of individual subjects. Although group results can be quite informative,

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

2.4 Methodological issues 55

they can also be misleading, concealing properties of individual grammars. For example, if a given group of L2 learners shows a 50% acceptance rate for some structure, this could be due to half the subjects accepting all sentences and half rejecting all, or it could be the result of all subjects accepting half of the sentences. Since the claim that the interlanguage grammar is (or is not) UG-constrained is a claim about individual linguistic competence, it is crucial to determine what is going on at the individual level.

A further methodological advance concerns the use of control groups. It has been fairly standard for many years to include native speakers of the L2 as controls. Research described in this chapter has also included a more recent development, namely the use of native speakers of the L1 (e.g. Dekydtspotter et al. 1997 – see section 2.1.2; Kanno 1998b – see section 2.1.1). This allows the experimenter to determine whether or not interlanguage performance could be accounted for in terms of properties of the L1 grammar.

However, the use of native-speaker control groups, whether they are speakers of the L1 or L2, raises the issue of the so-called comparative fallacy (Bley-Vroman 1983). Bley-Vroman remarked that ‘the learner’s system is worthy of study in its own right, not just as a degenerate form of the target system’ (1983: 4). A number of researchers have emphasized the need to consider interlanguage grammars in their own right with respect to principles and parameters of UG (e.g. du Plessis, Solin, Travis and White 1987; Finer and Broselow 1986; Martohardjono and Gair 1993; Schwartz and Sprouse 1994; White 1992c; Lakshmanan and Selinker 2001). As Birdsong points out with respect to grammaticality-judgment data: ‘the relevant data are learners’ judgments – not their similarity to or deviance from natives’ judgments’ (1989: 119). This focus on the interlanguage grammar remains the current perspective: much of the research described in this chapter (as well as later in the book) is committed to this position. That is, the crucial question is whether or not interlanguage grammars are UG-constrained, rather than whether or not they are native-like.

Nevertheless, avoiding the comparative fallacy does not require the experimenter to exclude native-speaker controls altogether. First of all, control groups are nec- essary simply to ensure: (i) that the tasks devised by the experimenter in fact are successful in testing what they are supposed to test; and (ii) that the facts in question are indeed as the experimenter supposes them to be (or as claimed in the theoretical linguistics literature). For example, in Christie and Lantolf’s study (section 2.3.1), the performance of the native speakers was unexpected, given the researchers’ assumptions about what properties of reflexives should cluster. Secondly, there are legitimate reasons for asking whether the learner in fact shows unconscious knowledge of principles and parameters relevant to the L2. What is problematic is when certain conclusions are drawn on the basis of failure to perform like native

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

56 Principles of Universal Grammar in L2 acquisition

speakers. Failure to acquire L2 properties may nevertheless involve acquiring prop- erties different from the L1, properties of other natural languages, properties that are underdetermined by the L2 input. Such failure does not necessarily imply lack of UG, as we have seen.

A related problem is how to interpret significant differences between native speaker and L2 learner performance on some task. Consider, for example, the re- sults of the study by Dekydtspotter et al. (1997) on process and result nominals (section 2.1.2.). Their beginner and intermediate subjects showed a significantly higher acceptance rate of multiple de-phrases with process nominals (ungrammat- ical in French). Thus, it might be claimed that their grammars are not constrained by the relevant principles governing argument structure. It has been suggested in this chapter that a more appropriate comparison is to look at whether certain sentence types are treated significantly differently from other sentence types by the same group of learners (c.f. Grimshaw and Rosen 1990). If such distinctions are found, this suggests that the interlanguage grammar represents the relevant distinction (whatever it may be), even if the degree to which they observe it differs from native speakers. This was in fact the case in Dekydtspotter et al.’s study: all groups distinguished between process and result nominals in terms of acceptability of multiple de-phrases.

2.5 Conclusion

In conclusion, this chapter has considered evidence that interlanguage grammars are constrained by principles of UG. While earlier research on UG and L2 acquisition concentrated largely (though certainly not exclusively) on L2 English, many other L2s are now being investigated, including, in the experiments described here, L2 Spanish, Japanese, French and Chinese. Results from several experiments suggest that learners of a variety of L2s demonstrate unconscious knowledge of subtle distinctions that are unlikely to have come from the L2 input (including instruction) or from the L1, consistent with the claim that principles of UG constrain interlanguage grammars. The claim that interlanguage grammars are sometimes wild (hence, not falling within the bounds laid down by UG) has also been considered. We have seen that analyses adopted by L2 learners may in fact be true of natural language, even if they happen not to be appropriate for the L1 or L2 of the learners in question. Of course, the reason why L2 learners should arrive at such alternative analyses still requires explanation.

In the next chapter, a different issue is discussed, namely the nature of the initial state in L2 acquisition. We will consider the extent to which the L1 grammar determines properties of the early interlanguage grammar, and whether or not UG constitutes the initial state, as in L1 acquisition (see chapter 1).

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use

Suggestions for additional reading 57

Topics for discussion

� Is it ever possible to eliminate the L1 as a source of knowledge of princi- ples of UG or is this a non-issue? According to Dekydtspotter et al. (1997), if one adopts the Minimalist Program whereby computational principles are universally the same and invariant, it no longer makes sense to distin- guish between direct access to UG or indirect access via the L1.

� According to Schwartz and Sprouse (2000a, b), too much current L2 research uses linguistic theory to provide relatively sophisticated and detailed analyses of interlanguage data, without considering the logical problem of L2 acquisition. They suggest that such research does not help us to understand the nature of L2 acquisition or interlanguage compe- tence. In contrast, Hawkins (2001b) argues that many researchers are overly preoccupied with the logical problem of L2 acquisition. Instead, he suggests, a better way to reach an understanding of L2 acquisition is to focus on differences between native speaker and L2 learner grammars (described in terms of current linguistic constructs). Are these positions incompatible?

� In this chapter we have seen several examples where problematic interlan- guage data are reanalysed in terms of some other theory. To what extent does this render the claim that UG constrains interlanguage grammars unfalsifiable?

� Linguistic theory is constantly changing and undergoing development; proposals as to the precise nature of UG have changed considerably over the years. What are the implications for L2 acquisition research, and particularly for theories that assume a role for UG? (See Schwartz and Sprouse (2000b) and White (1995a) for discussion.)

� Davies and Kaplan (1998) and Lantolf (1990) advocate group grammat- icality judgments, where learners (in pairs or groups) discuss test items together, in order to arrive at decisions about their grammaticality. Why is this approach problematic?

Suggestions for additional reading

� For more detailed discussion of the logical problem of L2 acquisition, see Bley-Vroman (1990), White (1989: chapter 2) and Schwartz and Sprouse (2000a, b).

EBSCOhost - printed on 4/6/2020 6:15 AM via UNIVERSITY OF ESSEX. All use subject to https://www.ebsco.com/terms-of-use