Topic-drop versus pro-drop: null subjects and pronominal...

Topic-drop versus pro-drop: nullsubjects and pronominal subjects inthe Spanish L2 of Chinese, English,French, German and JapanesespeakersJuana M. Liceras University of Ottawa andLourdes Díaz Univesitat Pompeu Fabra

Recent developments within the so-called Principles and Parameters modelof acquisition argue for a clear-cut separation of Universal Grammar (UG)principles from parametric options and locate all parameters withinfunctional categories (Borer, 1984; Lebeaux, 1988; Chomsky, 1991). This hasled Tsimpli and Roussou (1991) to propose that adult L2 (second language)learners have access to UG principles but do not reset the parameters ofthe L2, which amounts to saying that null subjects in the adult Spanish L2may or may not have the same status as native Spanish null subjects,depending on the speakers’ L1 (first language) and the UG principles atstake. In the case of L1 acquisition, Rizzi (1994) and Hyams (1994) providea competence account of null subjects in early child English which relatethem to adult English Diary Drop and German-style topic-drop rather thanto Spanish-style pro-drop. They specifically argue that these missing subjectsare restricted to the first position of non-wh root clauses and that fixing thenull subject parameter will consist of incorporating the ROOT=CP principleinto this grammar. In this paper, we analyse the Spanish L2 oral spontaneousdata produced by adult L1 speakers of pro-drop and topic-drop languagesin an attempt to provide a competence account of null subjects in adult non-native Spanish. Our data show that, unlike early English grammars, all theSpanish non-native grammars contain null subjects both in matrix andsubordinate clauses, and that this is the case at the early and advancedstages. It also shows that many non-native pronominal subjects do not havethe same value as native Spanish subjects and that subject pronouns areused for identification purposes. It is suggested that these data provideevidence for a model of L2 acquisition where adult non-native grammarconstruction resorts to a default licensing procedure which allows nullpronouns provided they can be identified.

© Arnold 1999 0267-6583(99)SR139OA

Address for correspondence: Juana Liceras, Department of Modern Languages andLiterature, University of Ottawa, 70 Laurier St. East, Ottawa, Ontario, K1N 6N5, Canada;e-mail: [email protected]

Second Language Research 15,1 (1999); pp. 1–40

2 Topic-drop versus pro-drop

I Introduction

The purpose of this paper is to incorporate Rizzi’s (1994) andHyams’ (1994) proposals concerning the nature of null subjects inchild language in a model of L2 grammar which: (1) allows accessto UG principles; (2) provides re-structuring options for the inputdata; and (3) incorporates L1 parametric transfer. We will exploreto what extent the above model can: (1) account for theidiosyncrasy of interlanguage (IL) grammars with respect to bothnative adult and child grammars; (2) provide an explanation for thedifferences among the Spanish IL grammars of speakers of differentL1 backgrounds; and (3) provide an explanation for the individualdifferences within a given IL.

Recent developments within the so-called Principles andParameters model – the predecessor of the Minimalist Programme(Chomsky, 1995) – provide an interesting framework forinvestigating the nature of non-native systems. This model separatesUG principles from parametric options and locates all parameterswithin functional categories (Borer, 1984; Lebeaux, 1988; Chomsky,1991). This clear-cut distinction between principles and parametershas led Tsimpli and Roussou (1991) to propose that adult L2learners have access to UG principles but do not set the parametersof the L2.1 Namely, the triggering experience which leads to the‘growing’ of an L1 in the child (Lightfoot, 1991; Pinker, 1994) is notavailable in the case of adult L2 learners.

In order to illustrate this proposal, Tsimpli and Roussou (1991)resort to one of the most extensively studied parameters, the pro-drop or null-subject parameter. They propose that adult nativespeakers of Modern Greek, Spanish or Italian do not reset the[+null subject] option to the [–null subject] one when they learnEnglish. Instead, learners access UG principles to re-structure theirL1 on the basis of the L2 data in order to acquire a mentalrepresentation. This mental representation – the IL grammar –differs from native English, even though the native and the L2production may be – and indeed usually end up being – similar.This would suggest that null subjects in the Spanish IL of Englishand Chinese speakers, for example, are the output of differentgrammatical representations. We would like to explore theconsequences of this hypothesis in the light of our reformulation ofHyams’ (1994) and Rizzi’s (1994) account of early null subjects in

1 It has been previously suggested that adult L2 learners in institutional settings do notimplement all principles of UG (Felix and Weigl, 1991). However, no distinction was madebetween constructions involving UG principles and those which were the result of choosinga parametric option.

Juana M. Liceras and Lourdes Díaz 3

[–null argument] languages. We will proceed as follows: First, wewill outline our approach to the logical problem of non-primarylanguage acquisition (Bley-Vroman, 1990). Second, we will presentHyams’ (1994) and Rizzi’s (1994) account of null subjects in childlanguage and the consequences for providing a typology of nullsubjects in a number of languages. Third, based on Tsimpli andRoussou’s (1991) analysis of non-native null subjects, we willdiscuss the type of IL grammatical representations which wouldresult from transfer and from accessing UG principles to re-structure L1 representations without setting the Spanish [+nullsubject] option of the [null subject] parameter. Finally, we willprovide a grammatical account rather than a processing orperformance account (Hyams and Wexler, 1993) of the omissionand production of subjects in the Spanish ILs of L1 [pro-drop] andL1 [topic-drop] speakers.

II The logical problem of foreign-language learning

The logical problem of L1 acquisition consists of accounting for thefact that the linguistic data that are available to children areunderdetermined in terms of the complexities of the resultant adultgrammar. Following Chomsky (1965; 1981; 1986; 1991; 1995), it hasbeen systematically assumed that children are equipped with aninnate structure (Universal Grammar) which consists of principleswhose role is to limit the range of possibilities which would becompatible with the data. In addition, and in order to account forthe range of variation present across languages, it has beenproposed that there are also parametric options which are realizeddifferently depending on the language. Consequently, principles ofUG will guide the selection of the actual options, which will in factbe triggered by the data.

For researchers working within the Chomskian framework, thelogical problem of foreign-language learning has usually taken theform of a debate concerning the availability of UG to adult second-language learners. A number of researchers have assumed that – asin the case of L1 – UG was also available for adult L2 learners(White, 1985; Liceras, 1986; Flynn, 1987), whether directly orindirectly.2 Other researchers (Clahsen and Muysken, 1986;Schachter, 1988; Bley-Vroman, 1990) do not postulate a role forUG in the case of L2 acquisition. For the latter, the logical problemof L2 acquisition cannot be resolved as in L1 acquisition.

2 This is an important issue and the predictions made may be quite different. However, wewill not entertain this discussion here.


Clahsen and Muysken (1986) and Schachter (1988) argue againstthe availability of UG on the basis of a particular analysis of theirdata, while Bley-Vroman (1990) offers an articulated conceptualexplanation to his position. The arguments based on data are farfrom convincing because alternative accounts of the same dataprovide ammunition to argue for the availability of UG in L2 – forinstance the response of du Plessis et al. (1987) to Clahsen andMuysken (1986). With respect to Bley-Vroman’s (1990) assertionthat L2 learners cannot work out the properties of L2 which areunderdetermined by the data, White (1990), Martohardjono andGair (1993) and other researchers have provided evidence that L2learners are indeed able to work out properties of the L2 which areunderdetermined by the input data.3

Thus, the question remains as to whether it is the L1 thatsubstitutes for UG – as Bley-Vroman maintains – providing ameans to sort out those underdetermined properties, or whether L2learners not only implement UG principles but also activate thedomain-specific learning procedures proposed for child languagedevelopment rather than resorting to problem-solving mechanisms,as in Bley-Vroman’s (1990) dichotomy which we reproduce in Table1. One of the main difficulties encountered by researchers isdetermining whether UG principles are available directly or via theL1 because, by definition, they are realized in all natural languages.4This is the main reason why the blunt dichotomy access versus non-access to UG cannot lead to any conclusive results. Consequently,it would be desirable to address the issue of the differences andsimilarities between L1 and L2 acquisition from a differentperspective.

The separation of principles and parameters of UG and thedefinition of parameters as [+/–] features located in functional

Table 1 The logical problem of foreign-language learning (from Bley-Vroman,1990)

Child language development Adult foreign-language learning

A. Universal Grammar A. Native-language knowledgeB. Domain-specific learning procedures B. General problem-solving systems

3 Based on the results of the grammaticality judgements of German adolescents learningEnglish via formal instruction with only classroom exposure to the target language, Felix andWeigl (1991) hypothesize that implementation of some UG principles may be impeded byfactors having to do with the learning environment.4 Even though subjacency has been considered a UG principle which had differentrealizations in different languages, we can still say that all languages obey subjacency, butthat it is not restricted to the same bounding nodes. For work on subjacency and L2acquisition see, for instance, White and Genesee (1996) and references therein.


categories is a welcome development because, unlike UG principles,parametric features will differ depending on the language. And itis the domain-specific nature of parameterized features (the‘functional module’) that is at the core of Tsimpli and Roussou’s(1991) proposal that L2 learners do not set parameters to a new,L2 value.5 They assume that the functional component is the onethat is subject to maturation, which implies, that it cannot ‘mature’more than once. This view is consistent with the biological basis oflanguage acquisition which is at the core of the generative tradition(Piatelli-Palmarini, 1989; Lightfoot, 1991; Pinker, 1994) and whichhas to some extent been adopted by researchers outside thistradition (Bickerton, 1990).

Besides being consistent with the biological approach, this viewof the logical problem of second-language learning puts inperspective Chomsky’s position with respect to non-nativegrammars (Chomsky, 1986), in that they cannot be considered I-languages (‘Internal’ languages) because they do not reflectconsistent options in relation to a given parameter.6 What resultsfrom the above approach to the logical problem is that ILs shouldnot be considered I-languages even if they reflect consistentparametric options, the reason being that the process that leads tosuperficial consistency may be very different. What we maintain isthat L1 learners set parameters by letting an initial representationgrow out of exposure to triggering data in the adult L1 system. L2learners, on the other hand, do not grow initial representationsbecause what they possess are grown representations. This impliesthat they are not sensitive to the specific triggers which lead toparameter setting in L1 acquisition.

The question which is systematically raised when parameter-setting is excluded from L2 learning is whether one could still saythat ILs are ‘natural’ languages. At this point it is obvious to us thatthey are, indeed, ‘natural’ languages but they are not I-languages.We cannot ‘grow’ an organ twice in our lives but our brain takesthe form it does because of our genetic endowment. Consequently,it is not closeness to the initial state that makes a given linguisticsystem ‘natural’, but being a further development of the languagemodule. Ils are not I-languages because their functional moduledoes not ‘grow’ as it does in the case of native languages, but theyare ‘natural’ languages because they result from the developmentof the language-specific grown representations which serve as a

5 They base their proposal on the idea that functional categories form an independentcomponent of UG, the UG lexicon (Ouhalla, 1993).6 Chomsky (1986) argues that the French IL of Russian aristocrats was not an I-languagebecause it did not contain consistent options of a given parameter.


point of departure to develop non-native grammars.Consequently, we propose an alternative account to Bley-

Vroman’s approach to the problem of foreign-language learning(see Table 2), and we use recent developments in linguistic theory(Hyams, 1994; Rizzi 1994) and in L2 theory (Tsimpli and Roussou,1991) to illustrate how this model accounts for the presence of nullsubjects in the Spanish non-native grammars mentioned above. Thisproposal implies that UG principles and categories will beimplemented in the IL grammar via the L1 and, if applicable, theother L2 languages learned.7 L2 data will be accessed throughsecondary-level domain-specific procedures created as a result ofrepresentational redescription (Karmiloff-Smith, 1994).8 Thisimplies that L2 learners will have ‘intuitions’ about the L2,something that Bley-Vroman (1990) could not account for withinhis model. However, L2 intuitions will be different from L1intuitions because, unlike the L1 representations created viadomain-specific learning procedures, secondary-level represen-tations resulting from representational redescription will only beaccessible via secondary level domain-specific procedures: namely,the domain-specific learning procedures which create secondary-level representations and L2 representations are not sensitive toenvironmental triggers the way L1 domain-specific learningprocedures are (Liceras, forthcoming). Therefore, L1 and L2learners will use different information and procedures to constructtheir respective grammars. While parameter-setting may guide L1acquisition, L2 acquisition will proceed by re-structuring specific

7 What we propose is that the linguistic mental representation of an individual who hasacquired a language will differ from the linguistic initial state (UG as biological endowment)of the newborn. We also propose that the linguistic mental representation of an individualwho has acquired more than one language will differ from that of an individual who hasonly acquired an L1. We assume that UG principles are realized in all languages althoughthey may take different forms. This should facilitate re-structuring because UG principleshave already been implemented to carry out re-structuring procedures in L2 learning or L3learning.8 The secondary level of mental representation as defined by some psychologists (Karmiloff-Smith, 1994), is a possible characterization of the initial state for non-primary languageacquisition.

Table 2 The logical problem of non-primary language acquisition

Child primary-language acquisition Adult non-primary-language acquisition

A. Universal Grammar A. Previous linguistic experienceB. Domain-specific learning procedures B. Secondary level domain-specific

learning proceduresC. General problem-solving systems


sections or units of their previous linguistic representations.9 Inprinciple, L2 re-structuring would proceed without implementingthe same procedures which lead to parameter-setting in the case ofL1 acquisition.

III The null argument parameter

In order to illustrate what parameters should be like, Chomsky(1981) chose the pro-drop parameter which had been described indetail by Rizzi (1982) and Jaeggli (1982). This parameter – inessence a formalization of Perlmutter’s (1971) surface structurefilter – was defined as a cluster of properties which determine twotypological groups of languages: [+/– pro-drop], depending onwhether they allow:

• null subjects as in (1a):

1) a. [e] hemos encontrado el libro.b. [e] *have found the book

‘We have found the book.’

• free inversion in simple sentences as in (2a):

2) a. [e] Ha comido Juan.b. [e] has eaten Juan

‘Juan has eaten.’

• ‘long wh-movement’ of subject as in (3a):

3) a. El hombre i que me pregunto a quién [e]i había visto.b. the man i that I wonder whom [e]i had seen

[‘the man x such that I wonder who x saw’]

• empty resumptive pronouns in embedded clauses as in (4a):

4) a. Esta es la chica i que me pregunto quién cree que [e]i lo hizo.b. this is the girl i that I wonder who thinks that [e]i did it

‘This is the girl that I wonder who thinks she did it.’

• apparent violations of the *[that-t] filter as in (5a):

5) a. ¿Quiéni crees que [e]i se irá?b. whoi (you) think that [e]i will leave?

‘Who do you think (that) will leave.’

9 It is a fact that the actual concept of parameter-setting is under constant revision. For arather complex but extremely interesting account of L1 parameter-setting see, for instance,Clark and Roberts’s (1993) proposal.


According to Chomsky (1981), who follows Taraldsen (1980), theseconstructions are possible in pro-drop languages because AGR(agreement) – a feature of INFL (inflection) – governs the emptycategory in each case. The intuitive idea is that subjects can bedropped when there is overt agreement.10

While linguists do not agree with respect to whether allproperties listed above are related to the pro-drop parameter,accounting for null arguments in adult and child language hasbecome a central issue for linguistic and psycholinguistic research.

Rizzi (1986) argued that in [+pro-drop] languages, pro has to belicensed and identified. Licensing takes place under head-government. Thus, in the examples above, the null element insubject position is governed by INFL and is identified through therich agreement specification (f-features) as shown in (1c).

1) c. CP

Spec C'

C(OMP) INFL

Spec I'

I(NFL) VP

Spec V'

V NP

proi hemosi encontrado el libro

have found the book

Huang (1984) argued that Chinese (the same can be said ofJapanese) is a [pro-drop] language in spite of the fact that its INFLlacks overt agreement features entirely. According to Huang (1984)Chinese pro is licensed by a discourse-bound operator in thespecifier of the root – the Spec of CP – and identified by null topics.Thus, INFL is not the only available licensing head for nullsubjects.

10 There are also mixed systems, which is the case of Hebrew.


1 Null subjects in child language

In the case of L1 acquisition, and in order to account for the factthat children learning [–pro-drop] languages such as Englishproduce null subjects, Hyams (1986) proposed that [+pro-drop] wasthe default option for the [pro-drop] parameter and that bothEnglish and Italian children (as well as children acquiring otherlanguages) started with that option.11 While some accounts ofchildren’s null subjects have attempted to explain them on the basisof processing and other performance mechanisms, Hyams andWexler (1993), Hyams (1994) and Rizzi (1994) argue in favour ofHyams’ (1986) original proposal that null subjects in child languageare a genuine grammatical option and cannot be reduced toextragrammatical factors. However, rather than comparingchildren’s null subjects in English to null subjects in adult nativeItalian or Spanish, Rizzi (1994) maintains that early null subjects inEnglish only appear in Spec-Root, a configurational constraintwhich is also found in adult systems such as the English and Frenchdiaries mentioned by Haegeman (1990). Rizzi (1994) observes thatnull subjects present the following pattern:

• They occur in matrix sentences such as (6):

6) a. ___ want moreb. ___ is brokenc. ___ boit café

‘drinks coffee’d. ___ est tombé

‘has fallen’

• They do not occur in wh questions when wh is fronted such asin (7), even though nominal subjects can appear as in (8):

7) @ Where ___ go?12

8) Where daddy go?

This is not the pattern which is found in the adult or the childgrammars of a null subject language such as Spanish, where theequivalent of (7) is possible, as shown in (9):

11 The default or unmarked option is the initial one chosen by all children regardless of thelanguage they may be exposed to.12 Even though this form was considered non-attested (@) in Rizzi (1994), Roeper andRohrbacher (1994) noted that Adam’s data (Brown, 1973) contained many instances offronted wh-questions with null subjects such as the one illustrated in (7). However, theycould not attest the inflected alternative, which lead them to propose that these null subjectswhich only occur in non-finite contexts are instances of Japanese-like pro as in Speas’ (1994)analysis.

9) A dónde va ___?where ___ goes?‘Where is he/she going?’

• If the wh word appears in situ, null subjects are possible as in(10):

10) a. ___ see what bear?b. ___ have what?c. ___ doing what?

• Null subjects do not occur in embedded clauses as indicated in(11):

11) a. ___ went in the basement # that what we do # after supperb. ___ know what I maked

Since a similar pattern is found in the case of diaries (Haegeman1990), Rizzi (1994) argues that these grammars do not implementthe principle ROOT=CP. This principle accounts for the intuitionthat we always use propositions when we produce utterances. If aroot sentence does not have to contain COMP, null subjects will bepossible in the case of [–pro-drop] languages.13

As we have seen above, licensing of pro is possible in Spanishbecause its INFL has [+strong] agreement as indicated in (12).

12) CP

Spec C'

C(OMP) INFL

Spec I' [+strong] AGR

I(NFL) VP

Spec V'

V NP

proi estudiamosi español

study Spanish


13 The pattern described by Rizzi (1994) does not occur in the case of child Spanish or childItalian. Valian (1991) noted that Italian children produce 70%, versus 30% produced byAmerican children. The production of null subjects by the two Spanish children whose datawe have analysed (Liceras, Valenzuela and Díaz,1998) is closer to that of the Italian childrenthan to that of the American children.

Null subjects are not licensed in English because it has [–strong]agreement, as indicated in (13).

13) CP

Spec C'

C(OMP) IP

Spec I' [–strong] AGR

I(NFL) VP

Spec V'

V NP

*pro/we study Spanish

However, a different type of null subject – which is not pro-nominal (pro is the empty option of pronominals such as we) butthe empty option of nominals, null constants (nc) – (the emptyoption of a null constant would be proper nouns such as John orJohn and Mary) – can occur when the CP projection is omitted asin (14).

14) CP

Spec C'

C(OMP) IP

Spec I' [–strong]

I(NFL) VP

Spec V'

V NP

__(null constant) study Spanish


Once principle ROOT=CP is operative, null subjects disappear.14

Following Lasnik and Stowell (1991), Rizzi (1994) argues that thenull subject in (14) is neither anaphoric PRO nor a variable (t), buta null constant, an empty category which is an R-expression withthe features [–anaphoric], [–pronominal], [–variable].15

From a somewhat different and complementary perspective,Hyams (1994) also argues for a grammatical account of children’snull subjects in English. Unlike her previous proposals whichassociated early null subjects first to Italian null subjects and laterto Chinese null subjects, Hyams (1994) proposes a null-argumentparameter according to which all languages can have null argumentsprovided they are both licensed and identified. The parameterdifferentiates languages according to whether licensing of pro16

takes place at Spec-COMP or at Spec-INFL, as in (15a) and (15b)below:

15) a. [topic-drop] b. [pro-drop]

CP CP

Spec C' Spec C'

C INFL C INFL

Spec I' Spec I'

I VP I VP

In languages such as Chinese, arguments (subjects and objects) arelicensed as in (15a), and identified by discourse topics. In Spanish,licensing occurs at the level of Spec-INFL – as in (15b) – and onlyin the case of subjects. Identification takes place through the f-features in AGR, as proposed by Rizzi (1986). German is a [topic-


14 ROOT=CP captures the intuition that we speak through propositions not throughfragments of propositions. The presence of this principle would also account for the followingfacts:(1) children can have fragments of CPs and adults too in very special contexts becauseother categories can act as root; (2) the in-situ wh-words (provided they are real questionsat LF) occur because there is no host (or landing site) for the wh-word;(3) whether functionalheads are available or not in child grammars, it is a fact that they are free to lexicalize ornot.15 Rizzi argues rather convincely that it is the same category that would account for nullsubjects – but not for null objects – in colloquial German.16 Hyams (1994) assumes that pro accounts for null subjects both in topic-drop languagessuch as German and Dutch and in pro-drop languages such as Spanish.She does not attemptto justify her proposal and specifically states that she is not assuming that Chinese nullarguments are also pro.

drop] language with pro licensed in Spec-COMP under Spec-headagreement thanks to the presence of V features in COMP (the sameas Dutch and other V2 languages). Null subjects occur in colloquialGerman because they are identified by a discourse-identifiedoperator.17

In order to account for children’s null subjects in English, Hyams(1994) proposes that they have pro in Spec-INFL, where they canbe licensed but not identified. Identification takes place by movingto Spec-C'. Thus, she proposes a hybrid explanation wherebylicensing occurs as in Italian and identification as in German.

This view of the argument parameter implies that all languages– including ILs – are null argument languages but not all canidentify. Thus, language learners set the parameter in terms of theposition where null arguments can be licensed and they have tolearn or ‘tune’ their grammar, to use Hyams’ term for the case ofEnglish children, in order to identify. Identification is based onlanguage specific factors which have to be learned: f-features in thecase of Spanish or Italian, and null topics in the case of Chinese orJapanese.

What we would like to propose is that identification alone cannotdefine a parameter. In our view, identification will guide L2acquisition because the adult L2 learner does not set parameters.However, identification will not be the driving force for L1acquisition. Following Liceras (1994), we propose that the [+/–strong] INFL features (Pollock, 1989) which differentiate Englishfrom Spanish and French in terms of V-movement, are responsiblefor whether pro can be licensed. Consequently, in [pro-drop]languages null subjects are licensed if the INFL features are[+strong] (French, Spanish), but they cannot be licensed if they are[–strong] (English). The identification requirement will determinewhether null subjects are actually realized (as in Spanish) orwhether they do not occur because they cannot be identified (as inFrench).18 In [topic-drop] languages (such as Chinese and Japanese)as well as in V2 languages, pro may or may not be licensed in Spec-C' position depending on a [+/– strong] feature located at that leveland provided there is an appropriate identifier.19 Consequently, null


17 Null objects are also possible in German but only in the third person. Hyams (1994) andRizzi (1994) provide different explanations for this phenomenon.18 Authier (personal communication) suggests that because only thematic pro has to beidentified,it would follow from this proposal that French, as Standard German, could haveexpletive pro. Since, as we have already mentioned, there is no agreement as to whetherGerman expletive pro follows from the pro-drop parameter (Hyams, 1994; Rizzi, 1994), wewill not pursue this issue here.19 We are assuming that in the case of V2 languages such as German the V carries the[+/–strong] feature to COMP. However, a discussion of the differences among the variousV2 languages is far from the scope of this paper.

subjects are possible in Chinese and Japanese because they arelicensed via the [+strong] feature located at Spec-C' level and theyare identified because there are null topics in these languages.Standard German, on the other hand, cannot license pro becauseit has a [–strong] feature at that level.

2 The null argument parameter in L2 acquisition

Research in L2 acquisition related to the pro-drop parameter hasfollowed the developments of linguistic theory and L1 acquisition.Early research (White, 1985; Hilles, 1986; Phinney, 1987; Liceras,1988; 1989) was based on the formulations of the pro-dropparameter within the Government and Binding framework(Chomsky, 1981; Jaeggli, 1982; Rizzi, 1982) and on Hyams’ (1986)work on L1 acquisition. Issues such as establishing the defaultoption, whether properties were simultaneously acquired, ordetermination of triggering effects were all discussed whenconsidering the role of L1, thus transfer (in general) and parametrictransfer (in particular) were always at stake. The reformulation ofthe pro-drop parameter in terms of the Morphological UniformityPrinciple (Jaeggli and Safir, 1989) – which stems from Huang’s(1984) analyses of null arguments in Chinese and Rizzi’s (1986)proposal concerning licensing and identification conditions forpresence of pro – enhanced the framework for cross-linguisticcomparison. Research aimed at determining whether access to UGguided the acquisition of pro-drop properties in L2 acquisition(Hilles, 1991; Lakshmanan, 1991; O’Grady, 1991), which draws onwork in the acquisition of L1 intended to determine the triggeringlocus for establishing the [+/– pro] setting (Roeper andWeissenborn, 1990; Hyams, 1989), also deals with parametrictransfer and the role of age in parameter-resetting. Finally,parametric transfer has also played a major role when investigatinghow the L1 [+/–topic-drop] could influence the setting of the L2[+/–pro-drop] parameter (Díaz and Liceras, 1990; Register, 1990:Jin, 1994).

Tsimpli and Roussou (1991) propose that the grammaticalrepresentation which underlies a sentence such as (16) in theEnglish IL contains a pro because the [+pro-drop] option of Greekis transferred.

16) ___ lives in Sitges (= he lives in Sitges)

On the other hand, the null subject in IL sentences such as(17a) or (17b) is PRO because the IL does not have the f-features


which would account for the identification of pro.

17) a. ___ is dancing (= she is dancing)b. ___ Thursday come to school (= On Thursdays, I come to school)

Since both options (parametric transfer and the use of an emptycategory from the inventory provided by UG – PRO in this case)are possible options for the L2 learner, the IL may be an instanceof any of the two representations. Thus, according to Tsimpli andRoussou (1991), (18) would be the appropriate representation for(16):

18) AGRP

Spec AGR'

AGR VP

Spec V'

V PP

pro -s live in Sitges

In the case of (17), the appropriate representation would be(19).

19) TP20

Spec T'

T VP

Spec V'

V PP

a. PRO is dancingb. PRO come to school


20 Tsimpli and Roussou (1991) adopt the so-called split-INFL hypotheses first proposed byPollock (1989) according to which INFL should be split as AGR and TENSE. They providea structure with AGRP for (16),and one with TP for (17) due to their assumption that AGRmight not be projected in the IL.

Since the [+pro-drop] option is supposed to be the only oneavailable in the English IL of Greek speakers, examples such as(20), which occur at a more advanced stage, will not correspond tothe native-English grammatical representation in (21a) or theSpanish ‘equivalent’ in (21b).

20) We eat a lot of fish.

21) AGRP

Spec AGR'

AGR VP

Spec V'

V NP

a. we eat a lot of fishb. proi -mosi comer mucho pescado21

What Tsimpli and Roussou (1991) actually propose is that a re-structuring process which consists of reanalysing the L2 subjectpronouns as f-features will create the IL representation shown in(22), so that both pro and the subject pronoun coexist.

22) AGRP

Spec AGR'

AGR VP

Spec V'

V PP

proi wei eat a lot of fish

This grammatical representation is similar to the one that has beenproposed to account for the clitic nature of subject pronouns in


21 The f-feature is located in AGR because this is a possible representation. There is notagreement with respect to the actual status of these morphemes. For instance, Sportiche(1996), proposes that they head their own projection.

French and other languages (Roberge, 1986) as shown in (23)–(24).22

23) Il habite à Sitges.He lives in Sitges.

24) AGRP

Spec AGR'

AGR VP

Spec V'

V PP

proi ili habite à Sitges

However, it would be very difficult to show that subject pronounsare clitics in the English IL of Modern Greek speakers, or Spanishspeakers for that matter, unless we propose that a reanalysis of thepronominal system as AGR does not necessarily lead to thecliticization of subject pronouns. Liceras (1994) suggested thatevidence for the reanalysis of subjects pronouns as AGR may comefrom sentences such as (25) which are systematically produced bysome native speakers of Spanish.

25) The girl that she is very nice.

Torrego (personal communication) suggests that re-structuringcould account for the fact that native speakers of [+pro-drop]languages systematically use he as the only third person pronounwhen learning English.23 In fact, this would show that he doesnot have a referential value but a structural value. However,for the time being we will have to take these examples asisolated phenomena, unless we can show that the whole range of


22 If the Greek-based English IL is in fact a null-subject language with clitic subject pronouns,it should also behave like French in the relevant ways. However, the fact that it probablydoes not, might not be enough evidence to discard such a representation.23 An anonymous reviewer has asked what evidence does Torrego have that native speakersof pro-drop languages systematically use ‘he’ in IL English. In fact, this phenomenon wasalready reported in the early literature on interlanguage systems. And even though this maynot be the case for the English non-native grammar of speakers of all pro-drop languages,many English teachers would be ready to confirm that it is a common feature of the Spanish-based non-native grammars of English.

properties accompanying the clitic nature of subject pronounsoccur.24

IV Null subjects in non-native Spanish

We have proposed that adult non-native speakers do not set theSpanish [+pro-drop] option of the pro-drop parameter because theyare not sensitive to the abstract [+/–strong] features of thefunctional categories. What these learners actually do is re-structurethe L1 grammatical representation and identify null subjects via theperson markers of the Spanish verb, the Spanish subject pronounsor null topics. Some of the hypotheses which follow from thisproposal are that:

1) The ROOT=CP principle will not be relaxed in the case of theadult non-native grammars because the conditions which aremet by the ‘grammar of diaries’ (Haegeman,1990),neither applyin the case of the narratives nor in the case of the interviews.

2) There will be differences in the production of null subjectsversus overt subjects both in matrix and subordinate clausesdepending on the level of competence. If total transfer from theL1 determines the presence of null subjects in the early stage,the French/English beginners will not produce null subjectswhile their advanced counterparts will.

3) There will be differences in the production of null subjectsversus overt subjects both in matrix and subordinate clausesdepending on the language background. Specifically, nativespeakers of [–pro-drop] and [–topic-drop] languages (such asEnglish and German) will produce more overt subject pronounsthan native speakers of [+topic-drop] languages (such asChinese and Japanese). In the case of the former, pronominalsubjects will tend to be used as identifiers (more at the earlystage than at the advanced intermediate stage). Identificationvia null topics will only occur in the case of the latter. IfFrench is a [+pro-drop] language, the production of null andovert subject pronouns will be closest to the native group.Excellent command of French or other [pro-drop] languagesmay also influence the production of null versus overt subjectpronouns.


24 Another consequence of this proposal (Authier, personal communication) is the predictionthat speakers whose grammatical representation might be as in (24) would never producepronouns conjoined with full NPs. We do not think that this is the case for the English IL.However, it would still be possible to propose that the IL had two different values for thesubject pronouns. In fact, in Liceras (1985) it is suggested that the IL clitic pronouns havetwo different values.

In order to explore the implications of these proposals forexplaining the grammatical representation which may underlie theIL production of non-native speakers of Spanish, we have analysedtwo different types of spontaneous production data:

1) Data produced by five adolescents (12 year olds) and sixuniversity students (20–25 year olds) who have had only had 50hours of classroom exposure to Spanish.25 The data was elicitedduring half-hour interviews. Subjects had to answer generalquestions about their school and family life and had to tellstories about the characters and actions depicted in comic strips.The same interviews were conducted with 3 native speakers ofSpanish who were studying at the University of Ottawa. Thecontent of the interviews was the same in all cases.26

2) Data obtained from speakers of 5 different languagebackgrounds (Chinese, English, French, German and Japanese).Fifteen advanced intermediate non-native speakers27 (they hadhad an average of 250 hours of formal instruction in Spanish)were asked to tell a story based on one of their favourite films.They were asked to discuss specifically the characters and theplot and were given half an hour to do so. Their speech wasrecorded. These speakers were chosen because they were at alevel where some degree of re-structuring should already havetaken placed. Three native speakers of Spanish were asked toperform the same task.

1 Early null subjects: French/English speakers

Tables 3 and 4 show the overall pattern of production of nullsubjects versus pronominal subjects at the early stage. It is clearthat all these subjects produce null subjects both in matrix andsubordinate clauses. The only speaker who produces a very lownumber of null subjects in the subordinate clauses is SC1. This


25 These data are part of a longitudinal study intended to investigate L2 development ofSpanish and English in the case of children, adolescents and adults who are learning theselanguages in classroom settings. For a description of the overall research project see Liceras,Díaz, Maxwell, Laguardia, Fernández and Fernández (1997). All subjects had an excellentcommand of both French and English but their previous linguistic experience was notuniform.Determination of previous linguistic experience was based on our interpretation ofspecific questions elicited from a questionnaire.26 We analysed data from the story-telling part of the three interviews.27 The level of competence was determined by a placement test published by SGEL whichwas given to all subjects. All subjects had learned Spanish in institutional settings. TheGerman subjects and the Chinese and Japanese subjects had studied Spanish in theircountries of origin and had been at the Universities of Madrid or Barcelona between oneand a half and three months when they were tested. The French and English subjects werestudents at the University of Ottawa and had studied Spanish for three to four years.

speaker also produces a high percentage of subject pronouns. Thisis interesting because, even though we classified subject SC1 asEnglish/French bilingual, we were not able to determine which ofthe two languages, English or French, is the dominant one for him.For the other speakers, English is never the dominant language.Thus, it is possible that the ‘initial state’ for SC1 is the Englishgrammatical representation. This would explain why a substantialnumber of the subject pronouns produced by this speaker wouldbe redundant in native Spanish. It looks as if identification viasubject pronouns is important for this subject. The rest of thesubjects did not produce any subject pronoun in the case of thesubordinate clauses.

In the case of the adult speakers (Table 4), subject UB5 alsoproduces a rather high number of subject pronouns. It would betempting to attribute this to the fact that this speaker’s dominantlanguage is English. However, she also happens to have native-likecompetence in Portuguese, which could constitute the actualgrammatical representation for constructing the Spanish grammar.This, in fact, could be the case not only because, unlike SC1, she


Table 3 Early stage: French/English speakers (adolescents)

Null subjects Subject pronouns

Speaker Matrix (%) Subord (%) Matrix (%) Subord (%)

SC1 27 14 14 29SC2 71 100 8 0SC4 79 87 3 0SC5 46 100 21 0SC6 30 50 10 0

NoteSC = secondary

Table 4 Early stage: French/English speakers (adults)



UB1 81 92 0 0UB3 77 100 6 0UB4 58 80 0 20UB5 50 66 8 33UB6 65 71 0 0

NoteUB = University beginner

produces a higher percentage of null subjects in the subordinateclauses, but also because her subject pronouns are never redundantfrom a native speaker point of view.

It is interesting to notice that null subjects are abundant in thisearly interlanguage, which indicates that identification via discourseor f-features overrides identification through subject pronouns. Thisseems only natural considering:(1) the large amount of null subjectsin the Spanish input; (2) the difficulty which is involved in learningthe subjective, the objective and the oblique inventory of Spanishpronouns; and (3) the fact that French is the dominant language ina number of cases.

The same interview was conducted with 3 adult native speakersof Spanish. The results (Table 5) show a clear difference betweenthe production of null subjects in subordinate clauses and in matrixclauses. These data show that the production of subject pronounsremains rather constant across subjects and in terms of theirdistribution between matrix and subordinate clauses.

An analysis of variance was conducted on the proportion of nullsubjects from the total number of clauses of a given type (matrixor subordinate) produced by the subjects (Table 6). The main effectof Group (adolescent versus adult) was not significant, nor was theSyntactic Position per Group interaction. However, the effect ofSyntactic Position was statistically significant at F (1,10) = 2256.87;p = 0.0028. As seen in Figure 1, both of the French/English groups


Table 5 Control Group: Spanish speakers (interviews)



NC1 68 82 12 8NC2 67 85 15 3NC3 57 92 22 7

NoteNC = Narrative control

Table 6 Null subjects: early stage (adolescent/adult groups)

Spanish versus: Diff Crit Dif f P-value

English/French adolescents –14.767 41.880 .3354English/French adults –1 .167 41.880 .9379

NotesBonferroni/Dunn (Control). Effect: Group. Error term: Type III sum of squares forsubject (Group) Dependent: syntactic position. Significance level: 0.05

behaved in a similar way in that they produced more null subjectsin the subordinate than in matrix clauses. A post-hoc Bonferronitest revealed no significant difference between the experimentalgroups and the native-speaker control group.

Our hypothesis 1 (see the start of this section) is confirmed inthat this production does not present the pattern of the grammarof diaries or lack of implementation of the ROOT=CP principlebecause there are null subjects in both matrix and subordinateclauses. Our hypothesis 2 is not confirmed because these subjectsproduce null subjects at this early stage. If these subjects takeFrench as the initial representation but do not identify via subjectpronouns but rather via the agreement features, our hypothesis 3is confirmed (recall that French is the dominant language for mostof the adolescent subjects).

An analysis of variance was also conducted on the proportion ofsubject pronouns from the total number of clauses of a given type(matrix or subordinate) produced by the subjects (Table 7). Themain effect of Group (adolescent versus adult) was not significant,nor was the Syntactic Position per Group interaction or theSyntactic Position per se.

Figure 2 shows that the adolescent Group was closer to the


matrix

embedded

French/English French/English Spanishadolescent adult

Group

NotesInteraction plotEffect: Category for syntactic position/GroupDependent: syntactic position

Figure 1 Null subjects: early stage (French/English speakers)

Control group in terms of the distribution of subject pronouns inmatrix and embedded clauses. In the case of the adult Group, thedistribution went in the opposite direction: more subject pronounswere produced in the embedded than in the matrix clauses. This iswhat we predicted for speakers of English or German if subjectpronouns were used as identifiers. This may be due to the fact thatFrench is the dominant language for all but one subject in theadolescent group while English was the dominant language for allthe speakers in the adult group. A post-hoc Bonferroni test revealedno significant difference between the experimental groups and thenative-speaker control group. This is due to the fact that, in the caseof the adult group, only one subject was responsible for the


Table 7 Overt pronouns: early stage (adolescent/adult groups)


English/French Adults –4.467 15.914 .4392English/French Adults –2.667 15.914 .6409

NotesBonferroni/Dunn (Control). Effect: Group. Error term: Type III sum of squares forsubject (Group) Dependent: syntactic position. Significance level: 0.05

matrix

embedded

French/English French/English Spanishadolescent adult

Group

NotesInteraction plotEffect: Category for syntactic position/GroupDependent: syntactic position

Figure 2 Subject pronouns: early stage (French/English speakers)

comparatively large number of subject pronouns in matrix clausesand two subjects for all the subject pronouns produced insubordinate clauses (see Table 4). In the case of the adolescentgroup, one subject was responsible for the production of all subjectpronouns in subordinate clauses (see Table 3).

2 Advanced intermediate speakers

a The Spanish IL: French speakers The [+strong] nature ofFrench INFL allows licensing of pro at Spec-I'. Since the identifieris the pronominal subject clitic, superficially it looks as if French isa [–pro-drop] language.28 Native speakers of French will transferthe pro option to Spanish and will either analyse subject pronounsas identifiers (if they are salient in the input) or will look for otherpossible identifiers. The appropriate identifier – the person markersof the Spanish verb – should not be hard to detect. The fact thatthey may have problems with verb morphology does not preventFrench learners from using null subjects systematically, as theprevious data shows. Consequently, the advanced intermediatesubjects should have already had enough input data to detect thatit is not subject pronouns but the f-features of the verb which serveas identifiers for pro and they should not produce more subjectpronouns than the subjects in the early group. In fact, as shown inTable 8, one French subject does not produce any subject pronounat all and the other two produce a low percentage. If we comparethe production of subject pronouns by French and native speakers(Table 9), we see that the French subjects produce fewer subjectpronouns than the control group. If French is a pro-drop languageand no parameter-resetting (identification of an abstract featurewhich is not present in the L1) is involved in the L2 acquisition ofSpanish, one would expect these results. However, this is what thequantitative analysis shows because if we look at the actual


28 But see Pierce (1992) for a comparison of null subjects in child and adult French versuschild and adult English.

Table 8 French advanced intermediate (FA)

Speaker Null subjects Subject pronouns

Matrix (%) Subord (%) Matrix (%) Subord (%)

FA1 52 37 8 27FA2 52 50 2 6FA3 80 74 0 0

examples we find two interesting facts: (1) most of the subjectpronouns produced by subject FA1 are redundant and do not havethe contrastive, disambiguating or emphatic value of the subjects’pronouns produced by the native subjects; and (2) FA1, preciselythe subject who seems to have access to a French-like grammaticalrepresentation in terms of using a considerable number of subjectclitics, also fills in the subject clitic position in sentences such as(26):

26) me voy a hablar.me (1st person singular) go to speak[ – voy a hablar (yo)]‘I am going to speak.’

This is not produced by the other two subjects. The other twoFrench subjects produced fewer subject pronouns than the nativespeakers.

As expected, the difference between production of pro in matrixversus subordinate clauses is very similar to that of native speakersand is far from showing the pattern of the English grammar ofchildren or the ‘grammar of diaries’. In fact, unlike English children,these speakers produce null-subject pronouns with inflected formsand fronted wh-questions, as shown in (27) and (28):

27) ¿Dónde eres __ ?where are (you, 2nd person singular)?‘Where are you from?’

28) ¿Dónde vives __?where live (you, 2nd person singular)?‘Where do you live?’

b The Spanish IL: English speakers In English, another [–pro-drop] language according to the ‘null-subject parameter’, subjectpronouns are not licensed because AGR is marked as [–strong].


Table 9 Control group: Spanish speakers (narratives)



NC1 12 14 7 4NC2 24 15 7 4NC3 41 29 15 17

NoteNC = Narrative control

Thus, subject pronouns will be present obligatorily. If nativespeakers of English learning Spanish do not reset the parameter,we could hypothesize that pro will never be licensed and that theremay be an initial period of transfer where the subject position isalways filled. Instances of overgeneration of subject pronouns havebeen systematically attested but so have null subjects (Phinney,1987; Liceras, 1988; 1989; Almoguera and Lagunas, 1993; Díaz andLiceras, 1995). These English subjects are far from overgeneratingsubject pronouns. In fact, as is shown in Table 10, if the productionof subject pronouns were an index of their level of competence, onecould say that these data show that they have already abandonedpronouns as identifiers. Thus, it is very probable that learnersimplement a re-structuring mechanism by licensing and identifyingvia any of the available options in UG. For instance, they may resortto the default option for licensing pro (all languages can licensesubject pronouns at the Spec-V' level) and identify null subjects viathe Spanish f-features. It is obvious that they do not drop theROOT=CP principle – as in the English grammar of children, ‘thegrammar of diaries’ or the grammar of ‘colloquial German’ –because if they did, they would only produce null subjects in matrixsentences. Our data (Table 10) shows that this is not an option andit does not seem to be an option for specific individuals or forearlier stages of the IL because it has not been attested in any ofthe L2 studies conducted (Phinney, 1987; Liceras, 1988; 1989;Almoguera and Lagunas, 1993; Díaz and Liceras, 1995). In fact,there are no significant differences in the distribution of nullsubjects in matrix and subordinate clauses. Thus, if English speakersproduce null subjects from the early stages both in matrix andsubordinate clauses, and if pro cannot be licensed because theynever reset the parameter to the [–strong] feature, what is the statusof the null subjects in this IL? We could propose, as Tsimpli andRoussou (1991) do, that the IL has PRO because AGR is notprojected, but if AGR is not projected what are the IL f-features?It seems to us that Hyams’ (1994) proposal could explain these datain that we could maintain that pro is an option for all languages,including interlanguages. The default option (with identification via


Table 10 English advanced immediate (EA)



EA1 61 52 5 0EA2 26 20 5 20EA3 50 44 0 12

discourse) is always possible. However, the choice of the defaultoption does not have syntactic consequences for adult L2 learnersin terms of licensing: they are not sensitive to the abstract featurewhich determines whether a language is [+/–pro-drop] or [+/– topic-trop]. Consequently, it is the identification procedure which shapesboth the different non-native grammars at a given developmentalstage.

Note, however, that our proposal implies that choosing thedefault option does not mean that the matrix null subjects producedby non-native speakers are null subjects in diaries, namely, instancesof ‘simplification’ of the principle ROOT=CP. If this were the case,we would have to find a significant difference in the production ofnull subject in matrix versus subordinate clauses. However, morenull subjects in the matrix clause could always be an option iflearners rely on discourse identification, the identificationprocedure which is possible in the case of coordinate clauses inEnglish.

c The Spanish IL: German speakers According to Hyams’ (1994)proposal, German is a [topic-drop] language with pro licensedunder Spec-head agreement in COMP. In V2 languages, COMPinherits the INFL features that the verb has picked up from AGR.Thus V2 languages and colloquial German have null subjects andthird-person null objects in root position. Following Huang’s (1984)analysis, Rizzi (1994) argues that null subjects in V2 languages suchas colloquial German are also null constants. He maintains that theyare not licensed by discourse operators and should rather bepartially assimilated to the early English system. If we adoptHyams’ (1994) proposal, the German learners will transfer the[topic-drop] parameter to the Spanish IL, in which case we wouldonly expect null subjects in matrix sentences. We do not have datafrom the early IL, which implies that a process of re-structuring willhave already taken place. Table 11 shows that two subjectsproduced more null subjects in subordinate than in matrix clauses,while one subject (GA2) produced a higher percentage of nullsubjects in matrix clauses. However, in absolute terms he produced


Table 11 German advanced intermediate (GA)



GA1 26 53 0 6GA2 61 38 23 33GA3 16 33 0 0

exactly the same number in both cases. This seems to indicate thatthere is a tendency to produce more null subjects in subordinateclauses.

If German learners kept pro in topic position and no pro wereavailable in subordinate clauses, we would have total transfer andnull constants (a similar language to colloquial German with proidentified by a discourse operator). But there are many null subjectsin subordinate clauses. To this we have to add that the productionof subject pronouns (Table 11) is very low in the case of subjectsGA1 and null in the case of GA3.A possible interpretation of thesedata is to say, à la Hyams (1994), that the parameter has been resetfrom licensing pro in the Spec-C' position to licensing pro in theSpec-I' position. If no parameter resetting takes place, we wouldhave to maintain that null subjects are licensed at Spec-C level andidentified via the f-features: in other words, the opposite of whatHyams (1994) proposed in the case of early English. Thus, wepropose once more to separate identification (a UG principle) fromlicensing, a process dependent on a feature which is located in afunctional category (Liceras, 1994).Only children access this featureto set a parameter.

d The Spanish IL: Chinese and Japanese speakers If Chinese andJapanese are [topic-drop] languages whose null subjects are licensedby a discourse null operator which is present in the specifier of theroot and they are identified through a process of topic-chaining,then the Spanish L2 Chinese and Japanese based grammars willlicense null subjects. When these null subjects are identified bynull topics they originate a degree of ambiguity which nativeSpanish does not allow. An instance of such a degree of ambiguityoccurs in sentence (29) produced by one of the Japanese speakers(JA1).29

29) Alex es un joven, ___ trabaja en la empresa de ropa, ___ no es muygrande y ____ es de su tío. ____ Es bastante competente pero ___ notiene mucho interés en su trabajo. [JA1]

Alex is a young man, ____works in a clothing store,_____ is not toolarge and _____ belongs to his uncle. ____is rather competent but ____isnot too much interested in his work.


29 Díaz and Liceras (1990) suggest that setting the [+pro-drop] option of Spanish will beeasier for Chinese, Japanese and Korean speakers than for English and French speakersbecause the fact that they have already fixed a more ‘pragmatically oriented parameter’(identification relies more on discourse than on morphology) will facilitate their task.However, it is also suggested that null subjects in these Spanish ILs may actually be variablesidentified through topic-chaining , which would account for the attested degree of ambiguity.

If re-structuring takes place and, even though null subjects cannotbe licensed at the Spec-I' level, these speakers may choose adifferent type of ‘identifier’: namely, the f-features of SpanishAGR may account for the identification requirement and, if this isthe case, no instances of ambiguity of the type shown in (29) willoccur.

None of the Chinese subjects produced ambiguous sentences, inspite of the fact that two of them produced a high percentage ofnull subjects in both the matrix and the subordinate clauses; thisseems to suggest that the f-features are fulfilling their identifyingrole. Note that the production of subject pronouns is virtually nullin the case of matrix clauses (Table 12). It is interesting to note thatthe Japanese speakers produce a rather large number of subjects incomparison to the Chinese speakers (Table 13). One couldspeculate that these subjects are transferring their knowledge ofEnglish to the Spanish IL or that both subject pronouns andf-features are candidates for identification of the null category.However, since the Japanese speakers in the study use clitic-doubling – while very advanced English speakers seldom producethis construction (Liceras and Díaz, 1985; Liceras, Laguardia,Fernández, Fernández and Díaz, 1998) – the overgeneration ofsubjects may have a different explanation altogether.

e Null objects Null objects should not occur in the Spanish IL ofFrench and English speakers because pro-drop languages do notlicense null objects and the Spanish data to which they will be


Table 12 Chinese advanced intermediate (CA)



CA1 0 30 0 1CA2 78 56 0 15CA3 53 60 0 0

Table 13 Japanese advanced intermediate (JA)



JA1 47 34 16 26JA2 20 57 57 10JA3 41 40 23 26

exposed will not contain null objects.30 Thus, null objects would alsonot occur via transfer. Null objects have not been reported inprevious studies and we have not found null objects in these data.

In the case of German, 3rd-person [–human] null objects arepossible. Even if their existence (Rizzi, 1994) is not related to thenull-argument parameter, these types of null objects could betransferred to the Spanish IL. We have not found any in the datathat has been analysed.

The Spanish IL of Chinese, Japanese and Korean speakers couldcontain null objects if the topic-drop option of the parameter weretransferred. However, if null objects are universally infrequent(Wang et al., 1992), it would be natural to avoid them unless theyare part of the input. We have only found two instances of nullobjects. However, what is interesting to note is that there areinstances of clitic-doubling such as those in (30)–(34):

30) y cuando él les visita a ellos [CA2]and when he them visits them(y cuando él los visita)‘and when he visits them’

31) sus hijos tenían que abandonarlos a los ancianos a un bosque [JA1]their children had to abandon them to the old people in the woods(sus hijos tenían que abandonar a los ancianos en el bosque)‘their children had to abandon their olders in the woods’

32) no quiso abandonarla a ella en el bosque [JA1]3rd p. sing. did not want to abandon her to her in the woods(no quiso abandonarla en el bosque)‘he did not want to abandon her in the woods’

33) por eso también lo quiere a Alex. [JA2]that is why also him 3rd p. sing. loves to Alex(por eso también quiere a Alex)‘That is why she also loves Alex’

34) a ella también lo quiere [JA2]to her also him 3rd p. sing. loves(ella también le quiere)‘he also loves her’


30 We do not think that the possible presence of arbitrary null objects in Spanish – (a) or(b) below – could be taken as relevant input.

a. Ese pescado __ envenena __ en tres minutosThat fish __ poisons __ within three minutes‘By eating that fish you can get poisoned within three mintues’

b. Esa luz (le) ciega (a uno)That light (to him) blinds (to one)‘That light blinds you’

These instances of clitic-doubling are redundant or ungrammaticalin the case of native Spanish.31 In fact, very advanced French andEnglish speakers fail to use clitic doubling when it is obligatory innative Spanish.32

It should be pointed out that all cases of object duplication are[+human], which is the type of clitic-doubling favoured in Spanish.It is precisely the [+human] objects which are not favoured inGerman and other languages which allow null objects.33 We wouldlike to hypothesize that, in this, Spanish IL identification takes placevia f-features rather than via null topics. The AGR features – in thiscase the object agreement features – act as identifiers. However, theactual paradigm may be difficult to internalize because there is notonly gender but this is also case-marking distinction. Furthermore,the actual input may be misleading because different speakers havedifferent case-markings depending on the variety of Spanishspoken. In (34) there is confusion with respect to gender (lo insteadof la). Thus, tonic objects are copied systematically when clitics arenot given a semantic interpretation. This could also explain theoverproduction of subject pronouns in the case of the Japanesespeakers. These speakers achieved similar results as the otherspeakers in terms of their overall performance in the placementtest, but the structure of their IL seems to have reached a moreadvanced stage.

f Topic-drop versus pro-drop group An analysis of variance wasconducted on the proportion of null subjects from the total numberof clauses (matrix or subordinate) produced by the advancedintermediate subjects. Neither the main effect of Group, nor theSyntactic Position per Group interaction or the effect of syntacticposition were statistically significant. A post-hoc Bonferroni testrevealed no significant difference between the experimental groupsand the native-speaker control group (Table 14). It is interesting tonote that the results of the French group are almost statisticallysignificant, since the French group produced many more null


31 Clitic-doubling of direct objects occurs in the Spanish of Rio de la Plata but these subjectshad studied Spanish in Spain.32 Clitic-doubling is obligatory in Spanish in examples such as those in (a) and (b) below:

a. Le di el libro a él / *Di el libro a élHim gave the book to him‘I gave the book to him’

b. A nosotros siempre nos dan las gracias / *A nosotros siempre dan las graciasTo us always us (they) give the thanks‘They always thank us’

33 This is the case of German that we have mentioned above, and it is also the case oflanguages such as Akan (Saah, 1994).

subjects than the rest of the groups. However, a close look at Figure3 shows that both the French group and the Spanish groupproduced less null subjects in subordinate than in matrix clauses.This is also the case for the English group. The opposite occurs inthe case of the L1 [+topic-drop] L1 groups (German, Chinese andJapanese), since all three produced more null subjects in matrixthan in subordinate clauses; this result is as expected.

An analysis of variance was conducted on the proportion ofsubject pronouns from the total number of clauses (matrix orsubordinate) produced by the subjects. Neither the main effect ofGroup, or the Syntactic Position per Group interaction, or the effect


Table 14 Null subjects: advanced intermediate stage


German 15.333 42.523 .2778Japanese 17.333 42.523 .2230English 19.667 42.523 .1705Chinese 23.667 42.523 .1048French 35.000 42.523 .0234

NotesBonferroni/Dunn (Control). Effect: Language. Error term: Type III sums of squaresfor subject (Group) Dependent: syntactic position. Significance level: 0.05

matrix

embedded

NotesInteraction plotEffect: syntactic position/LanguageDependent: syntactic position

Figure 3 Null subjects: advanced intermediate stage

of syntactic position were statistically significant. A post-hocBonferroni test revealed no significant difference between theexperimental groups and the native-speaker control group (Table15). Note that the results for the Japanese group are close to beingstatistically significant. This seems to confirm our findings inrelation to the overproduction of object pronouns in the case ofJapanese speakers. As shown in Figure 4, production of subjectpronouns seems to follow the Spanish pattern. Japanese speakersseem to be different from Chinese speakers in that they chooseovert subject pronouns as identifiers, which may indicate that these


NotesInteraction plotEffect: Category for syntactic position/LanguageDependent: syntactic position

Figure 4 Subject pronouns: advanced intermediate stage

matrix

embedded

Table 15 Subject pronouns: advanced intermediate stage


German –6.333 22.096 .3840Japanese –2.000 22.096 .7802English –1.833 22.096 .7981Chinese –1.333 22.096 .8523French 17.330 22.096 .0293

NotesBonferroni/Dunn (Control). Effect: Language. Error term: Type III sums of squaresfor subject (Group) Dependent: syntactic position. Significance level: 0.05

two languages are not similar with respect to the [topic-drop]parameter.

V Conclusion

We have argued that interlanguages (ILs) are not instances ofcoherent parametric options as native languages are. This wouldexplain the systematic presence of subject pronouns but the scarcepresence of object pronouns in the IL of the speakers of topic-droplanguages, as well as the non-native value of overt subject pronouns.Further evidence comes from other properties attributed to thenull-subject parameter. For instance, from the special status ofsubject/verb inversion in the Spanish IL (Liceras, 1988; 1989; 1994)and from the lack of morphological consistency which occurs withnull subjects, pronominal subjects and nominal subjects, and doespersist at this advanced intermediate level (Liceras, Díaz andMaxwell, 1998).

We have also argued that the production of null and subjectpronouns in the early and advanced Spanish interlanguages that wehave analysed is best explained by assuming that the adult ILgrammar has a default licensing procedure which is responsible forthe production of null subjects provided they are identified. Wehave argued that this must be the case for the following reasons:

1) The grammar of our interlanguage speakers does not provideevidence for a simplification procedure which would relax theROOT=CP, thus allowing null subjects only in Spec-rootpositions.

2) Neither the early nor the advanced interlanguage provideevidence of a clear-cut transfer of the licensing/identificationprocedures which forbid null subjects in French, English andGerman but allow them in Chinese and Japanese (the fact thatthe French beginners do not differ from the rest of theinterlanguage speakers provides further support for positing adefault licensing procedure). In fact, we have proposed that it isthe identification procedure that undergoes re-structuring; forinstance, from the discourse identification procedure which leadsto the production of ambiguous sentences in the case of theJapanese speakers, they move to a subject-pronoun identificationprocedure which,in turn,leads to an overuse of subject pronounsthat resembles the clitic-doubling strategy found in the case ofthe object pronouns.

3) While licensing could not explain the individual differences thatwe have found in the production of subject pronouns, these


differences can be accounted for if we assume that they are theresult of the identification procedures chosen by the individualsubjects.

4) These data do not provide evidence for proposing a defaultoption which is consistent with the licensing/identificationprocedures of neither the subjects’ native languages nor forSpanish. If Spanish were in fact the default procedure, thesesubjects would not produce null subjects nor subject pronounswhich are not possible in native Spanish.

All the above points to the presence of local re-structuringoptions. Such a procedure would also account for the fact that theIL subject pronouns may have a clitic-like structure in someconstructions but an English-like structure in other constructions.The spontaneous data that we have analysed provides evidenceneither for the clitic status of the subjects pronouns produced bythe French speakers or any other speaker, nor against it. However,as we have indicated above, subject pronouns may be analysed asidentifiers without having a clitic status.

It could be argued that counting null and overt subjects inspontaneous speech does not provide an insight into the speakers’competence; however, it does provide an insight into actualdiscourse patterns. While the same could be said in the case of manyL1 studies, we would like to emphasize that our counting has onlybeen used to differentiate matrix versus subordinate clauses, and asa point of comparison between the production of null and overtsubject pronouns. In fact, as in the case of early grammars, we wouldlike to emphasize that ‘naturalistic data is used only as a source ofclues through which to develop a model that addresses the primaryquestion: how is acquisition possible under any assumptions?’(Roeper, 1993: 75). There is not doubt that besides longitudinal datawe need data from specific experimental tests to achieve a betterunderstanding of the nature of null arguments as well as overtpronouns (subjects and objects) in the Spanish ILs.

Acknowledgements

Earlier versions of this paper were presented at GURT 1995(Georgetown, WA), GASLA 1995 (CUNY, New York) and LARS1995 (Universiteit Utrecht, Utrecht). This research has beensupported by grants from the Social Sciences and HumanitiesResearch Council (SSHRC) of Canada, the Secretary of State forMulticulturalism–Canadian Heritage (Canada), the Faculty of Artsof the University of Ottawa, Ottawa, the Delegación General de


Ciencia y Tecnología (DGCYT) of the Ministry of Education ofSpain, and the Faculty of Translation of the Universitat PompeuFabra, Barcelona. We would like to thank the students from theEscuela Central de Idiomas of Barcelona, the University of Ottawaand the Lycée Claudel of Ottawa for participating in the study. Weare specially grateful to Begoña Soloaga, the Spanish teacher fromthe Lycée Claudel, for her enthusiastic and unconditional help withour research projects. This paper has greatly benefited from thecomments and suggestions of two anonymous SLR reviewers.Needless to say, only we are responsible for all remaining errorsand/or inconsistencies.

VI References

Almoguera, R. and Lagunas, C. 1993:El parámetro pro-drop en el españolno nativo. In Liceras, J.M., editor, La lingüística y el análisis de lossistemas no nativos, Ottawa: Dovehouse, 140–49.

Bickerton, D. 1990: Language and species. Chicago: The University ofChicago Press.

Birdsong, D. 1992: Ultimate attainment in second language acquisition.Language 68, 706–55.

Bley-Vroman, R. 1990: The logical problem of foreign language learning.Linguistic Analysis 20, 1–49.

Borer, H. 1984: Parametric syntax: case studies in Semitic and Romancelanguages. Dordrecht: Foris.

Brown, R. 1973. A first language:the early stages. Cambridge, MA:HarvardUniversity Press.

Chomsky, N. 1965: Aspects of the theory of syntax. Cambridge, MA: MITPress.

–––– 1981: Lectures on government and binding. Dordrecht:Foris.

–––– 1986: Knowledge of language: its nature, origin, and use. New York:Praeger.

–––– 1991: Some notes on economy of derivation and representation. InFreidin, R., editor, Principles and parameters in comparative grammar.Cambridge, MA: MIT Press, 417–54.

–––– 1995: The minimalist program. Cambridge, MA: MIT Press.Clahsen, H. and Muysken, P. 1986: The availability of Universal Grammar

to adult and child learners:a study of the acquisition of German wordorder. Second Language Research 5, 1–29.

Clark, R. and Roberts, I. 1993: A computational model of languagelearnability and language change. Linguistic Inquiry 24, 299–345.

Díaz, L. and Liceras, J.M. 1990: Formulación de parámetros y adquisiciónde lenguas extranjeras. In Martín Vide, C., editor, Proceedings of theV Congress and Natural and Formal Languages. Barcelona:Universidad de Barcelona, PPU.


–––– 1995: Las gramáticas de la interlengua española. CuadernosHispánicos de Amsterdam 17, 135–56.

du Plessis, J., Solin, D., Travis, L. and White, L. 1987: UG or not UG, thatis the question: a reply to Clahsen and Muysken. Second LanguageResearch 3, 56–75.

Felix, S. and Weigl, W. 1991. Universal Grammar in the classroom: theeffects of formal instruction on second language acquisition. SecondLanguage Research 7, 162–81.

Flynn, S. 1987: A parameter-setting model of L2 acquisition: experimentalstudies in anaphora. Dordrecht: Reidel.

Haegeman, L. 1990: Non-overt subjects in diary contexts. In Mascaró, J.and Nespor, M.editors, Grammar in progress. GLOW essays for Henkvan Riesdijk . Dordrecht: Foris.

Hilles, S. 1986: Interlanguage and the pro-drop parameter. SecondLanguage Research 2, 33–52.

–––– 1991:Access to Universal Grammar in second language acquisition.In Eubank, L., editor, Point/Counterpoint. Universal Grammar in thesecond language . Amsterdam: John Benjamins, 305–38.

Huang, C.L. 1984: On the distribution and reference of empty pronouns.Linguistic Inquiry 15, 531–74.

Hyams, N. 1986: Language acquisition and the theory of parameters.Dordrecht: Reidel.

–––– 1989: The null subject parameter in language acquisition. In Jaeggli,O. and Safir, K.,editors, The null subject parameter. Dordrecht:Reidel.

–––– 1994:V2 null arguments and COMP projections. In Hoekstra, T. andSchwartz, B., editors, Language acquisition studies in generativegrammar. Amsterdam: John Benjamins, 21–55.

Hyams, N. and Wexler, K. 1993: On the grammatical basis of null subjectsin child language. Linguistic Inquiry 24, 421–59.

Jaeggli, O. 1982: Topics in Romance syntax. Dordrecht: Foris.Jaeggli, O. and Safir, K. 1989: The null subject parameter. Dordrecht.

Reidel.Jin, H.G. 1994: Topic-prominence and subject-prominence in L2

acquisition: evidence of English-to-Chinese typological transfer.Language Learning 44, 101–22.

Karmiloff-Smith, A. 1994: Innate constraints and developmental change.In Bloom, P., editor, Language acquisition. Cambridge, MA: MITPress, 563–90.

Lakshmanan, U. 1991: Morphological uniformity and null subjects in childsecond language acquisition.In Eubank,L.,editor, Point/counterpoint.Universal Grammar in the second language. Amsterdam: JohnBenjamins, 389–410.

Lasnik, H. and Stowell, T. 1991: Weakest crossover. Linguistic Inquiry 22,687–720.

Lebeaux, D. 1988: Language acquisition and the form of grammar. PhDdissertation. University of Massachusetts, Amherst, MA.

Liceras, J.M. 1985: The value of clitics in non-native Spanish. SecondLanguage Research 1, 4–36.


–––– 1986: Linguistic theory and second language acquisition: the Spanishnonnative grammar of English speakers. Tübingen: Gunter Narr.

–––– 1988: Syntax and stylistics: more on the pro-drop parameter. InPankhurst, J., Sharwood-Smith, M. and Van Buren, P. editors,Learnability and second languages, Dordrecht: Foris.

–––– 1989: On some properties of the pro-drop parameter: looking formissing subjects in non-native Spanish. In Gass, S. and Schachter, J.,editors, Linguistic perspectives in second language acquisition.Cambridge, MA: Cambridge University Press.

–––– 1994: La teoría gramatical y los principios que regulan la adquisicióndel orden de palabras del español. Paper read at the Primera MesaRedonda de Lingüística Española. Universidad AutónomaMetropolitana de México, México.

–––– forthcoming: La adquisición del lenguaje: a la búsqueda de losprincipios y mecanismos que regulan la adquisición del lenguaje nonativo. Madrid: Visor.

Liceras, J.M. and Díaz, L. 1998. On the nature of the relationship betweenmorphology and syntax:inflectional typology, f-features and null/overtpronouns in Spanish interlanguage. In Beck,M.L.,editor, Morphologyand its interfaces in second language knowledge. Amsterdam: JohnBenjamins, 307–37.

Liceras, J.M., Díaz, L. and Maxwell, D. 1998: Null arguments in non-nativegrammars: the Spanish L2 of Chinese, English, French, German,Japanese and Korean speakers. In Klein, E. and Martohardjono, G.,editors, The development of second language grammars. A generativeapproach. Amsterdam: John Benjamins, 113–49.

Liceras, J.M., Díaz, L., Maxwell, D., Laguardia, B., Fernández, Z. andFernández, R. 1997: A longitudinal study of Spanish non-nativegrammars: beyond parameters. In Pérez-Leroux, A. and Glass, W.,editors, Contemporary perspectives on the acquisition of Spanish.Sommerville: Cascadilla Press, 99–132.

Liceras, J.M., Laguardia, B., Fernández, Z., Fernández, R. and Díaz, L.1998. Licensing and identification of null categories in Spanish non-native grammars. In Lema, J. and Treviño, E., editors, TheoreticalAnalyses on Romance Languages, Amsterdam: John Benjamins,263–82.

Liceras, J.M., Valenzuela, E. and Díaz, L. 1998. Morphological paradigmsand syntax: null and overt arguments in L1 and L2 acquisition ofSpanish. Proceedings of GASLA ’97. McGill Working Papers inLinguistics, McGill University, Montreal.

Lightfoot, D. 1991: How to set parameters: arguments for language change.Cambridge, MA: MIT Press.

Martohardjono, G. and Gair, J. 1993. Apparent inaccessibility insecond language acquisition: misapplied principles or principlemisapplications? In Eckman, F., editor, Confluence: linguistics, L2acquisition, and speech pathology. Amsterdam: John Benjamins,79–104.

O’Grady, W. 1991: Language acquisition and the pro-drop phenomenon:


a response to Hilles. In Eubank, L., editor, Point/counterpoint.Universal Grammar in the second language. Amsterdam: JohnBenjamins, 339–49.

Ouhalla, J. 1993: Functional categories and parametric variation. London:Routledge.

Perlmutter, D. 1971: Deep and surface structure constraints in syntax. NewYork: Holt, Reinhart and Winston.

Phinney, M. 1987: The pro-drop parameter in second language acquisition.In Roeper, Th. and Williams, E., editors, Parameter setting. Dordrecht:Reidel, 221–38.

Piatelli-Palmarini, M. 1989: Evolution, selection, and cognition: from‘learning’ to parameter setting in biology and the study of language.Cognition 31, 1–44.

Pierce, A. 1992: Language acquisition and syntactic theory. Dordrecht:Kluwer.

Pinker, S. 1994: The language instinct. New York: William Morrow.Pollock, J.-Y. 1989: Verb movement,UG and the structure of IP. Linguistic

Inquiry 20, 365–424.Register, N. 1990: Influences of typological parameters on L2 learners’

judgments of null pronouns in English. Language Learning 40,369–85.Rizzi, L. 1982: Issues in Italian syntax. Dordrecht: Foris.–––– 1986: Null objects in Italian and the theory of pro. Linguistic Inquiry

17, 501–57.–––– 1994: Early null subjects and root null subjects. In Hoekstra, T. and

Schwartz, B., editors, Language Acquisition Studies in GenerativeGrammar. Amsterdam: John Benjamins, 151–76.

Roberge, Y. 1986: Subject doubling, free inversion and null argumentlanguages. Canadian Journal of Linguistics 31, 54–79.

Roeper, Th. 1993: The ‘least-effort’ principle in child grammar: choosing amarked parameter. In Reuland, E. and Abraham, W., editors,Knowledge and language I: From Orwell’s problem to Plato’s problem.Dordrecht: Kluwer, 71–104.

Roeper, Th. and Rohrbacher, B. 1994. Null subjects in early child Englishand the theory of economy of projection. University of PennsylvaniaWorking Papers in Linguistics 2, 83–119.

Roeper, Th. and Weissenborn, J. 1990: How to make parameters work. InFrazier, L. and de Villiers, J., editors, Language processing andlanguage acquisition . Dordrecht: Kluwer, 147–62.

Saah, K. 1994: Language acquisition and language processing in Akan.PhD dissertation. University of Ottawa.

Schachter, J. 1988: Second language acquisition and its relationship toUniversal Grammar. Applied Linguistics 9, 219–35.

Speas, M. 1994. Null arguments in a theory of economy of projection.University of Massachusetts Occasional Papers in Linguistics 17,179–208.

Sportiche, D. 1996. Clitic constructions. In Rooryck, J. and Zaring, L.,editors, Phrase structure and the lexicon. Dordrecht: Kluwer, 213–76.

Taraldsen, K.T. 1980: On the nominative island condition, vacuous


application and the that-t race filter. Bloomington, IN: IndianaLinguistics Club.

Tsimpli, I.-T. and Roussou, A. 1991: Parameter-resetting in L2? UCLWorking Papers in Linguistics 3, 149–69.

Valian, V. 1991. Syntactic subjects in the early speech of American andItalian children. Cognition 40, 21–81.

Wang, Q. et al. 1992: Null subject versus null object: some evidence fromthe acquisition of Chinese and English. Language Acquisition 2:221–54.

White, L. 1985: The pro-drop parameter in adult second languageacquisition. Language Learning 35, 47–62.

–––– 1990: Another look at the logical problem of foreign languagelearning: a reply to Bley-Vroman. Linguistic Analysis 20, 50–63.

White, L. and Genesee, F. 1996. How native is near-native? The issue ofultimate attainment in adult second language acquisition. Studies inSecond Language Acquisition 12, 233–65.


Topic-drop versus pro-drop: null subjects and pronominal...

Documents

Transcript of Topic-drop versus pro-drop: null subjects and pronominal...