How tolerant is universal grammar?: essays on language learnability and language variation 9783111634777, 9783484303096

Over the past few decades, the book series Linguistische Arbeiten [Linguistic Studies], comprising over 500 volumes, has

237 74 16MB

English Pages 329 [336] Year 1994

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Introduction
Raising Questions: Formal and Functional Aspects of the Acquisition of Wh-Questions in German
Possible Domains for Individual Variation in Early Developmental Stages
Variations on “Variation”: On the Acquisition of Complementizers in German
Learnability Meets Development: The Case of Pro-Drop
A Theory of Null Objects and the Development of a Brazilian Child Grammar
Activating Passives in Child Grammar
Variation in Grammar and First Language Acquisition: A New Concept of Parameter in Universal Grammar
Intermodular Synchronization: On the Role of Morphology in the Normal and Impaired Acquisition of a Verb-Second Language
How Do Children Cope with Variation in the Input? The Case of German Plurals and Compounding
Variation in the Acquisition of German Plural Morphology by Second Language Learners
Inference and Learnability in Second Language Acquisition: Universals vs. Language-Specific Phenomena in the Domain of Idiomatic Expressions
List of Contributors
Recommend Papers

How tolerant is universal grammar?: essays on language learnability and language variation
 9783111634777, 9783484303096

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Linguistische Arbeiten

309

Herausgegeben von Hans Altmann, Peter Blumenthal, Herbert E. Brekle, Gerhard Heibig, Hans Jürgen Heringer, Heinz Vater und Richard Wiese

How Tolerant Is Universal Grammar? Essays on Language Learnability and Language Variation

edited by Rosemarie Tracy and Elsa Lattey

Max Niemeyer Verlag Tübingen 1994

Die Deutsche Bibliothek - CIP-Einheitsaufnahme How tolerant is universal grammar? : Essays on language learnability and language variation / ed. by Rosemarie Tracy and Elsa Lattey. - Tubingen : Niemeyer, 1994 (Linguistische Arbeiten ; 309) NE: Tracy, Rosemarie; GT ISBN 3-484-30309-3

ISSN 0344-6727

© Max Niemeyer Verlag GmbH & Co. KG, Tübingen 1994 Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Printed in Germany. Druck: Weihert-Druck GmbH, Darmstadt Einband: Hugo Nadele, Nehren

Table of Contents

Introduction Rosemarie Tracy Raising Questions: Formal and Functional Aspects of the Acquisition of WA-Questions in German

vii

1

Zvi Penner Possible Domains for Individual Variation in Early Developmental Stages

35

Franz-Josef d'Avis & Petra Gretsch Variations on "Variation": On the Acquisition of Complementizers in German

59

Maaike Verrips Learnability Meets Development: The Case of Pro-Drop

Ill

Mary Aizawa Kato A Theory of Null Objects and the Development of a Brazilian Child Grammar

125

Agnes Fritzenschaft Activating Passives in Child Grammar

155

Beatrice Primus & Katrin Lindner Variation in Grammar and First Language Acquisition: A New Concept of Parameter in Universal Grammar

185

Chris Schaner-Wolles Intermodular Synchronization: On the Role of Morphology in the Normal and Impaired Acquisition of a Verb-Second Language

205

Ira Gawlitzek-Maiwald How Do Children Cope with Variation in the Input? The Case of German Plurals and Compounding

225

Heide Wegener Variation in the Acquisition of German Plural Morphology by Second Language Learners

267

Elsa Lattey Inference and Learnability in Second Language Acquisition: Universale vs. Language-Specific Phenomena in the Domain of Idiomatic Expressions

295

List of Contributors

315

Introduction Language variation is a topic that has received much attention and been the subject of a great deal of fruitful research in the field of sociolinguistics. That different ways of expressing oneself function in delimiting groups in geographic space and in society has long been recognized. These ways of expression are subsumed under such terms as dialect and sociolect. Even individual characteristics of style and expression have been acknowledged and given status by being named: thus, the term idiolect. It is clear that children acquire language within a particular geographic space, i.e. they grow up speaking a particular dialect. It is also obvious that their environment includes particular social groupings, which give rise to their sociolect. Given sufficient attention, a learner's speech will reveal idiosyncratic features as well, certainly in the phonological domain, and often also in syntax and lexicon. What has received considerably less attention thus far than the individuality apparent in the acquired adult language variety is the individuality in the path taken to get there. It is this developmental variation that is of concern to the authors of the articles in this book. They see variation not just as a sociolinguistic fact but as a call for psycholinguistic explanations. Since the paths children take on their way to the target grammar are determined by what they contribute to this task (their linguistic knowledge and problem-solving or learning strategies) plus the input they receive, a careful look at developmental data should provide independent evidence for theoretical claims about possible grammars. The title of our book asks to what extent principles of Universal Grammar (henceforth also UG) constrain intermediary grammars and thereby shape possible acquisition paths. The question as it is phrased here arose in the context of a linguistic paradigm which explicitly attributes particular psychological relevance to the axioms of linguistic theory (cf. Chomsky 1965): In this framework universale are interpreted as the predisposition which imposes a priori restrictions on the learner's hypothesis space and provides him/her with concepts that make acquisition possible. While the existence of universal constraints can explain - in principle - why the grammars (target and interim grammars) of learners of widely differing languages and cultures exhibit remarkable similarities, it still leaves the problem of variation (especially in children acquiring the same target) unaccounted for. There is no question that most investigators of language acquisition would like all children to go through the same stages on their way to adult competence, with each stage clearly delimitable and the overall sequence, except for individual speed, invariant. For ease of analysis, we would also like children to produce sufficient numbers of minimal pairs and to keep their thumbs or candies out of their mouths while they talk into our microphones. To make our lives easier yet, we would also like them to be conservative and to refrain from overgeneralizations. Needless to say, this is not what we find. Although these wishes all refer to problems that confront researchers at various stages of their investigations, they are clearly not all of the same type. Some are troublesome, but theoretically irrelevant. The fact that children temporarily overgeneralize (though not all in the same way) and subsequently recover from it without the help of systematic negative feedback is, however, a serious challenge for any current acquisition theory and has received due attention (cf. Bowerman 1988, Clark 1987, Pinker 1989, Randall 1990). On the other hand, the fact that our common-sense assumption about children's going through identical and

viii

Κ Lattey & R. Tracy

well-definable stages is put to a hard test by observed variation has been given less consideration, and those who have addressed the issue have focused mainly on very early developmental strategies (cf. Bloom et al. 1975, Peters 1977, Kaltenbacher 1990). In this respect, then, psycholinguistics has yet to catch up with developmental psychology, where stage definition and the issue of falsification have been discussed intensively (cf., for instance, Flavell 1980, Brainerd 1978). The hypothesis of clearly delimitable and interindividually invariant stages, while intuitively appealing and quite useful to start with, may well stand in the way of better ideas. Progress with respect to these issues requires careful weighing of qualitative and quantitative aspects. Two considerations are particularly important. There is, first of all, the problem of the accident of sampling: The smaller the developmental steps, the easier it is to miss them. What is more: We have to take into account the possibility of what Roeper called the "silent stage", i.e. Much of the acquisition of grammar is covert. [...] Therefore it is not unnatural to use apparently minor overt signs of systematicity as possible indications of fundamental underlying principles. The primary argument in behalf of an acquisition theory is its intrinsic logic and deductive power, not the claim that every predicted stage is overt and open to inspection by researchers. The goal, therefore, is to develop a theory which can, in principle, articulate the micro-structure of acquisition decisions. (1992:340) It is the major tenet of this book that in looking for developmental steps, variation is not just noise to abstract from. But taking variation seriously does not mean giving up the search for invariance or for central tendencies. Conversely, it also does not mean holding UG or the individual grammars derived from it responsible for every utterance. Taking variation seriously means searching for restrictive theories, for theories of UG as well as for complementary theories of processing and learning, that explain, rather than ignore, uncomfortable data. Indeed, there may be a lot to be gained from apparently messy data, such as a deepened understanding of the dynamics of the overall developmental process or a clearer view of the interaction of the child's competence with factors of performance. We believe that language acquisition research is ready for this empirical and theoretical challenge. Just about twenty years ago, Roger Brown wrote in his preface to A First Language (1973: xi): "Developmental psycholinguistics is today a very lively and promising research field. It is my deepest wish for it that it will leave behind a clear increment to psychological knowledge." In the twenty years since, language acquisition research has indeed continued to flourish. If Brown were to attempt another survey today, just a listing of investigations into what he called "the early stages" would fill many pages. This list would contain languages for which, at the time, acquisition data were underrepresented (like German, which is amply represented in this book) or completely non-existent (like Portuguese, which is discussed here in the paper by Kato). Simply in terms of empirical coverage, then, Brown's wish for an increment in knowledge has come true, his own ideas contributing to this in substantial ways. To this day investigators of child language compare their data to "Brown's stages" and use his procedure for establishing mean length of utterance values for assessing the child's overall linguistic level. The distinction he drew between the early development of grammatical relations and the subsequent acquisition of closed-class morphemes foreshadowed the recent

Introduction

ix

distinction between a lexical-thematic and a functional stage (cf. Radford 1990). In addition, Brown's acquisition criteria, which take into account several types of "contextual obligation" (cf. 1973:260), have set methodological standards. He also pointed out that there are phenomena for which obligatory contexts cannot be easily calculated. This is, for instance, the case for some of the data discussed in this book, like passives, subordinate clauses, questions, and idioms. What is most important for our current concerns: Brown repeatedly drew attention to the extent of variability encountered in the data and to the gradual way in which children reach his acquisition criterion, at least for grammatical morphemes (90% or better of occurrence in obligatory contexts in six consecutive sampling hours, cf. 1973:12). Of course, when Brown expressed his wish for an increase in knowledge, he did not simply have in mind a wider range of cross-linguistic data but a better understanding of what makes language acquisition "tick," as it were, i.e. what makes development possible and determines its direction. Here the gain in our knowledge is harder to determine. After all, what one is prepared to acknowledge as a "clear increment" depends on one's own theoretical perspective and also on one's personal "tolerance in matters epistemological," to quote from the title of an article by Feyerabend (1968:12). One gain in knowledge on which most investigators, regardless of theoretical bias, would probably agree is a greater understanding of what has become known as the learnability problem: How is it that we know so much on the basis of so little and despite the lack of systematic negative evidence (corrections, misunderstanding, etc.)? The overgeneralizations and subsequent retreat (without systematic correction) that we mentioned above are part of the issue. The twenty years that separate this book from Brown's have seen rapid development within linguistics: changes in what is attributed to adult competence (the targets to be acquired) and changes in the axioms of linguistic theory. Universal Grammar of the Standard Theory type (cf. Chomsky 1965) and its revisions gave way to modular theories, where the well-formedness of sentences is determined by the constraints imposed by principles of independent modules, and where the grammars of target languages differ along constrained parameters (cf. Chomsky 1981, Chomsky & Lasnik 1991). This led to a view of language acquisition as the fixing of the parameters of a constrained hypothesis space, with languagespecific values "triggered" by input (cf. Roeper & Williams 1987). Also, as a consequence of the advances in linguistic theory, the data from earlier language acquisition studies - like the data that Brown had already re-cast on the basis of different frameworks in his book - were re-interpreted in a new light, for instance in terms of X-bar theory, the approach adopted by many articles in this book. New questions arose in the course of these events. Those who no longer needed to be convinced that some help from UG was necessary to guarantee learnability asked whether UG principles are continuously available, constraining each intermediary system of linguistic knowledge, or whether there are temporary states that fall outside UG because universale have to mature (cf. Felix 1992) or need to be activated ("triggered") along the way. There is a whole continuum of possible answers, and there is no reason why only one answer should hold across the board, i.e. for all universals. And then there are questions that cut across acquisition types, e.g.: Are universals still accessible to second language learners (cf. White 1989)? Or, if grammars are organized in a modular fashion, can we predict selective deficits in individual modules (cf. Clahsen 1988)?

χ

Κ Lattey & R. Tracy

One intriguing learnability puzzle became known as the "bootstrapping" problem (cf. Pinker 1984): Since UG principles operate on abstract representations, we have to account for how learners could possibly cut through the continuous stream of speech they hear and establish those initial representations that reach the necessary degree of abstractness for UG to apply in the first place. The bootstrapping problem calls for explaining not just the availability of principles but their accessibility. This also touches on the question of what additional principles and mechanisms are needed beyond UG. That is, what help does UG in turn need from other cognitive systems? What evidence does it need from the input in order to function? Other questions arise: If language acquisition indeed involves parameter setting, how much evidence is needed for the learner to identify the correct value for his/her target language? Are there default settings? Is it possible that parameters can be reset? All these issues await further research. That is why our question with respect to what UG will allow is a real, open question. With this background in mind, we can now turn to the current volume. Most of the ideas discussed in the individual contributions were in one way or another presented in a workshop - with the same title as the book - which was held at the 1992 conference of the Deutsche Gesellschaft für Sprachwissenschaft (German Linguistics Society) in Bremen, the overall theme of which was "Linguistic Universals." The papers and the discussion centered on one or more of the following issues: (a)

Inter-individual variation: How can we reconcile our belief in restrictive and, moreover, domain-specific a priori knowledge with the observation that children proceed along different paths? How theoretically relevant are these differences? Can we identify strategies that manifest themselves again and again in the way learners tackle different aspects of the overall acquisition task?

(b)

Intra-individual variation: How can we account for apparent inconsistencies, i.e. for the spectrum of co-existing structures, the entropy, within individual learner systems at one time? How can we reconcile the observation of gradual change with the idea of abrupt changes predicted by most parameter-setting or rule-based approaches?

(c)

Different types of acquisition: e.g. second language acquisition in childhood and in adults, and development in language-impaired children. Are there acquisition strategies which hold across first and second language learners?

The first three papers of this volume deal with phenomena which have received a lot of attention lately: word order and properties of main and subordinate clauses in V2 languages like German (including Bernese Swiss German, in the case of Penner). The contribution by Rosemarie Tracy, "Raising Questions: Formal and Functional Aspects of the Acquisition of WA-Questions in German," focuses on intra-individual variation. She sketches the development of constituent questions from early gap formats without wA-operators, through verb-end clauses with wA-operators, to adult-like verb-second sentences and suggests systeminternal conflicts which could lead to restructuring. We decided to start the book with this paper because it also raises questions concerning the more general issue of competing linguistic analyses of a target system. In his paper "Possible Domains for Individual Variation in Early Developmental Stages," Zvi Penner presents a theory which ties inter-individual variation to the emergence of the CP

Introduction

xi

level. He discusses various asymmetries between root and subordinate clauses as a syntactic bootstrapping device and suggests a three-stage model for the acquisition of COMP in V2 languages. The contribution by Franz-Josef d'Avis & Petra Gretsch, "Variations on 'Variation': On the Acquisition of Complementizers in German," deals with variation in both a more general and a concrete sense. As for the latter, they argue, like Penner and Tracy, that the left periphery of subordinate clauses becomes accessible gradually. The authors develop a building-block model that predicts a limited number of developmental strategies and test it against data from four monolingual German children. They also propose a formal definition of developmental sequence. The paper by Maaike Verrips, "Learnability Meets Development: The Case of Pro-Drop," also confronts us with the conflict between the (apparent) gradual nature of change on the one hand and the abruptness implicit in parameter-setting approaches on the other. She proposes a model, MAX (maximize input), which supplements UG: "For every input string, create as many UG-allowed representations as possible," and she shows how the learner could solve the problem of subsequent retreat. While Verrips illustrates her model with subject-drop data, which have received a lot of attention in language acquisition research, Mary Kato develops a complementary theory of null objects, which have been neglected for adult grammar as well as for child language. In "A Theory of Null Objects and the Development of a Brazilian Child Grammar," Kato identifies various kinds of null objects in Brazilian Portuguese (null name, null pronominal, variable, and null VP) - considering both synchronic and diachronic data - and traces the order in which they are acquired by children, sketching a cumulative process of acquisition. The paper by Agnes Fritzenschaft, "Activating Passives in Child Grammar," also takes up a phenomenon for which, to this day, only few longitudinal case studies are available. She draws attention to criticial differences between the passive in English and the passive in German and claims that the emergence of the latter crucially depends on the acquisition of the auxiliary paradigm, not - as is generally assumed to be the case for English - on properties of the participles. In the particular case study presented, passives are bootstrapped into the system by early single-argument ergatives. In "Variation in Grammar and First Language Acquisition: A New Concept of Parameter in Universal Grammar," Beatrice Primus & Katrin Lindner propose a Generalized Hierarchy Approach as a parameter of its own. The values of this hierarchy arise from the various correlations of mutually irreducible relational concepts (case, thematic relations, grammatical relations), which are organized on hierarchies of their own. Variation among children or between children and adults (or among different languages) is a function of different correlations. The paper by Chris Schaner-Wolles, "Intermodular Synchronization: On the Role of Morphology in the Normal and Impaired Acquisition of a Verb-Second Language," returns to the acquisition of German main clauses in unimpaired children and compares this development with learners who suffer from Down's syndrome. She shows that there is a considerable amount of variation in the unimpaired children, i.e. they produce non-finite V2 structures as well as finite VE structures in main clauses. What happens in the Down's syndrome children is seen not as a qualitative difference but as a "disharmony in synchronization" of different modules.

Ε. Lattey & R. Tracy

xii

Two papers deal with morphology proper, and with a domain which has often been considered particularly difficult or downright intractable in terms of rule formation: the German plural. Ira Gawlitzek-Maiwald asks "How Do Children Cope with Variation in the Input? The Case of German Plurals and Compounding." She tests the model of a three-level lexicon against data from German-speaking adults and children, points out various shortcomings in it, and suggests improvement via the postulation of gender-based plural rules. In "Variation in the Acquisition of German Plural Morphology by Second Language Learners," Heide Wegener proposes a number of plural rules which simplify the target system. She studies the acquisition of these rules in two different groups of child second language learners (Turkish children and children of Aussiedler families in Germany) and links her results to the manifestation of two strategies, an analytic-cognitive and a holisticformulaic strategy. The very last paper, by Elsa Lattey, raises an issue which in both first and second language acquisition awaits more explicit investigation: the acquisition of idiomatic expressions. "Inference and Learnability in Second Language Acquisition: Universals vs. LanguageSpecific Phenomena in the Domain of Idiomatic Expressions" reveals a number of subtle syntactic, semantic, and pragmatic aspects which are part of the native speaker's knowledge and confront us with the questions of how they were learned and how they can be taught to second language learners. What unites these papers is the belief that investigating variation is a non-trivial pursuit and that it indeed reveals something about both the dynamics of developmental processes and the restrictiveness of universal grammar. Also, the variety of interests and answers offered throughout the book shed light, we hope, on the general intellectual climate in which acquisition research is currently conducted. Those who find this spectrum of variation bewildering should take comfort. If developmental psycholinguistics has continued to be a lively field, as was wished for by Brown, it is most likely due to this diversity of opinions and perspectives. Moreover, if we trust Feyerabend, we should take this as a rather promising sign for the future, the future of the people involved and the future of the field: Unanimity of opinion may be fitting for a church, for the frightened victims of some (ancient, or modern) myth, or for the weak and willing followers of some tyrant; variety of opinion is a feature necessary for objective knowledge; and a method that encourages variety is also the only method that is compatible with a humanitarian outlook. (1968:33)

* * *

This book took shape in truly modular fashion, its (we hope) overall well-formedness being possible only because of the help of many. We would like to express our gratitude to those who provided the components - the authors - and also to the editors of the series, especially Richard Wiese, for constructive comments. For their intolerance of typing errors and formatting inconsistencies we are indebted to many, especially to Cleo Becker, Diana Gierling, Petra Gretsch, Susanna Herwig, Svenja

xiii

Introduction

Kuhfuß, Dagmar Lalla, Petra Schulz, and Stephanie Schwartz.

Ira Gawlitzek-Maiwald and

Patrick Schindler assisted with the graphics. Production of the final print-out w a s managed by U w e Τ. Ruckgaber of CSR Computers, Tübingen. Finally, w e thank the children from the Tübingen language acquisition project for the drawings, w h i c h add a kind of variety that even those w h o disagree with whatever else is said about variation in this book should be able to enjoy! E. L. & R. T.

Bibliography Bloom, Lois, Patsy Lightbown & Lois Hood (1975): Structure and Variation in Child Language. - In: Monographs of the Society for Research in Child Development 40, Serial No. 160. Bowerman, Melissa (1987): "Commentary: Mechanisms of Language Acquisition." - In: Brian MacWhinney (ed.), 443-466. Bowerman, Melissa (1988): "The 'No-Negative-Evidence' Problem: How Do Children Avoid Constructing an Overly General Grammar? - In: John A. Hawkins (ed.): Explaining Linguistic Universals (Oxford: Blackwell) 73-101. Brainerd, Charles J. (1978): "The Stage Question in Cognitive-Developmental Theory." - In: The Behavioral and Brain Sciences 2, 173-213. Brown, Roger (1973): A First Language: The Early Stages. - Cambridge, Mass.: Harvard University Press. Chomsky, Noam (1965). Aspects of the Theory of Syntax. - Cambridge, Mass.: MIT Press. Chomsky, Noam (1981): Lectures on Government and Binding. - Dordrecht: Foris. Chomsky, Noam & Howard Lasnik (1991): "Principles and Parameters Theory." - Ms. - To appear in: Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld & Theo Vennemann (eds.): Syntax: An International Handbook of Contemporary Research. Clahsen, Harald (1988): Normale und gestörte Kindersprache: Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie. - Amsterdam: Benjamins. Clark, Eve V. (1987): "The Principle of Contrast: A Constraint on Language Acquisition." - In: Brian McWhinney (ed.), 1-33. Felix, Sascha (1992): "Language Acquisition as a Maturational Process." - In: Jürgen Weissenborn, Helen Goodluck & Tom Roeper (eds.): Theoretical Issues in Language Acquisition: Continuity and Change in Development (Hillsdale, N.J.: Erlbaum) 25-51. Feyerabend, Paul K. (1968): "How to Be a Good Empiricist - A Plea for Tolerance in Matters Epistemological." - In: Peter H. Nidditch (ed.): The Philosophy of Science (London: Oxford University Press) 12-39. Flavell, John H. (1980): "Structures, Stages, and Sequences in Cognitive Development." - In: Andrew W. Collins (ed.): Minnesota Symposia on Child Psychology, Vol. 15 (Minneapolis: University of Minnesota Press) 1-28.

xiv

Κ Lattey & R. Tracy

Kaltenbacher, Erika (1990): Strategien beim frühkindlichen Syntaxerwerb. - Tübingen: Narr. MacWhinney, Brian (ed.) (1987): Mechanisms of Language Acquisition. - Hillsdale, N.J.: Erlbaum. Peters, Ann (1977): "Language Learning Strategies." In: Language 53,560-573. Pinker, Steven (1984): Language Learnability and Language Development. - Cambridge, Mass.: Harvard University Press. Pinker, Steven (1989): Learnability & Cognition: The Acquisition of Argument Structure. - Cambridge, Mass.: ΜΓΓ Press. Radford, Andrew (1990): Syntactic Theory and the Acquisition of English Syntax: The Nature of Early Child Grammars of English. - Oxford: Blackwell. Randall, Janet (1990): "Catapults and Pendulums: The Mechanics of Language Acquisition." - In: Linguistics 28, 1381-1406. Roeper, Tom (1992): "From the Initial State to V2: Acquisition Principles in Action." - In: Jürgen Μ. Meisel (ed.): The Acquisition of Verb Placement: Functional Categories and V2 Phenomena in Language Acquisition (Dordrecht: Kluwer) 333-370. Roeper, Tom & Edwin Williams (eds.) (1987): Parameter Setting. - Dordrecht: Reidel. White, Lydia (1989): Universal Grammar and Second Language Acquisition. - Amsterdam: Benjamins.

Rosemarie Tracy

Raising Questions: Formal and Functional Aspects of the Acquisition of WÄ-Questions in German1

1.

Introduction

Children are seldom at a loss for an answer. That goes even for questions which would silence most of us, like the one below: Father (39;10): Lara (2;06):

Now, Lara, what's your view on language acquisition? I ate it all up.

The child's reaction in this episode was not just a clever ploy or default solution in the face of a question which could hardly be considered a serious request for information. She would, with a big smile, give the same response to other questions where lack of understanding was indeed out of the question, i.e. queries concerning the whereabouts of her brother or toys, in each case assuring proper pronominal reference (/ ate him all up vs. I ate them all up). Obviously, Lara knew that questions not only call for the filling of gaps in our knowledge; they are invitations to a game which involves many levels of form and function, and which can be fun to play. Rather than deal with children's answers, this paper focuses on their initiatives in this game, especially on their constituent questions. Section 2 summarizes relevant syntactic regularities of the German target system and addresses some controversies which have arisen within language acquisition studies over the way in which children acquire these regularities. In sections 3 and 4, I present data from a longitudinal case study, relating it to other investigations and identifying some explananda for acquisition theory. Section 5 returns to descriptive problems of the target system, this time from a more theoretical perspective, and section 6 offers some ideas on how the learner could get by with a little help from universal grammar.

1

This paper grew out of several talks: at the Glow conference in Leiden and the University of Durham in 1991, at the Max-Planck-Institute in Nijmegen in 1992, and at the University of Stuttgart in 1993. I am indebted to many participants for stimulating discussions and to the Max-Planck-Institute for a research stay which gave me the time to get it all written up. I also want to thank my colleagues in the Tübingen acquisition project who helped with the collection and the analysis of the data and who commented on an earlier version of this paper. Thanks also to David Reibel and my co-editor for valuable suggestions. I am especially grateful, of course, to Valle, for asking his questions in the first place. I have adopted the English wA-notation here, although in German, the questions dealt with are, of course, w-questions.

R. Tracy

2.

Some (relatively) uncontroversial facts about German and some controversial beliefs concerning their acquisition

2.1

German clause structure

Despite considerable disagreement about what a descriptively adequate grammar of German looks like, an issue to which I will return in section 5 below, the basic distributional facts relevant for w/i-questions can be stated in a relatively uncontroversial manner. Figure 1 presents a 'topological' picture of German clause structure (cf. Duden 1973), with the finite verb appearing in either the left or right member of the so-called Satzklammer ('sentence brace'). Figure 1:

A topological view of German clause structure

Satzklammer ('sentence brace') Vorfeld ('forefield')

(1)

ΧΡ(+Λ>νΑ)

Mittelfeld ('middle field1)

Nachfeld ('postfield')

Vj+finitt} (=V1/V2)

Vf.finiB}

(2a)

CONflV*/,)

V (+finite ) (=VE)

(2b)

wh- + COMP(daß)

V {+finite) (=VE)

d- + wo

The pattern in (1) captures the verb-second structure (henceforth V2) typical of German root clauses, with the initial position (the forefield) occupied by one and only one constituent, including wA-phrases, as in constituent questions like Welches Buch hast du gelesen? 'Which book have you read?', Was hat er gesagt! What has he said? 'What did he say?' Whenever the forefield remains unfilled, the result is a verb-first pattern (VI), as in yes/no questions (Kommst Du später? Come you later? 'Will you come later?') and in imperatives (Mach nicht soviel Krach! Make not so much noise! 'Don't make so much noise!'). This pattern can also be the result of discourse-licensed ellipsis or topic drop: (Was) Machst'n du? (What) make-PRT 2 ? 'What are you doing?'; (Das) Kommt gar nicht in Frage.' (That) comes not-at-all in question! 'Absolutely not!'; (Das) Hab' ich schon gesehen. (That) have I already seen. 'I've already seen it.' The postfield hosts constituents which have been extraposed

2

Here and in the following text, PRT stands for particles.

3

Raising Questions

(exbraciated) from the middle field, as in Ich habe schon gesehen, was er gefunden hat. I have already seen what he found has. Ί already saw what he found.' (2a) shows that in sentences introduced by a complementizer (including, for the time being, relative pronouns and the wA-elements of indirect questions), the finite verb remains in the position of the right sentence brace (= VE; Er weiß nicht, was er morgen machen will. He knows not what he tomorrow to-do wants. 'He does not know what he wants to do tomorrow'. Er zögerte mit der Antwort, da er niemandem wehtun wollte. He hesitated with the answer because he nobody to-hurt wanted. 'He hesitated to answer because he did not want to hurt anybody1). This also holds for VE clauses which are independent of any matrix context, as in exclamatives like Wie man nur so naiv sein kann! How one PRT so naive be can! 'How in the world can one be so naive!'). As a corollary, there are matrix verbs like glauben ('believe') and wissen ('know') which allow V2 complements (Jch weiß, er wohnt nebenan. Ί know he lives next door'). (2b) captures South-German subordinate clauses which allow the co-occurrence of whelements with the complementizer daß 'that' (cf. a dialect in Baden-Württemberg: Weesch wann deß derre ihr Gschäft uffmacht? Know-you when that her shop up-opens? 'Do you know when her shop opens?') and relative clauses introduced by both relative pronouns and the relative particle wo 'where', or by just the particle (Derre Fraa, (die) wo du ogrufe hosch... Thedat woman (who) where you up-called have...' (To) The woman whom you called up,..:). The V2 format is the unmarked interrogative pattern for root constituent questions, but questions come in other forms as well, among them those that leave the wA-element in situ, for instance echo questions (Du gehst WANN in die Stadt? You go WHEN into town? 'WHEN are you going into town?', cf. Wunderlich 1986). One question pattern which will assume importance later on in this paper correlates a request for confirmation with a VE pattern, as in the following adjacency pair, where B. quotes the propositional content of A.'s question, with the intention of checking whether s/he understood what A. had asked. B.'s question is no longer a constituent question, but one calling for a yes/no answer.

A: Wo

bist du

die ganze

Zeit

Where have you the whole time 'Where have you been the whole time?'

B: Wo

gewesen? been?

ich gewesen bin?

Where I been have? '(You mean,) Where I have been?' or 'Where have I been?' We would expect this uptake-securing strategy to be particularly frequent in conversations where speaker and hearer are contending with channel problems (as on the telephone), but also in situations where, at least as far as syntax is concerned, partners are not equally matched, as in child-adult discourse. I will later present examples from the input to children. Some children, like the protagonist of the case study in section 3, appear to be particularly receptive to this pattern, even though its functional load may elude them for a while.

R. Tracy

4

2.2

On to more controversial issues: language acquisition

The developmental pattern for German that most researchers would be likely to agree on can be summarized as in Figure 2 (cf. Mills 1985, Clahsen 1982, 1988, Tracy 1991). Ignoring the period of single-word utterances, it is generally assumed that children start with a preference for non-finite VE structures (= A). At time B, around the age of two, when agreement/finiteness markers appear on the verb, V1/V2 patterns emerge, with verbs now also including auxiliaries and modals. After that, at time C, children begin to tackle complex sentences. Figure 2:

Developmental story (highly simplified) examples

sentence brace

A.

V(.finite)

bus reinsitzen bus in-sit 'sit in bus'

mama

ball werfen

mummy ball throw B.

C.

Vl/V2{+finiie)... (MOD/AUX)

V{-finite)

COMP

V{+fmite)

ich geh mal rein da I

go PRT in

there

wenn ich groß bin when I big am 'when I am big'

In accordance with advances within linguistic theory (i.e. the differentiation of various types of phrase-projecting heads), and consonant with results on the acquisition of other languages, it has been suggested that underlying the distinction between A, on the one hand, and B/C, on the other, is a decisive qualitative move from an early, purely lexical stage to a functional stage, the latter characterized by the emergence of closed-class grammatical categories, i.e. articles, pronouns, complementizers, case and agreement morphology (cf. Guilfoyle & Noonan 1988, Radford 1990, various articles in Meisel 1992, Ouhalla 1993). Depending on when and how they believe functional categories to become available (by maturation, triggering, or step-wise reconstruction), researchers draw a continuous or discontinuous picture of the developmental process. Proponents of strong continuity hypotheses, for instance, recognize no real qualitative difference between A, B, and C, claiming that, at least as far as phrase structure is concerned, children's clauses can be analyzed from the beginning on the basis of representations made available by the target grammar (cf. Poeppel & Wexler 1993). Figure 2 is, of course, highly simplified. It does not reflect the extent to which development looks more like a series of "crises" (cf. Tracy 1991) than like a sequence of neatly delimitable steps. Most relevant to my current concerns is that it does not accommodate intra-individual variation, especially the co-existence of different structures.

5

Raising Questions

At the same time, for instance, when German-speaking children produce simple predicateargument structures as in A, they also produce expressions of more or less formulaic and idiomatic character which already simulate the V1/V2 pattern of B. 3 Given the existence of these expressions, it is difficult to draw the line between A and B. Similar problems arise with respect to B: Alongside well-formed V1/V2 patterns Germanspeaking children may have available a number of formats which are ruled out in the target language, for instance V3 questions as in was ich kann machen (what I can do? 'What can I do?'), or the types of questions which will be discussed in detail below. As for C, while it was believed for a long time that verb placement in embedded clauses is adult-like from the start, we know today that deviant orders do occur, and that, moreover, different kinds of deviant patterns may co-exist with well-formed patterns in the same child (cf. Fritzenschaft et al. 1990, Gawlitzek-Maiwald et al. 1992, d'Avis & Gretsch, this volume; for bilingual children cf. Müller 1993). The data to be discussed below cast additional doubt on the assumption that main clauses are all worked out and 'set' before embedded clauses appear. It will be shown that embedded sentences can reach an impressive level of complexity before root questions are fully mastered. I shall argue that it is precisely the competition between co-existing structures which leads the child again and again to search for, and eventually establish, a coherent overall grammatical system.

3.

A developmental story (told backwards)

3.1

A note on the child

Valle, the little boy whose wA-questions I discuss in some detail below, is one of ten children who took part in a research project designed to study the development of complex sentences in monolingual and bilingual children. 4 Valle is a monolingual native speaker of Swabian, the local dialect of his home town and the native dialect of both his parents. The relevant data stem from seven months of the total observation period and range from the first recording, when Valle was 1;11 (one year, eleven months), to the age of 2;7 (two years, seven months). He was audio- and/or videorecorded every other week, with each recording lasting at least one hour. In presenting the data, I will turn the clock back, so to speak; that is, we meet Valle at a time when the syntax of his main and subordinate clauses leaves little to be desired and then go back to earlier developmental periods.

3

For proposals concerning the gradual cracking of these formulae see Tracy Kaltenbacher (1990).

(1991);

cf.

also

4 The data were collected in a DFG-Project located at the University of Tübingen from 1988 to 1993. For an overview of the spectrum of variation found in the monolingual children, see Fritzenschaft et al. (1990) and the articles by d'Avis & Gretsch, Gawlitzek-Maiwald, and Fritzenschaft, this volume.

R. Tracy

6

3.2

The data5

3.2.1

Consistent adult-like main and subordinate clauses

Already at the age of 28 months, and beyond, Valle produces formally and functionally wellformed simple clauses and well-differentiated embedded and subordinate clauses, cf. (l)-(7). Case marking includes the dative, which is usually late to emerge (cf. Clahsen 1984, Tracy 1986), and there are spontaneous passive sentences, cf. (8). Among the few constructions still missing from his complex sentences are infinitival complements. (1)

warum kann dann sein daß ich rausfalh why can then be that I out-fall 'why could I then happen to fall out?'

(2)

wir kriegen wenn ich größer bin dann krieg-mer dann krieg we get when I bigger am then get-we then get ich entweder ne katze oder nen hamster oder einen hund\ I either a cat or a^c hamster or a.xc dog 'when I'm bigger I'll get either a cat or a hamster or a dog'

(3)

ich muß schauen ob man das gartenbeet von hier sieht\ I must look whether one the flower bed from here sees Ί have to see whether one can see the flower bed from here'

,(4)

ich weiß nicht wo die sind\ I know not where they are Ί don't know where they are.'

(5)

der winkt den autos daß die da nicht drüberfahrn dürfen he waves thedat cars that they there not over-drive may weil da eine bausteile is\ because there a construction site is 'he is waving to the cars to signal that they should not drive over this (barrier) because there is a construction site there'

(6)

V.: gefällt das den kiihenl pleases that thej at COWSdat 'do the cows like that?' Α.: Was denn? 'What PRT'

5

V. = Valle, A. = adult interlocutor, M. and F. = Mother and Father. Utterances are printed in italics, capital letters indicate heavy stress. The original phonetic transcription has been replaced by standard orthography except where I wanted to maintain features of the child's pronunciation; / and \ indicate sentence-final intonation patterns. Utterances are interlinearly translated and, wherever necessary, glosses are provided. I have drawn attention to evidence for morphological case marking by subscripts.

7

Raising Questions

V.: daß der das heu aufpiekst\ that he the hay up-spikes 'that he picks up the hay' (7)

(8)

als ich lachen verboten gesagt hab hab-ma when I laughing forbidden said have have-we dann auch mit der lokomotive gespielt/ then also with the locomotive played 'that time when I said 'laughing forbidden', did we then also play with the locomotive?' V.: die trompete muß da irgendwo hingebringt the trumpet must there somewhere taken 'the trumpet must be taken somewhere there'

werden\ get

Α.: Und von wem wird die da hingebracht? And by whom is it there taken 'And by whom is it taken there?' V.: von mir\ by mej« Apart from morphological overregularizations as in the participle hingebringt ('taken', instead of hingebracht) in (8), there are very few deviant structures. In the following example, for instance, Valle might just have been misled by the immediate input into producing an inappropriate expression for a thematic source (aus was, 'out-of what', instead of woher, 'from where'): (9)

F.: Valle, wer hat 'V. V.:

angerufen?

who (has) called'

Christine\

F.: Die

Christine?

V.: ja\ F.: Wer ist die 'Who is (the) V.: 'a

Christine? Christine?'

einefrau\ woman'

F.: Aus der Kindergruppe? 'From the play group?' V.: nein ich weiss nicht so genau aus was die is\ no I know not so well out (of) what she is 'no, I don't know what she is made of instead of: '...where she is from'

8

R Tracy

Although Valle has trouble construing the reflexive and its antecedent in the adult's utterance in (10), interpreting die Mama ('the mummy') as object instead of subject, one could certainly not improve on the form of his wA-question:6 (10)

Α.: Bevor der Papagei jetzt zur Ute fliegt trocknet sich noch die Before the parrot now to Ute flies dries herself still the Mama ab. mummy off 'Before the parrot flies to Ute, Mummy dries herself off V.: wer trocknet die mama ab\ who dries the mummy off

There is one episode where a question does not take its target-like V2 shape but appears in a VE pattern, cf. was ich dadrin sag/ (what I there-in say) in (11). (11)

V. looking at tape-recorder: is da JETZT was draufl is there now s.th. on-it Α.: Nee, jetzt is da noch nix drauf. No, now is there still nothing on-it 'No, there's nothing on it yet' Aber nachher, nach 'ner Weile [...] But later, after a while... V.: DANN is da dann was draufl then is there then something on-it Α.: Da bist dann du drauf. There are then you on-it 'You'll be on it, then' V.: was sag ich dann immer/ what say I then always 'what do I always say then' A: Das, was du jetzt auch sagst. That which you now also say 'What you're saying now' V.: was ich dadrin SAG/ what I in-there say 'what I'm saying there?'

This episode stems from a test sequence where Valle was asked to act out various stimulus sentences.

9

Raising Questions

Α.: Da hört man dann das, was wir jetzt gespielt haben. There hears one then that which we now played have One can then hear what we have played now' V.: was denn/ what then I have quoted this episode in full length in order to show that it is difficult to decide whether Valle is asking a direct question or whether he is asking for confirmation, i.e. for a yes/no reaction from his interlocutor. (Note, by the way, that he does not get it!) In any case, this wh...VE example remains the only one throughout this period and should be kept in mind when we take a step back to a time where patterns like was ich dadrin sag occur without the context which suggests that they could be instances of an uptake-securing strategy. 3.2.2

Two co-existing wA-options

Already a few weeks earlier, at the age of about 27 months, Valle impresses the hearer with most of the achievements mentioned above: His simple declaratives are without syntactic flaws, case marking includes the dative, and there are passives as well as various types of complex sentences. (12)

ich will dem die zähne putzen\ I want hirridai the teeth brush Ί want to brush his teeth'

(13)

das leg ich jetzt dahin bis der bauer mit seinem bagger fertig is\ that put I now there until the farmer with hisdat bulldozer done is 'I'll put it there until the farmer has finished with his bulldozer'

(14)

weil der bagger in den stall gebaggert werden will because the bulldozer into theacc barn bulldozed get wants 'because the bulldozer wants to get bulldozed into the barn'

As for constituent questions, however, two options co-exist, namely (a) and (b): (a)

XP(+Wh) V2(+rinitci ...

(b)

XP(+wh)

VE(+(iniie}

While in this particular corpus a clear majority of fourteen questions follows the V2 pattern, as in (15)-(16), (15)

V. looking at picture of a tiger: wo-s die zunge vom tiger\ where-s the tongue of-thedat tiger

R Tracy

10

(16)

V. looking at microphone: was kann man mit DEM machen what can one with that^, make 'what's that for?'

six are realized as VE. So alongside the (a) examples in (17)-(19) we find those under (b). (17)

(a) was what (b) was what

is das\ is that DAS is\ that is

(18)

(a) wer who (b) wer who

is DAS/ is that DAS is/ that is

(19)

(a) was der gerne WILL! what he gladly wants (b) was will DER denn\ what wants that-one then 'what would he enjoy doing?'

There is no doubt as to their status as spontaneous questions: None of them occurs with a preceding context which would lead us to suspect an uptake-securing function.7 As (20) shows, adults treat them as direct questions and react accordingly: (20)

V. looking at toy barn: was DAS is\ what that is 'what is that?' Α.: Das is'n Stall. 'That is a barn.'

(21), finally, unites both question patterns. There is a brief pause but only one overall prosodic contour. This example indicates not only that two question formats are available, but that there may even be a certain amount of competition between them. (21)

was is denn da weggegangen ... da weggegangen is\ what is then there away-gone ... there away- gone is 'what has left there?'

So having taken one step back from where we started to an earlier developmental period, we found that the V2 pattern is not the only vehicle for constituent questions. Valle's VE 7

Also, no semantic difference could be inferred. For a suggestion that semantic differences (some universal interrogative reading vs. a referential interpretation) could play a role here, cf. Penner (1993:16).

11

Raising Questions

structures divide into those which can be analyzed as adult-like subordinate clauses and those which are, for all intents and purposes, spontaneous w>A-questions. At this point in time, then, his treatment of wA-questions can at best be called inconsistent, falling into two finite patterns: V2 and VE. 3.2.3

A single ivA-pattern with consistent form-function correlation

Going further back, to a time when Valle is 24 to 26 months old, we find evidence for two main clause patterns, but now each one has a clear pragmatic function associated with it: Declaratives are V2,8 wA-questions are VE. The forefield does not host wA-elements but is otherwise available for topicalization, cf. (22)-(24). Note, too, that Valle's clauses already include the passive, and there is evidence for complex sentences, cf. (25)-(28). As (27)-(28) show, the complementizer may be missing, making these sentences what have been called preconjunctional embedded clauses (cf. Rothweiler 1993, Fritzenschaft et al. 1990, Penner, this volume, Penner & Müller 1992, d'Avis & Gretsch, this volume). (22)

viele dächer soll das haben\ many roofs should that have 'that should have many roofs'

(23)

aber bei dem schneeauto kann man das lenkrad nicht sehn\ but of the

total / Q

183 250 174 144 255 214 126 274 173 168

/ / / / / / / / / /

1 2 — 1 — 1 14 27 19 26

VE{finite}

total

9 34 8 22 25 46 20 38 32 23

subordinate questions other + C/-C +wh / -wh

... / ... 7 / — 1 / 4 3 / —

. . . / ... 27/ 13 / 35/ 31 / 23 /

— — — ?1 —

— / 6 — / 18 1 / 1 9 / 9 14/11 7 / 12 4 / 2 1 / —

... / ... ... / ...

3 9 1 1 —

1 1 2 — —

Recall that (11) was the only candidate for a confirmation question interpretation. Even in later corpora we find no evidence for uptake strategies in VE guise. It therefore seems reasonable to conclude that (11) is no more than a remnant of Valle's earlier direct wh. ..VE questions.

R Tracy

20

At this point, at least two preliminary conclusions can be drawn: (a)

Valle does not conform to the developmental path outlined in Figure 2 in 2.2 above. His root clauses are not 'all set1 before complex sentences with complementizers appear. Rather, for several months, his main clauses are asymmetrical, i.e. whquestions show no V2 effect. The fact that the integration of wA-elements goes hand in hand with the emergence of complementizers suggests a connection.

(b)

This root asymmetry does not delay his overall development. At the age of 2;4 Valle produces structures which are often believed to become productive only between the ages of three and four, if not later (see Mills 1985, Clahsen 1982, 1988). We could even argue that his root VE structures, which mimic subordinate clauses, provide him with a useful syntactic bootstrapping device.

4.2

Other studies

Valle's path is not a lonely one. Both Grimm (1973) and Wode (1976) found wh... VE(+ß„i,e} patterns in their longitudinal and cross-sectional studies.17 At the time, the significance of this observation failed to make an impact because linguistic theory had not yet provided the theoretical framework which allowed one to tie the structural properties of the left periphery of different sentence types together. Even among the small population of monolingual children studied in our project in Tübingen (N=5), one other child (Lisa) follows the same pattern as Valle, and two more children (Benny and Max) produce VE questions alongside other formats.18 In another case study by the present author, one of four children whose input was a Northern variety of German temporarily produced VE constituent questions (Tracy 1991:331ff.)· The fact that children from different dialect areas hit upon the same strategy rules out the possibility that Valle's Southern German input could have played a decisive role.19 In a recent overview, Penner (1993; cf. also Penner, this volume) reports that before the establishment of target-like questions, three intermediary patterns can be found, with some children actually choosing more than one: particle questions, zero questions and VE questions. Particle questions are illustrated in (55) (an example from Penner's Bernese Swiss data) and in (56) (from Tracy 1991). (55)

d'isch das

(56)

[vo:zd] ball

PRT-is this

'what is this?'

[where's-the] ball

17

Grimm (1973:104) discusses the possibility that VE questions could be intermediary steps.

18

Benny, for instance, produces questions of a type not at all documented in Valle: V3 questions like warum weitergeht nicht? This has been dealt with elsewhere, cf. Fritzenschaft et al. (1990), Gawlitzek-Maiwald etal. (1992).

19

Wode's subject was from the North as well. Grimm's subjects came from a dialect area in the South which shares features with Valle's dialect, for instance the pattern 2b of section 2.1.

Raising Questions

21

In Tracy (1991) I have claimed that [vo:zd\ is a monomorphemic element which cannot be analyzed as wA-pronoun plus inflected copula (cf. also Wode 1976).20 Penner's zero questions correspond to what I have called gap formats: expressions interpretable as constituent questions although there is no wA-operator. This may result in two orders: in a Vl-pattem, as in 1st das? ('(What) is that?'),21 and in the earliest question pattern observed in Valle, which in short utterances like die mama macht? (the mummy makes?, '(what's) mummy doing?') is ambiguous as to a V2 or VE analysis. According to Penner, VE questions with wA-elements are "root questions in subordinate disguise" (1993:15).22 It is possible to add a fourth pattern to this list: V3 questions, as in warum weiter geht nicht? (why further goes not? 'why doesn't it go further?') or was ich kann machen? (what I can make? 'what can I do?') which may co-exist with the above or with normal V2 questions, (cf. Gawlitzek-Maiwald et al. 1992, Müller 1993, Weissenborn 1992). However, this V3 pattern was not documented for Valle. Unfortunately no earlier data from Valle's first word combinations are available. We simply do not know whether there was a time when he produced more or less holistic question particles like [vo:zd] (where's-the).23 At the time at which observations start, he produces zero (i.e. gap) questions, which are eventually replaced by verb-end patterns. Valle differs from the children discussed in Penner (1993) in that his gap formats and VE questions still occur at a time when V2 is well established in declaratives, and in that his VE questions persist alongside embedded clauses. For a more complete picture, the emergence of yes/no questions ought to be taken into consideration as well. It has been noted above that the distinction between yes/no questions and wh-questions is not always easy to draw (cf. Felix 1980; Penner 1993). On the one hand it is a common belief lhaiyesino questions can already be identified on the basis of intonation patterns in single-word utterances (cf. Wode 1976:267). On the other hand, it seems that at least the syntactic properties of yes/no interrogatives are late to emerge (cf. Penner 1993, Tracy 1991).24 Valle's yes/no questions are very slow to appear in their adult form. Table 1 shows that at a time when Valle produces plenty of constituent questions, there are hardly any VI questions. Frequencies rise from zero at the age of 2;0.10 to 1 at the age of 2;2.25 and 2 at 2;0.15 to 27 at 2;4.27. Given the fact that all that seems to be required to form yes!no 2

® Penner (1993) considers the wA-element part of a verbal head (the copula). For other particle questions cf. also examples like [vedddj gehört das (where/who belongs that where does that belong?') and oh izd dd backer heißt (oh is the baker called-is. what's the baker called?') in Tracy (1991:335f). Note. too. that example (54) from Valle could be analyzed as a particle question.

21

For ambiguities resulting from this pattern cf. also Felix (1980)

22

Cf. Penner (1993. and this volume) for an analysis along the lines of adult cleft sentences and relative clauses.

23

When we first met Valle he was about 20 months old. Although we established at the time that he did not yet produce complementizers, we did not. unfonunately. pay attention to what his questions looked like. His parents, when questioned later, reported that they had not observed any of the holistic question particles like \vo:dd\ / [ro.ra]. For yes/no questions in various languages cf. Wode (1976). for their late appearance in Finnish cf. Bowerman (1973).

R Tracy

22

questions in German is to leave the forefield unfilled when the finite verb appears at the left periphery, this calls for an explanation.25 Taken together, these various observations suggest that there is more to the acquisition of the left periphery of German clause structure than was known until recently, and a more differentiated picture is called for than that embodied in Figure 2, which does not say anything about relationships between types of root clauses or root and non-root clauses. There are intermediate systems which need accounting for, and there is considerable interindividual and intra-individual variation, involving both formal and functional aspects. It has been known for a long time that English-speaking children have problems with structures requiring inversion, producing, for instance, V3 patterns (what he can do?) and structures with several auxiliaries (Can I can come?) (cf. Brown et al. 1973, Labov & Labov 1976, Radford 1990, Roeper 1992).26 Cross-linguistic evidence would then support the assumption that it is not an easy task to work out the details of question formation. It may involve more than the discovery and integration of wA-operators, namely the creation of syntactic positions which did not exist before. 4.3

Evidence for gaps

It might be appropriate to pause for a moment and consider what evidence there is for what I have called gap formats (Penner's zero questions). Why should we believe that Valle's utterances call for abstract representations which have more than meets the ear? Recall that we are talking about times when declarative clauses are well-formed, with various types of constituents occupying the forefield. While it is true that initially the verbs in questions are predominantly copular verbs (sein, heißen, 'be', 'be called'), there are also sentences with transitive verbs like machen ('make'). One finds ample evidence for these verbs in declaratives, displaying fully saturated argument structures. We can rule out the possibility that the delay of wA-elements could be due to the lack of the cognitive concepts encoded by wA-words. Valle's declaratives contain sufficient arguments and adjuncts referring to persons, objects, places, manner and time. Like Felix (1980) in his study on the development of question words, I therefore conclude that there must be grammar-internal reasons for the missing wA-operator, its emergence and the convergence to V2 in all root clauses. As we have seen, Valle's earliest and shortest gap formats are, at least superficially, ambiguous with respect to a V2 or VE analysis. This can be seen in Figure 4 below, an attempt to map one of Valle's sentences onto possible topological representations. These representations go beyond the single-level analysis of Figure 1, suggesting that constituents which eventually end up in the forefield and in the position of the left sentence brace originate in VP in the middle field. The analysis in (a) reflects the original D-structure, ignoring for the moment any further distinction between uninflected and inflected verbs. In On the basis of their verbal and non-verbal contexts the VI structures which occur before 2;4, sometimes as many as 31 per corpus, were interpreted as imperatives or declaratives, not asyeslno questions. Recent experimental studies, looking at long-distance extraction phenomena, found that English-speaking children fill medial positions, which is not allowed in adult English but ok in German: Was sagt er wen er gesehen hat? (»What does he say whom he saw?) Wen sagt er wen er gesehen hat? (»Whom does he say whom he saw?), cf. Roeper & de Villiers (1992).

23

Raising Questions

(b) the subject vacuously moves into the forefield, and in (c) in addition to subject movement the finite verb ends up in the position of the left sentence brace. The "f" refers to traces left behind and coindexed with their antecedents. Depending on one's preference for any one of these representations for a simple sentence like der mann heißt? (the man is-called?, 'what's the man's name?'), we find ourselves committed to various types of empty elements, besides the gap in the middle field. Figure 4:

Analyzing Valle's gap formats: V2 or VE? sentence brace middle field

forefield I 1

r

(a) (b)

der maruii

(c)

der marnij

heißt;

GAP

heißt

GAP

leißt Ι 1 t\

GAP

Questions which would unambiguously reveal that the verb in gap formats occupies the final position, i.e. instances like der mann dem kind GAP gegeben hat? (the man the da[ child GAP given has? 'What has the man given to the child?') are not documented initially. There are, however, as Table 1 shows, nine non-question gap formats in one of the early corpora, some of which contain main verbs and auxiliary verbs (valle macht hat, valle made has, '(look at what) valle did'), which support a VE analysis. As for questions, it is only after wA-elements have appeared that we find convincing evidence for VE formats, as in examples like was da dann rauskommt? (what there then outcomes? 'What comes out of there?'), wo der Peter hinsitzt? (where the Peter down-sits?, 'Where does Peter sit?1), where the parts of prefixed verbs show up as a continuous constituent or in examples where the presence of adverbials provides positional cues, as in was der gerne will (what he gladly wants?, 'What would he like?'). The question of how many and what kind of empty elements are involved in early gap formats is therefore notoriously hard to answer. For the adult system, it is assumed that whquestions contain a variable which must be bound by an operator in a scope-assigning position. 27 As for Valle, there is, initially, no overt operator, and it is not clear whether it makes sense to suggest that there is an abstract one. Indeed, it might be possible to suggest that abstract operators are superfluous as long as the gap resides in the strict subcategorization domain of verbs like heißen 'to be called', machen 'make', or the copula. In these cases the gap is governed, therefore locally recoverable on the basis of the argument structure of the verb. But apart from these

If not in the syntax (as in German) then at least at the level of Logical Form, cf. v. Stechow & Stemefeld (1988). Roeper & de Villiers (1992) have recently claimed that children initially treat wA-traces as null constants instead of as variables, i.e. they have problems assigning the appropriate distributive reading to wA-elements. Cf. also the discussion in Penner (1993).

R. Tracy

24

considerations, one could argue that there is no room for operators anyway. If the subject in (b) and (c) in Figure 4 ends up in the forefield, there is, at least in a conservative topological map, no place for operators, no matter whether abstract or overtly realized. 28 There appear to be two ways in which necessary space can be created in the syntax: (1)

Phrases occupying the forefield have to be 'demoted', i.e. they should not undergo movement and remain where they are in the first place, i.e. as in (a) in Figure 4.

(2)

Something could tell the child that more structure, i.e. an additional layer on top of (b)-(c), is needed. This is where complementizers could come into play.

But crucial developments do not just involve reanalysis at the left sentence periphery. They also involve decisions concerning the right sentence brace as well. These considerations require a more finely grained approach to clause structure than a topological map can provide, and I shall return to this issue in section 5 within the context of current linguistic theory. As for Valle's gap questions, there does not seem to be any external motivation (like negative evidence) for giving it up. Adults invariably treat Valle's gap formats as something calling for a particular kind of response. Because of the position of the verb, questions without κΆ-elements are not mistaken for VI yes Ino questions. Therefore it is not likely that any need to disambiguate question types for the sake of his listeners forces Valle to call in the wA-operator. Nevertheless it can be assumed that simple positive evidence in the input helps him to discover that he is acquiring a language in which gaps are overtly licensed. This does not just apply to gaps in questions. I have pointed out that Valle's early corpora contain other VE constructions which, although they do not function as questions, share their gap character. Take (57)-(58), for example, the former apparently modeled on embedded sentences of the form Guck mal, was Valle gemacht hat! (Look, what Valle done has!, 'Look, what Valle did!'). (57)

V. pointing: valle macht hat\ valle made has 'what valle did'

(58)

V. has put on a hat, addressing his mother: valle aufhat\ valle on-has 'what V. is wearing' M.: Was du aufhast? What you on-have 'What you are wearing?'

ΛΟ

For interesting hypotheses relating topic drop to some notion of discourse binding, cf. Rizzi (1991). Penner (1993) draws on these proposals for the interpretation of zero questions.

Raising Questions

25

The mother's response in (58) is particularly interesting. On the one hand, she provides Valle with positive evidence concerning the obligatoriness of the wA-operator; on the other hand, though, she offers a sentence pattern which is acceptable for her own confirmation question (calling for a yes/no answer) but which does not qualify as a model for direct questions! In fact, as we have seen in later corpora, Valle will adopt both the operator and the syntactic format, thereby slightly 'missing' the target. As for the integration of wA-elements, we could expect two factors to conspire: the knowledge made available by universal grammar (UG) that gaps must be licensed, and the language-specific discovery that this licensing has to be of an overt operator-variable variety. With respect to the latter, input as in (58) could serve a useful function. 29

5.

Gaps in our linguistic knowledge

5.1

Some open questions for linguistics

As I have argued elsewhere (cf. Tracy 1991), describing child language is like walking a tightrope between the fallacies of overinterpretation and reduction. One major methodological stumbling block lies in the fact that we do not have systematic access to young children's intuitions, and however young we feel, we cannot "go native" and develop the intuitions of a three-year-old. But even if we had access to these intuitions we would still be faced with the dilemma that there remains ample room for widely differing conceptions of what the target system looks like. While I have so far steered clear of descriptive problems, relying on a relatively uncontroversial topology-oriented metalanguage, it is now time to look at a more differentiated and therefore necessarily strongly theory-driven picture. 5.2

Topology meets X-bar

Despite the differences to be addressed shortly, there is also some remarkable consensus as to the basic architectural design underlying phrase structure in general. More or less all current approaches which are rooted within the generative paradigm assume that phrase structures are projected from endocentric heads, the most crucial distinction being that between head and non-head. Movement is severely restricted by principles of structure preservation, with heads only moving to head positions, and maximal phrases only to positions of the same status. There is also a fair amount of agreement that movement is induced by morphological considerations, i.e. the need to collect or, rather, check features for compatibility (cf. Chomsky 1992). Nevertheless when it comes to deciding on the structure of the simple clause in German, especially with respect to how many heads there are and where they are positioned, differences become apparent. Figure 5 presents just a small sample of possible ways of

Although adults may drop the w/t-element from the forefield in V2 patterns, which results in expressions like Machst'n du? (Make-PART you? 'What are you doing?'), they are consistent in their use of wA-phrases in VE patterns. I have found no example where adults dropped a Wi-element in a confirmation or echo question, i.e. the following invented dialogue should not occur: A: Was willst du machen? (What want you do? 'What do you want to do?') B: *Ich machen will? (I do want? '(What) I want to do?').

26

R. Tracy

analyzing a simple declarative of Valle's. I will assume without further discussion that the subject originates within VP. The given example is characterized via the solid lines. In subordinate clauses, some of these analyses differentiate the traditional right sentence brace into two verbal positions: the position of the non-finite verbal head of the VP, and 1°, a head-final category containing agreement and tense features (dotted lines).30 Likewise, in C and D there are two alternative positions for what on the surface appears as V2 (cf. solid vs. dotted lines). Figure 5:

Four alternatives for analyzing root clauses sentence brace middle field

forefield I 1 viele dächer soll

A.

[CP

B.

[ip XP b

XP [c



η η das

fo.

haben

[cpXPtc



or

-, Γ)]]] (if finite)

VP

I' or

C.

| VP

XP [ r

[IP X P

1 [R 1 °

]]]]

VP

(if SVO) Γ D.

or

or-

[ c p X P t c C 0 [τρ XP [r (if wA-question)



[p XP [r

VP

1

I 0 ]]]]]] (if finite)

The analysis in A takes a so-called symmetrical approach (cf. Vikner & Schwartz in press, Grewendorf 1988), deriving both main and subordinate clauses from a single underlying representation. In main clauses, the verb leaves VP for Γ in order to pick up morphological features and travels on to C°. SpecC, which corresponds to the traditional forefield of the topological map, catches constituents which have been moved out of VP, including whphrases. If there is a complementizer, the finite verb remains in 1°, since the complementizer occupies the only left head position the verb could otherwise move to. The analyses B-D represent various asymmetrical alternatives. In B, V2 clauses are just IPs with I0 at the left of its complement VP (cf. Reis & Rosengren 1991, Brandt et al. 1992). This is what I had in mind when I mentioned above that relegating the verb to the VE position in Figure 4 leaves a number of questions unsettled.

Raising Questions

27

The analyses B-D represent various asymmetrical alternatives. In B, V2 clauses are just IPs with Γ at the left of its complement VP (cf. Reis & Rosengren 1991, Brandt et al. 1992). Topicalization and wA-movement target SpecIP. Since there is no CP, Β does not accommodate subordinate clauses and would have to be complemented by a second clausal schema, which could, for instance, take the shape of the solution in C. Just like B, C offers a head-initial IP. Subjects surface in SpecIP. This analysis makes available a CP for those main clauses in which non-subjects are topicalized.31 In subordinate clauses inflectional features get lowered into VP or are base-generated in VP and checked. Therefore, there is no need for the verb to leave VP in subordinate clauses, and consequently, the analysis can do without a head-final F. D offers an additional phrase-structure layer on top of IP but below CP. This is a topic phrase (=TP), which provides a landing site for subjects and other topicalized elements except for wA-phrases (deriving SVO and 0(-Wh)VS orders), while root SpecCP is reserved for whphrases (cf. Müller & Stemefeld 1991). 32 The list in Figure 5 could be extended by adding approaches of the split-INFL variety (cf. Pollock 1989, Chomsky 1992). Any proposal which adds structural layers offers new landing sites, and involves more ordering decisions, and therefore creates the potential for further asymmetries. The more asymmetries in root clauses, the more the original idea of some unified V2 turns into a mere surface artefact. This has serious implications for learnability theory since we have to ask how a child could ever come to connect superficially identical V2 strings to various underlying structures. In all the approaches mentioned so far, wA-elements are representatives of maximal phrases and therefore in root clauses (unless they remain in situ) they move to the Spec position of the highest head. This movement, it has been suggested (cf. Rizzi 1991), is licensed by moving the verb to the corresponding head position, resulting in Spec-Head agreement. The status of wA-elements in non-root clauses is more difficult to decide. If we assume that they, here too, move to the highest specifier position, SpecCP, we have to explain why the verb does not, as in main clauses, move along with them. One of the most appealing answers is that in the case of subordinate clauses the topmost head, C°, is selected by a matrix verb, therefore hosts the corresponding features, and consequently cannot any longer serve as a landing site (cf. Rizzi 1991). On the other hand, it has also been suggested that wA-elements behave differently in root and non-root clauses, ending up in C° in the latter (cf. Bayer 1984 and the discussion of various options in Grewendorf 1988). Although this would violate the structure-preserving Travis (1984) bases her arguments on the observation that not all maximal constituents can undergo topicalization. Weak pronouns as, for instance, in *es kenne ich nicht (it know I not, Ί don't know it') are excluded. In Travis' analysis, main clauses with SVO order are IPs with only subjects reaching the preverbal position, whereas non-subject topics in OVS orders move to another maximal position at the left (SpecCP), the verb following to C°. Neat evidence for two verbal positions to the left of VP in root clauses can be found in Zwart (1993), who shows that in Dutch, depending on its position (pre-subject vs. postsubject), the verb takes a different second person inflection: jij (...) hebt ('you (...) have') vs. heb jij ('have you'). 32

For further interesting analyses which cannot be dealt with here cf. Weerman (1989), who only allows whelcments in SpecCP, and especially Hoekstra's recent (1993) analysis of CP-recursion in various Dutch dialects. Cf. also Penner, this volume.

28

R. Tracy

constraint on movement, which disallows a maximal category in a head position, it offers a clear proposal as to why there is no V2 effect in subordinate clauses: The potential landing site for verbs is occupied by either a lexical complementizer or a relative or wA-pronoun. The approaches on which Α-D are based all have their supporters among investigators of child language (cf. Meisel 1992 for an overview of various positions). The majority sides with a symmetrical perspective along the lines of A (or, more recently, a split-INFL variant), either for the target and for all intermediate stages, or at least for the target, in this case often accepting a solution like Β as a transitional step (cf. Clahsen 1988). Those who claim that there is an initial phase without CP have pointed out that the following CP-related phenomena are lacking: There are no complementizers, no topicalization (inversion) and no whmovement (cf. Meisel & Müller 1992, Penner 1993, also this volume). 33 Returning to Valle at the time at which his declaratives appear to be target-like, we see that the theories behind A-C lead us to expect wA-elements to behave just like other topicalized elements. But, as we have seen, this expectation is not borne out. Valle treats subjects and other [-wA] topics in one way (revealing that there must be at least one landing site for verbs and arguments above VP), [+wh] elements in another. While D, which reserves a structural layer for wA-elements and complementizers, predicts that wA-elements behave differently from other maximal phrases, it still does not tell us why this difference should manifest itself first in a complete lack of wA-operators and, after their emergence, in a lack of V2.34 From this brief look at alternative linguistic descriptions, it is clear that there is no obvious answer to questions concerning the status of wA-operators in the target system or concerning the precise nature of the relationship between complementizers, verbs and wA-operators, all of which are elements that compete for positions at the left sentence periphery. While disagreement with respect to the target system is frustrating from the point of view of acquisition theory, it is also quite suggestive. The very fact that there does not seem to be an easy way to resolve these issues empirically or theoretically, converges with what we arrived at in a more or less inductive way, i.e. by tracing micro-developmental changes in the data: Whatever the details of the task, it is not easy for the child either.

6.

How to get by with a little help from UG

In Valle's data, complementizers and wA-elements appear in close temporal succession. Afterwards, within a few weeks of the firm establishment of the former, r'A-questions give up their alliance with VE in direct questions. In the following I shall sketch three possible scenarios of what could be behind this change.

33

3
Subject > Verb > Object. The fronted adverb can be either jetzt

'now', dann 'then', or da

'here'. Müller proposes that these V 3 s derive via IP adjunction. (c) V 2 patterns of the type si si gange (s)'paziere they are gone walk 'They have gone for a walk'

(S. 2;1)

are derived by V P R and subject scrambling to the left side of A U X . (d) V I patterns of the type chöme (F)leisch usenää can-one meat take out 'Can one take meat out'

(S. 2;0)

are derived by VPR without subject scrambling. In this sense the various patterns o f word order in IP d o not constitute genuine variation.

Bibliography Baker, Colin (1970): "Notes on the Description of English Questions." - In: Foundations of Language 6, 107219. Bates, Elizabeth & Brian MacWhinney (1987): "Competition, Variation, and Language Learning." - In: Brian MacWhinney (ed.): Mechanisms of Language Acquisition (Hillsdale, N.J.: Erlbaum) 157-194. Bayer, Josef (1984): "Towards an Explanation of Certain that-t Phenomena: The COMP-Node in Bavarian." In: Wim de Geest & Yvan Puseys (eds.): Sentential Complementation: Proceedings of the International Conference Held at UFSAL (Dordrecht: Foris) 23-32. Berman, Ruth & Jürgen Weissenborn (1991): Acquisition of Word Order. - Final scientific report. Berwick, Robert C. (1985): The Acquisition of Syntactic Knowledge. - Cambridge: ΜΓΓ Press. Browning, Margaret (1987): Null Operator Constructions. - Ph.D. Diss. MIT. Chomsky, Noam (1986): Knowledge of Language: Its Nature, Origin, and Use. - New York: Praeger. Clahsen, Harald (1988): Normale und gestörte Kindersprache: Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie. - Amsterdam: Benjamins. de Villiers, Jill (1991): Defaults in the Acquisition of WA-Movement. - Talk given at GLOW, Leiden. de Villiers, Jill, Tom Roeper & Anne Vainikka (1990): "The Acquisition of Long-Distance Rules." - In: Lyn Frazier & Jill de Villiers (eds.), 257-298. den Besten, Hans (1992): Verb(projektions)anhebung im Afrikaans: Ein Forschungsvorhaben. - Paper presented at the 12th Groninger Grammatikgespräche. Felix, Sascha (1980): "Cognition and Language Development: A German Child's Acquisition of Question Words." - In: Dieter Nehls (ed.): Studies in Language Acquisition (Heidelberg: Groos) 91-109. Frazier, Lyn & Jill de Villiers (eds.) (1990): Language Processing and Language Acquisition. - Dordrecht: Kluwer.

Ζ Penner

56

Fritzenschaft, Agnes, Ira Gawlitzek-Maiwald, Rosemarie Tracy & Susanne Winkler (1990): "Wege zur komplexen Syntax." - In: Zeitschrift für Sprachwissenschaft 9, 52-134. Gawlitzek-Maiwald, Ira, Rosemarie Tracy & Agnes Fritzenschaft (1992): "Language Acquisition and Competing Linguistic Representations: The Child as Arbiter." - In: Jürgen Μ. Meise) (ed.), 139-179. Gazdar, Gerald, Ewan Klein, Geoffrey Pullum & Ivan Sag (1985): Generalized Phrase Structure Grammar. Oxford: Blackwell. Geilfuß, Jochen (1991): Verb- und Verbphrasensyntax. - Arbeitspapier 11. SFB 340. - University of Tübingen. Haegeman, Liliane (1992): Generative Syntax: Theory and Description. A Case Study from West Flemish. Cambridge: Cambridge University Press. Haverkort, Marco & Jürgen Weissenborn (1991): Cliticization in Early Child Language. - Talk given at the DFG-Conference "Crossing Boundaries: Formal and Functional Determinants of Language Acquisition," University of Tübingen. Hoekstra, Jarich & Liszlo Maracz (1989): "On the Position of Inflection in West-Germanic." - In: Working Papers in Scandinavian Syntax 44, 75-88. Hyams, Nina (1991): The Genesis of Functional Categories. - Talk given at GLOW, Leiden. Kaltenbacher, Erika (1990): Strategien beimßrühkindlichen Syntaxerwerb. - Tübingen: Narr. Kiparsky, Paul & Carol Kiparsky (1970): "Fact." - In: Manfred Bierwisch & Klaus Heidolph (eds.): Progress in Linguistics (The Hague: Mouton) 143-173. Labelle, Marie (1990): "Predication, WA-Movement, and the Development of Relative Clauses." -

In:

Language Acquisition 1,95-119. Longobardi, Giuseppe (1992): Proper Names and the Theory of N-Movement in Syntax and Logical Form. Ms. University of Venice. Mazuka, Reiko & Barbara Lust (1990): "On Parameter Setting and Parsing: Predictions for Cross-Linguistic Differences in Adult and Child Processing." - In: Lyn Frazier & Jill de Villiers (eds.), 163-205. McCloskey, James (1991): "Clause Structure, Ellipsis and Proper Government in Irish." - In: Lingua 85, 259302. Meisel, Jürgen Μ. & Natascha Müller (1992): "Finiteness and Verb Placement in Early Child Grammars: Evidence from Simultaneous Acquisition of French and German in Bilinguals." - In: Jürgen Μ. Meisel (ed.), 109-138. Meisel, Jürgen Μ. (ed.) (1992): The Acquisition of Verb Placement: Functional Categories and V2 Phenomena in Language Acquisition. - Dordrecht: Kluwer. Müller, Gereon & Wolfgang Sternefeld (1991): Extraction, Lexical Variation, and the Theory of Barriers. Arbeitspapiere der Fachgruppe Sprachwissenschaft 34. - University of Constance. Müller, Natascha (1992): "Parameters Cannot Be Reset: Evidence from the Development of COMP." - Ms. University of Hamburg. Müller, Natascha (1993): Komplexe Sätze: Der Erwerb von COMP und von Wortstellungsmustern

bei

bilingualen Kindern (Französisch/Deutsch). - Tübingen: Narr. Pasch, Renate (1989): "Adverbialsätze - Kommentarsätze - Adjungierte Sätze." - In: Linguistische Studien 194, 141-158.

Domains for Variation

57

Penner, Zvi (1990): "On the Acquisition of Verb Placement and Verb Projection Raising in Bernese Swiss German." - In: Monika Rothweiler (ed.), 166-189. Penner, Zvi (1992): "The Ban on Parameter Resetting, Default Mechanisms, and the Acquisition of V2 in Bernese Swiss German." - In: Jürgen Μ. Meisel (ed.), 245-281. Penner, Zvi (1993a): The Acquisition of DP in Bemese Swiss German: The Earliest Stage. - Arbeitspapier 30. Institut für Sprachwissenschaft, University of Beme. Penner, Zvi (1993b): "WA-Morphology in the COMP System of Bernese Swiss German and the Licensing of Empty Operators in the Prefield Position." - In: Werner Abraham & Josef Bayer (eds.): Dialektsyntax (Opladen: Westdeutscher Verlag) 201-212 (=Linguistische Berichte, Special issue 5). Penner, Zvi (1993c): "Asking Questions without CPs? On the Acquisition of WA-Questions in Bemese Swiss German and Standard German." - In: Teun Hoekstra & Bonnie Schwartz (eds.): Language Acquisition Studies in Generative Grammar: A Collection in Honor of Kenneth Wexler from the 1991 GLOW Workshop (Amsterdam: Benjamins). Penner, Zvi (in prep.): An Articulation Model of Parameter Setting. - Habilitation Thesis University of Berne. Penner, Zvi and Thomas Bader (1991): "Main Clause Phenomena in Embedded Clauses: The Licensing of Embedded V2-Clauses in Bemese Swiss German." - In: The Linguistic Review 8, 75-95. Penner, Zvi & Thomas Bader (in prep.): Issues in the Syntax of Subordination: A Comparative Study of the Complementizer System in Germanic, Romance, and Semitic Languages with Special Reference to Bemese Swiss German. Penner, Zvi & Natascha Müller (1992): "On the Early Stages in the Acquisition of Subordinate Clauses: The Syntax of the So-Called 'Preconjunctional Subordinate Clauses' in German, Swiss German, and French." In: Geneva Generative Papers, Vol. 0:1-2,163-181. Penner, Zvi & Martina Schönenberger (1992a): Cross-Dialectal Variation in Swiss German: Doubling Verbs, Verb Projection Raising, Barrierhood, and LF Movement. - Submitted (based on a talk given at the 7th Workshop on Comparative Germanic Syntax, Stuttgart). Penner, Zvi & Martina Schönenberger (1992b): The Distribution of DP Agreement Features in German Dialects: Expletive Dets and the So-Called Weak/Strong Asymmetry. - Talk given at the 8th Workshop on Comparative Germanic Syntax, Tromsa. Penner, Zvi, Rosemarie Tracy & Jürgen Weissenborn (1992): Scrambling in Early Developmental Stages in Standard and Swiss German. - Talk given at the DFG Conference, Düsseldorf. Pesetsky, David (1987): 'WA-in-situ: Movement and Unselective Binding." - In: Eric Reuland & Alec ter Meulen (eds.): The Representation of (ln)deflniteness (Cambridge, Mass.: MIT Press) 98-129. Phinney, Marianne (1981): Syntactic Constraints on the Acquisition of Sentential Complements. - Ph.D. Diss. University of Massachusetts, Amherst. Platzack, Christer & Anders Holmberg (1989): "The Role of AGR and Finiteness in Germanic VO Languages." - In: Working Papers in Scandinavian Syntax 43,51-76. Reis, Marga (1985): "Satzeinleitende Strukturen im Deutschen: Über COMP, Haupt- und Nebensätze, h>Bewegung und die Doppelkopfanalyse." - In: Werner Abraham (ed.): Erklärende Syntax des Deutschen (Tübingen: Narr) 271-311.

Ζ Penner

58

Rizzi, Luigi (1991): "Residual Verb Second and the WA-Criterion." - In: Technical Reports in Formai and Computational Linguistics 2, University of Geneva. Roeper, Tom (1990): How the Least Effort Concept Applies to the Acquisition of Head Movement, Copying, and Cyclic WA-Movement. - Ms. University of Massachusetts, Amherst. Roeper, Tom (1991): Parametric Variation in IP and the Shift from Adjunction to CP. - Talk given at GLOW, Leiden. Roeper, Tom, S. Akyama, L. Mallis, M. Rooth (1984): The Problem of Empty Categories and Bound Variables in Language Acquisition. - Ms. University of Massachusetts, Amherst. Roeper, Tom & Jill de Villiers (1992): The One-Feature Hypothesis. - Ms. University of Massachusetts, Amherst. Roeper, Tom & Jürgen Weissenborn (1990): "How to Make Parameters Work." - In: Lyn Frazier & Jill de Villiers (eds.), 147-162. Rothweiler, Monika (ed.) (1990): Spracherwerb und Grammatik: Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie. - Opladen: Westdeutscher Verlag. (=Linguistische Berichte, Special issue 3). Rothweiler, Monika (1993): Nebensatzerwerb im Deutschen. - Tübingen: Niemeyer. Stemefeld, Wolfgang (1990): Beheaded Barriers. - Arbeitspapier 14. Fachgruppe Sprachwissenschaft, University of Constance. Tracy, Rosemarie (1991): Sprachliche

Strukturentwicklung:

Linguistische

und

kognitionspsychologische

Aspekte einer Theorie des Erstspracherwerbs. - Tübingen: Narr. Valois, Daniel (1991): The Internal Syntax of DP. - Ph.D. Diss. University of California. Vikner, Sten (1990): Verb Movement and the Licensing of NP Positions in the Germanic Languages. - Ph.D. Diss. University of Geneva. Weissenborn, Jürgen (1990): "Functional Categories and Verb Movement: The Acquisition of German Syntax Reconsidered." - In: Monika Rothweiler (ed.), 190-224. Weissenborn, Jürgen (1992): Constraining the Child's Grammar: The Development of Verb Movement in German and French. - Talk given at the Conference on Syntactic Theory and First Language Acquisition: Cross-Linguistic Perspectives, Cornell University. Weverink, Meike (1991): "Inversion in Embedded Clauses." - In: Papers in the Acquisition of Wh. UMOP (University of Massachusetts Occasional Papers, Special edition) 19-42. Whitman, John B., Kwee-Ock Lee & Barbara Lust (1990): "Continuity of the Principles of Universal Grammar in First Language Acquisition: The Issue of Functional Categories." - In: Proceedings of NELS 21, 383398. Zanuttini, Rafaela (1991): Syntactic Properties of Sentential Negation: A Comparative Study of the Romance Languages. - Ph.D. Diss. University of Pennsylvania. Zwart, Jan-Wouter (1992): Verb Movement and Complementizer Agreement. - Ms. University of Groningen and MIT.

Franz-Josef d'Avis & Petra Gretsch

Variations on "Variation": On the Acquisition of Complementizers in German

0.

Introduction1

The recent literature concerning the acquisition of subordinate clauses has focused on a phase characterized by the coexistence of utterances adequate to the language and utterances that are systematically deviant, which signal a change in the grammatical system of the learner. A prominent feature of this phase are the so-called preconjunctional clauses, which look like ordinary subordinate clauses with a missing complementizer:2 ( l ) 3 Adult: Warum soil's ich nicht machen? Max22 (3;06.11) 'Why shouldn't 1 do it?' Child: ich alleine kann\ I alone can I can do it by myself.' '[Because] In addition to this special type of deviant but - as we will assume - systematic construction we find a collection of similarly interesting phenomena accompanying a child on his/her way to fully fleshed-out subordinate clauses. Nevertheless, no special attention has been paid in the literature to these other phonetically, syntactically or semantically deviant preconjunctional structures. (At this stage of investigation we use the term 'semantics' to refer to the lexical meaning of the respective complementizers.) The first section of this article deals with the different reflexes of the phenomenon of preconjunctionality and looks at the variation among its precursor structures. In 1.4 we

1

We would like to thank our colleagues from the Tiibingen DFG project "Erwerb der komplexen Syntax" for critical discussions and encouragement. We are grateful to Marga Reis, Janet Randall, Jürgen Weissenborn and Uwe Mönnich for their insightful comments. Particular thanks go to the editors, Elsa Lattey and Rosemarie Tracy, for patient support and help with the translation. All the data referred to in this article originate from the above-mentioned DFG project.

2

The term preconjunctional clause was coined by Rothweiler (1989:47ff) for unintroduced subordinate clauses, where the subordinate status is indicated by a systematic verb-end placement. In subsequent work the same phenomenon has also been labeled preconjunctional embedded clauses (PEC) and preconjunctional subordinate clauses (PSC), cf. Penner (this volume) and Penner & Müller (1992). The time span where these constructions occur is the preconjunctional phase, cf. again Rothweiler (1989).

3

We adopt the following notational conventions: To more closely represent the actual utterance, schwa, the glottal stop and slight modifications of standard orthography have been added where necessary; INT stands for interruption, FILLER for unidentified and undifferentiated elements (similar to the universal particles in Penner & Müller 1992), PRT for untranslatable but identified particles. Utterance-final intonation patterns are indicated by / and \; (...) indicates omitted parts of a discourse, [...] is used for added interpretations, (...) indicates scarcely intelligible parts of an utterance, and an apostrophe indicates attached clitics.

F.-J. d'Avis & P. Gretsch

60

propose a developmental model for the description of this variation and of the different learner types that make use of individual segments of the spectrum of variation. This building-block model accounts for the variation in the data on the basis of a highly modular architecture whereby the acquisition of a new structure depends on the acquisition of individual constituents, the building-blocks. There is, figuratively speaking, a box of blocks for every structure that gets built by the child during the acquisition process. In order to acquire a structure a child must assemble all the blocks. The way the blocks get assembled depends in each case on the learning strategies applied and the learner types involved. A more general concept of this model may describe not only the acquisition of complementizers but also other domains of acquisition. In the second section we apply the building-block model to the divergent acquisitional paths taken by four children, thus illustrating a range of variation for the acquisition of complementizers and subordinate clauses. The data presented can be used as a detailed acquisition document and as an empirical challenge for alternative models - apart from functioning as a justification for the proposed building-block model. In the third section we look at the prominent notion of sequence in language acquisition and explore the possibility of embedding our model in a learning-theoretic framework. The terminology introduced informally in sections 1 and 2 (i.e. productive usage, interim grammars, etc.) will be explicated in more detail there. In summary, then, the three sections of this article are in themselves variations on the basic theme of individual developmental variation, each from a different perspective.

1.

Precursor structures and the building-block model

1.1

The phenomenon of precursor structures

The structures preparing for and accompanying the acquisition of complementizers found in the analyzed corpora can be classified according to an informal catalogue of what here will be called precursor structures :4 •

unintroduced verb-end sentences with subordinating function and salient semantic connection (for example [+ conditional]):5

(2)

da draußen hat's nich scherben\ there outside has'it not broken glass 'there's no broken glass outside' dds immer hinfällt hat's draußen scherben\ that always down-falls has'it outside broken glass '[Whenever] that falls down there's broken glass outside.'

Valle04 (2;00.10)

4

See Fritzenschaft et al. (1990) and Gawlitzek-Maiwald et at. (1992) for comparison. We understand the term precursor as a twofold relational expression: On the one hand, precursors have to be seen in relation to the construction under consideration (here: complementizer-introduced subordinate clauses); on the other hand, they stand in relation to the individual grammatical system of the learner.

5

According to Rothweiler (1989), these constructions provide evidence for the preconjunctional phase.

61

Variations on "Variation"

(3)

Adult: Warum gehört das so? 'Why does it have to be that way?' Child: dds nich rausfällA this^ not out-falls '[So that] this can't fall out.'

Max22 (3;06.11)



verb-end sentences introduced by phonologically undifferentiated filler syllables with subordinating function and salient semantic connection (cf. Rothweiler 1989:47):

(4)

Adult: Warum - was ist da passiert? 'Why - what has happened?' Child: ?dndßsch TOT ist\ FILLER fish dead is '[Because] the fish is dead.'



incomplete or interrupted sentences which are introduced by complementizers, sometimes with a systematic change of topic. It is this systematic nature which makes us believe that they can be assigned precursor status:

(5)

(talking about the building of two bridges) Paul06 (3;06.12) wenn man so ein so (INT, going on in a different voice) lebendig fußgänger\ if one so a so - alive pedestrians



paratactic strings with several verbs and one uninterrupted prosodic contour, which could ease the acquisition of complement clauses:

(6)

(Paul acting as Mickey Mouse) oh ich guck es brennt da drin\ oh I am-looking it burns there inside 'Oh, I am looking how it is burning inside.'



paratactic strings with salient semantic connection (here [+ temporal]):

(7)

(Valle is sitting in the bathtub) Valle04 (2;00.10) volle will dann weitermalen\ valle fertig malen hat\ valle fertig badet hat\ valle wants then on-draw. valle finished draw has. valle finished bathed has. 7 'Valle wants to go on drawing when Valle has finished bathing.'

Maxi 1 (3;00.16)

Paul08 (3;06.27)

^

It might be that there is also a daft ('that') involved in das, but since the vowel is reduced, which is typical for pronouns, and features of stress that often go hand in hand with an early productive usage of complementizers are absent, we assume that das has to be interpreted as a pronominal element.

7

valle fertig malen haA is actually a production error and Valle corrected himself to valle fertig badet hat\ Interestingly, he keeps the finite verb in final position, indicating the special semantic connection.

F.-J. d'Avis & P. Gretsch

62



deviant verb positions in clauses introduced by complementizers:

(8)

Adult: Wann hast denn du das letzte Mal Frisbee gespielt? Benny09 (3;01.04) 'When was the last time that you played frisbee?' Child: wenn hab ich geburtstag habt dann han ich nein nein ich ganz au mal burtstag habt dann krieg ich dds au mal pispi\ when have I birthday had then have I no no I very too once birthday had then get I this too once frisbee 'When I have my next birthday then I'll get a frisbee, too.'



missing finite verbs in sentences introduced by complementizers:

(9)

Adult: Warum willst'n Papa naß machen? 'Why do you want to make daddy wet?' Child: weil er zu mir ganz bös\ because he to me very naughty 'Because he [is] so naughty to me.'



quotations and fragments with an unproductive complementizer:8

(10)

i han mein werkzeug\ wart mal ob - ob de (INT)\ I have my tools, wait PRT whether - whether the (INT)



use of lexical complementizers in positions which are not target-like:9

(11)

Adult: (talking about a dog) Paull3 (3;09.12) (...) und wir bringen dir auch immer was zum Fressen, wenn du auf uns aufpaßt. '(...) and we will always bring you something to eat if you keep an eye on us.' Child: (acting as the watchdog) ich paß aber ob bei dir aber auf\ I watch but whether at your-place but PREFIX



adoption of the complementizer function by lexical items from other categories (overgeneralizations):

(12)

Adult: Warum heißt ein Handtuch "Handtuch"? 'Why is a hand cloth ('towel') called a "hand cloth"?' Child: ja wegen's'n handtuch ist\ wegen's zum abtrocknen ist\ well because-of it'a hand cloth is. because-of it to off-dry is 'Well, because it's a towel, because it's for drying.'

Benny09 (3;01.04)

Benny08 (3;00.19)

Paul37 (5;01.24)

® The term unproductive refers to complementizers that are only memorized in holistic phrases or in citations. 9

In this case there is often no exact topological interpretation possible, as the example (11) illustrates.

63

Variations on "Variation"



complex dialogue patterns with each interlocutor contributing a part of the overall complex structure:

(13)

Adult: Wenn man den Luftballon aufbläst 'If you blow up the balloon -' Child: dann fährt des auto los\ then drives the car off '- then the car drives off.'



complementizers without a target-like semantic interpretation:

(14)

jetza will ich dds wenn du mich anhörst\ now (dialectal) want I this if/when you to-me listen 10 'Now I want you to listen to me.'

Paul06 (3;06.12)

Vallel2 (2;04.14)

It depends on the grammatical system of the child - and can actually only be decided post hoc - whether the respective string can be considered a precursor in the sense intended. In each case the researcher must consider possible interference from performance factors and influences from the particular recording situation as well as search for a systematic pattern in the individual deviations in the overall acquisition process. It is therefore necessary to present examples with their respective contexts, as the verbalization of complex content can initially be stretched out over several turns in dialogues, as in (13). Especially those complex sentences which are unintroduced and those which are introduced by undifferentiated phonological filler syllables (FILLER)11 can be identified only on the basis of contextual analysis.12 Moreover, these phenomena appear to call for more than a monadic interpretation of the category complementizer. It appears that small, flexibly acquired entities together make up the C head. 13 In interpreting the data one is constantly confronted with the problem of determining the borderline between precursor structures and acquired structures or between precursors and performance errors. In each case one must decide carefully on the basis of one's knowledge of the overall system. Even later, when a large part of the inventory of complementizers has been acquired, and when the child produces almost exclusively target-like subordinate clauses, we may still find phenomena (interruptions, undifferentiated filler syllables, deviant word order, etc.) which can then, however, no longer be interpreted as reflexes of construction-specific acquisition problems. A comparison of different developmental paths in a combined longitudinal and cross-sectional study should enable the researcher to arrive at a reliable assessment of the data.

10

dds ('this') is interpreted as a pronominal element for the reasons explicated in footnote 6.

11

The notion of undifferentiated FILLER has to be distinguished from the German particles (PRT), which are phonetically differentiated but cannot be translated as listed in footnote 3.

12

Contexts are omitted when the interpretation is clear.

13

Since we restrict ourselves to an analysis of embedded clauses, we will ignore potential C features in main clauses.

64

F.-J. d'Avis & P. Gretsch

The list of precursor structures above can be characterized by certain deficits (more or less grave) with respect to phonetic, syntactic and semantic components of the grammar. While we cannot discuss the structure of complementizer-introduced clauses in German in any detail here, it is obvious that the interaction of different levels is relevant both for a phenomenological description as well as for an explanation of the acquisition process.^ If learners have difficulty with the realization of lexical complementizers, this may not just be due to phonetic factors, but may also have syntactic reasons, for instance those described in Fritzenschaft et al. (1990) and Penner & Müller (1992). On the other hand, subcategorization requirements of matrix verbs with respect to complement clauses more likely point to the domain of the lexicon or possibly to semantic aspects of the complementizers. On the basis of clauses which paratactically follow each other, and which we have interpreted with the help of a salient semantic feature, cf. example (7), one can conclude that the grammatical prerequisites for target-like hypotactic constructions are still missing. To what extent the domain of general cognition and the interface between grammar and cognition could be involved will not be discussed here. In 1.2 we present a division into several phases which will serve as a basis for comparing different developmental paths. 1.2

Division into phases

As points of reference for comparing individual developmental paths we will define three phases, which can be called initial grammar, interim grammar, and target grammar,15 The initial grammar describes the language of the child before the appearance of precursor structures. The precursor structures are generated by interim grammars. The target grammar refers to the system after the acquisition of the respective construction has been completed. This division is not only relevant for the acquisition of complementizers but can be applied to other developmental analyses in which segmentation into a phase before, during, and after the acquisition process appears justifiable. (15)

Definition of the phases:

PHASE (A)

is the period immediately before the appearance of complementizer-introduced strings or of precursor structures. It is the system-internal point of departure, the initial grammar.

PHASE (B)

covers the period during which precursor structures like those in examples (1)(14) are attested.

PHASE (C)

begins at the point at which complementizers are used productively in accordance with the target grammar. This is also the point at which the children have reached their construction-specific developmental goal.

For a detailed description of German clause structure cf., among others, Reis (1985), von Stechow & Stemefeld (1988), and Brandt et al. (1992). For an overview cf. Gawlitzek-Maiwald et al. (1992). The notion phase describes a period that can be defined via distinctive features. This is not to be confused with the fixed phases of acquisition in Clahsen (1988).

65

Variations on "Variation"

Any exact delimitation of these phasal boundaries depends on the interpretation of the data. (16)

Exemplary schema for individual phases16 A PA-C" PA-AC PA-CC PA-RC

B

C

-—I

1—— , Γ .A·.,..... Γ A'

B'

A'

. 1

: C

11 ι 1 B'

'

C'

,

1

c

.

Age The unification of all (B1) phases is called superordinate phase (B). This captures the idea that the description of the acquisition of complementizers is the product of the combined developmental phases of adverbial, complement and relative clauses. On the one hand, the acquisition of complementizers can be divided into the acquisition of complement clauses, of adverbial clauses and - with some qualification - of relative clauses. On the other hand, the connection with an assumed C head, or rather the CP level, suggests that it is not until all the relevant phenomena have been acquired that we can actually talk about the completion of the acquisition of complementizers. The first documented precursor therefore marks the beginning of superordinate phase (B), the moment at which the building of the CP level becomes manifest. The end of superordinate phase (B) is defined by the replacement of precursors by target-like productive strings in all relevant structural domains - that is, the level of CP has been completely acquired. In searching for more finely grained diagnostics for the acquisition of the complementizer level one could also further subdivide the class of adverbial and complement clauses according to different distinctive features (cf. Penner & Müller 1992 for such an approach). There are, however, two reasons why we prefer a less finely grained schema which differentiates only adverbial and complement clauses. First, the focus of this paper lies on the representation of inter-individual variation, which means that the horizontal cut takes precedence over the vertical, intra-individual dimension. Second, the more finely grained the analysis, the less data is available, even in relatively large corpora. Take appositive relative clauses, for example. There may be so few instances of these in individual learners' data that they fall below the threshold of significance. We will include relative clauses in our discussion where their inclusion appears necessary, but exclude them for the time being from our acquisition model because the characterization of the C position is fraught with additional problems in this case. We realize that within our phase conception the interaction between construction-specific properties and linguistic theory remains problematic. However, it is a first approach that seems plausible on the basis of available data. PA-C° PA-AC PA-CC PA-RC

= = = =

phase of phase of phase of phase of

acquisition acquisition acquisition acquisition

of of of of

the C head/C level adverbial clauses complement clauses relative clauses

F.-J. d'Avis & P. Gretsch

66

Before we explain our building-block model in greater detail we will briefly discuss some general problems which the parameter concept creates for language acquisition theory. 1.3

Problems with the concept of parameter

Any attempt to explain the phenomenon of precursor structures within a parameter model is problematic. The conception of parameters and their triggers does not yield any immediately applicable proposals for the description of a developmental process that is continuous and rich in variation. One attempt to include precursor structures is the approach by Penner & Müller (1992), which, like the building-block model presented below, makes use of features. The authors deal with two instantiations of precursor structures, unintroduced preconjunctional subordinate clauses and subordinate clauses introduced by phonologically undifferentiated filler syllables. Penner & Müller (1992) shift the treatment of the developmental problem to the discussion of the acquisition of licensing conditions', they do not provide a detailed account of the assumed parameters and triggers, however. According to Verrips (1990, this volume), the question of how punctual parameter settings can be reconciled with a continuous developmental process remains unanswered within the currently dominant Principles and Parameters approach (cf. Chomsky & Lasnik 1991). The following points can be added to Verrips' dilemma: As for the concept of parameter, there is first of all the problem of number and range of applicability, which has been dealt with in the literature under different names. There is a tendency to replace parameters as carriers of far-reaching generalizations with more finely grained choices, down to parameters which concern individual lexical items. Compare especially Safir (1987), Nishigauchi & Roeper (1987), Webelhuth (1989), Müller (1993). Borer & Wexler (1987) discuss a related problem, namely that concerning the order in which different parameters are to be fixed. If we assume some fixed ordering among them, how can we then explain the phenomenon of temporal and learner-specific variation in language acquisition? If we assume, on the other hand, that ordering among parameters is free, how can we then explain the similarities across learners that have been found in diverse empirical investigations? The next problem concerns the concept of trigger and the basic question of how a trigger is to be defined. Various explications of this notion can be found in Roeper & Williams (1987), among others. Directly related is the triggering problem as it has been described in Borer & Wexler (1987): How is it that "triggering data" only become active after a certain time? In addition there is the question of how much input must be available before a language learner can recognize a relevant trigger (cf., for example, Tracy 1990). Is a single example enough, or does the child evaluate the input statistically? What happens in the case of ambiguous or deficient input? What about multilingual language acquisition? The final problematic domain we would like to address pertains to the problem of autonomy of the relevant linguistic module. This has caused a lot of controversy: How much and what kind of extra-linguistic interference does the language acquisition device (LAD) allow? Recently, there has been an increasing tendency in language acquisition theories to relax the opposition between the maturational (cf. Felix 1987, Borer & Wexler 1987) and the

Variations on "Variation "

67

continuity account (cf. Pinker 1984, Hyams 1986). Several hybrid models, i.e. combined approaches, as for example the structure-building hypothesis by Guilfoyle & Noonan (1988), have been proposed. Within their framework functional categories mature whereas lexical categories are assumed to be "inborn". Following this idea, Müller (1993) proposed a model for the acquisition of German subordinate clauses where "the CP" matures and "the IP" is accessible from early on. As this list of open questions shows, the notions parameter and trigger appear to be too undifferentiated to capture the various developmental aspects connected with the overall spectrum of precursor structures. Moreover, it is unclear how the necessary interaction between data-driven (bottom-up) and goal-driven (top-down) processes can be accounted for by the parameter model. In 1.4 we will try to model this developmental aspect in such a way that the model will guarantee consideration of a) the appropriate degree of differentiation, b) the bidirectionality of the overall process, c) the variety of individual choices among different developmental strategies, and d) the observance of modularity.

1.4

The building-block model

As the range of different precursor structures suggests, the complete acquisition of complementizers is the consequence of different interlocking developmental steps. We assume that the acquisition of complementizers manifests itself on different levels, with each successful developmental step consisting of the complete integration of the properties on the various levels involved. These properties can then be described as features of the C head, i.e. as feature-value pairs on the different levels. As a working hypothesis we will start with the following levels, which contain at least the listed feature specifications.

F.-J. d'Avis & P. Gretsch

68

(17)

Type-relevant information: Type phonetic level: 17

[PHON: X]

syntactic level:

[CAT: C] [BAR: 0] [SELECTION: S] [SUBORDINATION: +] [V2: -]

semantic level:

(18)

[CLASS OF CONNECTORS : Y]

Token-relevant information (shown for the example weit): Token phonetic level:

[PHON: vail]

syntactic level:

[CAT: C] [BAR : 0] [SELECTION: S] [SUBORDINATION: +] [V2: -] [ITEM-SPECIFICS: V2 possible]

semantic level:

[CLASS OF CONNECTORS : causal]

The developmental task can now be described as follows: the child must a) recognize which features are relevant for a certain construction and b) match individual features up with the respective target-language values. One could assume that there is some universal pool of features from which the child can draw. The identification could be helped along by the 17

This means that a learner must be able to distinguish words on the phonetic level.

Variations on 'Variation "

69

development of extralinguistic cognitive factors. The bundles of features on the different levels then together constitute the C head. In contrast with the common dichotomy 'presence vs. absence' of the C level we can thus model a cumulative developmental process. The schema in (17) can be understood as a template. The relationship between this template and a new lexical item corresponds to that between type and token. A type contains all relevant feature-value pairs with individual values replaced by variables. A token is, in contrast, the specific instantiation of some type. On the basis of this differentiation there are principally two possible developmental styles: Lexically-oriented learners (bottom-up learners) initially acquire single tokens and generalize across them in the course of development in order to derive a type. Top-down learners, on the other hand, begin with a type where place holders for individual values have to be replaced by specific values. One prediction would be that a bottom-up learner needs more time to master phase (B) than a top-down learner because s/he is repeatedly confronted with the acquisition of redundant information, like, for instance, categorial information. The top-down learner, on the other hand, only needs to recognize complementizers as such and to map the available type onto information connected with specific lexical items. This prediction is confirmed in our comparison below of Valle, who as a type learner masters phase (B) in three months, with Benny, a token or bottom-up learner, who takes twenty-one months. Further evidence for the existence of type learners is found in preconjunctional utterances. Unintroduced verb-end strings, as in examples (2) and (3), can be interpreted as the spelling out of some underspecified type. In a similar fashion verb-end strings which are introduced by undifferentiated phonological filler syllables can be interpreted as underspecified types with the phonetic component only hinted at. Furthermore it is possible to interpret a specific token initially as a type, which leads to the phenomenon of overgeneralization. For the learner to reach phase (C), s/he must have available both a target-like type and a representative selection of related tokens. In addition, there must always be the possibility for adding new tokens, because an end-point of the acquisition of new lexical items is impossible to determine. Our building-block approach models a developmental process non-deterministically within the framework provided by interim grammars, i.e. allowing for variation. The relatively free assembly of information units takes place in phase (B) and is reflected in different precursor structures. The types of precursor structures found suggest yet another distinction of developmental styles: Top-down learners as well as bottom-up learners have the option of starting their instantiation of feature values on different levels. The range of different precursor structures shows that the feature-value matching can start on any of the three levels (phonetic, syntactic, semantic). If a top-down learner orients her- or himself initially towards the syntactic level, s/he produces unintroduced preconjunctional subordinate clauses. A bottom-up (lexical) leamer, who starts with the phonetic level, utters single complementizers without taking into account the semantics of the lexical item or the syntactic behavior of the category. Because of the interaction of all these levels bootstrapping effects cannot be limited to a single level. The proposed learner types have to be understood as idealizations, and there may be mixed types. Nevertheless, an isolation of a prominent strategy is possible at least in our data (cf. the learner Paul in section 2). In this way it is possible to capture differentiated developmental styles which are the result of a combination of process orientation (top-down vs. bottom-up) and level orientation. More is being described than just the complete acquisition of a new structural level, the CP, because

70

F.-J. d'Avis & P. Gretsch

the changes in the intermediary grammars can be seen as necessary prerequisites for acquisition. Therefore, individual variation within phase (B) is not only possible but also expected. The interim grammars capture the transition during the feature-matching process, whereas the sequence of the superordinate grammars only takes into account "construction completely acquired" vs. "construction not yet acquired", cf. section 3. The criterion for complete acquisition is productive usage of the construction, which of course has to be empirically confirmed. 1 8 This free variation clashes with the requirement of learnability theory that language acquisition can only be explained in a deterministic fashion. The data appear to be compatible both with a deterministic process and with a non-deterministic process rich in variation. This paradox is resolved by subdividing the developmental process into superordinate grammars and subordinate interim grammars. It is on the level of feature acquisition, or rather feature matching (that is, at the level of the interim grammars), that variation occurs, whereas the level of the superordinate grammar, of complete developmental steps, is subject to rigid determinism. How can a language learner arrive at a new feature instantiation? Of course, the process of feature instantiation could be interpreted along the lines of traditional parameter fixing just as any parameter can be represented as a mapping between feature name and feature value. In that case, though, one is susceptible to the same criticism that has been raised against the lexical parameterization hypothesis (cf. Wexler & Manzini 1987) and related proposals: As individually proposed parameters have come to correspond to narrower differences between languages, the problem of promiscuous descriptiveness has become more apparent. (Safir 1987:78) As we have already mentioned above (cf. 1.3) the identification of parameters with features would go against the basically generalizing quality of parameters. We will therefore refrain from interpreting feature instantiation as the fixing of parameters. To what extent the C head as a functional category should be considered as given 1 9 can be left aside for the moment. Irrespective of whether one interprets the growing accessibility of the CP level as development along the lines of some maturational theory or as "triggered" according to the continuity hypothesis, it is clear that the accessibility changes in the course of time and this change should be accounted for by an adequate model. The building-block model offers the possibility of joining together apparently irreconcilable strategies within the acquisition process: The language learner can proceed both analytically with respect to a given template of the type, and synthetically with respect to the assembly of building blocks. The condition of learnability, which requires a deterministic model, is fulfilled by the ordering of the construction-specific grammars with respect to each other, while at the same time we find non-determinism in the realm of the interim grammars, which allow some limited spectrum of variation. This necessarily more flexible conceptualization of a developmental model tries for the synthesis of so far conflicting positions and is 11A 0 19

The status completely acquired should be assigned with caution, allowing for production errors, citations and so on. In the sense of a strict continuity hypothesis (Pinker 1984) or a full competence hypothesis (Poeppel & Wexler 1993).

71

Variations on 'Variation "

therefore more adept at handling the data than previous models which will be described below. For a more detailed description of the building-block model and its implications for syntactic structures and developmental models, cf. Gretsch (1993). A more precise definition of interim grammars will be given in section 3, where we discuss the notion of sequence, which lies at the heart of the building-block model.

2.

The range of variation in the data

In this section the spectrum of variation which we found with respect to the acquisition of complementizers will be discussed on the basis of data from four monolingual children (Valle, Benny, Paul, and Max). The perspective taken will be multi-dimensional rather than linear, that is, we will characterize some restricted developmental space rather than attempt to identify a central tendency with deviations. This developmental space can be used in different ways by individual children, and each developmental path is in general equally possible. We find, for example, inter-individual divergence with respect to the approach taken by the child and the time required for the acquisition of complementizers. Taken together, these yield for each child a characteristic picture of the acquisition process, a picture which nevertheless falls within the boundaries set by universal grammar (cf. Chomsky 1986). This account will receive some corrections based on the data presented below. We will occasionally refer to the acquisition of the passive in comparison with the development of complementizers in order to round off the respective developmental pictures (cf. Fritzenschaft, this volume). The same developmental styles or strategies 20 appear to manifest themselves in different domains of acquisition. 21 The consistent or variable use of these strategies then constitutes a domain of variation of its own. Since the initial systems of the four children have been discussed in greater detail elsewhere, cf. Fritzenschaft et al. (1990), we can restrict ourselves to the following overview over the distribution of verbs and finiteness features, where VE = verb end. (19)

Table of verb positions during phase (A):

LEARNER

Valle

Benny

Paul

Max

V POSITION

V2/VE fin

V2/V3/VE fin

V2 rin

V2/VE fin /VE nonfin

This inter-individual variation in the initial systems has to be taken into account in the analysis of different precursor structures. The following labels for phases have to be seen relative to the acquisition of complementizers. While we occasionally refer to the acquisition of other structures, such as the passive, we will not attempt to propose any structure-specific phases for that particular construction.

For a discussion of the differences between styles and strategies, cf. Kaltenbacher (1990:18). 21

Cf. Tracy (this volume), and Gretsch (1993).

F.-J. d'Avis & P. Gretsch

72

2.1

Valle

The relevant period of investigation (overlapping with the period discussed by Tracy, this volume) during which Valle proceeds from phase (A) to (C) covers three months, from 1 ;11 to 2;2. (20)

Temporal schema for Valle's developmental pattern:

PA-Ce

:

Β

[-

C

PA-AC

- >

> >·

PA-CC



PA-RC

Age

>

A'

- >

-Ψ.

4-

1

B

A'

C

- >

h-

:

> H 1;10

1 1;11

1 2;0

1 2;1

1 2;2

1 2;3

1—> 2;4

At the beginning of phase (B), Valle prefers a function-dependent correlation since [+w]structures (i.e. questions) correlate with VE, and [-w] with V1/V2 (cf. also Fritzenschaft et al. 1990, Tracy, this volume). It is possible to suggest for Valle at this time two different structural types with just this feature-form correlation. Consequently, he has no problems placing the finite verb in subordinate clauses, since this position is already in use. The first structures which we would like to regard as precursors of subordinate clauses are already target-like, which, of course, makes it difficult for us to argue that they are actually precursors. But later on in his course of acquisition difficulties arise which make us believe that the term precursors for those early utterances is correct. The earliest relevant utterances in Valle's data are three occurrences of the relative particle wo ('where'), which suggest at least for the type of relative clauses some direct move from phase (A) to (C) even though actual relative pronouns are still missing. In contrast to this, the examples for complement clauses (compare (23) and (24)) appear to belong to the category of quotations because the complementizer ob ('whether') is not documented again until four months later. Since some previous investigators of the acquisition of subordinate clauses in German have so far assumed that the acquisition proceeds in the order adverbial clause, complement clause, relative clause, the early appearance of relative clauses is noteworthy. 22

22

Cf. Rothweiler (1989:83): "Relevant sind für die früheste Phase des Erwerbs der Verbendstruktur sowohl Komplementsätze als auch Adverbialsätze, während Relativsätze so gut wie gar nicht auftreten (...)."

Variations on "Variation"

73

(21)

(V. labels an overturned basket as a bam and sits down on it) dds is ein dal (=Stall) wo man drauf sitzen kann\ this is a barn where one on sit can 'this is a bam on which one can sit'

Valle03 (1;11.25) (1 .RC) 23

(22)

(talking about the bam) wo so die kühe rein können\ wo die kühe rein können\ where so the cows in can. where the cows in can 'where the cows can go in like this'

Valle03 (1 ;11.25)

(23)

Adult: Pfeift das nicht?'Doesn't that whistle?' Valle: valle hat biat ob des pfeift\ Valle has tried whether that whistles

Valle03 (1;11.25) (1 .P-CC)

(24)

(Valle is shaking his coin purse) hat siittelt ob kein geld dine is\ has shaken whether no money inside is 'V. has shaken the purse to see whether there is any money in it'

Valle03 (1; 11.25)

(25)

{laudwilla} dat dich mal ditzeh loudly-wants-he that you PRT tickle 'he loudly wants me to tickle you'

Valle04 (2;00.10) (1 .CC)

In Valle04 we find a complement clause, cf. (25), although the beginning of the utterance is hardly intelligible, as indicated by the curly braces. Since Valle produces a target-like complement clause again in the next corpus, the interpretation of (25) as a precursor of a complement construction can be justified. The combination of matrix verb and lexical complementizer in (24) could be considered possible, but slightly deviant. In any case, it seems that verb placement in complement clauses has been acquired. This early acquisition of verb placement in subordinate clauses suggests that Valle is a top-down learner with syntactic orientation. The appearance of target-like daß ('that') complements as well as the occurrence of ob ('whether') show that he also has available individual instantiations, i.e. individual complementizer tokens. For adverbial clauses, a similar developmental picture emerges: The first documented adverbial syntagma (26) is well-formed except for the participle, which we will ignore at this point.

To clarify the progression of acquisition in the examples, we introduce the following abbreviations for the first occurrence of a construction in the data: l.P-AC = first occurrence of a precursor for adverbial clauses l.P-CC = first occurrence of a precursor for complement clauses l.P-RC = first occurrence of a precursor for relative clauses 1.AC - first occurrence of an adverbial clause l.CC = first occurrence of a complement clause l.RC = first occurrence of a relative clause

F.-J. d'Avis & P. Gretsch

74

(26)

(Valle is sitting in the bathtub)

Valle04 (2;00.10)

valle badet dn weiter bis du fertig malt hast\

(1 .AC)

Valle bathes then on until you finished drawn have 'Valle will go on bathing until you have finished drawing' The claim that Valle is in phase (B) is supported by three examples, cf. (27) to (29), which, because they are unintroduced subordinate clauses, suggest that he has not yet acquired the relevant adverbial complementizer but that he already has the type available. This fits in with his overall top-down procedure and his syntactic orientation. (27)

(28)

Valle: dds glas geht nicht put (=kaputt)\ 'That glass won't break' Adult: Doch das kann kaputtgehen. 'Yes it can break' (Valle takes the glass into his hands) Valle: {dds nich} hinfdllt\ that not down-falls 'so that it doesn't fall'

Valle04 (2;00.10)

(Valle is sitting in the bathtub)

Valle04 (2;00.10)

(l.P-AC)

valle will dann weitermalen\ valle fertig malen hat\ valle fertig badet hat\

(=(7))

Valle wants then on-draw, valle finished draw has, valle finished bathed has 'Valle wants to go on drawing [when] Valle has finished drawing [corrected by Valle himself to:] [when] Valle has finished bathing' (29)

dds immer hinfällt hat's draußen scherben\ that always down-falls has'it outside broken glass '[If/whenever] that falls down there's broken glass outside'

Valle04 (2;00.10) (=(2))

These utterances are particularly striking because so far in Valle's system VE sentences correlated significantly with the feature [+w], compare also Tracy (this volume). (27), in contrast, has a purpose or temporal interpretation, (28) a temporal one, and (29) a conditional reading. It should be noted, in particular, that (29) carries an overall uninterrupted intonation contour. To summarize: It can be concluded that Valle acquires the building block of VE placement in complementizer-introduced clauses early. His first precursors therefore do not appear particularly articulate. In the further course of development more and more lexical complementizers, that is, single tokens, are added. He also increases the set of his matrix verb-complementizer combinations, cf. (30). (30)

valle schaut zu wie des bärle klettert Valle is-watching PREFIX how the little-bear climbs

Valle05 (2;00.24)

As for adverbial complementizers, he acquires the lexical items weil ('because'), wenn ('whenyip, and purpose daß ('so-that'), which are used according to the target system. There are seventeen well-formed adverbial clauses in VallelO (2;02.24), partly elicited by wAyquestions (seven instances), as illustrated in (31), partly spontaneously uttered (ten instances),

Variations on "Variation "

75

as in (32) and (33). There are seven correct relative clauses, including those with d- pronouns as in (34). We find only four adverbial clauses in this corpus which are incomplete because Valle interrupts himself. However, the ratio of seventeen to four allows us to conclude that he no longer has problems with subordinate clauses. In this corpus there is only a single complement clause, namely (35), but in the following corpora we find complement clauses in combination with the matrix verbs zugucken wie, zuschauen wie, gucken wie, gucken was, gucken ob, as in (36), all roughly corresponding to 'watch how', and wissen wie ('know how'), which all conform to the target system. (31)

Adult: Warum steigt der ab?'Why is he getting off?' Valle: weil'er nimmer (=nicht mehr) baggern will·, because'he not-anymore dig wants 'Because he doesn't want to dig anymore'

VallelO (2;02.24)

(32)

Valle: dann nimmer (=nicht mehr) fahren\ then not-anymore drive '[he wo]n't drive anymore' Adult: Nee! 'No!' Valle: wenn der ein Unfall macht\ if that-one an accident makes 'if that-one causes an accident'

VallelO (2;02.24)

(33)

macht der bauer die türe zu daß es nich in den bulldog regnet\ makes the farmer the door shut so-that it not in the bulldozer rains 'the farmer closes the door so that it doesn't rain into the bulldozer'

VallelO (2;02.24)

(34)

da sin alle legos die ich ausgesüttet hab\ there are all legos that I dumped have 'There are all the Legos that I dumped'

VallelO (2;02.24)

(35)

Adult: Was denn? Was soll'n wir denn gucken? 'What PRT? What shall we look at?' Valle: ob der bauer nonich (-noch nicht) ichtig (-richtig) is\ whether the farmer yet-not correct is 'Whether the farmer isn't correctly [placed] yet1

VallelO (2;02.24)

(36)

dda guckt grad ob das eine decke ist\ that-one is-looking just whether that a blanket is\ 'That-one is just looking whether that is a blanket'

Valle 11 (2;04.14)

F.-J. d'Avis & P. Gretsch

76

(37)

Summary of Valle's developmental process:

Duration of phase (B):

from 1; 11 to 2;02, i.e. three months

Strategy:

top-down learner, because he acquires the 'functional' verb position before the other building blocks Valle pursues a syntactic orientation since he does not

Orientation:

produce errors of verbal placement but rather preconjunctional structures adverbial, complement, and relative clauses appear at the

Course of development:

same time Valle produces hardly any quotations; no escape to other

Further characteristics:

constituents can be found In addition, his phase (B) for complementizers correlates with first precursors of passives and a first infinitival complement. 2.2

Benny

For Benny, the period relevant for our discussion starts at 2;9 and lasts until nearly 4;5, that is about a year and a half (cf. also Gawlitzek-Maiwald et al. 1992). (38)

Temporal schema for Benny's developmental pattern:

PA-C"

PA-AC

: >*}-

-B-

>

1

I —

C

>

' C'

PA-CC

C

..

>

1

PA-RC

Age

A "

>

: >

I I I I I I I I 1 I I I I > 2:8 3;0 3:4 3;8 4;0 4;4 4;8

Examples (39) and (40) show the beginning of Benny's phase (B), when there is evidence for sentences introduced by complementizers. In both cases, an uninterrupted intonation contour suggests that we are dealing with a hypotactic sequence of syntagmas as an instantiation of complement clauses: (39)

Benny: weiß es net was ich mache\ know it not what I do Ί don't know what to do'

Benny04 (2;09.19) (1.P-CC)

77

Variations on "Variation"

Adult: Dann laß dir doch mal was einfallen. 'Then try to think of something' (40)

(talking about a puzzle) Benny04 (2;09.19) Adult: Mach's mal kaputt, die Ira soll mal probieren, ob sie's auch kann. 'Destroy it, let Ira try if she can do it, too.1 Benny: gucke wann anfänge\ watch when start '[I want to] watch when [she] starts'

The status of the utterance (41), which belongs to the same corpus, is more difficult to determine. Are we dealing with a paratactic or with a hypotactic structure? Except for a short pause before du ('you'), the intonation contour is uninterrupted. Quite possibly, du ich böse is a quote with some uncertainty as to the correct shifter pronoun. (41)

mama schimpfe du ich böse\ mommy scold you I naughty 'Mommy scold[s] you (=me?) when I'm naughty'

Benny04 (2;09.19)

Similar structures occur in the next two corpora. Compare (42) and (43), where there is also no break between clauses. This prosodic integration makes these utterances qualitatively different from the exclusively paratactic strings which are characteristic of the initial phase (A). The fact that we find finite verbs in final position even in matrix clauses - compare (43) to (45) - suggests that so far Benny has not established any systematic connection between the VE pattern and complementizer-introduced clauses as we would expect from his variable verb-placement pattern in phase (A), cf. table (19). His precursor constructions therefore take a different shape than the well-known preconjunctional structures of other studies. (42)

(Benny walks into the flat and is looking for a ladder)

Benny05 (2; 10.29)

i han guckt keine leiter han\24 I have looked no ladder have Ί have looked; we have no ladder' (43)

pipistengele halte dann ich pipi mach\ little-dicky hold then I peepee make '[If you] hold [my] little-dicky then I will pee'

Benny06 (2;11.21) (1 .P-AC)

(44)

was du machen/ {wenn dich} überßhrn\ what you make, when you knock down 'What will you do if I knock you down?'

Benny06 (2;11.21)

nein du hehefährt ganz ganz schnell

Benny06 (2; 11.21)

(45)

dann ich ganz schnellfähr\ no you by-drives very very fast, then I very fast drive 'No, [when] you drive by very fast, then I'm going to drive very fast' The Swabian han can be l.pers.sg. as well as 1.pers.pl.

F.-J. d'Avis & P. Gretsch

78

If one looks at each example individually, several alternatives of interpretation suggest themselves. But they can all be credited with a semantic complexity that constitutes a systematic pattern. Since some of these examples contain no overt indicators such as verb placement or lexical complementizers, Benny's orientation appears to be primarily semantic. He can be considered a token-oriented learner, using a bottom-up strategy. The following one-and-a-half months go by without particularly remarkable precursor structures, but also without complete subordinate clauses. There are, however, several possibly quoted occurrences of weil ('because'). This shows that he is already receptive to that lexical item in the input. It is in corpus Benny08 (3;00.19) that we find more precursors - the first one, (46), which is quoted, 25 and a second one, (47), an apparently intended complement clause which is realized paratactically. To what extent (47) can actually be included among precursor structures cannot be decided. The same corpus also contains two combinations of a matrix clause and a complement clause. One of them, (48), exhibits a C+V2 pattern, 26 and one, (49), shows the verb in an apparently target-like position, modeled after the immediate input. (46)

(Benny is fetching a tool box)

Benny08 (3;00.19)

i han mein werkzeug\ wart mal ob - ob de\

(=(10))

Ί have my tools, wait PRT whether - whether the' (47)

(48)

ich kann schon alleine\ möcht ich zeigen dir\ I can already alone, want I show you Ί want to show you that I can already do it by myself

Benny08 (3;00.19)

(Benny talking about the mermaid figure in a fishing game)

Benny08 (3;00.19)

will die meerjung/rau haben daß du has net die meerjungfraiA want the mermaid have so-that you have not the mermaid Ί want to have the mermaid [figure] so that you don't have the mermaid' (49)

25

Benny08 (3;00.19)

The quoted character of the utterance is supported by two ob ('whether') contexts in which ob is still absent: (i) (Benny is to repeat a question) Bennyl2 (3;02.26) Adult: Und was wollte der hier? 'And what did he want here?' Benny: der willfragen de {darfer} auch mit einlade/ that-one wants ask FILLER may he also with invite 'He wants to ask whether he may invite too' (ii)

26

Benny: was macht du hietsrichter (=schiedsrichter)\ what makes you referee 'What are you doing, referee?' Adult: Ich paß auf, daß keiner beschummelt. 'I'm making sure that nobody cheats'

m ich muß aber au mal probieren paßt der auch da rein/ FILLER I must PRT too PRT try fits that too there in Ί must also try whether that fits in there too'

Benny 14 (3;05.00)

Cf. Fritzenschaft et al. (1990) for a more detailed description of Benny's "superficial" V3 structures, e.g. warum weiter geht nichts (Benny06, 2:11.26).

Variations on "Variation "

79

Benny: er paßt auf daß keiner schummelt\ he watches out that nobody cheats 'He's making sure that nobody cheats' In Benny09, we continue to find interesting word order patterns. There is one case of wenn+V 1, cf. (50), and a particularly elegant solution for Benny's verb-placement dilemma: a case of weil without finite verb, cf. (51). Example (52) illustrates another adverbial precursor, here with a conditional reading, which resembles example (43) from Benny06. It is this coexistence of overt complementizers and precursor structures, as in Benny's conditional or temporal utterances, which makes us believe that for him the semantic level still has priority. (50)

Adult: Wann hast du denn das letzte Mal Frisbee gespielt? 'When was the last time you played frisbee?'

Benny09 (3;01.04) (=(8))

Benny: wenn hab ich geburtstag habt dann han ich nein nein ich ganz au mal burtstag habt dann krieg ich dds au mal pispi\ when have I birthday had then have I no no I very too once birthday had then get I this too once frisbee 'When I have my next birthday then I'll get a frisbee, too' (51)

Adult: Warum willst'n Papa naß machen? 'Why do you want to make daddy wet?' Benny: weil er zu mir ganz bös\ because he to me very naughty 'Because he [is] so naughty to me'

Benny09 (3;01.04) (=(9))

(52)

nichtpispi spielen dann du was anderes\ Benny09 (3 ;01.04) not frisbee play then you something else '[If we] don't play frisbee then you name something else we can play'

As of BennylO (3;01.13), he uses weil ('because') in the variations weil+Vl, weil+V2 and weil+VE. This word-order variation in combination with weil persists for about one month until Bennyl2 (3;02.26). The following examples (53), (54), and (55) illustrate one structural format each. (53)

Adult: Warum?'Why?'

BennylO (3;01.13)

Benny: weil dann kann du mich nicht heben\ because then can you me not lift 'Because then you can't lift me' (54)

Adult: Warum net?'Why not?'

Benny: weil sie dann noch ein kleines mädchen war\ because she then still a little girl was 'Because then she was still a little girl'

Benny 11 (3;02.00)

(1 .AC)

F.-J. d'Avis & P. Gretsch

80

(55)

Adult: Warum hat das geweint?'Why did that cry?' Bennyl2 (3;02.26) Benny: weil hat du dds so fest dran getut darum hat dds geweint\ because have you that so tight thereon put therefore has that cried 'Because you put it on so tight, that's why it cried'

After Benny 13 (3;03.21), he restricts this spectrum to the target-like wei'/+V2 and weil+VE. As of Benny 15 (3;06.02), he further restricts weil structures to an exclusive weil+VE. The same development of structural narrowing applies to wenw-clauses, as demonstrated in (56)(57). (56)

du wenn däs dreht sich was tut's dannl wenn's dds dreht! you v o c if that turns itself what does'it then, if it that turns 'Hey, if that turns what does it do then, if that turns?'

Bennyl2 (3;02.26)

(57)

doch des brauch ein haus wenn 's mal regnetl Benny 18 (3; 10.20) und mal schnee kommt\ dann macht's dann macht's sich ja flugzeug dreckig but that needs a house if it PRT rains, and PRT snow comes, then makes'it then makes'it itself PRT airplane dirty

The fact that only the lexical items weil and wenn occur suggests some lexical token-by-token strategy. It is only later that these tokens are assigned a type, as shown by the late correlation with the VE position. The superordinate nature of this type reveals itself in the final (over-) generalization of VE patterns to all we/Z-clauses. This appears to conclude Benny's phase (B) with respect to adverbial clauses. As for complement clauses, in Bennyl2 (3;02.26) we find examples for daß ('so-that') in correlation with VE patterns, cf. (58) with a non-finite verb in Benny's last turn,28 whereas his complement-introducing ob ('whether') does not occur until Bennyl5 (3;06.02), cf. (59). (58)

Adult: Was soll ich ihm versprechen? 'What shall I promise him?'

Bennyl2 (3;02.26)

27

Note the doubly filled subject position ('s and flugzeug), which might indicate that Benny still has problems with clitics or pronominal elements.

28

Two further examples support this assessment of Benny's development: (i)

(playing with puppets where the figure Benny 12 (3;02.26) spoken for by Benny is to repeat everything) Adult: Ich freue mich, wenn es nicht regnet. 'I'm happy when it isn't raining' Benny: ich freue mich daß nicht das regnet\ I am-happy REFLEXIVE that not that rains 'I'm happy that it isn't raining' Adult: (repeating herself) Ich freue mich, wenn es nicht regnet. 'I'm happy if it isn't raining' Benny: ich freue mich daß es nicht regnetS I am-happy REFLEXIVE that it not rains 'I'm happy that it isn't raining' (ii) ich möchte bitte daß du aufstehst Benny 18 (3;10.20) I want please that you stand-up Ί want you to stand up, please'

Variations on "Variation"

Benny: daß du nicht da kucheti ganz alleine eßt\ that you not the cake completely alone eat 'That you won't eat the cake all by yourself Benny: (...) da stimmt doch nicht was/29 '(...) there's something wrong' Adult: Was stimmt denn da nicht? 'What's wrong there?' Benny: daß doch der nicht ganz alleine kuchen essen\ that PRT he not completely alone cake eat 'That he doesn't eat [the] cake all by himself (59)

81

(1 .CC)

(playing with a knight's castle) Bennyl5 (3;06.02) du! eigentlich jetzt guck mal de wauwau rein ob de ritter hoch drinne is\ you voc / actually\ now look PRT the doggy in whether the knight high-up inside is 'Hey you, well, now the doggy looks [to see] whether the knight is high up inside'

During the next months Benny further expands his inventory of complementizers. In Bennyl5 (3;06.02) he adds bis ('until') to his adverbial complementizers. As of Bennyl8 (3;10.20), we find the matrix predicates sehen daß ('see that'), möchten daß ('want that'), wissen wie ('know how'), gucken ob ('look whether') and in Benny20 (4;02.01) wissen daß ('know that'). This variety of combinations consisting of matrix predicates plus complementizers suggests that Benny has established a target-like type. After the type has been established, the integration of new lexical complementizers is no longer problematic. There is only one deviant word order among these complement examples, cf. (60): (60)

nein dann weiß ich nicht wie geht da wieder zusammen\ no then know I not how fits there again together 'No, then I don't know how it fits together again'

Benny 18 (3; 10.20)

As was the case with Valle, Benny's adverbial and complement clauses develop in parallel. However, the two learners differ with respect to the acquisition pattern of relative clauses. Benny does not acquire relative clauses until after he uses his other subordinate clause types productively. The first precursors, see (61) and (62), as well as two complete relative clauses, (63) and (64), are documented in Benny21 (4;04.30). At the same time he already uses various kinds of passives as well as infinitival complements. (61)

(Benny is talking about his video cassettes) Benny21 (4;04.30) und dd wo jumbo hab ich auch - hab ich auch wo - wo ich herbie drauf hab\ (1 .P-RC) and [the/there] where Jumbo have I also - have I also where - where I Herbie on have 'And I also have a cassette with Jumbo and with Herbie.'

(62)

Adult: Die haben - was halten die denn da? 'They have - what are they holding?'

Benny21 (4;04.30)

It should be noted that the word order is deviant with respect to the negation. The target-like equivalent would be: da stimmt doch wo? nicht (was = short for etwas 'something').

F.-J. d'Avis & P. Gretsch

82

Benny: die da dds wo vorher\ und da kann er fliegen und {die xxx}\ those there that where earlieA and there can he fly and {the xxx} 'They're holding that which earlier (was held by others?), and there he can fly and ...' (63)

(talking about the picture book "Pooh the Bear") Benny21 (4;04.30) Adult: Und der Pooh hat immer Hunger. Was versucht der hier? 'And Pooh is always hungry. What is he trying to do here?' Benny: honig rauszulecken\ ha wo dem bienen gehört\ (1 .RC) honey out-to-lick\ PRT where to-the bees belongs 'To lick up the honey that belongs to the bees'

(64)

Adult: Mensch, ist der dumm. Der weiß nich mal, was ein Benny21 (4;04.30) Fernlenkauto is. Sag ihm das mal. 'Oh dear, he's so dumb. He doesn't even know what a remote-controlled car is. You tell him.' Benny: ein fernlenkauto is ein auto wo eine fernbedienung hat un wo man dann vorwärts rückwärts hupe und licht anmachen kann\ a remote-control car is a car where a remote control has and where one then forward backward horn and light switch-on can Ά remote-controlled car is a car which has a remote control and where one can manipulate forwards, backwards, the horn, and the lights'

(65)

Summary of Benny's developmental process:

Duration of phase (B):

from 2;9 to 4;6, that is, 21 months bottom-up learner, who abstracts a type on the basis of

Strategy:

several tokens Orientation:

semantic orientation with various deviant patterns of verb placement

Course of development:

relative clauses appear significantly later than adverbial and complement clauses

Further characteristics:

2.3.

difficulties in verb placement

Paul

Paul's phase (B) does not begin until he is three-and-a-half years old (Paul05 (3;05.27)), and lasts about nine months until he is four (4;02.15).

83

Variations on "Variation"

(66)

Temporal schema of Paul's developmental pattern:

PA-Ce

:



PA-AC

:

>

PA-CC

:

> ·

PA-RC

:

Β

*

Β'

1 A'

β;

J

C

ι

c

Β1

A'

l· Age

:

>

I I I I I I I I I I I I I 3;4 3;β 3;8 3;10 4;0 4;2 4;4

>

The first precursor structures, (67) to (69), are documented relatively late, that is, at the age of almost three-and-a-half, and are restricted to adverbial structures. produces one target-like weil+VZ

In addition, he also

structure, cf. (70). Since at the beginning of his precursor

phase (B) Paul constructs his main clauses exclusively with V 2 (see the verb position table at the beginning of section 2), Fritzenschaft et al. (1990:85, cf. also Gawlitzek-Maiwald et al. 1992) assumed that he not only needed to establish the functional correlation between V E placement and subordinate clauses but that he had to develop the V E format as such. Examples (67) to (70) support this interpretation. (67)

Adult: Warum denn nicht?'Why not?' Paul:

Paul05 (3;05.27) (1 . P - A C )

η deda is doch gar keine tür\ F I L L E R there is P R T at-all no door '[But/because] there's no door there at all1

(68)

Adult: Warum stellst du das Signal dahin?

Paul05 (3;05.27)

'Why are you putting the signal there?' Paul:

9d 9d

schranke auj\

F I L L E R F I L L E R barrier open 'To open the barrier' OR: 'Because I want the barrier open' (69)

(an adult commenting on the house Paul is building)

Paul05 (3;05.27)

Adult: Ein grünes Haus und ein rotes Dach. Das ist aber ein biBchen schmal, das Haus? Ά green house and a red roof. Isn't it a bit narrow, the house?'

Paul:

η nein des ist nich zu weil hoch\ F I L L E R no that is not too because high 'No it's not too [narrow] because it's high'

(70)

(Paul points to a place where the train is not allowed to go)

Paul05 (3;05.27)

da fahrt die Eisenbahn\ weil da raus darf man nichtfahren gell/ there goes the train, because there out may one not go, isn't-it-so 'The train goes this way, because it isn't allowed to go in that direction'

F.-J. d'Avis & P. Gretsch

84

The utterances in (67) and (68) illustrate a typical pattern for Paul: As long as he has not acquired some new construction type, he tends to repeat himself, to break off and to change the topic, and to produce undifferentiated sounds at the left periphery. 30 Of course, it is possible to respond to a wAy-question with a V2 matrix clause without any explicit weil. Such utterances would not have been included among the precursor structures, if it had not been for his systematic production difficulties, which seem to indicate that he is trying to express more than he is actually able to say. It therefore seems justified to consider (67) and (68) as adverbial clauses which are introduced by undifferentiated filler syllables. (69) is an example of a completely reduced adverbial clause fragment. (70), however, qualifies as a well-formed weil+\2 sentence. Paul is the only learner in the group discussed here, whose utterances suggest some connection between complementizers and prepositions in the acquisition of complex clauses. 31 Besides example (71) from the first corpus in which we found precursor structures, one-and-a-half years later, in Paul37 (5;01.24), there are causal wegen+VΕ ('because-of) sentences. This suggests that the relevant lexical item wegen has been classified in a way which differs from the target. (71)

(Paul is explaining the function of a barrier)

Paul05 (3;05.27)

bei zu darf 9d darf man nich fahren bei die schranke\ bei zu schranke darf man nich fahrn\ at closed may FILLER may one not go at the barrier, at closed barrier may one not go 'When/if the barrier is closed one is not allowed to go' Conditionality is expressed here with bei ('at'), Paul being the only child in our group who produces conditional prepositions. The exact status of the two occurrences of zu ('closed') remains unclear. The next two corpora also exhibit exclusively precursor structures: One interrupted wenn clause with following change of topic, (72), one possible relative clause precursor, (73), two descriptions of complex events, (74), and the precursor of a complement structure, (75): (72)

wenn man so ein so (INT, continuing in a different voice) lebendig fußgänger\ if one so a so - alive pedestrian

Paul06 (3;06.12) (=(5))

(73)

(Paul is looking at holes in his tow truck) die wo die die da sind {nikas}dranknoten\ these where these these there are {nikas} on-knot 'There where these are [the] {nikas} [should be] knotted on'

Paul06 (3;06.12) (1 .P-RC)

30

Such examples are not exceptions but are part of the standard repertoire of Paul's structures, cf. Fritzenschaft et a!. (1990:84).

31

For the relevance of prepositions to the acquisition of complementizers, cf. Pinker (1984), Roeper (1986), and Müller (1993).

85

Variations on "Variation"

(74)

Adult: Das ist ein Luftballonauto. Wenn man die Luftballons aufbläst 'That's a balloon car. If one blows up the balloons - '

Paul:

Paul06 (3;06.12) (=(13))

dann fährt das auto los\

then drives the car off 'Then the car drives off Adult: Und wenn der Luftballon leer ist? 'And in case the balloon is empty?'

Paul: dann Jährt das auto nich los\ then drives the car not off 'Then the car doesn't drive off (75)

du weißt du wieso ist da eine matratze drin/ y o u v o c know you why is there a mattress in-there 'Hey, do you know why there's a mattress in there?'

Paul07 (3 ;06.13) (1 .P-CC)

The utterance in (75) exhibits one overall prosodic contour, thus pointing to a hypotactic structure, even though the final rise and the verb placement are deviant. The following corpus contains the first complete adverbial clause with VE position, (76). But there are still two precursors ending in self-interruptions. The first case, (77), could be an attempted complement structure, where wie is ambiguous ('how1 vs. wieso 'why'), which is, similar to (75), changed into a direct question. There are two clearly delimitable intonation contours. The second example, (78), shows an interrupted adverbial clause with change of topic and a deviantgucken+\2 ('look') complement (again with closed intonation). (76)

Adult

Paul

(77)

(addressing Paul in the role of a bear): Paul08 (3;06.27) Ah ja, ich helf dir. Warum bist du nicht lieb zu meiner Maus? Du Mickeymaus! 'Oh yes, I'm going to help you. Why aren't you nice to my mouse? You Mickey Mouse!1 (acting as Mickey Mouse): weil ich ärgern will· (l.AC) because I annoy want 'because I want to annoy [you]'

(Paul discovering a grandfather's clock) Paul: da is ne uhr\ 'There is a clock' Adult: Da hast du deine Uhr wiedergefunden. 'There you have found your clock again'

Paul08 (3;06.27)

Paul: samandu wieddwieduwiev - warum ist des ne uhrl say y o u v o c how/why FILLER how/why FILLER how/why - why is that a clock 'Tell me, why is that a clock'

F.-J. d'Avis & P. Gretsch

86

(78)

Paul

(acting as Mickey Mouse): Paul08 (3;06.27) ich geh mal gleich mal rein weil wei wei wei (INT) (=(6)) ja oh ich guck es brennt da drin\ I go PRT soon PRT in because because because because (INT) yes oh I look it burns there inside 'I'll just go inside because (INT) - oh I am-looking how it is burning inside'

In the next two corpora, Paul09 (3;07.11) and PaullO (3;08.01), there is only one weilutterance, (79), with both the undifferentiated beginning and the interruption characteristic for this particular learner. In Pauli 1 (3;08.12), he produces another complement clause precursor, (80), one wei/-sentcnce, (81), and one utterance with the finite verb in last position, (82). It is to be noted that he constructs this example - the second documented VE pattern after (76) - without an overt complementizer, although one could interpret the η as an undifferentiated complementizer or as a universal particle, cf. footnote 3 and example (67). (79)

Adult: Warum ruft'n der Peter den Papa? 'Why is Peter calling for daddy?' Paul: 9d ?d weil η regen (INT) FILLER because FILLER rains 'Because it is raining'

PaullO (3;08.01)

(80)

Paul

(acting as the figure 'Peter'): mag auch raus\ want also out Ί want to go out, too' Adult: Der Peter will auch raus? 'Peter wants to go out, too?' Paul: der peter sagt auch raus\ the peter says also out 'Peter says that he wants to go out, too'

Paulll (3;08.12)

(81)

Adult: Der braucht - du brauchst'n großen Parkplatz. 'He needs - you need a big parking space' Paul: weil weit weg park ich bei den schafen\ because far away park I near the sheep 'Because I park far away near the sheep'

Pauli 1 (3;08.12)

(82)

du der will will er heute η schon wir nich gemacht haben\ Pauli 1 (3;08.12) you v o c he wants wants he today FILLER already we not made have 'He wishes today that we hadn't done that'

Like Benny, Paul has problems with verb placement, but in contrast to Benny's variable word-order patterns, Paul seems to experience systematic production problems and therefore tries to avoid certain structures. The undifferentiated filler syllables in sentence-intial position suggest a tendency of top-down learning with an undifferentiated type, together with an orientation towards the phonetic level. But in contrast to Valle and Benny, Paul does not appear to be restricted to a particular strategy. His complex utterances without fillers or universal particles suggest an additionional orientation towards the semantic level. Two

87

Variations on "Variation"

weeks later we find, beside some more interruptions, 32 one adverbial clause with a typically undifferentiated onset, cf. (83), and a first target-like relative clause (84). (83)

Adult: Wie soll das dann sein? 'How should that be?'

Paul:

Paull2 (3;08.26)

η η du ?θ wenn die wenn die 9d die lokomotive da nam die seite re hoch fahrt dann muß da ein kurzes stück hin\ FILLER FILLER y o u v o c FILLER when the when the FILLER the locomotive there FILLER the side FILLER up goes then must there a short piece go 'If the locomotive goes up on this side then we need a short piece here'

(84)

(Paul is looking for a cow without a bell)

is des ?d dd a die und die kuh die keine Glocke hat/

Paull2 (3;08.26)

(1 .RC)

is that FILLER the FILLER the and the cow that no bell has 'Is this the cow that doesn't have a bell?' As for complement clauses daß ('that') gets added to the inventory of complementizers, as (85) illustrates. The position of the finite verb is not clear: Either this is a VE construction with an extraposed draußen ('outside'), or we are dealing with a case of V3. The intonation contour is closed. In the same corpus, Pauli3, we find a combination of aufpassen ob ('watch whether1) with deviant word order and a doubled aber ('but'), cf. (86). (85)

(Paul takes a sheep from the trailer)

Paull3 (3;09.12)

und die und die papa hat'nen kleinen schäfchen\ and the and the daddy has-a little sheep (Paul places the sheep near the bam)

und dd will daß zwei raus schlafen draußen\ and he wants that two out sleep outside 'And he wants two of the sheep to sleep outside' (86)

32

Adult (in a play): Paull3 (3;09.12) Hallo, du bist unser liebster - liebstes Hundetier. (=(11)) Du darfst da neben dem Stall wohnen, hier, und wir bringen dir auch immer was zum Fressen, wenn du auf uns aufpaßt. 'Hello, you are our favorite doggy. You can live there next to the bam, here, and we will always bring you something to eat, if you keep an eye on us.'

These structures resemble interruptions, but they appear to be complete for Paul. The intonation contour and the systematicity of these utterances suggest an actual competence rather than a performance problem, as the following dialogue demonstrates: Paul: ja jetzt kommt der traktoA wat (=weit?) jetzt kommt der traktot\ Paul 12 (3:08.26) yes now comes the tractor, because (?) now comes the tractor Adult: Warum kommt der große Traktor? 'Why is the big tractor coming?' Paul: weih because. Adult: Weil warum? 'Because why?' Paul: weih because. Adult: Was machen wir denn mit dem großen Traktor? 'What are we doing with the big tractor?' Paul: (releasing the trailer) η ne ddda soll der anhänger absein\ FILLER FILLER there should the trailer off-be T h e trailer should be off there.'

F.-J. d'Avis & P. Gretsch

88

Paul

(acting as the watchdog): ichpaß aber ob bei dir aber auj\ I watch but whether at your-place but PREFIX 'But I will watch whether [something happens] to you'

Although Paul at this time produces individual well-formed adverbial relative and complement clauses, we still find systematic precursor structures, a coexistence which we also find in Benny's data. For this reason one cannot yet consider Paul's acquisition of subordinate clauses to be completed: (87)

Adult (acting as a parrot in a play): Paull3 (3;09.12) Kannst du mir sagen, was is das, wo ich da fest - wo ich mich grad festhalt'? Was is das? 33 'Can you tell me what this is, that I'm holding onto? What is it?1 Paul: dnd ede nevddudu hälst da fest ddda geflieg ich wieder runter\ FILLER FILLER {no-if/where-you-you} hold there onto {then there} fly I again downwards 'No, if you hold on there, I'm going to fly down again'

This coexistence of precursor structures and well-formed target-like constructions continues for several corpora, that is, from Paull5 (3;10.10) to Paull9 (4;00.04). There are two more examples with overgeneralizations as in Paul07 and a significant increase in his own whyquestions, with which he elicits adult subordinate clause answers in a sort of language game. 34 The corpus Paul20 (4;00.18) points to a developmental break: From this time on relative clauses are well-formed; it is interesting that there are two cases in which the VE pattern is used in a matrix clause context, as in (88). With sixteen well-formed wenn-clauses, preferably used with dann ('then') or da ('since') consequence, the acquisition of wen/i-clauses seems to be concluded (cf., for instance, (89)). There are three new damit ('so that') +VE constructions, cf. (90), and for complement clauses we find angucken wie here in (91) and one interrupted schau mal wie, which both translate as 'look how'. (88)

Paul

(looking at a picture book): schau - und dann der weint\ look - and then that-one cries

Paul20 (4;00.18)

(89)

Paul

(P. is talking about his sisters): un un un wenn 'se so so groß wie die luisa ist dann dann kommt'se in die schule\

Paul20 (4;00.18)

Note the deviant word-order pattern in the adult utterance, where an embedded interrogative is constructed with V2. 34

This significant increase in the number of wAy-questions during phase (B) can be found in other corpora too. It suggests that the language learner is able to exercise some control over the input. For Paul we find the following: 6 wAy-questions in Paul 17 (3; 11.14) 10 wAy-questions in Paull8 (3;11.21) 32 wAy-questions in Paull9 (4;00.04)

Variations on "Variation"

89

and and and when'she as as tall as the luisa is then then comes'she into the school 'And when she is as tall as Luisa then she'll go to school' (90)

Paul

(talking about skiers in a book):

Paul20 (4;00.18)

schaumal schaumal der skijahrer boh schaumal drdkt der drückt da runter da macht'er da macht der kei - damit er kein um/all macht\ look-PRT look-PRT the skier INTERJECTION look-PRT press he presses there down there makes'he there makes that-one no - so that he no accident makes 'Look there, the skier, he presses down there so that he won't cause an accident' Adult: Der kommt da runter, ja. 'He comes down there, yes.'

Paul: ja\ damit er kein um/all macht\ yes, so-that he no accident makes 'Yes, so that he won't cause an accident1 (91)

Adult (looking at a picture book): Was macht denn der mit der Kamera? 'What's he doing with that camera?'

Paul20 (4;00.18)

Paul: ja was angucken wie die leute da hoch gehen\ yes something look-at how the people there up walk 'He is watching the people walk up there' Although in earlier corpora Paul's use of we//-utterances appeared target-like, his use of weil in Paul20 (4;00.18) is pragmatically deviant. Confronted with Warum fwhy')-questions, he reacts with ja weih ('yes because') or only weih without anything else, as in (92), or with a following V2 sentence, as in (93). The intonation contour suggests that these are not just cases of interruptions, but that the lexical item weil represents a complete answer in elicited contexts. 35 On the other hand, non-elicited wei/-utterances appear in V2 or VE structures and are therefore target-like, cf., for instance, (94). (92)

Adult (looking at a book): Warum lachen die denn? 'Why are they laughing?' Paul: η ja weih FILLER yes because

Paul20 (4;00.18)

(93)

Adult (asking why it isn't possible to use that stove): Warum nicht? 'Why not?'

Paul20 (4;00.18)

Paul: ja weih der is schon alt der kochherd\ yes because, that is already old the cooking-stove 'Yes, because the stove is already old'

35

Cf. footnote 32. We find the same "abbreviated use" of the weil item in Max's data and in the corpus of a bilingual child, Adam, analyzed in Gretsch (1993), so this seems to be a common pattern.

F.-J. d'Avis & P. Gretsch

90

(94)

Adult

(looking at the toys): Der große gehört dir? 'The big one is yours?1 Paul: ja der\ 'yes that one' Adult: Aha! Ί see!'

Paul20 (4;00.18)

Paul: weil ich so bißchen größer bin da gehört'er mir\ because I such a-little-bit taller am therefore belongs'he to-me 'Because I'm a little bit taller therefore he belongs to me' In the following corpus, Paul21 (4;01.04), adverbial precursor structures coexist with wellformed target-like utterances. In this corpus als ('when') joins the adverbial complementizers. As for complement clauses, he continues to construct V2 complements beside examples as in (82) and (91), which are VE. Matrix verbs are restricted to the class of verba dicendi. One month later, Paul seems to have overcome his difficulties of verb placement in complement clauses. Even the holophrastic form guck mal now occurs with verb-final complement clauses, cf. (95). His use of weil is now target-like. (95)

Paul

(playing with a toy car):

Paul25 (4;02.23)

guckmal wie man's machen kann\ look how one'it do can 'Look how one can do it' About a year later, we find wegen as sentence introducing complementizer for a longer period beside weil, also in combination with a cliticized pronoun wegen's, cf. (96). (96)

Adult (talking about composites): Warum heißt ein Handtuch "Handtuch"? 'Why is a hand cloth called a "hand cloth"?'

Paul37 (5;01.24) (=(12))

Paul: ja wegen dsn handtuch ist\ wegen's zum abtrocknen ist\ well because-of it'a hand cloth is, because-of it to off-dry is 'Well, because it's a towel, because it's for drying' Even though this item-specific deviance from the target system could be considered a precursor structure, the development of the C head has already been accomplished. Paul's consistent VE format in complement, adverbial, and relative clauses shows that he has differentiated the syntactic level and thereby completed his type. Since only the particular lexical item wegen is concerned, we assume that this is a purely lexical phenomenon. In addition, the developmentally independent character suggests that the differentiation of the type remains untouched while the lexical entry of a token is temporarily miscategorized. To conclude, we found that Paul belongs to the slow learners of subordinate clauses with particular difficulty in acquiring VE placement. We observed a long period during which interrupted constructions and clauses with undifferentiated filler syllables coexist with targetlike syntactic patterns. At the same time as he is working his way through phase (B), Paul also develops passive structures (cf. Fritzenschaft, this volume) and first infinitival complements.

Variations on 'Variation "

(97)

Summary of Paul's developmental pattern: late phase (B), from 3;5 to 4;2, i.e. lasting about nine

Duration of phase (B):

months top-down learner with undifferentiated filler syllables in

Strategy:

initial position mixed orientation, with both the semantic and the

Orientation:

phonetic/prosodic level playing an important role Course of development:

parallel acquisition of adverbial, complement, and relative clauses

Further characteristics:

different evasive strategies (for instance, change of topic); deviant use of weil for several months; developmentally independent overgeneralization of the preposition wegen as C head

2.4

Max

Among the four learners, Max is the learner with the longest phase (B). He takes almost two years to get from first precursors at the age of 2; 11 to complete acquisition at 4; 10. (98)

Temporal schema of Max's developmental pattern:

PA-C"

::

>*,

PA-AC : >

Β

*f

?

*

PA-CC : PA-RC : Age

A1

?

-

-\—B'/C?

> >

>H—I I I I I I I I I I I I I > 2;10 3;2 3;β 3;10 4;2 4;6 4:10

Max's first precursors in the corpus Max08 (2;11.01) show initial steps towards complement clauses: (99)

(Max tries to place 'Peter' in the tractor) Adult: Doch der paßt, glaub ich. 'PRT he fits, I think' Max: paßt glaub ich auch\ fits think I too Ί too think that he fits'

Max08 (2;11.01) (l.P-CC)

92

(100) Max: die aufgeht ich probiern\ this opens I try 'I'll try if/whether that one opens'

F.-J. d'Avis & P. Gretsch

Max08 (2;11.01)

In example (99) it is not possible to clearly distinguish between a real complement and a parenthetical remark. Moreover, the utterance appears to be an imitation. The status of the complement clause in (100) is less doubtful, with the positioning of the complement in front of the V2 matrix clause being particularly remarkable. Two months later we find precursors of adverbial clauses with undifferentiated filler syllables, all of them constructed with VE patterns, cf. the examples (101) to (104): (101) Adult: Warum - was ist da passiert? 'Why - what happened there?' Max: 9dndfisch TOTis\ FILLER fish dead is '[When/because the] fish is dead' Adult: Warum?'Why?' 9 Max d tot is\ FILLER FILLER dead is 'Because it's dead'

Maxll (3;00.16) (=(4)) (l.P-AC)

(102) Adult: Der Fisch ist tot. Warum ist der tot? 'The fish is dead. Why is it dead?' Max: ndnd schnellepozistn (=polizisten) ho holn miissen\ FILLER quick policemen get must '[Because] we must quickly call for the police'

Maxll (3;00.16)

(103) Adult (talking about low-flying aircrafts): Warum macht'n der des? 'Why does he do that?' Max: ndndnd laut is\ FILLER loud is '[Because/so that] it is loud'

Maxl5 (3;02.17)

(104) Adult (talking about instruments): Was willst du spielen? 'What do you want to play?' Adult-2: Gitarre? 'Guitar?' Max: 9d des so laut is\ FILLER that so loud is '[Because] it is so loud/noisy'

Maxl5 (3;02.17)

Among our subjects Max is the child with the most remarkable undifferentiated preconjunctional structures. Like Paul he can be considered a top-down learner with a phonetic orientation. As for the syntactic level, his rigid VE pattern indicates a target-like differentiated type. Shortly after the beginning of phase (B), there is one weil+Yl structure, (105), which is possibly an imitation, and another precursor with a typical FILLER, cf. (106).

Variations on "Variation"

93

(105) Adult (in a play): Maxl2 (3;01.07) Also, jetzt frag ich den Verkäufer: "Wieviel wollen Sie denn für ein Kilo Löwenzahn, weil ich muß doch meinen Löwen füttern." 'Now I ask the salesman: "How much do you want for a kilo of dandelion because I have to feed my lion.'" Max: ja\ 'yes' (misinterpretation of the question) Adult (asking again): "Wieviel wollen Sie dafür?" 'How much is it?' Max: bal der ißt löwenzahn\ (misinterpretation again) (bal=weit) 'because he eats dandelion' (106) Adult: Du schießt auf den Mann, der im Bett ist? Warum denn? 'You are shooting at the man who is in bed? Why that?'

Max:

Maxl3 (3;01.20)

9

n dd schnell so - schießen so\ pch\

FILLER that-one quickly so - shoot so. pch (imitating a shot) In the following months, there are also adverbial precursors without undifferentiated filler syllables, that is, clauses of the preconjunctional type, as in (107) to (109), which are very similar to Valle's first adverbial precursor structures. The fact that the undifferentiated sounds disappear indicates that the phonetic level is becoming somewhat less important, and that Max now relies solely on the verb position for indicating subordinate clause function. This implies that he has shifted from a phonetic to a syntactic orientation. (107) Adult: Warum gehört das so? 'Why does it have to be that way?'

Maxl7 (3;04.00) (=(3))

Max: dds nich rausfällt\ this not out-falls '[So that] this won't fall out' (108) Adult: Warum soil's ich nicht machen? 'Why shouldn't I do it?'

Max22 (3;06.11) (=(1))

Max: ich alleine kann\ I alone can '[Because] I can do it by myself (109) Adult: Ah, du, ich hab's vergessen: Max27 (3;09.07) Warum schläft der so? Warum schläft der da? 'Oh, I have forgotten: Why does he sleep that way? Why does he sleep there?'

Max: da der so machen kann\ there he so make can '[Because] he can do thus there' (110) Adult: Ich hab' gedacht, das wär ein Löwe. Ί thought that was a lion.'

Max: ja\ dddach - ich hab dddach des wär ein schwein\ yes thought - 1 have thought that were a pig 'Yes, thought - 1 thought it was a pig'

Maxi 7 (3;04.00)

94

F.-J. d'Avis & P. Gretsch

The second half of Max's phase (B) is characterized by difficulties with the item weil (cf. (Ill)), which is reminiscent of problems encountered by Paul (cf. footnotes 32 and 35). Primary stress and falling intonation indicate that there is no interruption after weil. In addition, he pursues some evasion strategies: If he is pushed to produce some more advanced syntactic construction, he changes the topic or ignores the question, as in (112).36 Here, too, one recognizes parallels to Paul. ( I l l ) Adult: Und warum schläft der da vor der Kiste? 'And why does he sleep there in front of the box?1 Max: weih 'Because.'

Max27 (3,09.07)

(112) Adult: Und warum kracht da das Nilpferd durch? 'And why does the hippopotamus break through there?' Max: weil\ 'Because.' Adult: Weil? Weil's? 'Because? Because it?' Max: bitsch batsch\ (onomatopoeic) Adult: Weil's zu dick ist? 'Because it's too fat?' Max: blitsch batsch\ (onomatopoeic)

Max25 (3;07.25)

As for complement clauses, the following developmental pattern emerges: A hardly intelligible complement to gick mal, (113), in Max25 (3;07.25) is followed one-and-a-half months later by a similar but more easily intelligible construction, cf. (114). (113) {guckmal was ich} kann\ look what I can 'Look what I can do'

Max25 (3;07.25)

(114) heiguckmal 9dn was ich da gemacht hab\ hey, look-PRT FILLER what I there done have 'Hey, look, what I did there'

Max27 (3;09.07) (1 .CC)

In the following corpora his complement clauses are restricted to [+w]-complements. We can see that he has some difficulty with the semantics of individual complementizers, cf. the deviant use of daß ('that') in (115). The appropriate choice should have been gucken wie ('look how'). Both his difficulty with the semantics and the fact that apart from complementintroducing [+w]-phrases he has no differentiated complementizer tokens indicate that he is pursuing a top-down strategy. His list of matrix verbs is increased by the combinations wissen wer ('know who'), cf. (116), and fühlen wo ('feel where'). The matrix predicates are often taken from a set of typical, sentence-introducing idioms like guck mal ('look PRT) with a [+w]-complement as in the previous examples wad schau mal ('look PRT') with [+w] in (117): A second typical example of this systematic evasive strategy with a subsequent change of topic is: Adult: Ja, warum denn? 'Well, why then?" Max: weißt' was hier drin is/ know what here in is 'Do you know what's in here?'

Variations on "Variation"

(115) Adult: Und was macht der jetzt auf dem Wachturm? 'And what is he doing now on top of the watchtower?' Max: guckt daß das angreift\ looks that this attacks 'He looks *that (intended: how) this attacks'

95

Max31 (4;06.01)

(116) du weißt dul weißt dud- weißt du wer der d Max31 (4;06.01) weißt du schon wer der - weißt du schon wer der michi is bei unserm kindergarten/ y o u v o c know you/ know you FILLER - know you who the FILLER - know you already who the - know you already who the michi is at our kindergarden/ 'Do you know who Michi is in our kindergarden?' (117) Adult Max:

(talking about a cat): Und was spielst du mit ihr? Max31 (4;06.01) 'And what do you play with it?' spiel auch mal mit der\ schaumal was ich schon kann\ schaumal was ich schon kann\ play too PRT with her, look what I already can, look what I already can '[I] play with her, too. Look what I can already do, look what I can already do!'

Max produces his first complete and free adverbial clause at the age of 4;6, cf. (118). The same corpus also contains the first infinitive precursor with following wenn+VE imitation, (119). (118) Max:

und des is dann ganz gut wenn'n pferd da hinreitet\ and this is then quite good when'a horse there rides 'And it's quite ok when a horse rides there' Adult: Aber wieso lassen die dann die Lichter an in der Nacht? 'But why do they leave the lights on during the night?' Max: vd dd dann sehen die besser\ FILLER (= because then) then see they better '[Because then] they can see better1

Max31 (4;06.01) (l.AC)

(119) Max

(hopping around in his bed): Max31 (4;06.01) im bettzu hopfen könnt ihr schon des! in bed to hop can you already that 'Hopping in bed, can you already do that?' Adult: Ich glaub', wenn wir des machen, dann schlagen wir uns oben den Kopf an. Ί think, if we do that then we'll hit our heads against the ceiling' Max: aber wenn ihr so macht but if you so make 'But if you do it this way' Adult: Des kann sein. 'That might be'

Max's data show that the difference between top-down and bottom-up learners is not the only relevant parameter, and that the way in which individual building blocks are put together

F.-J. d'Avis & P. Gretsch

96

influences the overall duration of the process. In the case of Max this means an expanded phase (B) of about two years. The length of his phase (B) may be due to the shift from a phonetic orientation to a syntactic orientation which we found between Maxl5 and Maxl7, around the age of 3;3. And since he, like Paul, employs several avoidance strategies, the assembly of individual building blocks may, as a consequence, be more difficult. (120) Summary of Max's developmental process: Duration of phase (B):

elongated phase (B), from 2;11 to 4;11, i.e. two years

Strategy:

top-down learner with undifferentiated filler syllables and unintroduced preconjunctional clauses

Orientation:

initially phonetic orientation, after five months shift to syntactic orientation

Course of development:

adverbial, complement, and relative clauses appear at the same time

Further characteristics:

problems with the semantics of complementizers; various avoidance strategies; deviant use of the item weil for a period of five months

2.5

Comparison of the developmental patterns

The comparison of the superordinate phases (B) of the four learners shows a considerable spectrum of variation with respect to both the beginning of phase (B) and its duration. (121) Schema of temporal comparison:

Valla's

PA-C" : > A

Benny's

PA-C e : >

Paul's

PA-C· :>

Max 1

PA-Ce:>

Age

|-B-|

5

A'

j_ _ „

> 1

A! [

:> I

I

I

I

-B-

I

I

£

>

C|

>

\

I

>

I

1—>

1;β 2;0 2;β 3;0 3;β 4;0 4;8 5;0 5;6 The following picture emerges for the four learners: a) Variation in the developmental sequence

Variations on 'Variation"

97

The data discussed show that there is no invariable sequence of emergence. This applies to the acquisition of individual construction types such as complement, adverbial, and relative clauses as well as to the acquisition of specific lexical items. For the children analyzed here the developmental patterns of adverbial clauses and complement clauses are closely related, which does not mean that there is no temporal ordering among these constructions. Order, however, can only be stated for an individual developmental pattern. Relative clauses do not occupy a fixed position within the overall sequence of emergence: For Valle this type is the first documented complementizer-introduced construction; for the others it emerges at the same time as the other constructions, or later (cf. the learner Benny). As far as individual lexical complementizers are concerned, there is a tendency for the most prominent representatives of the adverbials wenn and weil to appear first - as has been described in the literature. For the complement clauses, there exists a bias to subcategorized complementizers of verba dicendi and sentiendi. It must be stressed, however, that this preference could have discourse-pragmatic or other psychological causes rather than systeminternal ones. Moreover, there are also counterexamples, as Valle's sequence of complementizers demonstrates. b) Variation in the temporal pattern The spectrum of variation is very clear here, cf. the schema in (121). Benny and Max took more time to progress through phase (B) than Valle and Paul. c) Variation with respect to available strategies We have suggested two different basic developmental axes along which learners can be classified: On the one hand, learners can be differentiated with respect to some directional criterion. There are top-down learners, who proceed from types to individual tokens, and bottom-up learners, who abstract a type on the basis of individual tokens. On the other hand, there is an orientation towards at least three grammatical levels (the phonetic, syntactic and semantic levels), with each supplying individual building blocks for the construction to be acquired. It was suggested that the phonetic orientation can be identified on the basis of preconjunctional clauses with undifferentiated filler syllables at the left periphery as well as on the basis of frequently used quotations. Top-down learners with a syntactic orientation tend to use unintroduced preconjunctional syntactic structures and acquire the VE pattern first. Children with a semantic orientation tend to produce paratactic, yet clearly complex structures with salient semantic connections, possibly with one overall closed intonation contour and without other overt indices for subordinate clauses. With respect to this latter axis, that of the building blocks on individual levels, we can also find mixed forms and shifts in orientation. We assume that the dichotomous strategies37 of top-down vs. bottom-up both constitute necessary procedures for language acquisition. Nevertheless, a tendency to choose one of these procedures as a main course of action can be observed. So before a child can be

The notion strategy is used here in its conventional sense and deviates from the terminology in Kaltenbacher (1990).

98

F.-J. d'Avis & P. Gretsch

categorized as a top-down or bottom-up learner the quantitative use of the two strategies has to be carefully weighed. It seems that top-down oriented learners proceed more speedily than bottom-up learners, with some interference from the second axis, that is, the manner in which individual building blocks are assembled, as we could see in Max's acquisition pattern. d) Variation in the acquisition of subordinate clauses and complementizers The current practice in the acquisition literature to identify the acquisition of subordinate clauses with the acquisition of complementizers may be adequate for some individual developmental patterns, but does not seem to us to be differentiated enough. The occurrence of VE structures which indicate subordinate functions has to be seen in close connection with the acquisition of complementizers, but ought to be considered as an independent developmental step on the way to complex sentence formation. 38 This step is therefore also subject to individual variation. Children who produce unintroduced VE constructions as in (1) - (3) have obviously acquired the component of verbal placement early (see, for instance, Valle, or also the child Marianne in Rothweiler 1989). Children who, like Benny, initially produce complementizers as sentence introducers for structures without correlating VE position, acquire this main indicator for subordination late compared to the other building blocks. Another clear case of a token learner seems to be the child Ivar who is described in Müller (1993). Ivar abstracts a type from tokens quite late, i.e. at 4; 11, cf. Müller (1993: 237f). e) Variation in the initial structures The different initial verb-placement patterns of phase (A), which we did not deal with in any detail here, also exhibit variation. Compare, for instance, Benny's placement patterns with Paul's rigid V2 position (cf. also Fritzenschaft et al. 1990 and Gawlitzek-Maiwald et al. 1992). f) Variation in the acquisition of individual complementizers or subordinate clauses with respect to other constructions If we compare the acquisition of complex sentences with the acquisition of other constructions, for instance the passive, we find no invariant developmental sequence either, cf. Fritzenschaft (this volume) and for the development of wh-questions Tracy (this volume) and Penner (this volume). While Paul's data show some overlap between different precursor phases - that is, he produces precursors of passive constructions together with adequate passives before productive complementizers occur - these developmental steps remain completely separate in Benny's development. For him, first precursors of passives are not

The relationship between VE placement and subordinate clauses is problematic since we have unembedded VE-clauses in German: Ob sie wohl noch kommt? (whether she PRT still arrives 'whether she will still come?'), Daß du ja diesen Artikel liest! (that you PRT this article read! 'Be sure to read this article!'). Nevertheless we simplify this matter and refer to VE placement as a functional indicator of embedded clauses since target-like unembedded/independent VE-clauses seem to play no role during the relevant developmental period.

99

Variations on "Variation "

documented until a year after the complete acquisition of complementizers and subordinate clauses.

3.

Language development and the notion of sequence

As the data have shown, it is necessary to allow for intra-individual as well as inter-individual variation (cf. also Tracy 1991 for additional case studies). The question now is how to combine variation with a concept of sequence as a serial ordering of grammars. The term sequence captures the fact that the utterances of language learners change with respect to their grammaticality status (in relation to the target system) and in complexity, and eventually converge on the target grammar.39 Thus, we have a sequence S of grammars Gj, G 2 GT, which correspond to the child's linguistic competence at given points in time and generate different languages.40 In trying to define the notion of sequence more precisely one is confronted with the problem of how to characterize the first grammar G,. Instances for G, can be found in the first utterances which are interpretable on the basis of the target grammar.41 Whether there is a grammar before that time cannot be shown, i.e. to say that the utterances of the one-word stage are based on Gj is just a supposition.42 We will assume the existence of a 'first' grammar and a sequence S = L(Gj+i).

What do we gain with this definition? The first clause (i) guarantees that all grammars in the acquisition process lie at least partly within the range of the target grammar. The case of the empty intersection of L(G|) and L(GT) is excluded, since we postulated that the first grammar of the sequence, G,, is reflected in the first utterances interpretable on the basis of the target grammer. One of the empirical predictions made by (i) is that completely acquired target-like constructions are always kept in a process of accumulation of information, where the notion completely acquired is of particular importance, see below. The lesser-than-or-equal relation includes cases like the restriction of overgeneralizations mentioned earlier, where L(Gj) η L(GT) equals L(Gi+1) η L(GT). CI W e suppose that the set of constructions (in the sense of footnote 45) which can be generated is finite, in contrast to the set of sentences that can be generated. 53

This connection could be introduced via a subset relation, i.e. L(Gj) c L ( G j + i ) , but the problem of overgeneralization persists.

Variations on "Variation "

103

The clauses (ii) and (iii) regulate the relation between two successive grammars, (ii) compresses the information of definition (122'). In this case, the language gets extended, and L(G;) posits a real subset of L(G i+1 ). In clause (iii), we consider the possibility that the language which is generated from Gj gets restricted. Although the grammar itself gets expanded, the part g of G T that is added to the grammar does not enlarge the actual language L(Gj) but restricts it. We therefore modify (122') to (122"):

(122")

Def.: Let GJ be the grammar which generates the target language L(GT), (let GI be a grammar with L(G0 * L(GT)). S = is a Sequence , iff (i)

I (L(Gi) η L(G T )) I < I (L(G i + i) η L(Gr)) I, and either (ii) or (iii).

(ii)

L(Gi) c L(G i + i),

(iii)

L(Gj) => L(Gj+i).

Then for each Gj there should exist a gj e G T , such that L(Gj+g;) = L(G i+1 ), where each gj is provable through a construction Kj, with L(G;) υ K ; = L(G i+1 ) or L(G;)\K ; = L(G i+1 ). What does it mean for g; to be provable? Surely, it is not enough to point to a single occurrence of a construction in a corpus. Moreover, usual competence-performance problems apply so that not every performance error should be taken as evidence for the acquisition of a new construction. To capture the idea of the actual acquisition of a part of GT, namely which was introduced as the basis for a newly observed construction, the term productive use has been adopted. In our framework we speak of a productive use of a part gj of the target grammar G T if it can be empirically proved that (i) in the case of a positive extension of a grammar G; through g ; the child understands utterances which are based on g; properly, and (ii) in the case of a negative extension of the grammar G ; through gj the child considers this constraint in its production data. 54 The notion of a "negative extension" points to the fact that the addition of a constraint may reduce the language but nevertheless enlarge the grammar. Thus (122") gets altered to (122'"):

54

Cf. the experiments in Crain (1991). A productive use of a construction can, for example, be excluded if the relevant utterance can be shown to have citational character.

F.-J. d'Avis & P. Gretsch

104

(122"') Def.: Let G j be the grammar which generates the target language L(GT), (let G j be a grammar with L(Gi) Φ L(GT))· S = is a Sequence , iff (I)

I (L(Gi) Η L(GT)) I < I (L(G i + i) Η L(G T )) I, and either (ii) or (iii),

such that for each Gi there exists a gj e G j , with L(Gi + gj) = L(Gi+i) and g; is completely

(124)

(ii)

L(Gj) c L(G i + i),

(iii)

L(Gi) 3 L(Gj+i).

Def.: A component g of a target grammar GT is completely acquired iff g is productively

(125)

acquired.

used.

Def.: A component g of a target grammar G j is productively

used

iff it can be shown empirically that (i)

if g is a positive expansion, the language learner understands the utterance based on this expansion, or

(ii)

if g is a negative expansion, the language learner observes the corresponding restriction.

Now that we have tried to connect the term sequence with an empirical basis (parts which can be empirically proved), we turn back to the notion of interim grammars. The idea was to distinguish between two grammars, GJ and G I+1 , by means of a completely acquired g, with L(GJ+GJ) = L(G I+1 ). Within the space between the two grammars mentioned above one can find further grammars, which are distinct from GJ as well as from G I+1 but which generate the same language as Gj. This means that for those grammars no gj can be found that is completely acquired. The introduction of interim grammars can be motivated through the precursor structures and the data which was looked at in the first two sections. Two interim grammars which lie between G ; and G I+1 differ in that in each case just a part of g| gets added to the grammar. Therefore g, has to be considered as a complex pattern of smaller entities and not as a monadic item as it often is. Thus the step from GJ to G I+1 constitutes a sequence of its own in another dimension. We label this sequence of the interim grammars sequence'.

105

Variations on "Variation"

(126)

Def.: Let Gi,o and Gi+ι,ο be two successive grammars in a Sequence S. S' = < Gi,o, Gi,i,..., Gij, Gij+i,..., Gi,k, Gi+ι,ο > is a Sequence', iff there is a gj, L(Gj,o+gi) = L(Gj+io), gi is completely acquired, and for each Gj j , 0 < j < k, there is a gj', gj' e gi, and there is a gk' e gi, such that Gi,j + gj' = Gi t j + i , Gj,k + gk' = Gi+ι,ο and L(Gi>0) = L(Gi,i) = ... = L(Gi,k) * L(G i+ i, 0 ) and for each g', g' ist proveable.

Obviously, the empirical basis in (126) can be traced back to the term provable. We consider the precursor structures which were introduced in the first section evidence for the existence of g'. Hence, the definition of sequence in (122') can be modified to (127), taking the interim grammars into account. (127)

Def.: Let GJ.O be the grammar which generates the target language L(GT,O). (let GI(O be a g r a m m a r with L(GJ,o) * L(GR,o))·

S = o > is a Sequence , iff (i)

I (L(Gi>0) η L(Gr,o)) I < I (L(G i+ i, 0 ) π L(Gr, 0 )) I, is a Sequenz' and either (ii) or (iii). (ii)

L(Gi>0) c L(Gi+i,o),

(iii)

L(Gi,o)=>L(Gi+i,o).

Defining the term sequence on the basis of successive interim grammars provides us with means for handling the phenomenon of variation. The different types of precursor structures we came across in the data each hint at parts of g in the grammar so that g is completely acquired if all parts of g are added to the grammar. The way in which the parts get assembled can be attributed to the divergent strategies mentioned in section 1. Considerations about determinism vs. non-determinism lead to the proposal that the interim grammars constitute the non-deterministic component in language acquisition, thus allowing for variation. Therefore a distinction between deterministic and non-deterministic levels or, better, dimensions seems to be necessary. One way to look at this is illustrated in (128), where the sequence S is the deterministic level, and the sequence S' the non-deterministic one.

106

F.-J. d'Avis & P. Gretsch

(128)

< Gi,o, G j j

4.

(= S; deterministic)

Gjj, Gjj+i,...., Gjjc > (= S'; non-deterministic)

Summary

As the data have shown, there exists a significant range of variation which cannot be explained via deviation from a hypothetical standard developmental course. We found a variety of precursor structures apart form the so-called preconjunctional verb-end clauses. These precursors are indicative of strategies which can be classified following two developmental axes: The first of the proposed axes is concerned with the direction of the developmental process and is labeled top-down vs. bottom-up. The top-down strategy implies a functionally oriented process where the learner first acquires a construction-specific type and later on attributes individual tokens to that type. The bottom-up learner, in contrast, is a lexically oriented child, who abstracts a type on the basis of tokens which were acquired earlier. The second axis captures the orientation of a learner towards a specific grammatical level (phonetic, syntactic, and semantic). The construction-specific information of each level is broken down into individual building blocks, and the various ways in which the blocks can be assembled produce different precursor types. The interaction of the two developmental axes allows us to explain the variation found in the data in a systematic way. This is a categorization of tendencies, and at this point we are forced to allow for shifts in the processes and in the orientations, as the course of development must in the end accomplish the differentiation of all grammatical levels and of construction-specific types as well as attributed tokens. The proposed building-block model therefore (i) accounts for individual variation because of the different possibilities for proceeding; thus we are able to (ii) classify the variation among the learners via the two developmental axes leading to individual strategies. Moreover, (iii) the inherently procedural aspect of the model results from the necessity to assemble a complete entity - the type - from individual building blocks over time. Especially (iii) constitutes a well-known problem for parametric frameworks. A closer look at (iii) led us to a definition of sequence in acquisition, where variation is treated as an intrinsic feature of developmental theory on empirical as well as conceptual grounds. To explain variation within a deterministic acquisitional framework we described the developmental process in terms of deterministic transitions from G, to Gi+1, which are subclassified into variable interim grammars. So the precursors mirror the different interim grammars, which depend on the individual strategies of a learner. The adequacy of the building-block model could be tested in future work by comparing the development of different constructions both in the same language and cross-linguistically. We expect to find the same general patterns of variation. Especially the prediction

107

Variations on "Variation"

concerning the end of phase (B) and the notion productive use could be analyzed in more detail via specific production and comprehension tests. Reliable means of comparison could thus be designed to complement data available from children's spontaneous speech productions. Since inter-individual variation can be understood as an inherent feature of acquisition, the application of our model should reveal a wider range of variation in other domains, too.

Bibliography Bickerton, Derek (1984): "The Language Bioprogram Hypothesis." - In: The Behavioral and Brain Sciences 7, 173-188. Borer, Hagit & Ken Wexler (1987): "The Maturation of Syntax." - In: Tom Roeper & Edwin Williams (eds.): Parameter Setting (Dordrecht: Reidel) 123-172. Bowerman, Melissa (1987): "Commentary: Mechanisms of Language Acquisition." - In: Brian MacWhinney (ed.): Mechanisms of Language Acquisition (Hillsdale, NJ.: Erlbaum) 443-466. Brainerd, Charles J. (1978): "The Stage Question in Cognitive-Developmental Theory." - In: The Behavioral and Brain Sciences 2, 173-213. Brandt, Margareta, Marga Reis, Inger Rosengren & Ilse Zimmermann (1992): "Satztyp, Satzmodus und Illokution." - In: Inger Rosengren (ed.): Satz und lllokution. Vol. 1. (Tübingen: Niemeyer) 1-90. Chomsky, Noam (1986): Knowledge of Language: Its Nature, Origin, and Use. - New York: Praeger. Chomsky, Noam & Howard Lasnik (1991): "Principles and Parameters Theory." - To appear in: Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld & Theo Vennemann (eds.): Syntax: An International Handbook of Contemporary Research. Clahsen, Harald (1988): Normale und gestörte Kindersprache·. Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie. - Amsterdam: Benjamins. Crain, Steven (1991): "Language Acquisition in the Absence of Experience." - In: The Behavioral and Brain Sciences 14, 597-650. Felix, Sascha (1987): Cognition and Language Growth. - Dordrecht: Foris. Fritzenschaft, Agnes, Ira Gawlitzek-Maiwald, Rosemarie Tracy & Susanne Winkler (1990): "Wege zur komplexen Syntax." - In: Zeitschriftßr

Sprachwissenschaft 9, 52-134.

Fritzenschaft, Agnes (this volume): "Activating Passives in Child Grammar." Gawlitzek-Maiwald, Ira, Rosemarie Tracy & Agnes Fritzenschaft (1992): "Language Acquisition and Competing Linguistic Representations: The Child as Arbiter." - In: Jürgen Μ. Meisel (ed.): The Acquisition of Verb Placement: Functional Categories and V2 Phenomena in Language Acquisition

(Dordrecht:

Kluwer) 139-179. Gretsch, Petra (1993): Studien zum Erwerb der Komplementierer unter besonderer Berücksichtigung der interindividuellen Variation. - M.A. Thesis University of Tübingen. Grimm, Hannelore (1973): Strukturanalytische Untersuchung der Kindersprache. - Bern: Huber.

108

F.-J. d'Avis & P. Gretsch

Guilfoyle, Eithne & Mäire Noonan (1988): "Functional Categories and Language Acquisition." - In: Claire L-efebvre, John Lumsden & Lisa Travis (eds.): The Canadian Journal of Linguistics / La Revue Canadienne de Linguistique 37:2, Numero Special: Functional Categories, 241-272. Hyams, Nina (1986): Language Acquisition and the Theory of Parameters. - Dordrecht: Reidel. Kaltenbacher, Erika (1990): Strategien beim frühkindlichen Syntaxerwerb. - Tübingen: Narr. Lebeaux, David S. (1988): Language Acquisition and the Form of the Grammar. - Ph.D. Diss. University of Massachusetts. McCawley, James D. (1991): "'Negative Evidence' and the Gratuitous Leap from Principles to Parameters." In: The Behavioral and Brain Sciences 14,627-629. Mills, Anne E. (1985): "The Acquisition of German." - In: Dan I. Slobin (ed.): The Cross-Linguistic Study of Language Acquisition. Vol. 1. (Hillsdale, NJ: Erlbaum) 141-254. Müller, Natascha (1992): Parameters Cannot Be Reset. - Ms. University of Hamburg. Müller, Natascha (1993): Komplexe Sätze: Der Erwerb von COMP und von Wortstellungsmustern

bei

bilingualen Kindern (Französisch!Deutsch). - Tübingen: Narr. Nishigauchi, Taisuke & Tom Roeper (1987): "Deductive Parameters and the Growth of Empty Categories." In: Tom Roeper & Edwin Williams (eds.), 91-121. Penner, Zvi (this volume): "Possible Domains for Individual Variation in Early Developmental Stages." Penner, Zvi & Natascha Müller (1992): "On the Early Stages in the Acquisition of Finite Subordinate Clauses: The Syntax of the So-Called 'Preconjunctional Subordinate Clauses' in German, Swiss German, and French." - In: Geneva Generative Papers 0:1-2,163-181. Pinker, Steven (1984): Language Learnability and Language Development. - Cambridge, Mass.: Harvard University Press. Poeppel, David & Ken Wexler (1993): "The Full Competence Hypothesis of Clause Structure in Early German." - In: Language 69:1, 1-33. Reis, Marga (1985): "Satzeinleitende Strukturen im Deutschen: Über COMP, Haupt- und Nebensätze, wBewegung und die Doppelkopfanalyse." - In: Werner Abraham (ed.): Erklärende Syntax des Deutschen (Tübingen: Narr) 271-311. Roeper, Tom (1986): "How Children Acquire Bound Variables." - In: Barbara Lust (ed.): Studies in

the

Acquisition of Anaphora. Vol. 1. (Dordrecht: Reidel) 191-200. Roeper, Tom & Edwin Williams (eds.) (1987): Parameter Setting. - Dordrecht: Reidel. Rothweiler, Monika (1989; to appear 1993): Nebensatzerwerb im Deutschen. - Tübingen: Niemeyer. Rothweiler, Monika (ed.) (1990): Spracherwerb und Grammatik· Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie. - Opladen: Westdeutscher Verlag (=Linguistische Berichte, Special issue 3.) Safir, Kenneth J. (1987): "Comments on Wexler and Manzini." - In: Tom Roeper & Edwin Williams (eds.), 77-89. von Stechow, Arnim & Wolfgang Stemefeld (1988): Bausteine syntaktischen Wissens: Ein Lehrbuch der generativen Grammatik - Opladen: Westdeutscher Verlag.

109

Variations on "Variation "

Tracy, Rosemarie (1990): "Spracherwerb trotz Input." - In: Monika Rothweiler (ed.), 22-49. Tracy, Rosemarie (1991): Sprachliche

Strukturentwicklung:

Linguistische

und

kognitionspsychologische

Aspekte einer Theorie des Erstspracherwerbs. - Tübingen: Narr. Tracy, Rosemarie (this volume): "Raising Questions: Formal and Functional Aspects of the Acquisition of WhQuestions in German." Verrips, Maaike (1990): "Models of Development." - In: Monika Rothweiler (ed.), 11-21. Verrips, Maaike (this volume): "Learnability Meets Development: The Case of Pro-Drop." Webelhuth, Gert (1989): Syntactic Saturation Phenomena and the Modern Germanic Languages. - Ph.D. Diss. University of Massachusetts. Wexler, Ken & Maria Rita Manzini (1987): "Parameters and Learnability in Binding Theory." - In: Tom Roeper & Edwin Williams (eds.), 41-76.

Maaike Verrips

Learnability Meets Development: The Case of Pro-Drop1

0.

Introduction

This paper tries to make explicit the intuition that variation in language development is the result of the fact that children's grammatical representations are dynamic, in the sense that they are open to undergoing change. It tries to understand some types of variation as typically developmental phenomena. In the process of elaborating these ideas it became increasingly clear that, in order to understand anything at all as a developmental phenomenon, one needs a much more extensive and explicit developmental theory than is generally provided in the parameter-setting literature.2 This paper is an attempt to give a developmental interpretation of universal grammar (UG), reasoning on the basis of empirical findings in the acquisition literature. A full theory of language development that gives credit to both universal principles of language and to the features of language development in children is obviously beyond the scope of a paper like this, at least as much as it is beyond the scope of our present understanding. In most investigations into parameter-setting processes in children some attention is given to what is called 'the developmental problem of language acquisition.' This attention, however, is mostly limited to the question of whether changes in children's linguistic representations are primarily the result of interaction with the environment (continuity), or whether they are primarily the result of changes in the child that take place independently of the environment (discontinuity, maturation); for a recent example of such a discussion cf. Clahsen 1992. One problem with this state of affairs is that often one's developmental conclusions are so closely related to a particular idea about the nature and the content of UG that their scope becomes very unclear. Moreover, as I suggested in a previous paper (Verrips 1990), it seems unlikely that language development should be either fully continuous or fully discontinuous. A more sensible approach seems to be to try to understand how continuity and discontinuity can interact in such ways as to produce the attested results and the correct adult grammar. All in all, very little attention is given to developmental issues such as how the dynamic nature of children's grammars affects their linguistic representations, or how exactly the initial state interacts with the input so that an adult grammar is the end result of this process (for 1

This research was (partly) supported by the Foundation for Linguistic Research, which is funded by the Netherlands Organisation for research, NWO. I would like to thank the audience present at the DGfSColloquium 1992 in Bremen and in particular Hubert Haider, Zvi Penner, Jamal Ouhalla, and Harald Clahsen. For further comments and patience I would like to thank the editors of this volume and Geertje Leemans, Tom Roeper, Achim Stenzel, and Tonjes Veenstra. Obviously all errors are my own.

2

Variation within and between children plays an important role in the work by Tracy and her collaborators (cf. Tracy 1991, Gawlitzek-Maiwald et ah 1992). These analyses focus on more general features of sentence construction than are at issue here. For the sake of clarity then, I will not discuss the content of their proposals.

Μ. Verrips

112

exceptions to this, see Tracy 1991, Roeper 1987, Goodluck 1991). In what follows I will argue that as a result some important empirical questions for a parametric model of language development are not asked, let alone answered. Section 1 points out some more and less discussed developmental problems and section 2 presents an approach to language development that aims to incorporate these questions. Section 3 applies this model to null subject phenomena in early speech. The solution to the original variation problems has become so deeply embedded within the developmental model that it is no longer given a separate place in this paper.

1.

Development and the theory of UG

In order to understand how a theory of language development interacts with a theory of UG, it is important to be clear about the fundamental necessity of a theory of UG and about the general argument for the existence of innate linguistic structure. It will be clear that the postulation of UG is based on an argument about how language acquisition is possible in principle, rather than about how language acquisition proceeds in real time. I shall not make any controversial assumptions on this topic. The basic argument for the existence of UG has always been that native speakers' intuitions concerning structural ambiguity and ungrammatically are universally parts of linguistic competence that cannot be acquired via induction from positive evidence alone. On top of this, grammars seem to vary in constrained and principled ways. These considerations have been applied to the study of the range of variation between adult languages. This study in turn led to a modular theory of UG in which innate principles that vary in constrained ways interact with one another to produce a large though finite set of possible grammars. The various specific forms a universal principle may take are referred to as parameter settings. Supposedly learners are able to choose between these options on the basis of positive evidence in the input, thereby avoiding the need to encounter negative evidence. Applying such a parametric theory about the range and structure of variation between adult grammars to the study of grammar development, one could in principle take either of two tacks. The first tack one might take is to assume that the parametric constraints are properties of adult languages, therefore properties of the final state. One could argue that inherent properties of developmental change are responsible for the converging of adult grammars within the space defined by UG. This approach could be called constructivism.3 I shall not take this tack here, and will have nothing more to say about it, but more extensive discussion can be found in Tracy (1991) and in the debate between Chomsky and Piaget that is reported in Piattelli-Palmarini (1980). The second, more popular tack to take is to interpret Parameter Theory as a theory of the process of language acquisition. In such a process theory, language development is viewed as a process in which children choose between the settings of parameters that UG provides. This approach is taken in all the literature on parameter setting that I am aware of, if one abstracts 3

The notion constructivism does not refer to the idea that general linguistic development is in some sense secondary to general cognitive development (as Goodluck 1991 puts it). Constructivism here stands for the idea that properties of the constructive process in language acquisition (such as possible types of changes, self-regulatory processes and the like) necessarily lead to an end state that falls within the boundaries defined by UG. This is the sense in which the term constructivism is used in Piattelli-Palmerini (1980).

Learnability Meets Development

113

away from different opinions researchers have about the extent to which UG is available to children at the earliest stages. Let us call this process interpretation the parametric development approach. The proposal I make in this paper also fits into this general format, and is presented in section 2. A particularly popular approach within parametric development (since Hyams' 1983 dissertation) has been to describe non-adult-like stages of early grammars as misset parameters. In other words, the null subject stage in the acquisition of various languages has been described as "Italian," or "Chinese." For application of this idea to English cf. Hyams (1986, 1992), to German cf. Clahsen (1991) and Weissenborn (1990, 1992), to Dutch Weverink (1990) and De Haan & Tuijnman (1988). Throughout, the idea is that children may initially misset a parameter to a value that applies in a language other than the target. Some piece of positive evidence will tell them finally that in fact the target setting is different, and, having analysed this piece of positive evidence, they will reset their parameter to the adult value. The most popular alternative approach within parametric development is to regard nonadult-like stages as stages in which a particular part of UG is not activated yet, it being subject to maturation (for application to German cf. Felix (1992), to English cf. Radford (1990)). This approach, however, also implies the availability of triggers to set parameters - once they have matured - to the correct value. As noted in the introduction to this paper, the controversy centres on whether the nonadult-like stages are in fact the result of parameter missettings or of the initial absence of parts of UG. So far I have only summarized the general approach to the study of language development in children from a UG perspective. As noted above, in this discussion some important developmental questions are rarely addressed seriously. I will discuss two of these below and argue that in the parameter-missetting approach it is particularly hard to give them proper consideration. The rest of this paper focuses on a possibility to incorporate these issues. 1.1

Supersets

In the parametric development approach, the logical problem of language acquisition is resolved by postulating that the presence of particular features of the input language tells a child that the target selects a particular parameter setting. Where the language defined by one parametric option forms a subset of a language defined by another option, children will ideally entertain the narrowest hypothesis to account for the input data, and will only extend their grammars on the basis of evidence that cannot be handled within their overly constrained hypotheses. This developmental mechanism is called the subset principle, or conservatism (Berwick 1985, Wexler & Manzini 1987). However, in a number of well-studied areas of grammar development, children consistently seem to violate the subset principle. This is well-known, and I will only mention a few examples here to illustrate the point, all of which are cases of an obligatory rule or realization in the target language, which seems to be optional in early grammars. Any development from optionality to obligatoriness involves a violation of the subset principle in a strict sense.

Μ. Verrips

114

For example, a language that allows both overt and null subjects is a superset of a language that allows lexical subjects only. Therefore, the narrowest hypothesis for a learner would be that lexical subjects are obligatory in the target language. However all the available evidence points in the opposite direction: Children start out with grammars that allow both null and lexical subjects, regardless of whether the target language allows null subjects (Hyams 1986, 1992, Weissenborn 1992, Clahsen 1991, to name a few). Some other cases where optionality seems to precede obligatoriness are found in the cross-linguistic study of the acquisition of functional categories. Finite verb movement in German, though present, is not obligatory in early German (cf. the papers in Meisel 1992), wA-movement to SpecCP is optional in early Dutch before it becomes obligatory (cf. van Kampen 1989), lexical complementizers are optionally realized before they become obligatory (cf. Rothweiler 1993, Krikhaar 1992). It has been argued in the case of null subjects that the non-null subject language is not really a subset of the null subject language but an intersecting set. In other words, a grammar that does not allow null subjects will contain some other element that is not present in the null subject language (proposals include expletive pronouns, unstressed pronouns, properties of verbal inflection; cf. Hyams 1986, 1992). This element will serve as a trigger for the nonnull-subject option. This may well be true, but the question remains: Why should children follow this complicated path when abiding by the subset principle would avoid the initial wrong setting and the subsequent resetting. 4

1.2

Extended transitions

Another general finding is that in most studies children show productive use of the trigger elements for an extended period before they reset the parameter that the trigger is supposed to make them set. In the case of pro-drop again, Weissenborn (1992) shows that verbal morphology is used productively by children acquiring French and German several months before null subjects disappear from their speech.^ During this period, the frequency of null subjects decreases gradually, a fact that has led some (e.g. Clahsen 1991, Hyams 1992) to assume that the parameter resetting has actually taken place, even though its effects are somehow not in full force yet (cf. also White 1980). Hyams (1986:108) argues It is well-known that children's language behaviour does not change over night. That is to say that the various "stages" of acquisition [...] are idealizations. Between these stages there are "fuzzy" areas in which some utterances resemble the language of the previous stage, while others resemble the language of the stage to come. The fact that transitions extend over long periods of time and that they appear to be gradual rather than sudden and complete poses a problem for the parametric development approach.

4

Apart from that, it has not been possible so far to identify any triggering element that appears uniquely in either null-subject or non-null-subject languages.

5

In fact, the period over which the decrease takes place in Weissenborn^ data amounts to seven months in the case of French and six months in the case of German (Tables 12.1 and 12.2).

Learnability Meets Development

115

Whatever the reason is for the extended periods of transition, it is clear that the acquisition of a trigger is possibly a necessary but not a sufficient condition for a learner to rule out the incorrect representations.

2.

Developing a developmental model

In this section I will sketch a model of development that hopes to incorporate these developmental characteristics, while retaining the parametric model to explain the logical possibility of language acquisition. The safest way to go for the developmental theorist seems to be to assume that UG constrains language from the earliest stages on. Early grammars fall within the range of possibilities described by UG. Language development, then, is the process of finding the right evidence to set parameters to their correct values. However, from the discussion in sections 1.1 and 1.2 two things have become clear: First, the subset principle is empirically inadequate as a developmental principle. Second, parametric change is not sudden and complete. This leaves us with the question of what is going on between the child stage and the adult stage. What is the nature of the child's grammar in the period in which the non-adult representations are still part of the grammar because the child still produces them - but are at the same time ruled out - because the trigger has had the effect of making them slowly disappear? In other words, we are left with the developmental problem. 2.1

Introducing MAX

My proposal is to take the tack opposite from the subset principle. Let us assume that learners are subject to the following developmental strategy: (1)

MAX (MAXIMIZE INPUT): For every input string, create as many UG-allowed representations as possible.

MAX is a mechanism that makes a learner add all those representations to her grammar that provide an analysis of the input string (possibly within some limits of 'plausibility' in terms of the meaning attributed to the input string).6 In the case of the null subject parameter, an English child encounters null subjects in finite matrix clauses as in (2) and (3). (2)

(It) seems like you 're trying to get rid of me.

(3)

(I) made you a sandwich.

6

It may be necessary to assume that not all parameterized principles of language allow for co-existing settings during an extended period of time. For example, it may be that directionality parameters need to be fixed to a particular setting from the earliest stages onward. If this is so, it will surely facilitate the acquisition task of the child AFTER this setting. For now, however, let us assume that initially every parameter could be set to every value. The order in which parameters become fixed to a single value is then possibly determined by the accessibility of the relevant evidence.

Μ. Verrips

116

This learner will add every type of null subject to her grammar that could acount for these data. It may include both a topic chain and an Italian-type pro-drop. Acquiring the target grammar of a language is, then, in fact a process in which all the options generated by MAX are ultimately reduced to only the correct set. How this may work and what this means will be made more explicit below. First a little elaboration on the nature and function of MAX. First, note that MAX incorporates the constraints on UG, by allowing the learner a choice only between representations that are allowed by UG. As such, the outcome of MAX is in principle contained within the appropriate set of UG allowed grammars. Second, since MAX relates characteristics of input to UG, the prediction is that the outcome of MAX varies cross-linguistically. That is, although every child will build up UG-allowed representations generating a superset of the possible sentences in the target language, no child will have to entertain all the possibilities allowed by UG for long. In this way, MAX constrains the hypothesis space for children to a smaller space containing the correct target grammar, while at the same time allowing children the flexibility to entertain competing representations if no specific information is available to them for choosing between them. Third, it is easy to see how MAX could be functional for a learner. In resolving the learnability problem on the basis of positive evidence, a learner would be helped by some form of conservatism. On the other hand, the child's communicative needs lead her in the opposite direction, namely to use the small amount of language knowledge creatively to produce the largest possible set of utterances. Presumably, MAX would help a child to quickly build up a set of rules that generate a language that is close to the target language, thereby quickly constraining the UG-allowed space without the cost of eliminating creativity altogether. It will be clear that a function like MAX could possibly provide an explanation for the empirical finding that most children's grammars at some stage generate supersets rather than subsets of the target adult grammars. The stipulation of MAX replaces the stipulation of default parameter settings.7 What needs to be shown now is that it is logically possible for a learner equipped with both UG and MAX to acquire language, and, secondly, how a developmental model containing MAX leads to a better understanding of the apparent gradual transitions. The first of these questions is discussed in sections 2.2. and 2.3.8 2.2

Cutting down a grammar. How?

What needs to be shown then is that triggers could do this job in theory. A brief example will illustrate this point (but see also section 3). Let us assume that a German child, encountering null subjects in topic-drop constructions in the input language, applies MAX and assumes that both null topics and small pros could be allowed in German. The small pros could be licensed 7

This is advantageous, because - as Borer & Wcxler (1987) point out - the existence of default settings or initial states ideally should not be part of the linguistic representations, but rather of the developmental element in linguistic theory.

8

This paper was written almost immediately after the DGfS-meeting. After submission, it has been distributed, and several colleagues provided comments that led to some more precise ideas about the ideas presented in 2.2 and 2.3. Nevertheless, I have decided to leave this paper in its present form, rather than squeeze in some improvements. They will have to await future work. Even so, thanks are due to Cornelia Hamann, Nina Hyams, Eric Hoekstra, Jürgen Weissenborn and Frank Wijnen.

Learnability Meets Development

117

and identified in various - UG-specified - ways, and let us assume that MAX leads the child to initially assume that all these licensing and identification conditions might apply. In other words, the child has no way to choose between various possibilities of licensing the null subject (through morphological uniformity of the verbal paradigm, through nominative caseassignment by a head), nor to choose between various ways of identifying it (through a topic chain, or a strong INFL). This situation leads the child to entertain all these options as possibilities in her grammar. As the knowledge of the verbal paradigm increases, however, the child discovers, for example, that German INFL is not strong. Since her grammar is constrained by UG, null subjects in contexts in which they would be identified by a strong INFL are ruled out. This discovery leads the child to drop null subjects from contexts in which only strong INFL could identify them, for example from embedded clauses and from vv/i-questions. However, in contexts in which other identification mechanisms might apply, like sentence-initial position, the null subjects will still occur. In this way, the acquisition of verbal morphology indeed triggers a change in the child's grammar, but not with the effect of a total drop of null subjects. Rather, the number of null subjects in specific contexts drops to zero, whereas those in other contexts, in which other identification requirements hold, may still occur. There is quite some evidence in the literature that null subjects in early German are restricted to certain grammatical constructions quite early on, even though children at this stage may still produce null subjects more frequently than adults do (Hamann 1992, Weissenborn 1992). Having established that German INFL is not strong, the child will discard the analysis of any future input sentence in which strong INFL plays a role. Note that this scenario predicts a stepwise decrease in the use of null subjects in early German, with stepwise elimination of null subjects from particular types of contexts. The underlying idea is in fact that true gradualism is only apparent, and that closer scrutiny of the developmental steps will reveal stepwise decrease rather than gradual decreases of non-adult-like phenomena. It is of course an empirical question whether close scrutiny will actually reveal such patterns.

2.3

Cutting down a grammar. When?

With respect to timing, various theoretical options come to mind. For example, it may be that at a certain maturational stage, a child will simply stop looking for the most general representations and start looking for what constrains the adult grammar (and start fixing parameters). It may also be that there is an upper boundary to the application of the function MAX, maybe inherent in MAX itself. What seems most plausible to me now is that MAX and parameter setting apply simultaneously, as in the example above. To be more precise, every input string that is ambiguous to the child leads to a number of possibilities for MAX. As acquisition proceeds, and other parts of the grammar are acquired, like verbal morphology in the example above, the outcome of MAX becomes constrained. Crucially, this presupposes that a child can distinguish the representations that have been generated by MAX from the representations for which clearcut positive evidence has been available. To clarify, let us go back once more to the example of the German child ruling out null subjects in certain contexts. A learner like this must be able to distinguish the optional analysis of null subjects as identified by a strong INFL from her conclusion that the INFL in German is in fact not strong. The latter conclusion, as it is (presumably) based on positive evidence, must weigh more heavily, and will serve to rule out the initial option. In other words, once a child has

118

Μ. Verrips

analysed the evidence against a particular parametric option, that option must be 'erased' somehow from the inventory of possibilities. The grammar overall will become increasingly constrained and will converge towards the adult system. Let us assume that such an erasuretype operation is indeed available to children.

3.

The case of pro-drop

In the previous section, I illustrated the developmental mechanism I have proposed with examples from the pro-drop literature. The present section serves to pull these various threads together, and to present a more coherent discussion of these developmental phenomena. The discussion is based on some of the recent literature on this topic, mainly Clahsen (1991) and the papers in Weissenborn et al. (eds.) (1992). Clahsen (1991) finds that early German is characterised by dropped subjects. Since Clahsen does not distinguish in his paper between dropped subjects that would actually be grammatical in the adult grammar (those that appear in topic position in V2 clauses) and dropped subjects that would be ungrammatical in adult German, it is hard to evaluate whether Clahsen's finding means that early German is a superset of adult German. The frequency of dropped subjects in child German is much higher than what is estimated for adult German. However, if child null subjects occur exclusively in positions in which adult German allows them, this could be a matter of pragmatic rather than grammatical development (cf. Weissenborn 1992). Furthermore, Clahsen claims that the transition from the +pro-drop to the -pro-diop stage is due to the acquisition of some property of the person agreement paradigm on the verb. As pointed out by Weissenborn (1992), it is not so clear what this property could be. Besides, there is some evidence that German children still allow empty subjects long after full productive use of subject-verb agreement (Weissenborn 1992, Hamann 1992). Clahsen's data are not presented in a way that allows a re-interpretation here. What should be clear, however, is that neither the transition problem nor the superset problem are resolved satisfactorily in Clahsen's analysis. Hyams' (1992) analysis of null subjects in early English is based on the notion of "Morphological Uniformity," developed by Jaeggli and Safir (1989) to explain the distribution of null-subject phenomena cross-linguistically. Some examples of child English null-subject utterances in Hyams (1992):9 (4)

Want more apple See under there No play matches Is toys in here

As in her earlier work, Hyams notes that the initial grammar allows for null subjects, as opposed to adult English. According to Hyams (1992) this is due to the child's assumption that the subject-verb agreement paradigm is morphologically uniform. In fact, this means that

9

Hyams cites these examples from Bloom, Lightbown and Hood (1975).

Learnability Meets Development

119

with respect to morphological uniformity, the default setting is apparently [+morphologically uniform]. In other words, the superset-subset order is stipulated through a default setting. Hyams relates the disappearance of null subjects to the development of verbal inflection. That is to say, verbal inflection supposedly "triggers" the disappearance of null subjects. But the discussion of her Figure 1 (which shows percentages of lexical subject pronouns, inflectional morphology and auxiliaries for Adam) leaves very little of the parameter-setting logic intact when she writes: We see a sharp decrease (from 70% to 10%) in the use of null subjects during this 5-month period. At the same time the use of inflectional morphology increases from 30% to 75%. [...] Once the child realizes that English is a "mixed" morphological system, evidenced by the fact that his verbs are surfacing with inflectional morphemes, null subjects are no longer licensed. (Hyams 1992:255-256, emphasis mine) The figure itself shows a gradual decrease in the frequency of null subjects during this fivemonth period.10 Hyams does not discuss the fact that null subjects are licensed until five months after Adam's realization that English is a "mixed morphological system, evidenced by the fact that his verbs are surfacing with inflectional morphemes." Hyams too, then, gives a non-satisfactory solution to both the first step and the gradual-transition problem. Weissenborn (1992) discusses the development of null subjects in a French and a German child. Again, both the French child and the German child initially construct grammars that allow null subjects, although the null-subject option is much more productive for the German child than for the French child. I will discuss these frequency effects below. First some examples from child German and French: (5)

German Glaub nich Believe (lsg) not Brauche nich lala Need not pacifier Will lala habe Want pacifier have

(6)

French Peux le faire Can it-do Va chercher un avion Go to look (for) a plane Ai tout bu le verre moi Have drunk the whole glass me

Some of the problems with Clahsen's and Hyams' proposals noted above are presented by Weissenborn, notably the issue that not all null subjects in German are equal: When SpecCP is empty, a null subject is allowed by the adult grammar through the formation of a topic

She calculates the percentages on a monthly basis. The percentages during these five months are: 70%, 75%, 70%, 40%, 25%, 10%.

Μ. Verrips

120

chain. This option may be used under strict pragmatic constraints. For his German data, Weissenborn (1992:278) argues that "the omission of lexical subjects in children does not differ in nature from the one in adults". Weissenborn's explanation for this is that the only trigger that would lead a child to assume that subject-verb agreement (instead of a topic chain) identifies a null subject in a clause is an "unambiguous context" to this effect. Weissenborn claims that the presence of empty pronominal subjects in embedded clauses in Italian, which must be identified clause-internally through subject-verb agreement, would provide such a trigger. In Weissenborn's analysis, the absence of null thematic subjects in German embedded clauses means that the German child is never triggered to allow for morphological identification of the empty subject, and thus sticks with the adult-like topic chains. Weissenborn's basic claim, then, is that in the development of German, children do not violate the subset principle, but rather adhere to it strictly. His explanation of the facts depends crucially on the finding that German null subjects do not appear in environments in which they could not be identified through a topic chain. If null subjects would at some stage appear in these environments, there is no mechanism in Weissenborn's theory to cut them out. These environments would be either embedded clauses or matrix clauses containing a non-subject in SpecCP. In later research (Verrips & Weissenborn 1992) we found that there are some cases in the first four months of recordings of the German child studied by Weissenborn where these 'ungrammatical' null subjects do occur. Some examples: (7)

mir schmeckt nich to-me tastes not nein das krieg nicht no this get not zähne hat nicht teeth has not

The number of cases is very low, but so is the number of contexts in which they could possibly occur. It is therefore debatable whether we should take this evidence seriously or discard it as performance errors and leave it at that. Hamann (1992) also discusses late cases of null subjects in these 'harmful' environments, and Clahsen (p.c.) has claimed that he found some null subjects in finite embedded clauses. If we take this seriously, data like (7) are counter evidence to Weissenborn's developmental analysis. The developmental model proposed in this paper could account for them. We might say that the child's knowledge of the agreement paradigm leads her to limit her null subjects to contexts in which identification can proceed through topic-chain formation, whereas at the earlier stage the child may still assume that identification of the /vo-subject takes place through either topic-chain formation or subject-verb agreement. In Weissenborn's discussion of French the superset and the transitional problem apply differently. First of all, though French is like English in that it disallows null subjects in both the Italian and the German way, French children do drop their subjects (in the period of Weissenborn's investigation the frequency of null subjects ranges between 36% initially to around 5% finally of declarative matrix clauses). Weissenborn's analysis of the development of null subjects in French requires elaborate explanation, and I will not go into it here. It is important to note, however, that Weissenborn's

121

Learnability Meets Development

model of 'parameter interaction' is one of the very few notions in the literature that could make sense of the apparently gradual decrease of ungrammatical sentences over time. For a similar approach to interacting parametric decisions, cf. Roeper & Weissenborn (1990). This concludes the discussion of the developmental literature about null subjects. In these last paragraphs I would like to discuss some further evidence from the pro-drop literature for some of the assumptions I have made in the developmental model. First of all, I have postulated that attaining the adult grammar is a process in which a child slowly excludes various types of representation for a single adult phenomenon. This predicts that whereas in the earliest stages null subjects may occur in every possible context, later stages reflect that certain contexts, requiring certain licensing and identification mechanisms, are favoured over others. That is to say, ideally, that null subjects in German and English will disappear suddenly from specific contexts, once the relevant data have been analysed. Hamann (1992) produces evidence that the realization of subjects in child language is sensitive to syntactic position. According to her, German children's null subjects appear preferably in preverbal declarative position, and decreasingly in the other given positions. Some examples from Hamann (1992): (8)

preverbal declarative postverbal declarative embedded clause u>/i-question

Is gar nich dunkel Is not at all dark Jez is hier ganz dunkel Now is here totally dark Weiß nich, wa'um das macht hab Know not, why that done have Wa'um hab das ausgerechnet an? Why have that of-all-things on

Whether a full analysis of all of Hamann's data would ultimately show the developmental steps I have argued for is a question for further research. We may conclude, however, that so far analyses of the precise contexts in which null subjects appear all indicate a sensitivity to the target language constraints in the positions in which null subjects appear. In general these findings provide support for the developmental mechanisms I have outlined. A final persistent observation in the domain of null subjects is that children acquiring a grammar that allows null subjects produce null subjects more frequently than children acquiring a language that does not. These frequency effects have been reported by Valian (1989) and Weissenborn (1992), and discussed by Bloom (1990) and Goodluck (1991). It has been noted in the discussions that a developmental model that regards null subject English simply as Italian or Chinese cannot account for these frequency effects. Can we conclude from this that non-adult-like null subjects in early speech should be regarded as processing or pragmatic problems, rather than as a grammatical phenomenon? There are a number of problems with a non-grammatical characterization of early nullsubjects that I will not go into here (Weissenborn 1992, Hyams & Wexler 1991). I believe, however, that the frequency effect can find a natural and grammatical explanation in the developmental model I have argued for. Since it requires children to distinguish between parts of their grammar that resulted from the application of MAX, and parts that resulted from straightforward positive evidence (cf. section 2.3), it allows for a principled explanation of the different frequencies with which children use the parts of their grammar that they can be sure

122

Μ. Verrips

about and the parts that they can not. As such, early null subjects can be a grammatical phenomenon and at the same time show a sensitivity to the target grammar.

4.

Conclusion

This paper arose partly out of a deep dissatisfaction with the current state of theoretical discussion about the nature of language development. I have tried to show how some very general features of language development are given insufficient attention in the literature and how the solutions that have been put forward are inadequate. In reaction to this, I have made these developmental findings the central concern of the developmental approach I argued for. Adopting the present proposal would have serious repercussions for the parametric theory, especially for the notion 'trigger.' A trigger in this context is an element or a feature that signifies that a certain analysis is untenable. Rather than telling a child that she must incorporate some additional possibility in her grammar, triggers serve to show that certain possibilities are excluded. The motivation for such a reconception of the function of triggers is found in the actual developmental orders attested. It remains to be seen, however, that grammars can actually be written this way. There are many ways in which the proposed model could be wrong and insufficient as it stands. I believe, however, that it makes some interesting predictions about child language development, and that these predictions can be tested and possibly falsified by empirical fact. It is also clear that the model as such is not very specific about a number of developmental questions, mainly regarding the timing of parameter settings. Various interpretations are open, as I have indicated, and it is largely an empirical question as to how the timing proceeds. The fact that the developmental model makes use of a particular linguistic theory makes it hard to test whether falsification of particular developmental predictions are due to the developmental model itself or to wrong formulations in linguistic theory. This is the price one has to pay if one wants to apply a not fully established theory to a new set of data. It is a price, however, that one pays whether one has an explicit developmental theory of UG or none at all.

Bibliography Berwick, Robert C. (1985): The Acquisition of Syntactic Knowledge. - Cambridge, Mass.: MIT Press. Bloom, Lois, Patsy Lightbown & Lois Hood (1975): Structure and Variation in Child Language. (=Monograph of the Society for Research in Child Development 40). Bloom, Paul (1990): "Subjectless Sentences in Child Language." - In: Linguistic Inquiry 21:4, 491-504. Borer, Hagit & Ken Wexler (1987): "The Maturation of Syntax." - In: Tom Roeper & Edwin Williams (eds.): Parameter Setting (Dordrecht: Reidel) 123-172. Clahsen, Harald (1991): "Constraints on Parameter Setting: A Grammatical Analysis of Some Acquisition Stages in German Child Language." - In: Language Acquisition 2, 67-88.

Learnability Meets Development

123

Clahsen, Harald (1992): "Learnability Theory and the Problem of Development in Language Acquisition." - In: Jürgen Weissenborn, Helen Goodluck & Tom Roeper (eds.), 53-76. Felix, Sascha (1992): "Language Acquisition as a Maturational Process." - In: Jürgen Weissenborn, Helen Goodluck & Tom Roeper (eds.), 25-51. Gawlitzek-Maiwald, Ira, Rosemarie Tracy & Agnes Fritzenschaft (1992): "Language Acquisition and Competing Linguistic Representations: The Child as Arbiter." - In: Jürgen Μ. Meisel (ed.), 139-179. Goodluck, Helen (1991): Language Acquisition: A Linguistic Introduction. - Oxford: Blackwell. de Haan, Ger & Kees Tuijnman (1988): "Missing Subjects and Objects in Child Grammar." - In: Peter Jordens & Josien Lalleman (eds.): Language Development (Dordrecht: Foris) 101-123. Hamann, Cornelia (1992): "Late Empty Subjects in German Child Language." - In: Technical Reports in Formal and Computational Linguistics 4, 1-45. · University of Geneva. Hyams, Nina (1983): The Acquisition of Parameterized Grammars. - Unpublished Ph.D. Diss. City University of New York. Hyams, Nina (1986): Language Acquisition and the Theory of Parameters. - Dordrecht: Reidel. Hyams, Nina (1992): "A Reanalysis of Null Subjects in Child Language." - In: Jürgen Weissenborn, Helen Goodluck & Tom Roeper (eds.), 249-267. Hyams, Nina & Ken Wexler (1991): On the Grammatical Basis of Null Subjects in Child Language. - Ms. UCLA and MIT. Jaeggli, Osvaldo & Ken Safir (eds.) (1989): The Null Subject Parameter. - Dordrecht: Kluwer. van Kampen, Jacqueline (1989): De verwerving van vvA-vraagzinnen. - M.A.Thesis University of Utrecht. Krikhaar, Evelien (1992): Voegwoordloze bijzinnen in kindertaal. - M.A. Thesis University of Utrecht. Meisel, Jürgen Μ. (ed.) (1992): The Acquisition of Verb Placement: Functional Categories and V2 Phenomena in Language Acquisition. - Dordrecht: Kluwer. Piattelli-Palmarini, Massimo (ed.) (1980): Language and Learning. - London: Routledge. Radford, Andrew (1990): Syntactic Theory and the Acquisition of English Syntax: The Nature of Early Child Grammars of English. - Oxford: Blackwell. Roeper, Tom (1987): "The Acquisition of Implicit Arguments and the Distinction between Theory, Process, and Mechanism." - In: Brian MacWhinney (ed.): Mechanisms of Language Acquisition (Hillsdale, N.J.: Erlbaum) 309-344. Roeper, Tom & Jürgen Weissenborn (1990): "How to Make Parameters Work." - In: Lyn Frazier & Jill de Villlers (eds.) : Language Processing and Language Acquisition (Dordrecht: Kluwer) 147-162. Rothweiler, Monika (ed.) (1990): Spracherwerb und Grammatik: Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie. - Opladen: Westdeutscher Verlag. (=Linguistische Berichte, Special issue 3). Rothweiler, Monika (1993): Nebensatzerwerb im Deutschen. - Tübingen: Niemeyer. Tracy, Rosemarie (1991): Sprachliche Strukturentwicklung: Linguistische und kognitionspsychologische Aspekte einer Theorie des Erstspracherwerbs. - Tübingen: Narr.

124

Μ Verrips

Valian, Virginia (1989): "Children's Production of Subjects: Competence, Performance and the Null Subject Parameter." - In: Papers and Reports on Child Language Development 28, 156-163. - Stanford University. Verrips, Maaike (1990): "Models of Development." - In: Monika Rothweiler (ed.), 11-21. Verrips, Maaike & Jürgen Weissenborn (1992): "Routes to Verb Placement in Early German and French: The Independence of Finiteness and Agreement." - In: Jürgen Μ. Meisel (ed.), 283-332. Weissenborn, Jürgen (1990): "Functional Categories and Verb Movement: The Acquisition of German Syntax Reconsidered." - In: Monika Rothweiler (ed.), 190-224. Weissenborn, Jürgen (1992): "Null Subjects in Early Grammars: Implications for Parameter-Setting Theories." - In: Jürgen Weissenborn, Helen Goodluck & Tom Roeper (eds.), 269-299. Weissenborn, Jürgen, Helen Goodluck & Tom Roeper (eds.) (1992): Theoretical Issues in Language Acquisition: Continuity and Change in Development. - Hillsdale, N.J.: Erlbaum. Weverink, Meike (1990): "What's Missing in Dutch?" - In: Papers and Reports on Child Language Development 29, 125-132. - Stanford University. Wexler, Ken & Rita Manzini (1987): "Parameters and Leamability in Binding Theory." - In: Tom Roeper & Edwin Williams (eds.): Parameter Setting (Dordrecht: Reidel) 41-77. White, Lydia (1980): Grammatical Theory and Language Acquisition. - Dordrecht: Foris.

Mary Aizawa Kato

A Theory of Null Objects and the Development of a Brazilian Child Grammar1

1.

Knowledge representation and knowledge acquisition

The aim of this paper is to discuss the nature of empty objects with definite interpretation in Brazilian child language as part of a larger project on the development of Portuguese syntax in children. Two fundamental questions define the research program of Generative Grammar today (cf. Chomsky 1988a) and provide my point of departure: a. b.

What constitutes knowledge of language? How is this knowledge acquired?

Rizzi (1988:1) says that the second question "is logically more fundamental and in fact the whole paradigm of explanation in generative grammar," but claims that the first "is methodologically prior in that the question of acquisition cannot be successfully addressed in the absence of a precise characterization of the system that is acquired." It seems, however, that such a theory of the steady state will also benefit from the consideration of language acquisition facts, especially if such facts are used to constrain the hypotheses built for the adult I-language. 2 Kato (1993) presents a theory of the null object in Brazilian Portuguese, hereafter BP, which, in a revised version, will be presupposed as the representation of the acquired system. In consonance with my belief stated above, its form and argumentation were, to a great extent, inspired and constrained by the author's intuition about child data. The paper is organized in three sections. The first summarizes the author's theory of the null object in Brazilian Portuguese; the second presents an analysis of the nature of the null object in child language as reflected in a longitudinal study of a child (Rachel (R)) as well as supportive cross-sectional data (Tiago (T), and Luciano (L)); and the final one presents considerations concerning the basic theoretical questions in the field of language acquisition and linguistic theory.

1

I wish to thank Jairo Nunes, Sonia Cyrino, Eduardo Raposo and the audience at the Tübingen Conference on Language Acquisition: Crossing Boundaries, October 1991, for valuable comments on an earlier version of this paper. I also thank Prof. Claudia G. De Lemos and her team for letting me use the data-base of the Social-interactionist Project on Language Acquisition at UNICAMP, Campinas.

2 Cf. Chomsky (1986, Chapter 2) for the distinction between I- and E-language.

126

Μ. A Kato

2.

Null object languages

2.1

The concept of parameter

Once the existence of the much discussed null-subject parameter was established, linguists started considering the possibility that some languages may license an empty object, a property that could be used to describe cross-linguistic variation (Huang 1984; Raposo 1986; Cole 1987). As with the null subject, any null object should be subject to licensing and identification conditions. The licensing conditions for the occurrence of a null subject have been attributed mainly to morphology. The first attempt (Taraldsen 1978) was to suggest that a rich inflection licenses null subjects. For Roberts (1993) a rich inflectional paradigm is one that is functionally rich. A functionally rich system admits one null ending and one syncretism. Following the same reasoning it would be reasonable to say that a rich clitic system should warrant the presence of at least one null clitic. Jaeggli & Safir (1987), however, attributed the licensing condition to morphological uniformity in order to account for languages as different as Italian (full inflected paradigm) and Chinese (no inflected form), and the same can also be proposed for the licensing of null objects. Identification, on the other hand, would be parametrized. In the case of Italian the identification of the null subject would be a function of inflection features; in the case of Chinese it would be a function of an antecedent in topic position. In a parallel analysis, identification of the null object can be seen as a function of clitic features (Italian) or of the presence of an element in topic position (Chinese). The concept of parameter has also been associated with the lexicon. Parameters, however, can be said to be a function of closed-class morphemes such as inflectional affixes and closed-class words like complementizers, determiners, prepositions, and pronouns rather than of the lexicon in general. The cross-linguistic variation in the domain of binding items 3 - anaphors, or reflexives, pronouns and variables - is seen by Sportiche (1986) as a function of differences in lexicalizations in a matrix where locality and c-command are the primitive notions: Table 1: c-command required

3

locality condition

anaphors

antilocality conditions

pronouns as variables

c-command not required

referential pronouns

The Government and Binding theory (cf. Chomsky 1981 and passim) postulates three categories in its Binding Theory: a. the anaphor, which is bound in its domain (clause); b. the pronoun, which is free in its domain; c. the name, which is always free. It also proposes four empty categories: (a) the empty pronoun pro [+pronominal, -anaphor] identified by the morphology of the verb in subject position and by the clitic in object position; (b) the empty subject PRO of non-finite verbs, considered [+pronominal, + anaphor), (c) the trace of NP movement [-pronominal, +anaphor], and (d) the variable, or the trace of wA-movement [-pronominal, -anaphor].

127

Brazilian Null Objects

Sportiche shows that while English lexicalizes in the horizontal parameter, both the pronouns as variables and the referential pronouns being a single lexical item, Japanese lexicalizes in the vertical parameter, the anaphor and the bound pronoun having the same form. Actually, the Japanese pronominal form kare in a c-commanded position can only have a disjunctive reading (= deictic or exophoric) regarding its antecedent. Thus the slot 'referential pronouns' needs further lexicalization distinctions, which will be seen in Table 2 later on. In this paper it is thus assumed that at least for parameters in which the plus or minus values are in a nested relationship (Valian 1988, Clahsen 1989a), variation is a function of morphology, in the sense of Borer (1983) and Jaeggli & Safir (1987), and/or of lexical encoding, as in Sportiche (1988), Bickerton (1984), and Chomsky (1988b).

2.2

The null-object analysis for BP

Tarallo & Kato (1989) show that the progressive loss of the clitic system in BP explains its restrictive use of subject-verb inversion. In Kato (1993) I have argued that the ongoing disappearance of the clitic system is also responsible for the peculiar null-object phenomenon in BP. It was suggested that the null object in BP can be analyzed as having four different representations, rather than constituting a homogenous phenomenon as proposed for Chinese by Huang (1984), 4 for whom the null element is always a variable. In BP the empty object can be any of the following: a null name, a null pronominal, a variable, and a null VP.

2.2.1

The null deictic

Raposo (1986) describes the null object in European Portuguese (EP) as having either a pragmatic (situational) or a discourse antecedent. In either case he attributes the status of a variable to the empty category. This variable would be bound by an empty operator (OP) in COMP, which in tum would be associated with a discourse topic (D-Top). (1)

Ο Manuel trouxe 0 agora mesmo. the Manuel brought 0 now just

(2)

D-Topj ... [OPj

[O Manuel trouxe 0t agora mesmo]]

For Kato (1993), however, the null objects with a pragmatic or deictic interpretation should be separated from those with a discourse antecedent. 5 Example (1) can be uttered with the speaker pointing at some object (a cake, for example) in the speech act situation, both in European Portuguese and BP. But the question in (3a) is answered differently in these two languages:

4

Cole (1987) proposes a four-way typology regarding null objects: languages with only variables, languages with only pronominals, languages with both variables and pronominals, and languages with neither pronominals nor variables. The assumption here is that languages vary more quantitatively than qualitatively when this sort of nested parametric variation is at stake.

5

Perhaps the same lexicalization for deictic and non-deictic pronouns in Western languages makes linguists tend to reduce deictic and non-deictic pronouns to a single representation.

128

(3)

Μ. A Kato

(a) - Quern comeu ο bolo? - Who ate the cake (bl) - Comeu ο Manuel. - Ate the Manuel

(EP)

(b2) - Ο Manuel (que) comeu (ele) -The Manuel (that) ate (it)

(BP)

(with falling intonation)

The fronting of the verb in (bl) can be taken as an indication that the empty object has been moved out of the sentence (cf. Torrego 1984, and Ambar 1985), as in wA-questions, a residual V-second phenomenon (cf. Rizzi 1991): (4)

Que comeu ο Manuel? What ate the Manuel

In the Brazilian option (b2), on the other hand, it is the subject that is moved out of the sentence, and consequently the object position cannot be a variable. Kato (1993) proposed that the empty object in (b2) was a pronoun. The representation in (2), with an empty operator, seems adequate, therefore, not for (1), but for (3bl), where the antecedent is in the discourse and not in the situation. Sentence (1) with a deictic interpretation for the object should be interpreted as having a null deictic pronoun or demonstrative. In Kato (1993) it was pointed out that, while pronouns are postulated to be free only in the domain of their governing category, kare-like deictics in Japanese are always free, like names and demonstratives,6 and, therefore, subject to principle C of the binding theory. The null demonstrative in (1) is equally postulated to be under condition C, and, therefore, not bound to any antecedent. The null deictic object was shown to be often followed by a locative deictic, and restricted to root sentences: (5)

(a) Segura 0 aqui! Hold (this) here! (b) Eu pego 0 lä para voce. 1 (will) get 0 there for you.

The null deictic (or exophoric) pronoun, or exopro, has been shown to have a wide distribution in imperatives and to have always an inanimate third-person referent, corresponding to this or it in English. Null operators, on the other hand, have generally been proposed to be a subordinate phenomenon and not restricted to inanimate entities: 6

Like names, kare can appear with an antecedent in a non-c-commanding position. Compare a sentence in Portuguese (from Raposo) with a referential expression, and a Japanese sentence with kare in the same context: (i) Ο pai da Alexandra disse que a Alexandra estä doente the father of Alexandra said that the Alexandra is sick in bed (ii)

Kare-no titioya-ga Jun-wa biokide nete-iru to oshata. Thatone'Sg e n f a t h e r n o m Jun- t 0 p sick lying COMP said

129

Brazilian Null Objects

(6)

(a) The man! (OP! (she loves t t ) ) (b) He brought the bookj (for (OP, (the boys to read t,)))

Contrary to what is currently believed, the deictic null object in root sentences is widely distributed cross-linguistically,7 appearing even in languages which are marked negatively for the null object parameter, like English and French. The difference between BP and English lies in the fact that this type of null constituent appears only in imperatives in the latter, but also in non-imperatives in the former. In both languages it is a root phenomenon. (7)

(a) (b) (c) (d)

Push 0 here! Pitch in 0 ! Sent 0 by mail. (printed in a telephone bill) *I am going to send 0 by mail.

Translated into BP: (8)

(a) (b) (c) (d)

Empurre 0 aqui! Jogue 0 dentro. Envie 0 pelo correio. Eu vou enviar 0 pelo correio.

A deictic object may be expressed with full lexicalization (a), partial lexicalization ((b) and (c)), or null lexicalization (d): (9)

(a) (b) (c) (d) (e)

Hold Hold Hold Hold Hold

this this 0 this 0!

parcel here! 0 here! 0 here! 0 0!

In English the pronoun it lexicalizes both the endophoric and the exophoric third person; for this reason the null deictic occurs in alternation with it. (10)

Push it/0!

But it in endophoric function has no 0 variant like the deictic it, so that separate treatment is justified. (11)

The door closed before I pushed *(it).®

Brazilian Portuguese, on the other hand, has both the deictic and the endophoric null object, as in (8a) above, and (12):

7

Aguilar (1981) shows that a deictic object can be null in Spanish as well.

8 It seems that even this sentence is not totally impossible, depending on the context of utterance.

Μ Α Kalo

130

(8)

(a) Empurre 0 aqui!

(12)

A porta; fechou antes de eu empurrar 0 ; . the door; closed before I holdj n fj n 0;.

As both exophoric and endophoric uses of the pronoun are referential, further refinement is necessary in Table 1 to account for the differences seen above. Table 2: Referential object pronouns - deictic

English Japanese EP BP

+ animate him / her jibun / 0 ο / a ele / ela / 0

+ deictic - animate it 0 ο / a 0

+ animate him / her kare / kanojo ο / a ele / ela

- animate it / 0 0 ο / a / 0 0

While English and European Portuguese restrict the null category to only the [+deictic, -animate] objects, BP and Japanese have a null variant in all slots except one, the [+deictic, +animate] object. 2.2.2

The null third-person clitic

Huang (1984) and Raposo (1986) treat the null object as a case of variable in Chinese and European Portuguese respectively. Their motivation for this proposal follows from their observation that null objects cannot have antecedents in Α-position and do not occur in islands. (13)

(a) Joäoj perguntou John asked (b) Ο livrO;, Joäo The book, John

Mariaj se Mary if perguntou asked

(14)

(a) * 0 menino que The boy who (b) * 0 pirata partiu The pirate left

Pedro viu 0 , ^ . Pedro saw 0 . a Maria se Pedro viu 0,. the Mary whether Peter saw 0 .

trouxe 0 mesmo agora e brought 0 just now is para as araibas depois for the Caribbean after

meu afilhado. my godson, de ter guardado 0 . having hidden 0.

Huang's and Raposo's analyses require that all null objects be identified like the variable of a wA-phrase, which implies chain representation, with the head of the chain in A'-position. Kato's (1993) proposal that at least null deictics allow identification in situ, in the way that demonstratives are interpreted, seems more interesting for early stages in child language with null objects, as COMP presumably has not been activated yet. 9 To accept Huang's and 9

See, for instance, Clahsen (1989a), Radford (1990), Meisel & Müller (1990), among others.

131

Brazilian Null Objects

Raposo's analysis, on the other hand, implies a much stronger continuity hypothesis than the one being assumed here, namely that all categories are represented in the child's I-language from the beginning, even those referring to positions that are never lexicalized. However, BP is not restricted to empty objects with a deictic interpretation. Empty objects can have antecedents in the discourse and even in the same sentence, though not in the same clause. They may also appear in syntactic islands. Observing that BP can have null objects in islands, Galves (1989) and Farrell (1990) proposed that the null object in this language can be pro. Farrell proposes that BP has both a variable and a pro. According to him, the null object pro in BP would have the property of being intrinsically identified, though he does not explain what he means by intrinsic identification. In addition Farrell's study fails to explain the restricted distribution of pro the fact that it cannot appear in the object position of a complement clause 10 - and does not provide a satisfactory solution for the issues of licensing and identification. Along with the null deictic, which has a much wider cross-linguistic distribution (perhaps a universal one), Kato (1993) postulates a null third-person clitic for BP, thus splitting Farrell's pronominal category into deictic and non-deictic, only the latter of which is subject to principle Β (cf. fn. 4). Several empirical studies of BP (cf. for example Duarte 1986, Cyrino 1990) have attested to the ongoing loss of the clitic system in BP, the latter stressing the fact that the loss affects the third person much more than the other persons. The following phases can be proposed for the clitic system in Brazilian Portuguese (Table 3), disregarding gender and number differences. Table 3: Change in BP clitic system PHASE I eu tu ele

PHASE II metelo-/ o-

eu voce ele

PHASE III mete- / o 0-

eu voce ele

— — —

According to Cyrino the change form Phase I to' Phase II starts in the first half of the nineteenth century. 11 One can say that BP today is situated between phases II and III. 10

Maia (1991) argues that object pro can occur in complement clauses with the antecedent in Α-position in examples such as: (i) Joäo falou que Pedro ajudou 0 a fazer ο exercicio. John said that Peter helped 0 to do the exercices. (ii) A garfonete reclamou que ο fregues nem ao menos agradeceu 0 . the waitress complained that the customer didn't even thank 0. It seems, however, that the complements of these verbs are lexically restricted and have an arbitrary interpretation like the null objects in Italian studied by Rizzi (1986). In the examples above the arbitrary interpretation is present and the assignment of coreference seems to be much more a problem of inference than of grammar.

11

Actually it was proposed (Kato, 1993) that there would be a stage previous to phase I where clitics would be enclitic, but a recent work (Pagotto 1992) has shown that enclisis is a recent phenomenon in EP. It had also been proposed that between phase II and III there would be a phase with one syncretism (me-, 0 - , 0-). Here, I simplified the table and made phase II move directly to phase III.

Μ. A Kato

132

Cyrino (1990) shows that in the nineteenth century the clitic ceases to climb and behaves like an inflection (a prefix in this case) of the main verb. As a consequence, one can assume that the single-vowel clitic (or prefix) of the third person has been dephoneticized in phase II, although still forming a uniform paradigm with the first and second persons (me, te, 0) in Jaeggli & Safir's (1987) sense, and continues to be functionally rich in Roberts' (1993) approach. The pro is identified as third person because this zero clitic is like the zero inflection of verbs.12

As has already been pointed out, the null clitic is preferred for inanimate objects while for animate objects the non-clitic lexical pronoun is preferred. What this seems to indicate is a change from a masculine/feminine paradigm to a more iconic gender system of the he/she/it type. In that case, the last phase would have a third person paradigm of pronominals corresponding to the he/she/it system: Phase II eu voce ele/ela

Phase III me te 0 (='if)

eu voce ele ela 0

Che') ('she') ('it')

The addition of the concept of a zero-clitic with it-like properties makes Farrell's proposal of intrinsic identification more intuitive and workable. The null clitic has been proposed to license a pro, which like the lexical pronoun can be «»referential with some constituent outside its domain. But like the null deictic, the null clitic is more restricted in distribution than the lexical pronoun. It can only be associated with an NP in a non-c-commanding position. The following can be considered as non-c-commanding positions: A.

The antecedent is governed by an external head: 13

(15)

(a) Falando de boloj, ο Pedro 0-comeu. As regards the cake, the Peter ate (it). (b) Joäo disse sobre a pastaj que Pedro 0 r perdeu. John said about the bag that Peter lost (it). (c) Eu comprei ο casaco, sem 0j-experimentar. I bought the coat without trying (it) (on). (d) Joäo descascou a banana, e Pedro 0,-comeu. John peeled the banana and Peter ate (it).

For an interesting phonological treatment of this phenomenon, see Nunes (1993). We will ignore here the empty category in object position, co-indexed with the null clitic.

133

Brazilian Null Objects

B.

The antecedent is governed by a null head:

(16)

C.

(0 (o boloj)), ο Pedro 0 r comeu. (As for the cake), the Peter ate (it) The antecedent is a discourse topic, recoverable from a previous sentence or from a previous turn: 14

(17)

-Equanto ao bolo? - What about the cake? - Ο Pedro 0j-comeu ontem. - The Peter ate (it) yesterday.

As a clitic it is free to appear in islands: (18)

2.2.3

Quanto ao boloj, a cozinheira que 0,-fez proi e baiana. As for the cake, the cook who made pro is Bahian.

Variables

Since European Portuguese has a full clitic system, this analysis predicts that clitics but not null objects will appear in islands. Null deictics, on the other hand, are root phenomena and, therefore, are also unauthorized in island contexts. However, variables are not excluded in BP as the language evidences topicalized quantified NPs which leave a trace. This kind of movement can also occur with an overt complementizer: (19)

(a) So Only (b) So Only

os homensj the men os homensj the men

a Maria convidou tj. the Mary invited, que a Maria convidou t;. that the Mary invited.

Fronting of a ΗΆ-element can also appear with or without a complementizer: (20)

- Quenij que a Maria convidou tj? - Who that the Mary invited? - Ο Pedro (que) a Maria convidou 0 . - The Peter (that) the Mary invited.

In our analysis a discourse topic (codifying an aboutness relation) is any constituent that precedes the nullobject clause and is governed by a head. No intervening operator or chain is assumed, and the association is merely one of coreference. The left-dislocated position has to obey the case filter like any other NP, and its head, whether lexical or null, assigns case. See Calabrese (1986) for a syntactic treatment of sentences in a discourse through the notiön of T(hema)-domain.

Μ. A Kato

134

Such topicalization, or movement of a constituent to an A'-position, has generally to do with focus and not with discourse topics, which are presupposed and not focal. In our analysis discourse topics, when grammaticalized, appear as left-dislocated NPs and the sentenceinternal coreferent is a pronominal. Farrell (1990) shows that in variable contexts both the first and second persons are possible antecedents, but would be intermediated by a null operator, unlike the context of non-variables: (21)

(a) Mamäe Mother (b) Mamäe Mother

me\ me te\ you

deixou left deixou left

la there lä there

para todo mundo v e r 0 j . for everyone to see. para todo mundo ver 0 j for everyone to see.

The analysis of these constructions will not be discussed here and the null operator interpretation will be assumed. In terms of language acquisition the concomitant emergence of wA-questions, focal structures and final adjunct clauses would provide support for the null operator in these sentences.

2.2.4

A functional definition of null categories

Both the exopro (null exophoric object) and the pro object were shown to have a restricted distribution. But why should the empty categories be more restricted in distribution than their lexical counterparts? Government & Binding (GB) literature has considered two types of definition for binding items: categorial and functional. The foregoing analysis of null objects in BP suggests that, though the null categories have categorial features according to their position in the phoric paradigms, they are functionally interpreted due to the fact that their realization is the same phonologically empty form. We could even consider adding to this paradigm the type of null anaphor (or reflexive) found in examples such as Joäo 0-machucou, 'John hurt himself. Thus while an anaphor has a local c-commanding antecedent, a variable must have an A'c-commanding antecedent. On the other hand, a pro (null clitic) has a non-local, non-c-commanding antecedent, and the exopro has no antecedent. If c-command is an innate notion, the child will easily acquire these uses of the empty elements in BP. Table 4 shows how phoric items enter a revised version of Sportiche's (1986) twodimensional matrix:

Brazilian Null Objects

135

Table 4: Paradigm of 0-phoric items c-command

+antecedent

required

A-position

0-reflexive

A'-position

variable

0-clitic(pro)

anaphors

pronoun 0-deictic(exopro)

-antecedent

2.3

anti-c-command required

Apparent examples of null object

Raposo (1986) considers cases of VP deletion as apparent cases of null objects. (22)

- Voce

colocou ο

- (Did) you put - Coloquei 0 . - Put+past+1 .p.

vaso na

estante?

the vase on the shelf?

The answer to the question constitutes a case of VP deletion, and not of null object, as what is subsumed under the zero is not only the object. 1 5 The VP-deletion phenomenon is probably associated with the phenomenon of verb raising, but why and how languages vary with regard to this movement is still a controversial matter. What is relevant here is that certain apparent cases of null object are better treated as VP deletion. According to the analysis above, European Portuguese and Chinese can have a deictic null object but not a null clitic. Both of these languages, however, can apparently have a null object in a complement clause, as it does not constitute a syntactic island: (23)

Ο vidraceiro disse que colocou 0 ontem. The glazier said hat put+INFL yesterday

If the null deictic is a root phenomenon, the null object in this sentence cannot be a deictic. If it is not a deictic, and since it cannot be a pronominal, this has to be a case of VP deletion or a variable. If it is a variable it should have a pattern similar to that of a question when the object is the wA-element, the difference being that there would be an empty operator instead of a wA-word:

In Kato & Tarallo (1992) short answers in BP are analyzed. We proposed there that what appears in short answers is what gets moved to INFL. In English there is no verb movement to INFL. Therefore the answer is always an auxiliary or modal. Romance languages have verb and clitic climbing to INFL, and. therefore, the verb appears accompanied by the clitics. In BP the verb is supposed to move and the clitic can be zero. The analysis has problems, however, as European Portuguese does not exhibit the enclitic pronoun and BP does not exhibit the first- or second-person clitic either.

Μ. A Kato

136

(24)

(a) Que; disse ο vidraceiro que 0i-colocou ontem? What said the glazier that put+INFL 0 yesterday? (b) OPj disse ο vidraceiro que 0 r colocou ; ontem. OP said the glazier that put 0 yesterday

Since in (23) the verb is not fronted as in (24), it may be inferred that the empty object is not a variable. It seems that the last possibility - VP deletion - is the answer for (23). The discourse context could be: (25)

- Eu queria saber quando väo colocar ο vidro na janela. I like+INFL to-know when go+INFL to-put the pane in the window. - Ο vidraceiro disse que colocou 0 ontem. (=23)

We could then have three types of languages (Table 5): (a) Those that have variables and marginal exopro like English (E), (b) those that have variables, productive exopro, and VP deletion, like European Portuguese (EP) and Chinese (Ch), and (c) those that have all the above, as well as a null clitic (BP):16 Table 5: Types of languages

Ε

VP deletion*

pro

exopro

variables

peripheric

+

-

-

Ch EP

+

+

+

-

BP

+

+

+

+

* VP deletion that maintains the main verb, which moves to INFL

3.

Null objects in child language

3.1

Assumptions and hypotheses

The proposal in section 2 defended a lexical and morphological basis for linguistic variation along the line proposed by Jaeggli & Safir, on the one hand, and Borer, Chomsky and

Brazilian Portuguese seems to be changing from a wA-movement language to a the status of variables may be peripherica!.

wh-in-situ

language, and

Brazilian Null Objects

137

Sportiche, on the other. The contribution to the analysis of BP was to consider zero elements as part of the phoric paradigm.17 The general hypothesis regarding language acquisition to be presented here is consonant with the aforementioned view on language variation. It endorses Wexler & Manzini's (1987) and Nishigauchi & Roeper's (1987) proposals that language acquisition is lexical learning, but here grammatical growth is attributed only to closed-class items such as affixes, auxiliaries, pronouns, determiners, and complementizers. It also assumes, along with Clahsen (1989a), Lebeaux (1988), and Meisel (1990), that the child's I-language is a sub-part of the adult's Ilanguage, as grammar development is dependent on growth in the lexical repertoire of functional items. Words and affixes form paradigms where null items can have a definite value. Like inflectional paradigms, pronominal systems can exhibit a zero. The child has to leam the value of the zero pronoun in the same way that he/she learns the zero inflection, that is through its place in an n-dimensional matrix of forms. As shown above, Brazilian Portuguese actually presents a zero element for each category in the binding theory: the null deictic, or exopro (principle C), the null clitic, or pro (principle B), and also a null reflexive (principle A). The deictic null object, if universally distributed, should be the one to emerge first. As it appears only in root sentences, it falls within the degree-0 complexity domain, in the sense of Lightfoot (1989). Moreover, since the deictic null object is not bound, no COMP category has to have been activated, and it would constitute the unmarked empty category. Short answers require the pragmatic capacity to assent (agree or disagree), but, grammatically, these require the acquisition of INFL and of V movement. It will be proposed that short answers constitute the robust triggering data through which a child can fix the parameter regarding V movement. As for the null pronoun, we may hypothesize that the child will include it in his/her grammar as soon as he/she is able to associate things in a non-local domain, which explains why children acquire the endophoric use of pronouns in the context of narratives, as was shown in De Lemos' (1992) study. Developmentally, if we consider with Meisel that pronouns are spell-outs of AGR(eement),18 they should precede the appearance of COMP, as V movement to COMP presupposes V movement to AGR. 19 The development of variables requires activation of COMP, or, in the maturational view the capacity to establish chain relations (cf. Borer & Wexler 1987). The antecedent in this chain is often a focal element like the wA-word or the marked focus. Since focal movement

17

Williams (1987:introduction) believes that a "nominal or verbal paradigm, which is essentially an ndimensional matrix of forms, mapped in a systematic (but not necessarily bi-unique) way onto a set of mophological distinctions" does not lend itself to parametrization, but I am suggesting here that it is exactly the differences in paradigms of functional words and affixes that are responsible for parametrization.

18

In recent years the category Inflection (INFL) is seen as two distinct heads - Tense and Agreement - a separation that is interesting for languages like Portuguese, which has the infinitive inflected for agreement.

19

For De Lemos (1992) pronouns emerge in narrative situations, a finding that makes us suppose that Agreement and the grammatical notion of person emerge concomitantly with the symbolic capacity required by narratives.

138

Μ A Kato

requires knowledge of what is old and what is new to the interlocutor, the acquisition of COMP must also be concomitant to the development of textual awareness. The developmental pattern that can be expected is thus the following: Null deictic > VP deletion and null clitic > variable inflection We can also expect that the empty category for a certain function precedes the appearance of the corresponding lexical category, not only because of processing costs, but even more importantly because null categories have a more restricted distribution than their lexical counterparts. Thus, in the spirit of the subset principle, we can say that a system containing only null categories is smaller than a grammar containing lexical deictics and pronouns. Null deictic > lexical deictic

pro > lexical pronoun

This developmental pattern may be said to go parallel to Nishigauchi & Roeper's (1987) finding that children acquire referential items prior to non-referential items (he/she/it > itexpletive). The former has to do with phonetic content and the latter with semantic content. Thus the identification of four different representations for the null constituent(s) allows us to predict a developmental pattern concerning the use of empty categories and progressive replacement by non-null forms. 20 A modular view of knowledge representation can also be adopted for the acquisition of knowledge. Thus we might assume that the several modules of mind can develop independently and yet interact harmoniously instead of accepting the precedence of the pragmatic or semantic mode over the grammatical one, or endorsing the thesis that processing limitations constrain representational development. In her paper "From dialogue processes to narrative perspectives", De Lemos (1992) studies the acquisition of pronouns by a child, Rachel, covering the period from 1;6 to 2;8.2, and although her aims are more dialogic and textual than syntactic, some of her grammatical intuitions and observations are relevant for our study, which covers more or less the same period (1;6 to 2;6.8).21 Analyzing data from the same child and the same period may reveal more reasons for investigating the parallel development of several cognitive modules. 3.2

The four types of empty objects in child data

3.2.1

The null deictic or exopro

The presence of deictic pronouns in the data can be described as follows: 3.2.1.1 They are present from the earliest stage, according to our prediction, and their presence is constant from stage I to the latest stage studied.

20

See Duarte (1990) for a comparison between null subjects and null objects in Brazilian Portuguese acquisition. According to her, "from 1;6 till 1;9 the child's contribution to a still small subset of interactional situations seems on its way to acquiring a linguistic status" (p. 4).

Brazilian Null Objects

(26)

(27)

(28)

Child:

Child:

Child:

139

a ρδ pöe tela tela a tela

('put on' = infinitive) ('put on' = imperative) ('take off = imperative) ('take off = infinitive) ('take off = infinitive)

Liga aqui. Turn on here. Eu vou, eu vou pegar a ca/, hum, ta pesada. I go I go get the ca/ hum, (it) is heavy Pega, mäe, pega, mäe, tä pesada. Hold, mummy, hold, mummy, (it) is heavy

(R. =1;6.6)

(R. 1;9.8)

(R. = 2;02.14)

3.2.1.2 The imperative is also abundant in the adult input, and moreover, transitive verbs often appear without lexical complements. As the child complies with the request, he/she models his/her speech to the adult's immediate input, and the transitive verb takes on a finite form with a null object, interpretable as the situational element. In BP the second and third person have the same inflected form (the unmarked 0 ) , which will not be signalled in the English glosses. (29)

Adult: Adult2: Child: Adult:

Child: Adult: Adult2: Child: Adult2: Child:

Puxe! ('pull' = subjunctive imperative) Puxa! ('pull1 = indicative imperative) Coloca! ('put' = indicative imperative 2 2 ) Coica. Ah, vai colocar na caixinha? Ah, go put in the box? 'Oh, are you going to put (it) in the box?' ?ay ('go' = assent) Coloca, Tiago! ('put' (imperative), Tiago) Tiago, coloca! ?ay. Coloca! Cooca.

(T. 1;8.20)

3.2.1.3 The difference between the stages observed is only quantitative, the null deictic (exopro) being more frequent in the early stages. 3.2.1.4 The imperative, which is assumed to be the universal context for null objects, is one of the contexts where it appears first, but its use with the infinitives and the perfectives 2 3

22

With verbs that require a Theme plus an obligatory Goal theta role, empty categories are not possible in object position in English even when the form is imperative. Perhaps this impossibility has to do with the fact that the two complements form a small clause.

Μ. A Kato

140

shows that the child has already expanded the exopro domain to non-imperatives in view of the positive data found in the input, thus setting BP as a productive exopro language. (30)

Child: Adult: Child:

('took off) ('took off) Cwill(?) put on') ('put' = imperative)

Telo! Tiro! A po. Pöe.

(R. 1;6.6)

3.2.1.5 Lexical deictic objects appear later than null ones, which confirms the prediction that the null category is a forerunner of the lexical one. But before being filled with a full deictic pronoun it is first partially saturated by a deictic locative. The lexical deictic pronouns appear simultaneously with the demonstratives, which seems to indicate that the latter appear with the full projection of the NP. (31)

Child: Child: Adult: Child: Adult:

Zuda aqui. Help here Ta bicando. (It) is pecking Τä bicando ο que? (It) is pecking the what? Boquendo aqui. Pecking here Ta bicando ο que aqui? (It) is pecking the what here?

Child:

Aqui.

Child:

Here Butua aqui. Button up here

Child:

Abi aqui.

Child:

Open here Liga aqui. Turn on here

Child:

da

ziz

aqui. (R. 1:9,8)

(32)

Child: Child:

(33)

Child:

Eu vou puxar este. I am going to pull this Tila esse aqui. Take away this here.

(R. 2;00)

Eu quelo cortar esse aqui. I want to cut this here.

The perfective form seems to be assigned only an aspectual status by the child, with the meaning of 'completion of action'. I assume that Tense as a category has not been analyzed yet, as there is no contrast between present and perfect. The perfective is opposed by the child to the infinitive form, which seems to be consistently assigned future-action meaning.

141

Brazilian Null Objects

Child: Child:

Voche, viche, chegula aqui. You, you, hold here. Eu quelo tambem eche aqui. I want also this here.

(R. 2;01,16)

3.2.1.6 Lexical deictic pronouns as objects appear as the child chooses human objects in a system where the exopro stands for the neuter it. Thus when the object is inanimate the object is still null: (34)

Child: Child: Child:

3.2.2

A ell aqui. Look him here Väo pega eli. Let's get him Vila. Turn (it)

(R. 1 ;9.25)

The null pronominal

The use of the null pronominal can be described as follows: 3.2.2.1 It is not productive during the period studied; co-reference is more often established through repetition of the R-expression (referential NP) than through null or lexical pronouns. (35)

(36)

Child:

Child:

Esse aqui, eu vou por esse aqui. This one here, I am going to put this one here.

(R. 2;0.05)

Aminha blusa, mäe, pöe aminha blusa. My blouse, mommy, put my blouse (on).

(R. 2;4.11)

3.2.2.2 The null pronominal (pro) appears at a later stage than exopro, confirming the initial hypothesis. 3.2.2.3 When it does appear, it occurs in a left-dislocated structure, as proposed in Kato (1993): (37)

(38)

Child:

Child: Child:

(39)

Mother: Child:

Caiu popö. (-pacifier 1 ) Fell pacifier

(R. 1;8)

Popö pega. Pacifier pick Homi t i modendu. Man is biting = the man, the dog is biting (him).

(R. 1;9.25)

Que que a menina ta fazendo? What the girl is doing? Zogando bola. Throwing ball.

(R. 1;9,20)

(R. 1;8)

142

(40)

(41)

Μ A Kato

Child:

Child:

Cola, da pra mim. Glue, give (it) to me.

(R. 2;02.14)

Ha, vou, eu vou conta pa ela esse livrinho. Hah, I am going to tell her this little book.

(R. 2;02.14)

In (37) the noun follows the unaccusative verb, as expected. The example in (38) could suggest that word order is still problematic for the child. For Lebeaux (1989), who sees grammatical growth starting from lexical representation and moving progressively to thematic representation, order is not fixed until the child projects X° to its maximal projection. It is then and only then that case would be assigned and the order fixed. 2 4 But the clear production of INFL in (38), and its comprehension in (39) show that at this stage Rachel has already gone beyond the thematic structure. Utterances with the object preposed to the verb are instances of left-dislocation with a null pronominal. Popö, homi and cola are discourse

topics, which in Kato's (1993) analysis are codified as dislocated elements. 25 Example (41) can be considered a case of right dislocation, with the null inanimate object located adjacent to the verb - ( ( V Oi PP) NP,) - where it gets case. 3.2.2.4 Both as deictic and as anaphoric expressions, lexical pronouns always refer to humans, thus presenting a partial paradigmatic contrast with the null pronoun, which always refers to inanimates in deictic function and to both animates and inanimates in anaphoric function. 3.2.2.5 Third-person lexical pronouns do not appear with an antecedent until 2;2, again confirming that the null category is a precursor of the lexical one. Moreover, they appear first in subject position (example (42)), perhaps because subjects tend to be humans. In object position, where non-humans are more frequent, the pronominal is null until 2;6. Lexicalized co-reference in this position tends to appear as a repetition (example (36)):

24

Though his model is very attractive, it seems that the Brazilian children have INFL very early as the head selecting the verb. In Lebeaux's view the child's structure is a thematic structure with an unfixed order. But this thematic structure is already configurational, with the agent as the external element. If the thematic structure is a grid in Stowell's (1981) conception, the arguments would not be in asymmetric positions and free order should be natural at this level. The verbs would appear inflected in the lexicon either as third person indicative, infinitive, or third person past, until analysis took place. The presence of the auxiliary tä suggests, however, that INFL has been analyzed at least as a finiteness marker.

25

This is not a new proposal in the literature, as Gruber (1967) and Lebeaux (1988) have already suggested that the initial NP in child language is a topic. It is the characterization of discourse topic as a leftdislocation phenomenon that is new here. No operator is involved in Kato's analysis. In the operator analysis, however, the representation for Homi tä modendu would be: (i) D-Topicj, D-TopiCj... (OP: (pro; V tj)) In Chinese the null subject is a pro which has to be identified also by a Topic, as Chinese has no agreement to identify pro. If a Chinese sentence has both a null subject and a null object, its representation would have to contain two operators. (ii) D-TopiC;, D-Topicj... (OP; OPj (tjVtj)) If Rachel had not yet acquired agreement, her sentence would have to look like the Chinese representation. Besides the problem of identification, it seems improbable that such a simple sentence as Homi tä modendu has such a complex representation.

143

Brazilian Null Objects

(42)

(36)

Child:

Child:

Ο Zepeto, ele tä comsono. The Zepeto, he is sleepy

(R· 2;2.14)

Aminhablusa, mäe, pöe aminha blusa. My blouse, mommy, put (on) my blouse

3.2.2.6 Pronouns as arguments emerge at the same time as first-person agreement, as in R.'s speech a to vava (Ί am angry'), an utterance with which R. initiates a turn. This confirms Meisel's (1990) postulation that pronouns are spell-outs of AGR. As agreement is acquired, nominative case can be structurally assigned and the subject position can be filled with a maximal projection. What this seems to suggest is that case is assigned to the determiner head and not to the noun. 26 The unproductive use of 0-clitics at the early stage can be attributed to the scarce occurrence of clitics in the input. An examination of the adult data during the period studied revealed the presence of only seven clitics:27 Adult2: Adult2: Adultl: Adultl: Adultl: Adultl: Adultl:

Me ajudaa escolher 'Help me to choose' A Senhora näo vai me dar hoje, mamäe? 'You (fem., formal) are not going to give(it) to me today, mommy?' Tä querendo me gozar, e? 'Are you making fun of me?' Mamäe te da uma parte, o.' (o = olha 'look') 'Mommy will give you thus a part, o1 Eu vou te mostrar, viu? Ί am going to show you, hear?' Foi a pulga que te pegou, ne? 'It was the flea that got you, no?' Ce cai e se machuca. 'You fall down and hurt yourself

This means that the child can hardly find the necessary elements to build a paradigm of clitics.28 In order to develop a third-person 0-clitic one has to have developed the non-null members. 29 The data show further that in the beginning the parents use the non-clitic 26

This justifies Fukui's (1986) analysis of treating only functional heads as having maximal projections. Here both direct and indirect objects are used as clitics. Considering only direct objects the number is even smaller. What this seems to indicate is that the clitic system is being lost first in the direct-object function with a Theme role.

28

Moreover, Adult2 is a sporadic interlocutor, who pretends to be the child but does not simulate her speech very successfully, as she mixes a formal addressee form like Senhora with Mommy and uses 'me dar' instead of the more colloquial dar pra mim '('give to me'). If one takes only the spontaneous input into account, the occurrence of clitics will be reduced even more. At a later stage, not included in this study, R. presents a second-person clitic Child: Eu vo te azudä a arrumä a mala.... Ί am going to help you pack the bag...' (R. 3;0.24)

Μ. A Kato

144

pronominals for first, second, and third persons, thus providing a full paradigm of non-clitics in object position. (44)

Father:

Voce ja viu eu com ο meu paleto? 'Did you see Τ with my jacket?' Mother: Eu pego aqui pro ce, asso pro e dou na tua boca. '(I) get this for 'you' (no-clitic), (I) blow (it) and give (it) in your mouth' Mother: Entäo pöe ele de pe, uai. 'Then put 'he' on foot'

The inanimate deictic object, however, is consistently null, which justifies once again the separation of the null deictic pronominal from the non-deictic pronominal. 3.2.3

VP deletion in short answers

Raposo (1986) showed that VP deletion yields a surface form that is homophonous with null object constructions in European Portuguese. Our subject produces VP deletion exclusively in short answers. Before describing such answers, a brief parenthesis will show what can constitute a short answer in natural languages.30 Kato & Tarallo (1992), studying cross-linguistic variation in the form of short answers, show that not all possibilities in one language can be found in the other. Thus, a question containing a pre-verbal adverbial may have this adverbial as an answer in Portuguese, but not in English (E), French (F) or Japanese (J): (45) (a) (b) (c) (d)

- Voce ja comeu a pizza? - 'Have you already eaten the pizza?' -Ja. (BP) *- Already. (E) *- Dejä. (F) *- Moo. (J)

Quantifiers can appear as answers in some languages, but not in others: (46) (a) (b) (c) (d)

- Voce comeu tudo? - 'Have you eaten everything?' - Tudo. (BP) - Everything. (E) - ?Tous. (F) - *Zembu. (J)

In both cases, what is repeated (when it is repeated) is the focal element.

I will limit the presentation to what we may call bound answers', excluding cases of free illocutionary answers like That's a lie!, etc.

145

Brazilian Null Objects

When the I NFL node, with whatever it incorporates, is used as a short answer, there will also be cross-linguistic differences. But here the difference seems to be parametrically determined by what is moved to INFL or to some other functional head, be it AGR or Neg. (47)

- Is he eating the cake in the kitchen? - He be+INFL (yp eating the cake in the kitchen) - (Ele) estar+INFL (yp comendo bolo na cozinha)

(48)

- Did you eat the cake yesterday? - Yes, I do+INFL (yp(eat the cake) yesterday) - (Eu) comer+INFL (yp^ ο bolo) ontem)

The types of short answers observed in the child were the following: 3.2.3.1 The child starts answering by extracting part of the parental speech, namely the focal element: (49)

Adult: Child: Adult: Child: Adult: Child: Adult: Child: Adult: Child:

Tiro tudol 'Did you take off everything' Tudo. Everything Ce quer por ο microfone embaixo do gravador? '(Do) you want to put the mike below the recorder?' Bassu. Below Ce quer fazer sozinhal '(Do) you want to do it alone?' Sozinha. Alone Chega. Ja limpoul 'Enough. (Have you) already cleaned?' Po. Cleaned Vamos ver se a gente acha a cabe^a? 'Let's see if we find the head' Väo. Let's

(R. 1;8.25)

3.2.3.2 In the following stage the child learns to alternate the third-person (= second person) final vowel in the present tense with the first-person final vowel, but actually the interpretation he/she seems to be assigning to this morpheme is that of aspect/mood or finiteness [+F] rather than of person, or agreement, as it only appears in short answers with subject drop. Assuming that [+F]31 is a lower head than AGR, the verb in child language raises to it, but not any further. Compare (50) with (51): I am assuming, with Belletti (1988), that Tense is a lower head than Agreement, but before the child acquires time distinctions the Τ head has only the feature [+F], interpreted as 'assertion'.

Μ A Kato

146

(50)

Adult: Child: Adult: Child:

(51)

Adult: Child: Adult: Child:

Ah, ce vay por na caixinha? 'Are you going to put in the box?' ">Ay. Are going. Onde ce vai? Vai na escola? 'Where are you going? Are you going to school?' Vai. Are going. Vai pö na boneca? 'Are you going to put it on the doll?' Vo. (I) am going. Ce contal You tell+2nd p.s. Conto. Tell+lst p.s.

(R. 1 ;9.8)

(R. 1 ;9.8)

3.2.3.3 It is only after the present tense first-person vowel is used in its finite/aspectual value, as an assenting strategy, with no overt pronoun, that the inflection assumes the agreement status in non-answers with a lexical pronoun. Compare (51) and (52): (52)

Child: Child: Child:

Eu vou por, vou por aqui. I am going to put, am going to put here Eu vou jog a. I am going to throw Agora eu vou ίεςέ. Now I am going to close

(R. 1;10.23)

3.2.3.4 When first-person inflection appears in answers it alternates with third person (referring to the child) for some time: (53)

Child:

A le Vai le Ο νό le (ö = eu)

(A) to read 'Is going to read' Ί am going to read'

(R. 1 ;9.8)

Why should there be such alternation? One hypothesis that should be investigated in future work is that in the beginning children have only names as referential expressions and, thus, only one grammatical person. These names can refer to different discourse or semantic persons. Grammatical persons are considered determiners and are therefore functional categories which are acquired later. The a vowel that precedes verbs was considered in our first hypothesis as an embryonic auxiliary for vai, but it could also be considered Rachel's first expression to refer to herself as it contains the first vowel of her name. Thus for her, referring to herself may be done through a name (a, Ra, Quel) or through a spell-out of the verb inflection (ό). The vowel a seems to be a peculiarity of Rachel's. Tiago and Luciano have only the infinitive before they start producing the first person verbal form. But further comparative research is necessary to sustain such a hypothesis.

147

Brazilian Null Objects

One important difference is in the fact that it is in non-answers that the pronoun is spelled out. While in answers the inflection seems to be used by the child as a finite/aspectual element with an assertive function (which it is), in non-answers it receives an additional agreement function. The child even seems to turn agreement into a phonological harmony phenomenon, which is not far-fetched, considering that masculine-feminine agreement in Portuguese is basically phonological. Compare: (54)

(a) a mala ο bolo (b) 6 v IP > CP) one could say that not only does the child choose the smaller grammar in terms of parameter setting, but that his/her own linguistic representations grow in size. The child's analysis of pronouns as words was shown to be dependent on an analysis of AGR(eement) as an independent feature, but prior to this analysis comes the identification of INFL as an aspectual/finiteness marker related to its function as the head of assertion (or to its pragmatic function of assent), but with no tense value. One could use this fact as a counterargument for the continuity hypothesis, as the INFL node seems to change its nature. What actually happens, however, is that the INFL node acquires more features. In the beginning it

151

Brazilian Null Objects

is m e r e l y a

finiteness/aspectual

marker; in the s e c o n d s t a g e it b e c o m e s the locus

later o n t e n s e features are a d d e d .

o f person;

C o n s i d e r i n g that i n f l e c t i o n a l l a n g u a g e s o f t e n h a v e o n e

s i n g l e s e g m e n t w h i c h is m u l t i - f u n c t i o n a l , it is to b e e x p e c t e d that not all of its f u n c t i o n s are a p p r e h e n d e d at o n c e . T h e c h i l d learns o n e s i n g l e f o r m a l unit w h i c h is the s a m e as the adult's, but acquires its features in a c u m u l a t i v e f a s h i o n . T h e s e g m e n t a t i o n of the formal unit w i t h its primary f u n c t i o n , o n the other hand, w a s s h o w n to b e i n d u c t i v e l y a c h i e v e d first through the e x t r a c t i o n of a s e g m e n t f r o m the input question, and later by its r e p l a c e m e n t by a v o w e l - h a r m o n y r e s p o n s e . T h u s , a l e x i c o / m o r p h o p h o n o l o g i c a l h y p o t h e s i s of l a n g u a g e a c q u i s i t i o n h a s to deal w i t h the p r o b l e m o f l e x i c a l and m o r p h e m i c s e g m e n t a t i o n .

In this study o n e c a t c h e s a g l i m p s e of the

w a y in w h i c h R a c h e l d i s c o v e r s I N F L ( e c t i o n ) , as a f u n c t i o n a l unit.

It w a s the inspiring w o r k

of D e L e m o s ( 1 9 9 2 ) , w o r k i n g in a d i a l o g i c and textual p e r s p e c t i v e , that permitted

the

o b s e r v a t i o n of R a c h e l ' s l e a p f r o m a pre-parametric to a parametric stage.

Bibliography Aguilar, R. Cano (1981): Estruturas Sintacticas Transitivas en el Espanol Actual. - Madrid: Editorial Gredos. Ambar, Manuela (1985): "Sobre a estrutura dos constituintes inierrogaiivos: Govemo e inversäo." - In: Actos do 1° Encontro da Associafäo Portuguesa de Lingüistica (Lisboa) 247-268. Belletti, Adriana (1988): Generalized V-Movement. - Paper presented at GLOW, Bucharest. Bickerton, Derek (1984): Learnability and the Structure of Parameters. - Ms. University of Hawaii. Borer, Hagit (1983): Parametric Syntax - Dordrecht: Foris. Calabrese, Andrea (1986): "Pronomina: Some Properties of the Italian Pronominal System." - In: MIT Working Papers in Linguistics 8, 1-46. Chomsky, Noam (1981): Lectures on Government and Binding: The Pisa Lectures. - Dordrecht: Foris. Chomsky, Noam (1986): Knowledge of Language: Its Nature, Origin, and Use. - New York: Praeger. Chomsky, Noam (1988a): Language and Problems

of Knowledge:

The Managua Lectures.

- Cambridge,

Mass.: MIT Press. Chomsky, Noam (1988b): Prospects for the Study of Language and Mind. - Ms. Linguistics and Cognitive Science: Problems and Mysteries Session, Israel. Clahsen, Harald (1989a): Constraints on Parameter Setting: A Grammatical Analysis of Some Acquisition Stages in German Child Language. - Ms. University of Düsseldorf. Clahsen, Harald (1989b): "Creole Genesis, the Lexical Learning Hypothesis and the Problem of Development in Language Acquisition." - In: Martin Pütz & Rene Dirven (eds.): Wheels Within Wheels: Papers on the Duisburg Symposium on Pidgin Creole Languages (Frankfurt am Main: Lang) 55-79. Cole, Peter (1987): "Null Objects in Universal Grammar." - In: Linguistic Inquiry 18, 597-612. Cyrino, Sonia L. (1990): Ο objeto nulo no portugues do Brasil. - Ms. UNICAMP, Campinas.

Μ. A Kato

152

De Lemos, Claudia G. (1992): "From Dialogue Processes to Narrative Perspectives." - In: Mara S.Z. Paschoal & Maria A.A. Celani (eds.): Linguistica Aplicada: Da Aplicafio

da Unguistica para uma Lingüistica

Transdisciplinar (Säo Paulo: EDUC) 83-98. Duarte, Maria E.L. (1986): Varia?äo e sintaxe: Clitico Acusativo, Pronome Lexical e Categoria Vazia no Portugues do Brasil. - M.A. Thesis PUCSP, Säo Paulo. Duarte, Maria E. L. (1990): Α realizafäo do sujeito e do objeto na aquisi9äo. - Ms. UNICAMP, Campinas. Farrell, Patrick (1990): "Null Objects in Brazilian Portuguese. - In: Natural Language and Linguistic Theory 8:3, 325-346. Fukui, Naoki (1986): A Theory of Category Projection and Its Applications. - Ph.D. Diss. MIT. Galves, Charlotte C. (1984): "Pronomes e categorias vazias." - In: Cadernos de Estudos Linguisticos 7, 107-136. Galves, Charlotte C. (1989): "O objeto nulo no portugues brasileiro: Percurso de uma pesquisa." - In: Cadernos de Estudos Linguisticos 17,65-90. Gruber, Jeffrey (1967): "Topicalization in Child Language. - In: Foundations of Language 3, 37-65. Huang, C.T. James (1984): "On the Distribution and Reference of Empty Pronouns." - In: Linguistic Inquiry 15, 531-575. Jaeggli, Osvaldo & Ken Safir (1987): "The Null Subject Parameter and Parameter Theory." - In: Osvaldo Jaeggli & Ken Safir (eds.): The Null-Subject Parameter (Dordrecht: Kluwer) 1-44. Kato, Mary Aizawa (1993): "The Distribution of Null and Pronominal Objects in Brazilian Portuguese." - In: William J. Ashby, Marianne Mithun, Giorgio Paris Parissinoto & Eduardo P. Raposo (eds.): Linguistic Perspectives on Romance Languages (Amsterdam: Benjamins) (=Current Issues in Linguistic Theory 103) 225-235. Kato, Mary Aizawa (in prep.): The Category 'Subject' in Early Child Language. Kato, Mary Aizawa & Fernando Tarallo (1992): "Sim: Respondendo afirmativamente em portugues." In: Maria S. Z. Paschoal & Maria A.A. Celani (eds.): Linguistica Aplicada: Da Aplicafäo da Lingüistica para uma Linguistica Transdisciplinar (Säo Paulo: EDUC) 259-278. Lebeaux, David (1988): Language Acquisition and the Form of Grammar. - Ph.D. Diss. University of Massachusetts. Lightfoot, David (1989): "The Child's Trigger Experience: Degree-0 Learnability." - In: Behavioral and Brain Sciences 12:2, 321-334. Lightfoot, David (1991): How to Set Parameters. - Cambridge, Mass.: MIT Press. Maia, Marcus (1991): The Null Object in Brazilian Portuguese. - Ms. USC, Los Angeles. Meisel, Jürgen Μ. (1990): Verbal Functional Categories in Early Grammatical Development: Evidence from Simultaneous Acquisition of the First Language: French and German. - Ms. University of Hamburg. Meisel, Jürgen Μ. & Natascha Müller (1990): On the Position of Finiteness in Early Child Grammar: Evidence from Simultaneous Acquisition of Two First Languages: French and German. - Ms. University of Hamburg.

153

Brazilian Null Objects

Nishigauchi, Taissugue & Tom Roeper (1987): "Deductive Parameters and the Growth of Empty Categories." In: Tom Roeper & Edwin Williams (eds.), 91-121. Nunes, Jairo (1993): "Dire^äo de cliticizafäo, objeto nulo e pronome tönico na posi;äo de objeto em portugues brasileiro." - In: Ian Roberts & Mary Aizawa Kato (eds.): Portugues Brasileiro: Uma Viagem Diacrönica pelas Fases do Portugues Brasileiro (Campinas: Editora da UNICAMP) 207-222. Pagotto, E.G. (1992): Α Pos 19aο dos Cliticos em Portugues. - M.A Thesis. UNICAMP Campinas. Pollock, Jean-Yves (1989): "Verb-Movement, Universal Grammar, and the Structure of IP." - In: Linguistic Inquiry 20, 365-424. Radford, Andrew (1990): Syntactic Theory and the Acquisition of English Syntax: The Nature of Early Child Grammars of English. - Oxford: Blackwell. Raposo, Eduardo P. (1986): "On the Null Object in European Portuguese." - In: Osvaldo Jaeggli & Carmen S. Corvalan (eds.): Studies in Romance Linguistics (Dordrecht: Foris) 373-390. Rizzi, Luigi (1986): "Null Objects in Italian and the Theory of pro." - In: Linguistic Inquiry 17, 501-557. Rizzi, Luigi (1988): The New Comparative Syntax: Principles and Parameters of Universal Grammar. - Ms. University of Geneva. Rizzi, Luigi (1991): Residual Verb-Second and WA-Criterion. - Ms. University of Geneva. Roberts, Ian (1993): Verb andDiachronic Syntax. - Dordrecht: Kluwer. Roeper, Tom & Edwin Williams (eds.) (1987): Parameter Setting. - Dordrecht: Reidel. Sportiche, Dominique (1986): "Jibun." - In: Linguistic Inquiry 17, 369-374. Stowell, Tim (1981): Origins of Phrase Structure. - Ph.D. Diss. MIT. Taraldsen, Tarald (1978): "On the NIC, Vacuous Application and the that-trace Filter." - Indiana Linguistics Club, Bloomington. Tarallo, Fernando (1983): Relativization Strategies in Brazilian Portuguese - Ph.D. Diss. University of Pennsylvania. Tarallo, Fernando &

Mary Aizawa Kato (1989): "Harmonia Trans-sistemica: Varia^äo intra- e

inter-linguistica." Preedifäo 5. Torrego, Esther (1984): "On Inversion in Spanish and Some of Its Effects." - In: Linguistic Inquiry 15, 103129. Valian, Virginia (1988): Positive Evidence, Indirect Negative Evidence, Parameter Setting, and Language Learning. - Ms. Hunter College, New York. Wexler, Ken & Rita Manzini (1987): "Parameters and Leamability in Binding Theory." - In: Tom Roeper & Edwin Williams (eds.), 41-76. White, Lydia (1981): "The Responsability of Grammatical Theory to Acquisitional Data." - In: Norbert Hornstein & David Lightfoot (eds.): Explanations in Linguistics (London: Longman) 214-271. Williams, Edwin (1987): "Introduction." - In: Tom Roeper & Edwin Williams (eds.), vii-xix.

Agnes Fritzenschaft

Activating Passives in Child Grammar1

1.

Introduction

This paper deals with the acquisition of passive constructions in German. It pursues two main aims. First, it attempts to draw attention to the fact that theoretical frameworks provided in the literature on language acquisition are hardly able to capture the details a child learning German has to cope with in order to acquire this construction successfully. Second, it seeks to give a general account of the various steps a language learner has to take in order to acquire the adult-like forms of Zustandspassiv (static reading) and Vorgangspassiv (event reading) in German. In those acquisition studies which are based on the framework of generative grammar, the study of the development of passive constructions has been largely restricted to (a) analyzing data of English-speaking children (Horgan 1978) and (b) conducting experimental studies (de Villiers 1980). A major question in almost all of the empirical studies described in the literature is: What conceptual or universal devices must the learner have available in order to succeed in mastering this developmental task (Borer & Wexler 1987, Roeper 1987)? This issue has to be seen in the light of the ongoing debate of whether universal grammar (UG) is already available to a child from the very beginning and just needs to be activated by language input (hypothesis of continuity) - Pinker (1984) refers to this developmental step as lexical learning - or whether UG represents a cognitive faculty that matures in time (hypothesis of maturation; Borer & Wexler 1987, Felix 1987). As far as the acquisition of the passive is concerned, this controversy boils down to the question of whether a child has access to universal concepts such as 'Move Alpha' or 'Binding'2 from the very beginning or whether s/he first strictly relies on lexical information and applies the relevant (syntactic) processes or modules at a later stage, i.e. after they have been activated or have matured. The majority of the studies available to date capture only isolated stages without tracing the overall developmental process. Only a few recent studies concern themselves with an account of the actual developmental processes involved (Lebeaux 1988, Guilfoyle & Noonan 1988). However, even there one seems to concentrate largely on defining the point in time at which acquisition can be said to have taken place, thus trying to identify single clues in the input.

1

I would like to thank the following colleagues for comments, suggestions and other forms of help: Ira Gawlitzek-Maiwald, Diana Gierling, Petra Gretsch, Elsa Lattey and Rosemarie Tracy. The article is based on a project which has been funded by the German Research Foundation (DFG) since 1988 .

2

Both of these concepts capture the same phenomenon. They are invoked to account for relationships between two positions - an antecedent and its trace - in a structural configuration. Whereas the concept of 'Move Alpha' is a component of a linguistic theory which relies heavily on transformations, the concept of 'Binding' represents a module of a representational grammar model.

156

Α Fritzenschaft

In contrast to what has been claimed in the literature I propose that children do not acquire the passive construction(s) of their target language in an all or nothing fashion, but rather develop them over time, 3 by discovering and collecting relevant information at various levels or modules of the target grammar, i.e. pragmatics, semantics, syntax, lexicon, etc. For a language learner, it is thus not enough to identify just one grammatical property of the structure in question - for example the use of different auxiliaries for different types of passives in German. As individual learners may temporarily focus on different aspects of their overall task, variation is to be expected. This leads to the second major point I wish to make, namely that theoretical frameworks originally developed for describing the acquisition of English are not always adequate for dealing with the developmental data found in German speaking children, even though a considerable number of researchers proceeds in that way. German sentence structure differs from English in several features and acquisition research cannot afford to ignore them. 4 Hence I want to suggest that theoretical analyses of passives in German (Höhle 1978, Haider 1984, Toman 1986, Wilder 1989) provide a much clearer picture of the structural relationships a learner of German has to deal with in order to acquire this concept. The overall picture of the acquisition of passives which will be put forth can be summarized as follows: 1.

In contrast to English, the morphology (participles) of adjectival passives (Zustandspassiv) and verbal passives (Vorgangspassiv) in German exhibit identical properties. As a consequence, there is no need for language learners to distinguish between two types of passive at the syntactic level, i.e. s/he does not have to identify two separate derivational mechanisms. Operations such as NP movement and case absorption, which have been proposed within the framework of generative grammar, might thus play only a secondary role in the acquisitional process.

2.

Thus, passive participles have to be categorized as verb forms and passive constructions are to be regarded as members of the paradigm of auxiliary constructions in German. A child has to discover the regularities of this paradigm in order to be able to grasp and establish the concept of passive in his/her system.

3

Cf. d'Avis & Gretsch (this volume) for a similar approach with respect to the acquisition of complementizers.

4

To give just one example: The subject in a German main clause can, for instance, remain in the Mittelfeld, i.e. it is not obligatory that the subject precedes the finite verb, as is the (unmarked) case in English. Thus one finds passive constructions in which the subject (underlying object) occupies the same s-structure position (cf. (ii)) as the direct object in a corresponding active clause (cf. (i)). (i) Gestern hat meinen Chef die Polizei verhaftet yesterday has my l>ossacc 'he P°" ce nom arrested 'Yesterday the police arrested my boss' (ii) Gestern wurde mein Chef (von der Polizei) verhaftet yesterday got my boss n o m (by the police) arrested 'Yesterday my boss was arrested (by the police)'

Activating

Passives

157

3.

Parallels a learner discovers between ergative sentences and passives in the input can be used as an internal bootstrapping device to activate the structural relationships of the passive.

4.

Von phrases (by phrases) discovered in the input serve as an external trigger for the child to understand that passivized verbs can be associated with an external theta role realized as a von phrase at the syntactic level. This new knowledge helps him/her to make first hypotheses about the relationship between active and passive constructions.

The paper is organized as follows: In the following section I will briefly present the theoretical framework developed for passive constructions within generative grammar by Chomsky (1957-1981). In section 3 I sketch the analysis of passive constructions in German and discuss its implications for the developmental task. Section 4 is a critical evaluation of approaches which try to explain the development of passives in German in terms of acquisition models constructed on the basis of English data. In section 5 the acquisition of passives is described in the framework of a modular or level-oriented approach. Finally, longitudinal data from one child are described and discussed in detail. It will be shown that the developmental strategy of this child reflects and confirms the assumptions made within a level-oriented approach.

2.

Passives in English

English, it has been widely assumed by generative grammarians, distinguishes at least two types of passives: adjectival (or lexical, cf. (l)-(2)) and verbal (or syntactic, cf. (3)-(4)). (1) (2)

It is obvious that the car is redlstolen!badly damaged There is no doubt that our team is famous/beaten

(3) (4)

The car was stolen/badly damaged by joyriders The team was beaten by their own carelessness

This distinction is largely based on Wasow's 1977 observation that passive participles exhibit different distributional properties depending on whether they represent an adjective or a verb form. Whenever morphological changes, such as un-prefixation (cf. (5) below), which are typical for adjectives, result in a grammatical construction (cf. sentences (a)-(b)) the participles involved have to be classified as adjectives. If such a morphological process renders certain constructions ungrammatical (cf. sentences (c)-(d)), the participle employed has to be a member of the class of verbs. The same diagnostic holds for the distribution of passive participles at the syntactic level. On the basis of structural positions adjectives can occupy, participles in passive sentences are categorized either as adjectives (cf. examples (a) and (b) in (6)-(7)) or as verb forms (cf. examples (c) and (d) in (6)-(7)). In the first case the passive participles occupy positions typical for adjectives, which results in grammatical constructions. In the second case the passive participles cannot occupy such a position without rendering the structure ungrammatical. Thus it has to be concluded that the participle forms in examples (c) and (d) do not represent members of the class of adjectives.

Α Fritzenschaft

158

(5)

Semantic opposites (a) (b) (c) (d)

(6)

The man was ««broken/unhappy The food was untouched *The girl was uncombed by her mother *The man was ««frightened by the situation

Topicalization (a) (b) (c) (d)

(7)

Neglected/happy though the man was he still made an effort Untouched/hot though the food was the woman cleared the table *Combed though the girl was by her mother *Injured though the man was by his opponent he still didn't want to give up

Participles as prenominal modifiers (a) (b) (c) (d)

The unbroken/expensive: glass The untouched!hot food *The combed girl by her mother *The injured man by his opponent

Depending on the classification of passive participles as adjectives or verb forms, the derivation of these different lexical categories is - in the first case - the result of processes in the lexicon alone (therefore lexical passive). In the second case, i.e. the formation of the syntactic passive, the derivation includes the syntactic operation of NP movement. Syntactic passives in general describe events or ongoing activities and can therefore be characterized as [dynamic] at the semantic level. Since verb forms have a VP-internal argument position available, a direct object of the verbal head can occupy this position at a d(eep)-structural level. In the case of a passive sentence such as The car was stolen (by joyriders) the relevant structure would appear as follows: (8)

D:

e was stolen the car (T) (by joyriders (A)) 5

Some researchers assume that according to the internal properties of the (verbal) passive participle a thematic role is assigned to the argument in direct object position, whereas the subject position remains without a thematic role. At the same time the absence of a direct object has been linked to case theory, in that passive participles do not assign case.6 Baker et al. (1989) have claimed that these processes of theta role absorption and case absorption are Throughout the paper the following conventions are used: A = Agent, Τ = Theme, Ε = Experiencer, Β = Beneficiary, G = Goal, S = Source, e = empty category. ® The relationship that exists between the absorption of the external theta role and internal case features goes back to Burzio's Generalization, which says: (i) A verb which lacks an external argument fails to assign accusative case (Burzio 1986:178-179). (ii) A verb which fails to assign accusative case fails to theta-mark an external argument (Burzio 1986:184).

Activating

159

Passives

the result of the word-formation mechanisms in the lexicon which attach the passive participle morpheme \-ed\ to the verb stem, thus converting a verb categorized by the features [+V, - N] into a participle characterized by the feature [+V], As this passive morpheme \-ed\ is interpreted as an external argument, both the external theta role and features of the accusative have to be assigned to this morpheme by the verb stem to prevent the structure from being rendered ungrammatical by theta or case theory. Consequently, the subject position of the verb remains without a thematic role, and the internal argument does not receive any case features in its base-generated position. Arguments of a verb, however, need case in order to pass the case filter. 7 The internal argument of the passivized verb is thereby forced to move into a position where it can be case-marked. This results in movement of the direct object into subject position (NP movement, cf. (8')). (8')

D':

The car( (T) was stolen

t ; (by joyriders (A))

On the syntactic surface the constituent structure of adjectival (or lexical) passives is identical to that of verbal passives. Here, too, the internal theta role theme (T) corresponds to an argument in subject position, whereas the external theta role (A) can be optionally realized as a by phrase. In contrast to verbal passives, however, adjectival passives are not generated via a movement process. In the lexicon the participle, which exhibits categorial features of an adjective, subcategorizes only one argument, which occupies the position of the external argument (process of externalization, Williams 1981). As there is no VP-internal argument position available which could be occupied by an argument, the thematic role theme is directly assigned to subject position. The d-structural (cf. (9)) and s-structural representations (cf. (91)) of adjectival passives are therefore identical (cf. Levin & Rappaport 1986 for details). (9) (9')

D: His claim (T) is unsupported (by data (A)) S: His claim (T) is unsupported (by data (A))

In accordance with the semantic interpretation of adjectives, which generally refer to a state, this type of passive has a static reading and is thus assigned the semantic feature [static]. Both forms, verbal as well as adjectival passives, employ a form of the auxiliary be, which plays only a minor role in the derivational mechanisms posited for passive constructions in English. However, it might play a more important role when the acquisition of passives is at issue. As the examples in (8) and (9) above illustrate, both types of passive can occur with and without a by phrase and thus seem to license the occurrence of the two theta roles in both constructions (cf. Grimshaw 1990 for details). As a consequence, adjectival and verbal passives exhibit strong similarities regarding their constituent structure at the syntactic surface level. This actually raises an important question with respect to leamability. If one assumes that children acquire certain linguistic phenomena on the basis of triggers found in the input 7

It is a basic requirement of case theory that every single argument in a sentence has to be assigned case. The case filter is a mechanism which checks whether this requirement has been fulfilled. If an argument fails to receive case marking the whole structure is rendered ungrammatical (filtered out) unless this argument can move into a position where it can be case-marked.

Α Fritzenschaft

160

one cannot help asking whether they would actually see a need to draw a distinction between the two types in terms of their syntactic derivation from early on. The evidence provided by the input might suggest to a child that the participles in adjectival and verbal passives must be treated as representatives of the same lexical category.

3.

Passives in German

Passive constructions in the languages of the world exhibit great similarities with respect to the grammatical function their external and internal arguments fulfill in the surface structure.8 Hence it seems natural to make use of the theoretical framework developed by Wasow for the description of passive constructions in English and to attempt to generalize this framework to account for the passive in other languages. As a result, passive constructions in languages other than English are generally assumed to exhibit identical properties and are classified as either verbal or adjectival passives (cf. for instance Demuth 1988 for Sesotho). This also holds to some extent for the analysis of the passive in German (den Besten 1981, Fanselow & Felix 1987). However, a closer look at the facts shows that a more differentiated account is necessary. 3.1

Zustandspassiv and Vorgangspassiv

First of all, adjectival and verbal passives, or rather their supposed equivalents Zustandspassiv and Vorgangspassiv, can be distinguished on the basis of the auxiliaries they employ. In contrast to English, which forms both types of passive with the auxiliary be, there are two different auxiliaries for the two types of passive in German, namely a form of sein ('be') for adjectival passives and a form of werden ('get, become') for verbal passives, as illustrated in examples (10)-(11). (10)

Der Brief war (von mir) geschrieben (und nicht von Hans) the letter was (by me) written (and not by Hans) 'The letter was written (by me)'

(11)

Der Brief wurde (von mir) geschrieben the letter was (by me) written 'The letter was written (by me)'

The use of different auxiliaries with different types of passives9 allows a distinction not only on the syntactic level. It also implies differences in the semantic interpretation of the relevant structures: Passive constructions with sein ('be') describe results and therefore should be

8

To be precise, the internal argument that functions as direct object in active constructions always seems to occupy the position of the subject in passive constructions.

9

The use of different auxiliaries in passives is not restricted to the forms discussed here, but seems to be characteristic for all types of this construction, i.e. bekommen ('get') passive, erhalten ('receive') passive, etc.

Activating Passives

161

associated with the semantic property [static]·, passives employing werden ('get') refer to events or activities and are characterized by the semantic feature [dynamic]}® However, a somewhat different picture emerges with respect to differentiating between Zustandspassiv and Vorgangspassiv on the basis of the categorial status of their passive participles. Distributional analysis shows that passive participles do not exhibit specific properties of adjectives when employed in structural patterns typical for this lexical categoiy. 11 Thus it must be concluded that passive participles in German do not vary with respect to their inherent features, but rather have to be assigned the same categorial status of [+V] for all types of passives (cf. footnote 11). Moreover, if the distinction of two types of passive participle is not valid, there is no more need for two different derivational mechanisms. An analysis according to which German passive participles represent verb forms is put forth in various theories (cf. Höhle 1978, den Besten 1985, Haider 1984, 1986, Toman 1986, Wilder 1989). It has been suggested that: (a) (b)

German employs only one participle form in passives and the perfect. Passives have to be analyzed in relation to the paradigm of auxiliary constructions.

Auxiliaries exhibit selectional restrictions with respect to the argument structure of the participle forms they subcategorize. The auxiliary haben ('have') combines only with participles which have an external argument available (cf. (12) and (13)). 12 (12)

kaufen: (fy, θ ^ Der Mann hat einen Staubsauger gekauft the man has a vacuum cleaner bought 'The man bought a vacuum cleaner'

(13)

schlafen: (fy) Der Mann hat geschlafen the man has slept 'The man slept'

10 There seem to be exceptions to this. In German one finds constructions with sein and werden which at first glance seem to be interchangeable. Example (i) is taken from Höhle (1978) (translation mine, A.F.): (i) Das Regal ist/wird von vier Haken gehalten The shelf is held by four hooks' (ii) Der Garten ist/wird von einem Zaun umgeben The garden is surrounded by a fence' As these sentences contain specific verb classes with idiosyncratic semantic interpretations, it is however far from clear whether they represent true counterexamples to what has been claimed in the text. If passive participles are to be treated as adjectives the following constructions - among others - should be grammatical: (i) *Der Brief ist/wurde von mir ungeschrieben The letter is/was unwritten by me' (ii) *Der Brief ist/wurde von mir geschriebener The letter is/was more wirtten by me' (iii) *Der Brief ist/wurde von mir zu geschrieben The letter is/was by me to written' (iv) *Der Brief ist/wurde von mir geschrieben genug The letter is/was written enough by me' Höhle (1978) points out that constructions with bleiben ('remain') such as Diese Frage bleibt unbestritten (This question remains uncontested') are an exception. In sentences with bleiben participles have to be interpreted as adjectives. Following Williams (1981), external arguments are underlined.

Α Fritzenschaft

162

In constructions with sein ('be') and werden ('get, become'), participles are selected which have no external theta role available. This means they select verbs whose argument structure exhibits only one internal argument, as is the case with ergative verbs, 1 3 or where the external argument is suppressed as is the case in passives. If an internal theta role is provided by the theta grid (cf. (14) and (15) for ergatives, (16) and (17) for passives) it is assigned to subject position. (14)

Der Schnee (Τ) ist geschmolzen the snow is melted 'The snow has melted'

(15)

Das Geld (T) ist verschwunden the money is disappeared 'The money has disappeared'

(16)

Ein solcher Ausgang der Diskussion (Τ) war von a such outcome the discussion was by niemandem (Α) beabsichtigt nobody intended 'Such an outcome of the discussion was intended by nobody'

(17)

Ein solcher Ausgang der Diskussion (Τ) wurde von a such outcome the discussion got by niemandem (Α) beabsichtigt nobody intended 'Such an outcome of the discussion was intended by nobody'

In cases where verbs do not have internal arguments but undergo passivization - e.g. intransitive verbs - the structural pattern results in subjectless (impersonal) passives, cf. (18) and (19).

Ergativity is generally defined as a property of verbs that have only one internal argument, which occurs in subject position (cf. Grewendorf 1989 for details). This argument is categorized as internal because it exhibits similar properties as the objects of transitive verbs, cf. (i)-(iii). It can be concluded that it represents an argument which is generated VP-internally, cf. (iv)-(vi). (i) Der Mann hat ein Auto gekauft 'The man has bought a car' (ii) Das gekaufte Auto 'the bought car' (iii) Ein Auto gekauft hat er noch nie 'He has never bought a car' (iv) Das Geld ist verschwunden 'The money has disappeared' (v) Das verschwundene Geld 'the disappeared money' (vi) Geld verschwunden ist hier noch nie 'Money has so far never disappeared here' The parallels between passives and ergative sentences are based on two observations: (a) Both combine with the same types of auxiliary, i.e sein or werden, and (b) the theta role assigned to the subject position is an internal one in both constructions. If one assumes that the auxiliaries sein and werden select only participles which do not have an external theta role (or argument) available, one can conclude that the identical structural patterns are a result of this selectional property.

Activating Passives

(18)

Es wurde getanzt und gelacht bis in die frühen Morgenstunden it was danced and laughed until in the early morning hours 'There was dancing and laughing until the early morning'

(19)

Jetzt wird aber endlich geschlafen now is-being PRT 1 4 finally slept 'Now we're finally going to get to sleep'

163

One could argue that, from a theoretical point of view, the parallels between ergative and passive constructions are not as straightforward as suggested here. 15 Nevertheless, I will hypothesize that structural similarities between ergative sentences and forms of the sein passive could serve as a bootstrapping device for the child. Finally I would like to draw attention to the fact that in German, too, both the sein passive and the werden passive can occur with or without a von phrase at the surface level. (20)

Der Wagen ist gestohlen the car is stolen 'The car has been stolen' or 'That is a stolen car'

(21)

Der Wagen wurde gestohlen the car was stolen 'The car was stolen' or 'The car has been stolen'

(22)

Von mir wurde der Brief nicht geschrieben by me was the letter not written 'The letter was not written by me' or Ί didn't write the letter'

(23)

Von mir ist der Brief nicht geschrieben by me is the letter not written 'The letter was not written by me' or 'The letter is not a product of mine'

This raises the same question that emerged with respect to English: By means of what type of input would a child realize that s/he is supposed to arrive at two different derivational mechanisms for passives in German? Contrary to English, where one could argue that children associate the semantic feature [static] with adjectives and the feature [dynamic] with verb forms, the semantic interpretation of passives in German has to be associated with the auxiliary instead of the participle. Thus passive participles can neither provide any clues concerning the semantics of the relevant passives involved nor indicate what type of underlying derivation one would have to assume. However, if participles in German passives are categorized as representatives of the lexical category of verbs, a learner does not have to acquire two different structures. On the contrary, the fact that Zustandspassiv and Vorgangspassiv exhibit identical structural patterns at the syntactic surface, i.e. sentences with as well as without a von phrase, presents additional 14

PRT stands for particles. Höhle (p.c.) pointed out that parallels between ergative and passive structures have to be seen in the light of adnominal constructions rather than simple constructions with sein.

Α Fritzenschaft

164

evidence for the child that the derivational mechanisms for the generation of the two types of passives is the same. With respect to language acquisition, it can thus be suggested that: (a)

Learners are not forced to detect different types of participle forms. They can treat perfect participles and participles occurring in passive structures as a uniform lexical category.

(b)

Learners do not have to differentiate between derivational mechanisms for the two types of passives, but rather derive Zustandspassiv and Vorgangspassiv on the basis of the same hypothesis with respect to their internal structuring.

(c)

Learners do not have to associate different semantic interpretations, i.e. static vs. dynamic with a specific type of participle (adjective vs. verb). It is sufficient to distinguish between states and activities on the basis of the different auxiliaries employed.

(d)

Learners have to correlate the various combinations of auxiliaries and participles. They have to discover the selectional properties of auxiliaries, and they also have to gain insight into the argument structure of participles.

(e)

Learners have to realize that participle forms of transitive and intransitive verbs (which have an external argument available and usually select haben) can combine with auxiliaries (sein and werden) which do not realize external arguments in subject position.

With these hypotheses in mind we cannot regard the acquisition of the passive in German as an isolated phenomenon which is developed independently of other constructions in the grammar. Rather, in the framework proposed here, the development of passive structures is part of a general acquisition process which is closely linked to the instantiation of the paradigm of auxiliary constructions.

4.

Learnability

As mentioned before there is a noticeable tendency among investigators of language acquisition to adopt, for the analysis of German, theoretical approaches that have been developed on the basis of data collected from English-speaking children. In the light of the structural idiosyncracies of passive constructions in German such an approach is somewhat problematic.

4.1

The case of English (Borer & Wexler 1987)

As would be expected, experimental studies of the acquisition of the English passive (cf. Borer & Wexler 1987) try to establish whether children are capable of distinguishing between a static and dynamic reading. In a typical test situation, children are confronted with pictures representing either an ongoing activity or a situation where an activity has already been

Activating Passives

165

carried out (state). While s/he is looking at one of the pictures, the child is presented with a passive construction such as that given in (24). On the basis of this sentence the child has to decide whether the picture shown depicts a state or an activity. (24)

The doll was combed

(24) reflects a structural problem with English passives. The employment of the same copula in adjectival as well as verbal passives very often results in semantic ambiguity. Especially in a sentence such as (24), where no possible agent is mentioned, only the nonverbal context could provide the correct semantic interpretation, i.e. whether the clause refers to a state or an event. If learners give all or the majority of structures of the above type a static reading, one can conclude that they process adjectival passives only. If, however, they prove capable of distinguishing between a static and a dynamic interpretation it can be assumed that they have acquired both types of passive, adjectival and verbal, and derive the latter by movement. The analysis proposed by Borer & Wexler is based on the assumption that children at a very early age acquire knowledge about the mapping of semantic information onto lexical categories (cf. Pinker 1984; Roeper 1987) and therefore also distinguish between states and activities in that they interpret adjectives as typical representatives of states, whereas activities or events are typically associated with the lexical class of verbs. As soon as children are confronted with passives in English they are expected to associate the semantic feature [static] with the lexical category of adjectives and the feature [dynamic] with verb forms, thereby immediately distinguishing between adjectival and verbal passives. In their study Borer & Wexler found evidence that learners at an early stage in development gave all sentences a static interpretation. On the basis of this observation they came to the conclusion that children classify passive participles as adjectives because they are not yet in a position to form Α-chains, a property associated with passive participles belonging to the class of verbs. Children give sentences the static interpretation because binding principles have not developed (matured) to their full extent. This is why they initially use only adjectival passives, i.e. constructions which do not involve movement operations. I do not wish to discuss the issue whether or not there are learners who might actually approach the task of acquiring passives by exploiting the relationship between semantics and lexical categories in the way described by Borer & Wexler and others. However, it is far from clear whether this can be considered a general strategy that is uniformly applied by every language learner. 16 The ambiguity of passive expressions such as (24) might cause a learner to apply a different strategy than that proposed by Borer & Wexler. The similarity which adjectival and verbal passives exhibit at the syntactic level can lead a child to treat both as syntactically identical and to differentiate them merely on semantic grounds, at least early in development. This proposal does not stand in direct contrast to what Borer & Wexler have claimed. It could still be the case that there are children who make use of adjectival passives first before It can be argued that children at that stage of development have collected rather conflicting information about their target language. For instance, a child could already have acquired specific verb forms such as stand, sit, lie which do not denote an activity in the strict sense, but refer to results or states. The question then is whether a child would still strictly associate verb forms with a dynamic and adjectives with a static interpretation.

166

Α

Fritzenschaft

they go on to discover a second type, namely verbal passives. However, it casts some doubt on the assumptions (a) that children uniformly start off using adjectival before verbal passives and (b) that they distinguish these forms at the syntactic level as soon as they produce both forms. A point mentioned earlier (section 1), namely that both types of passives can occur with a by phrase, might actually support a child's first hypothesis that adjectival and verbal passives are derived on the basis of the same mechanism. A modular grammar provides a child with numerous possibilities for tackling a new structure. This leads one to expect variations in the developmental paths chosen by different learners (cf. also d'Avis & Gretsch, this volume). In contrast to the proposals discussed above, I assume that children might well "choose" to establish the concept of passive by using verbal passives first. Also, at a stage when both forms seem to be productive, some learners might still not have discovered that two derivational processes are employed in their generation. Although binding principles could already be available to the child, s/he need not yet relate them to passive constructions. 4.2

The case of German (Eisenbeiß 1990)

Borer & Wexler's hypothesis (1987) was the point of departure for one of the most recent studies on the acquisition of passives in German: Eisenbeiß (1990). In contrast to Borer & Wexler's proposal that certain principles of universal grammar have to mature in time, Eisenbeiß believes that the necessary components of universal grammar are already available to the child from the very beginning. They only have to be triggered by appropriate information which the child finds in the input. Experiments Eisenbeiß conducted show that at first children give all forms of passive a dynamic reading, even though only forms with sein ('be') are presented to them. At the same time, children use verbs with particles of the type illustrated in (25)-(26) to express a static relationship.17 (25)

Das steht da drauf this stands there on-it 'This is standing on it'

(26)

Das ist da ab this is there off 'This has been taken off

Eisenbeiß assumes that it is not until the copula werden ('get, become') is acquired that children realize that verbal passives (passives characterized by the feature dynamic) require a different pattern, i.e. a different auxiliary in German. According to Eisenbeiß, this process of lexical learning helps the learner to distinguish between the two types of passive and thus to differentiate the structures and their derivational history. At first glance Eisenbeiß' study seems to confirm what I have claimed above, namely that children start off employing verbal passives before they discover a second type, i.e adjectival passive. However, it is debatable whether werden actually serves as a trigger that enables the child to immediately associate a different categorial status (namely that of a verb form) with The examples given are mere illustrations and have not been taken from Eisenbeiß.

Activating

Passives

167

the passive participles. Experimental studies of the kind conducted by Eisenbeiß (or Borer & Wexler) help establish whether some learners are already aware of the fact that there are two different types of passive which can be distinguished by means of the features static and dynamic at the semantic level. But it is far from clear whether such an analysis allows us to draw conclusions concerning the underlying processes. My reasons are as follows: (a)

As illustrated above, an analysis of the passive in German shows that Zustandspassiv and Vorgangspassiv exhibit the same passive morphology at the syntactic surface level. Therefore they could be assumed to be generated by the same processes. Hence, an analysis that is largely based on the assumption that children have to discover differences in the categorial status of the passive participles employed does not capture the facts of passive constructions in German at a level of descriptive adequacy.

(b)

In contrast to English, in German the semantic differentiation into states and events is not linked to the participles in passive constructions. Rather, the semantics of Zustandspassiv and Vorgangspassiv have to be associated with the auxiliaries {sein vs. werden). If a child is able to distinguish between a static and a dynamic reading, s/he might do so on the basis of the different auxiliaries employed.

As a consequence, the lexical learning hypothesis put forward by Pinker (1984) and applied by Eisenbeiß to German data is not as straightforward as proposed. The discovery of werden in the input can help a child to apply the two forms in an adult-like fashion even though their inherent structuring might not be fully analyzed yet. Many grammatical constructions can be defined in terms of the different properties they exhibit at different levels of the grammar. This makes the acquisition task more complex than what it (sometimes) appears to be in the light of learnability theories. These generally assume that a child just has to detect positive evidence of some sort in the input in order to trigger the instantiation of a specific construction in his/her system. 18 I will argue below that at least the acquisition of the passive has to be considered a complex developmental task from the learner's point of view. This task is not limited to developments at one level but involves the activation and interaction of modules at various levels of the grammar.

5.

A level-oriented, modular approach

A level-oriented or modular approach to language acquisition implies that a learner does not simply identify the inherent grammatical properties of a specific construction in one instance, i.e. on the basis of a single clue (trigger) discovered in the input. Instead a child has to gradually identify the relevant properties of his/her target language at all levels of the grammar (semantics, syntax, pragmatics, lexicon, etc.) and add them to his/her own linguistic system. 19 This emphasizes the fact that the developmental process has to be described in terms of a transition where the child gradually moves from one grammatical stage (G) to Cf. Verrips (this volume) for a critical discussion of this issue. I will not discuss the issue of whether this gradual integration should be associated with a theory of continuity or of maturation, as both theoretical frameworks might offer devices to describe such a developmental process.

168

Α Fritzenschaft

another (G').20 It is hardly possible to describe each level of the grammar independently of all the other levels involved. Just as the various modules of the grammar interact in the constitution of a specific construction, a child acquiring a specific phenomenon of the adult language will draw on the modular character of his/her grammar. This means, that s/he will make use of the information discovered in one component of the grammar to predict relevant information in another. In light of the previous discussion and the characterization of passives as auxiliary constructions, a language learner will thus have to assemble the following properties at the levels of pragmatics (A), syntax (B), lexicon (C) and semantics (D) in order to develop a concept of passive that corresponds to that of the target language. (A)

At the pragmatic level a language learner has to realize that passives can be used as a focusing device in the target language. Although the discovery of the pragmatic function is certainly important in the developmental processes its treatment is nevertheless beyond the scope of this paper.

(B)

With respect to the syntactic level the child has to discover that sein and werden behave differently from haben, in that the latter allows the assignment of the external theta role to subject position and of accusative case to the internal argument (Burzio's Generalization 1986), whereas sein and werden do not. In contrast to haben, they express a relationship in which no theta role is assigned and both arguments receive nominative case.

(C)

On the basis of the information provided by the input, the learner has to associate the following feature specifications concerning accusative case and the external theta role with the respective auxiliaries in the lexicon.21 [theta role]

[case]

haben

+

+

sein

-

-

werden

-

-

In addition to analyzing the internal lexical properties of haben, sein, and werden, a child has to identify the various verb classes (transitives, intransitives and ergatives) and record their idiosyncratic argument structures in the lexicon.22

2

® This task can be described in terms of the features a child has to identify at various levels of the grammar. The current proposal could therefore be integrated into the framework developed by d'Avis & Gretsch (this volume) for the acquisition of subordinate clauses.

21

If an auxiliary allows the realization of a feature, this is indicated by (+); (-) indicates that the assignment of this specific feature is blocked.

Activating Passives

(D)

169

At the semantic level a learner has to discover that copulas can be divided into two types, depending on whether they refer to states {sein) or to activities {werden). These properties are maintained when copulas function as auxiliaries.

The information exchange between the various levels simplifies the acquisitional task and provides the learner with a powerful acquisition device. In the case of passives this could, for instance, look like this: After a child has realized that the verb haben allows^ whereas sein and werden block, the assignment of the external theta role and of structural case to the respective argument positions, s/he can transfer this information to the auxiliary patterns. S/he thus concludes that auxiliary constructions have to be divided into two main groups: into combinatorial patterns (haben + participle) which always require the realization of the external argument of the participle, and others which never occur with an external argument in subject position isein, werden + participle).23 The different participle constructions the child finds in the target language would then reflect the idiosyncratic lexical properties of haben, sein, and werden as full verbs and copulas,24 and the acquisition of the passive would follow from the gradual emergence of the auxiliary paradigm. In the following section I will present data which support this view of the developmental process.

6.

The data: the case of Paul

Paul is a monolingual speaker of German and the only son of a non-Swabian academic family of six. At the time of the first recording he was 3 years 3 months and 12 days old (3;03.12). At home he received mainly non-Swabian input, whereas in kindergarten, which he attended from the age of three onwards, he was also exposed to Swabian. Some aspects of the child's development have been dealt with before and will not be discussed in this paper (cf. Fritzenschaft et al. 1990; Gawlitzek-Maiwald et al. 1992, also d'Avis & Gretsch, this volume). The acquisition processes described below are based on a collection of 28 corpora with approximately 400 utterances each. All 28 corpora were completely analyzed, and we thus have a fairly coherent picture of Paul's development of passive constructions. The recordings were done on a biweekly basis over a period of 18 months. Apart from a few experimental 22

On the assumption that all lexical information is available to the child from the beginning, the acquisitional task is reduced to checking new information against the lexical entries provided by the lexicon. The question whether this is indeed the case or whether a child actually has to learn the argument structures of different verb classes individually shall not be discussed in this paper.

23

For a detailed discussion of the mechanisms that could explain such idiosyncratic properties of auxiliary constructions cf. Toman (1986), Czepluch (1987) or Grewendorf (1989).

24

The question of whether the mechanisms which suppress the external theta role of passive participles stem from lexical processes (second entry for passivized verb, i.e. "participle with a blocked external theta role"), from syntactic operations such as the formation of the participle morphology at the level of syntax (cf. Wilder 1989), or from percolation mechanisms (cf. Czepluch 1987) reflects the dilemma which the theory of passives faces and is a problem that cannot be resolved here (for further details cf. Fritzenschaft in prep.).

Α Fritzenschaft

170

studies on case, passive, temporal sequences and word order, the majority of the data represent spontaneous speech samples that were collected during play with the child. 6.1

Paul's system

At the beginning of the recordings Paul (3:03.12) produces well-formed main clauses with the finite verb in first or second position, but there are no complementizer-introduced subordinate clauses at this stage (cf. also d'Avis & Gretsch, this volume). (27)

soll ma für a feuerwehr erst eine garage machen/ shall we for a fire-brigade first a garage make 'Shall we build a garage first for a fire brigade?'

(PaulOl, 3;03.12)

(28)

des muß im auto bleiben\ that must in-the car stay 'That has to stay in the car'

(PaulOl, 3;03.12)

(29)

wo fährt die eisenbahn/ where drives the train 'Where does the train travel?'

(PaulOl, 3;03; 12)

Besides a great number of copula constructions with sein, sentences with the verb werden are also found in this first recording. (30)

da wird die anke (schranke)\ there gets the gate 'There the gate is being (built)'

(PaulOl, 3;03.12)

(31)

da wird die bim bam kirche\ there gets the bim bam church 'There the bim bam church is being (built)'

(PaulOl, 3;03.12)

(32)

da wird eine auf - ka - garage\ there gets a open - ga - garage 'There the garage is being (built)'

(PaulOl, 3;03.12)

All of the structures employing werden can be associated with a dynamic reading and look like verbal passives with the second part of the verbal complex, i.e. the participle, missing. These somewhat fragmentary "verbal passives" lead one to expect that this structure should soon turn up in a completed fashion in Paul's data and then be used on a regular basis. This assumption appears to be confirmed two weeks later, when the first forms of verbal passives are spontaneously produced, cf. (33)-(35). (33)

ηi) da wird eine treppe gebaut\ n8 there gets a staircase built "There a staircase is being built1

(Paul02,3;03.26)

171

Activating Passives

(34)

nn gleich wird die gefahrt\ nn soon gets this-one driven 'It will be driven soon'

(Paul02,3;03.26)

(35)

the locomotive is pushed off da wird abgeschoben\ gell/ there gets pushed-off PRT 'There (it) is being pushed off, isn't it?'

(Paul02,3;03.26)

At the same time, however, passives where the participle is missing, cf. (36)-(37), and constructions with werden as copula, as in (38), or full verbs (39), are found in the corpus too. (36)

P. is building soll man η shall one η 'Here a creek

with Legos da da wird ein bach\ there there gets a creek is (being formed)'

(Paul02,3;03.26)

(37)

P. is building with Legos da wird des fenster\ there gets the window (a) 'Here the window is (being built)' (b) 'This is where the window (goes or will go)'

(Paul02,3;03.26)

(38)

da wird auch griin\ gell\ there gets also green PRT 'That is getting green too, isn't it?'

(Paul02,3;03.26)

(39)

P. is pointing at sleeping doll die soll wach werden\ gell/ this-one shall awake get PRT 'She ought to wake up, isn't that so?'

(Paul02, 3;03.26)

No forms of adjectival passive are documented in the data at this time. On the basis of this observation only, one could come to the conclusion that Paul happens to be acquiring verbal passives first and that he only later realizes that there is a second type of passive, the adjectival passive. However, these few examples of verbal passives, cf. (33)-(35), and the structures with werden remain the only ones for quite a long time. Apart from one more werden structure in Paul03 at the age of 3;03.11 and one verbal passive at the age of 3;05.27 in Paul05, no further passive construction is found in the data for the next five months. Out of approximately 3200 analyzed utterances only one was a verbal passive, even though the forms of the passive found at the beginning of the study led us to expect that Paul would continue with the production of these structures. 2 5

As always, we can never rule out (he accident of sampling. Since the recordings were conducted on a biweekly basis, it seems fairly unlikely, however, that passive sentences should have been acquired by Paul without him applying even one of these structures during the recording sessions.

A Fritzenscha.fi

172

Instead, Paul draws on constructions with modal or copula verbs plus particles to distinguish between static ((41)-(42)) and dynamic ((40), (43)-(45)) relationships. (40)

jetzt muß die des runta\ now must this-onethat down 'Now she has to (take) it down'

(PaulOl, 3;03.12)

(41)

jetzt ist der klodeckel zu\ now is the toilet-lid closed 'Now the toilet lid is closed'

(Paul07, 3;06.18)

(42)

da ist ein honig drin\ there is a honey inside 'There is honey inside'

(Paul07, 3;06.18)

(43)

P. is closing door of cupboard un des soll zu sein\ and this should closed be 'And this should be closed'

(Paul07, 3;06,18)

(44)

P. is putting doll in box nein ein mensch soll rein\ da drin sein\ no a human being should in there inside be 'No, a human being should be put inside - be there inside'

(Pauli 1, 3;08.12)

(45)

jetzt kommt der da drauf\ now comes this-one here on-top 'Now this one goes on top here'

(Paul 10,3;07.25)

These findings support similar observations made by Eisenbeiß, who found that children used verbs with particles to express static relationships, in most cases resultatives. On the basis of these data I would, however, claim that Paul employs modal or copula structures as a kind of "surrogate" for both types of passive, adjectival and verbal. Examples (43) to (45), for instance, invite a dynamic rather than a static interpretation because Paul is commenting on what he is doing. In addition, if one takes into account that the passive represents a type of participle construction, Paul's generally rare auxiliary structures at this stage provide further evidence for my hypothesis that the internal structuring of adjectival or verbal passives has not been identified yet. The few participle structures produced by Paul within the first months of the investigation exhibit the auxiliary haben combined with a transitive or intransitive verb, as well as ergative verbs in combination with the auxiliary sein. However, as Figure 1 shows, they represent only a very small percentage of the total number of utterances. Figure 1 reads as follows: Across the top in bold print, the relevant corpora are listed. Underneath, in the same box, Paul's age (in parentheses) and the total number of utterances are given (e.g. Paul04 was recorded at the age of 3 years, 5 months and 15 days and it comprises a total number of 296 utterances). The leftmost column - again in bold print contains, first, the total number of participle constructions in each corpus, followed by the

173

Activating Passives

individual types of auxiliary + participle structures. The actual number of participle constructions that occurred in the data are given in italics, followed by the percentage of total utterances they represent (in parentheses). Figure 1: Occurrence of participle constructions in Paul's early data

Paul04 (3:05.15) 296

PaulO 6 (3:06.12) 365

Paul07 (3:06.18) 763

Paul08 (3:06.27) 301

Pauli 3 (3;09.12) 424

tot. part.

6 (2.03)

5(1.40)

4 (0.52)

10 (3.52)

11 (2.59)

haben

J (1.01)

2 (0.56)

1 (0.13)

2 (0.66)

4 (0.91)

sein (ergative)

2 (0.68)

2 (0.68)

1 (0.13)

2 (0.66)

3 (0.70)

sein (passive) werden

-



-





At the age of 3;08.12 adjectival passives become productive, i.e. are used on a regular basis. (46)

jetzt ist des abgemacht\ now is it off-taken 'Now it has been taken off

(Pauli 1, 3;08.12)

(47)

jetztn du jetzt ist da alle traktore ausgeladen\ now you now is there all tractors unloaded 'Well, now all the tractors are unloaded there'

(Pauli 1,3;08.12)

(48)

un der ist nicht richtig hingesetzt\ and this-one is not properly down-sat 'This one hasn't been set down properly'

(Paulll, 3;08.12)

Only two weeks later (3;08.26) verbal passives are found again, and they occur regularly from then on. Interestingly, none of them is used with a von phrase. (49)

du jetzt wird des reingeladen\ you now gets this in-loaded 'Hey, now this is loaded in(to it)'

(Paul 12, 3;08.26)

(50)

dran gemacht wirda\ to-it fixed gets-it 'It is being fixed to it'

(Pauli3, 3;09.12)

Α

174

Fritzenschaft

(51)

schau mal jetzt wird das kleine gefahren\ look PRT now gets the small-one driven 'Look, now the little one is being driven'

(Paull9, 4;00.04)

(52)

wie wird der reingesteckt/ how gets this-one in-put 'How does this one go in?'

(Paul25, 4;05.23)

6.2

The developmental process

What do those observations reveal with respect to the overall developmental picture? Initially, there is evidence for differentiation at the semantic level and in the lexicon. This covers the time from the first recordings to the point in Paul's development when passives have become productive. During this period Paul distinguishes states and activities by employing structures containing modals and the copula sein, and he observes the basic selectional restrictions of auxiliaries with respect to the argument structure of the corresponding participles. In the early stages, Paul also uses forms of werden as well as some idiomatic forms of verbal passives with a dynamic reading. According to Eisenbeiß (1990), the availability of auxiliary werden, in particular, provides enough information for lexical learning to take place, and, consequently, for an identification of the internal structure of verbal passives. However, the fact that in Paul's data verbal passives appear only in the beginning and then disappear again for a period of about 5 months speaks against the hypothesis that the lexical identification of the auxiliary serves as the crucial key to the internal structure of verbal passives. Apparently, no substantial connection between passive structures and the rest of Paul's system has been established at this early stage. On the contrary, Paul's use of fragmentary passives with missing participles and the low frequency of structures with participles in general show that the combination of werden + participle to establish the structural pattern of the passive has not been fully acquired yet and that Paul therefore has not analyzed the inherent properties of the passive construction either. This might be the reason why he does not develop the pattern any further. Instead, he returns to a verbal complex that is familiar to him, namely to modal and copula structures with particles, which enable him to express the same semantic relationships as passives (static vs. dynamic). This use of sentences containing modals or copulas, in fact, falsifies the assumption put forward in the literature that children simply have to draw on their inherent knowledge (made available by semantic bootstrapping) about the one-to-one relationship between adjectives and states, and verbs and activities. Paul's data provide evidence that this knowledge may not be exploited by a child acquiring German. In German, passives are assigned their semantic interpretations on the basis of the auxiliaries rather than the participles. The first occurrence of productive passive constructions at the age of 3;08.12 (Pauli 1) and 3;08.26 (Paul 12) fall into a period when the total number of participle constructions is still very small (cf. Figure 1). The employment of passive constructions is just an additional step in Paul's elaboration of an auxiliary system which contains all major types of auxiliary patterns. Although structures with haben and sein were not used extensively during the first

Activating

Passives

175

months of observation, I want to argue that they prepared the way for the integration of the passive into Paul's grammar. From the very beginning Paul's structures with the auxiliaries haben and sein are targetlike, i.e. no deviant combinatorial forms such as *Der Junge hat hingefallen (instead of Der Junge ist hingefallen 'the boy has fallen down') or *Der Junge ist geschlafen (instead of Der Junge hat geschlafen 'the boy has slept') are documented. This shows that Paul has already acquired the different selectional patterns that are associated with different auxiliary forms. Although Paul's auxiliary structures conform to what we find in the adult language, an analysis of his overall linguistic system raises the question what kind of underlying syntactic representation we can assign to these sentences with auxiliaries. Constructions with haben prove relatively unproblematic, as they represent instances of regular case and theta marking, properties Paul has already acquired for the derivation of simple sentences with transitive or intransitive verbs. The crucial question is whether Paul realizes that in sentences with auxiliary sein and werden it is an internal argument which functions as the subject of the clause. The data allow at least the following two competing interpretations: (a)

The fact that ergative verbs are consistently associated with the auxiliary sein from early on could be taken as an indication that Paul has already acquired the different selectional restrictions of participle and auxiliary and knows that sein only allows the realization of an internal argument in subject position. This knowledge is first transferred to adjectival and then to verbal passives. In the latter case, the information that werden represents the same type of auxiliary as sein with a different semantic interpretation is thereby inferred from the lexicon.

(b)

Despite his adult-like use of auxiliary patterns Paul is not aware of the fact that the argument that functions as subject in ergative and passive structures is base generated VP-internally and moved into subject position via the syntactic operation of NP movement. He associates the participles involved with an independent lexical entry representing a single-argument structure.

For the reasons given below I want to assume the latter, i.e. (b): It was pointed out above that during the early period in Paul's development his language exhibits a fairly high percentage of copula constructions with sein. If my reasoning (cf. section 4) that children transfer information about the grammatical properties of the verbs haben, sein, and werden to auxiliary constructions is correct, then Paul could assume, as a first hypothesis, that sentences with auxiliary sein have to be treated as instances of a predicative construction at the structural level. As a consequence, participles would be assigned the status of adjectives with a single argument. This amounts to saying that at a fairly early stage, Paul distinguishes between two participle forms: a verb form, which occurs with haben, and an adjective, which occurs only with sein. The first adjectival passives which occur in the data would have to be interpreted accordingly. At the syntactic level, they, too, exhibit a typical pattern of simple (i.e. mainclause) copula sentences with the finite verb in second position and a participle in verb-end position. And both ergatives as well as adjectival passives realize the thematic role theme in subject position. At the semantic level, both constructions have to be associated with a static reading. Thus it could be assumed that Paul treats ergative structures and adjectival passives

Α Fritzenschaft

176

as identical syntactic patterns. The fact that the two types of constructions require different verb classes is irrelevant at that stage because the learner does not associate participles with their corresponding verb forms yet, but classifies them as independent lexical items. Shortly after Paul uses adjectival passives for the first time, he realizes that there is another syntactic pattern for expressing the feature [dynamic] besides modal and copula structures, namely combinations of a copula verb with the type of adjective he already employs in the production of adjectival passives.26 Thus the pattern which Paul employs for ergative and passive constructions can be used to "bootstrap" another structure into his system. The only difference between his old and new forms lies in the fact that a dynamic interpretation requires an auxiliary referring to activities or events. This might explain why instances of verbal passives appear only two weeks after adjectival passives were first observed. Interestingly enough, sentences in which werden is used as a full verb reappear only after first forms of verbal passives are found in the data. (53)

jetzt wird das ein bißchen länger now becomes this a bit longer 'Now this gets a bit longer1

(Paull3,3;09.12)

(54)

(Pauli 3, 3;09.12) du das wird ein tablett da platscht der kakao raus you v o c 2 7 that becomes a tray there spills the cocoa out 'Hey, this is going to be a tray, onto which the cocoa can be spilled'

The fact that both types of passive occur in a target-like fashion before Paul makes use of copula werden could be taken as evidence that Paul's early forms of verbal passives, too, were based on individual lexical entries for participle forms, i.e. passivized verbs. 6.2.1

Further evidence

So far the main argument for interpreting Paul's early passives as predicative constructions is based on two observations: (a)

From an empirical point of view, one cannot ignore the fact that passive constructions emerge at a stage in development when the auxiliary system is established in Paul's grammar.

(b)

From a theoretical point of view, it is important to note that ergative constructions, which are found in the data shortly before first forms of passives appear, exhibit structural parallels with the passive with respect to their argument structure at the syntactic surface level.

26

Paul's having made use of this structure before - although only in a very limited way - might help him to reactivate this pattern.

27

voc stands for vocative.

Activating Passives

177

Of course, the data illustrated above also suggest an alternative interpretation of the underlying developmental process. As there are no deviant forms in the data which could point to particular acquisition problems, the developmental process appears to be a fairly straightforward one. One could therefore always claim that Paul knows from early on about the inherent processes involved in the derivation of auxiliary constructions. As a consequence, he associates passives with exactly the derivational mechanism required by the target grammar, for instance, a syntactic operation which derives ergative and passive constructions by applying NP movement in order to place the internal argument in subject position. However, there is additional evidence supporting an interpretation of Paul's passives as single-argument constructions if we once more consider the above-made assumption that children associate certain grammatical properties with haben, sein, and werden before they realize that these verbs can function as auxiliaries. In the case described it is not only the structural pattern auxiliary constructions exhibit at the syntactic surface which have to be taken into account, theta- and case-marking features also play an important role. 28 To be precise, if we assume that Paul treats ergatives and passives as simple predicative constructions, we have to prove that he has not analyzed the inherent case- and theta-marking features of copula werden and sein by the time he uses adjectival and verbal passives for the first time, even though he seems to be producing adequate constructions of the target language. The following observations support this proposal: (a)

Case-marking facts: The examples in (55)-(59) demonstrate that Paul does not obey the requirements of the adult system of morphological case at the stage when verbal passives occur on a regular basis in his data.

(55)

jetzt kitzel ich dir\ now tickle I you*dat 'Now I tickle you'

(Paull2,3;08.26)

(56)

A: Wer ist da alles im Zirkus? who is there all in-the circus 'Who all is there in the circus?'

(Paull6,3;11.00)

P: krokodile immer so un dann noch diesen bär\ crocodiles always so and then also this*acc bear 'Mostly crocodiles and then this bear' (57)

28

A: Und wer bekommt die Erdnuß hier? and who gets the peanut here 'And who is getting the peanut here?' P: dem peter\ the*dat peter 'Peter'

(Pauli 9,4;00.04)

It should be pointed out that children might come to their own conclusions with respect to the case- and theta-marking properties they associate with full verbs and the copula. As a consequence the properties they assign to these verbs do not necessarily have to conform to what is assumed for these constructions in the adult grammar.

Α Fritzenschaft

178

(58)

Α: Der Löwe Leo wird vom Peter gerettet. the lion Leo gets by Peter saved 'Leo the lion is being rescued by Peter' P: warum/ will II29 eßt der der/ why? wants II eats he no m he* n0 m 'Does he eat him?'

(59)

(Pauli 9,4;00.04)

or eats he*nom he n om?

du ich magder nich mehr spielen you v o c I like this*nom not anymore play 'Hey, I don't want to play this anymore'

(Pauli9,4;00.04)

He does not even overgeneralize certain case forms as reported in Clahsen (1984) and Tracy (1986), but randomly assigns them to argument-positions instead. He obviously believes that - according to case theory - every single argument has to receive case, but he does not seem to know about the conditions and mechanisms governing their distribution in German. Interestingly enough, he extends his deviant case marking to copula, cf. (60)-(63), as well as to passive constructions, cf. (64)-(65). (60)

den sind füsse\ gell/ these*acc are feet, aren't they 'These are feet, aren't they?'

(PaulOl, 3;03.06)

(61)

den besen soll auch in sandkasten sein\ this*acc broom should too in sandbox be 'This broom should be in the sandbox too'

(Paul04,3;05.15)

(62)

nein das ist doch unseren haus\ no that-one is PRT our*acc house 'No that's our house'

(Paul05,3;05.27)

(63)

A: Hei guck, was ist da noch drin? hey look what is there still in 'Hey look, what else is in there?'

(Paul05,3;05.27)

P: den schiff\ the*acc ship 'The ship' (64)

29

wo du v o c werd den Stoff gemacht/ where you gets the*acc material made 'Where does the material get produced?'

The symbol II indicates a new start.

(Paull2,3;08.26)

179

Activating Passives

(65)

Α: Der wird gefressen? this-one gets eaten 'He gets eaten'

(Paull9,4;00.04)

P: ja\ den wird gefressen\ yes this-one*acc gets eaten 'Yes this one gets eaten' These examples indicate that Paul has not identified the idiosyncratic lexical properties of transitive verbs and copulas with respect to case marking yet. Apart from target-like sentences both verb forms occur with deviant case marking: Copulas are used with an argument in the accusative, cf. (60) -(62), whereas sentences with a transitive verb have objects in the nominative, cf. (58)-(59). The data confirm the assumption made above, i.e. Paul does not yet know about the idiosyncratic lexical behavior of copula and auxiliary sein and werden with respect to theta and case marking. Instead he treats both forms as any other transitive verb which allows different case features to be assigned to the argument positions of the verb.30 It is thus debatable whether the child's early participle constructions with sein and werden have to be regarded as instances of true passives in the sense that the learner is already aware of all the processes employed in the derivation of these structures. In examples (52)-(62) Paul simply distributes case features in order to satisfy basic requirements of case theory. (b)

(66)

Experimental studies show that at the same time when first forms of passives are documented, Paul still has problems pointing out an agent when confronted with constructions of verbal passives. P. is asked to act out the following scene: A: Der Traktor wird vom Peter gefahren The tractor is by Peter driven 'The tractor is driven by Peter'

(Pauli 1, 3;08.12)

P: nein\ der hat nicht gefahren\ no, this-one has not driven 'No, he didn't drive' (67)

A: Von wem wird er denn gefahren? by whom is it then driven 'By whom is it (being) driven then?'

(Pauli 1, 3;08.12)

P: 33 vo von de vo von II schaumal der hat ne anhängerkupplung\ by by the by by II look this-one has a tow bar This lack of competence can be explained on the assumption that Paul treats passives as another type of auxiliary construction. If the learner classifies constructions with copula and participle as single-argument structures with an independent lexical entry, he does not need to Radford (1990) proposes a similar analysis for the data of English children who miscategorize copula forms as non-auxiliary verbs.

180

Α Fritzenschaft

associate another (external) argument with them. In a single-argument construction in which the only available argument occupies the subject position the theta criterion is fully satisfied. 6.3

Reorganizing lexical knowledge

If Paul's early passives, which are all agentless, are projections of single-argument structures, there has to be a phase in his development in which he realizes that these structures can be systematically related to another clause structure which is available in his target language, namely that of active constructions containing the same verbs. This requires a process of reorganization which includes the classification of passive participles as verbs with more than one argument. This reanalysis could be the result of the discovery that some participle constructions have to be associated with an additional theta role which can occur in a von phrase in the input. On the basis of this discovery, he concludes that what he had interpreted as an independent lexical entry before is actually a variant of the same lexical category he finds in simple sentences with transitive verbs. As soon as he associates the additional argument with the passive participle, the two separate entries converge. This way Paul infers that transitive verbs can combine with both auxiliary haben and copula sein or werden, which results, however, in a different mapping at the surface level in the latter case. (68)

Der Bauer hat den Traktor aufgeladen the farmer has the tractor up-loaded 'The farmer has loaded the tractor'

(69)

Der Traktor wird (vom Bauern) aufgeladen the tractor is (by the farmer) up-loaded 'The tractor is (being) loaded (by the farmer)'

The major difference between these two forms lies in (a) the mapping of thematic roles onto different syntactic positions, and (b) the employment of different auxiliaries. The passive has an argument in subject position that is assigned the thematic role theme plus a form of sein or werden, whereas participle constructions with have assign the thematic role agent to the external argument and the thematic role theme to the internal argument. At this point in the developmental process Paul has established the relationship between active and passive forms.

7.

Concluding remarks

The approach outlined in this paper rests on the assumption that language learners need to accumulate information at various levels of the grammar in order to evoke a new concept in their system, in the case described that of passive constructions. They succeed in doing so by drawing on similarities encountered between structural configurations with respect to their lexical, syntactic, and semantic representations.

Activating Passives

181

I have argued that the acquisition of the German Zustands- and Vorgangspassiv can be considered part of the construction of the paradigm of auxiliaries. As illustrated above, this largely involves the child's discovery of relationships that hold between auxiliaries and participles, including restrictions with respect to theta- and case-marking properties. The modular or level-oriented approach argued for accounts for variations in the developmental strategies applied by individual learners in the process of constructing an overall coherent grammar. In fact, it claims that learners can draw on the principles made available by UG and on the information presented to them in the input in a rather individual fashion. Depending on which components of their already established system children initially focus on, different developmental patterns can be observed. One learner type has been described in this paper. Other children might choose different approaches. At the same time, a level-oriented approach restricts the number of possible developmental paths a learner can choose from because it clearly defines the properties which need to be discovered before a specific construction is acquired. On the basis of the case study presented it can be concluded that Paul represents a type of (conservative) syntactic learner who relies on available syntactic patterns to either maintain the grammar he has already developed or to incorporate new structures into his system. This assessment is supported by the following observations: (a)

Paul employs well-known modal and copula constructions to distinguish between states and activities at a point in development when passives are not yet productively used.

(b)

The structural pattern which Paul employed for ergatives is transferred first to the adjectival and then to the verbal passive.

(c)

Both forms, Zustandspassiv and Vorgangspassiv, are agentless and therefore exhibit strong similarities at the syntactic surface level. The only difference between them lies in the employment of different auxiliaries indicating different semantic interpretations.

If a child treats the two passives as identical structural configurations distinguished only by different auxiliaries, the derivational mechanism associated with them does not have to be a complicated one, at least not early stages of development. It therefore seems likely that early forms of the passives are treated as projections of single-argument predicates where the only available argument is associated with the subject position in order to fulfil the requirements of the Extended Projection Principle (i.e. the requirement that sentences must have subjects, cf. Chomsky 1981). Paul's first hypothesis about the internal structure of passives is based on a parallel which he perceives between ergatives and passives at the lexical and syntactic level. As mentioned above, the relationship between ergative and passive constructions is certainly more complicated than it seems to be on the basis of the observations made here (cf. Haider 1984; Toman 1986 for details). Nevertheless, a learner might make use of these similarities, ignoring details in the beginning because s/he is not aware of all the structural properties involved. Thus a child who has already incorporated ergative constructions in his/her system might transfer information s/he has collected about their internal structure to other sentences which apparently exhibit an identical pattern at the syntactic surface level. This way ergative

182

Α Fritzenschaft

constructions can function as a system-internal trigger or bootstrapping device for the concept of passive. At the time when the first forms of passives already appear to be productively used, Paul does not even need to be aware of the fact that this construction differs from other auxiliary patterns. A derivational relationship with an active sentence is not likely to be established yet. The full implications of passive constructions emerge only later, i.e. at a time when passives containing a von phrase or impersonal passives which have no subject at all are identified. Only then might the child realize that the idiosyncratic property of the passive is linked to the suppression of the external theta role rather than to the assignment of the only available theta role to the subject position. Linking the development of passive constructions in German to the discovery of relationships between the argument structure of verbs and auxiliaries has certain consequences for the learnability theories proposed in the literature. In particular, the classification of passives as adjectival and verbal in the sense described by Wasow (1977) might not be relevant for the description of Zustandspassiv and Vorgangspassiv in German. Consequently, Borer & Wexler's proposal that learners have to distinguish between (a) passive participles reflecting the category of adjectives, and (b) participles reflecting the category of verb forms in order to acquire passives as a concept does not hold for German in the way it does for English. The child's ability to distinguish between the two features static and dynamic indicates that s/he has acquired two types of passive construction at the semantic level. However, it does not provide any evidence for a movement process involved in the generation of verbal passives, as is claimed by Wexler, Eisenbeiß, and others. The detection of a movement process and also, to some extent, properties of case absorption might only be secondary in establishing the concept of passive in German. Consequently, children like Paul can make productive use of what already look like target-like passives before they know about their internal structuring. Similarly, it appears to be rather problematic to describe the acquisition of the passive pattern at the syntactic level as a result of lexical learning (Pinker 1984, Roeper 1987). This would imply that the child, for instance, simply has to identify the inherent properties of werden for the structural realization of the Vorgangspassiv to follow suit. Paul seems to approach this task in a different way. Although he has "parts" of the structure available earlier on, he does not pursue its integration into his system any further. Instead he turns to developing a more general syntactic pattern first, that of auxiliary and participle. He finally exploits this pattern to "reactivate" the copula werden, drawing on its function in passive structures. A learner can infer information about single lexical items by relying on the complex structural patterns in which they occur, instead of identifying the lexical items first and then drawing conclusions concerning the internal organization of the sentences in which they are employed. Of course, one could argue that some kind of lexical learning is involved anyway, even though it might not strictly correspond to the mechanisms defined as lexical learning in learnability theory. This again throws new light on the issue of lexical learning in that it raises the following questions which, unfortunately, have to be left unanswered for the time being: (a) How powerful can lexical learning be, and (b) what does lexical learning really mean? Activating passive constructions in German clearly boils down to the children realizing that they are acquiring a language which requires a uniform lexical category, namely a verb form for both types of passive. The fact that single-argument clauses which could be

Activating Passives

183

interpreted as adjectival constructions might nevertheless be important for explaining developmental processes is a somewhat different matter. One could certainly argue that the interpretation of passives as adjectival forms is an option which is offered to the language learner by the structural properties of German. It therefore comes as no surprise that a child adopts an acquisition strategy that exploits adjectival constructions and their internal structure and infers from there other structures with similar properties. However, this should not be taken as the only developmental path a learner of German can take.

Bibliography Baker, Mark, Kyle Johnson & Ian Roberts (1989): "Passive Arguments Raised." - In: Linguistic Inquiry 20, 219-251. Borer, Hagit & Ken Wexler (1987): "The Maturation of Syntax." - In: Tom Roeper & Edwin Williams (eds.): Parameter Setting (Dordrecht: Reidel) 123-173. Burzio, Luigi (1986): Italian Syntax: A Government-Binding Approach. - Dordrecht: Reidel. Chomsky, Noam (1957): Syntactic Structures. - The Hague: Mouton. Chomsky, Noam (1965): Aspects of the Theory of Syntax. - Cambridge, Mass.: MIT Press. Chomsky, Noam (1981): Lectures on Government and Binding. - Dordrecht: Foris. Clahsen, Harald (1984): "Der Erwerb der Kasusmarkierung in der deutschen Kindersprache." - In: Linguistische Berichte 89,1-31. Czepluch, Hartmut (1987): Zur Theorie des abstrakten Kasus. - Habilitation Thesis University of Göttingen. Czepluch, Hartmut (1988): "Kasusmorphologie und Kasusrelation: Überlegungen zur Kasustheorie am Beispiel des Deutschen." - In: Linguistische Berichte 116, 275-305. d'Avis, Franz-Josef & Petra Gretsch (this volume): "Variations on "Variation": On the Acquisition of Complementizers in German." Demuth, Kathrin (1988): Maturation and the Acquisition of the Sesotho Passive. - Ms. African Studies Center, Boston University. de Villiers, Jill (1980): "The Process of Rule Learning in Child Speech: A New Look." - In: Keith Nelson (ed.): Children's Language. Vol. 2. (New York: Gardner Press). den Besten, Hans (1981): "A Case Filter for Passives." - In: Adriana Belletti, Luciana Brandi & Luigi Rizzi (eds.): Theory of Markedness in Generative Grammar (Pisa: Scuola Normale Superiore) 65-123. den Besten, Hans (1985): "Some Remarks on the Ergative Hypothesis." - In: Werner Abraham (ed.): Erklärende Syntax des Deutschen (Tübingen: Narr) 53-74. Eisenbeiß, Sonja (1990): Zum Passiverwerb in der deutschen Kindersprache. - Ms. University of Düsseldorf. Fanselow, Gisbert & Sascha Felix (1987): Sprachtheorie. Vol. 2: Die Rektions- und Bindungstheorie. Tübingen: UTB Francke. Felix, Sascha (1987): Cognition and Language Growth. - Dordrecht: Foris. Fritzenschaft, Agnes (in prep.): Der Erwerb des Passivs im Deutschen. - Ph.D. Diss. University of Tübingen.

184

Α Fritzenschaft

Fritzenschaft, Agnes, Ira Gawlitzek-Maiwald, Rosemarie Tracy & Susanne Winkler (1990): "Wege zur komplexen Syntax." - In: Zeitschrift für Sprachwissenschaft 9,52-134. Gawlitzek-Maiwald, Ira, Rosemarie Tracy & Agnes Fritzenschaft (1992): "Language Acquisition and Competing Linguistic Representations: The Child as Arbiter." - In: Jürgen Μ. Meisel (ed.): The Acquisition of Verb Placement:

Functional

Categories and V2 Phenomena in Language Acquisition

(Dordrecht:

Kluwer) 139-179. Grewendorf, Günther (1988): Aspekte der deutschen Syntax. - Tübingen: Narr. Grewendorf, Günther (1989): Ergativität im Deutschen. - Dordrecht: Foris. Guilfoyle, Eithne & Mäire Noonan (1988): "Functional Categories and Language Acquisition." - Paper presented at the 13th Boston Conference on Language Development. Grimshaw, Jane (1990): Argument Structure. - Cambridge, Mass.: MIT Press. Haider, Hubert (1984): "Was zu haben ist und was zu sein hat." - In: Papiere zur Linguistik 30, 32-46. Haider, Hubert (1986): "Fehlende Argumente: Vom Passiv zu kohärenten Infinitiven." - In: Linguistische Berichte 101, 3-33. Höhle, Tilman (1978): Lexikalische Syntax: Die Aktiv-Passiv-Relation

und andere Infinitkonstruktionen im

Deutschen. - Tübingen: Niemeyer. Hoekstra, Teun (1984): Transitivity: Grammatical Relations in Government-Binding

Theory. - Dordrecht:

Foris. Horgan, Dianna (1978): "The Development of the Full Passive." - In: Journal of Child Language 5,65-80. Lebeaux, David S. (1988): Language Acquisition and the Form of the Grammar. - Ph.D. Diss. University of Massachusetts. Levin, Beth & Melka Rappaport (1986): "The Formation of Adjectival Passives." - In: Linguistic Inquiry 17, 623-661. Pinker, Steven (1984): Language Learnability and Language Development. - Cambridge, Mass.: Harvard University Press. Radford, Andrew (1990): Syntactic Theory and the Acquisition of English Syntax: The Nature of Early Child Grammars of English. - Oxford: Blackwell. Roeper, Tom (1987): "The Acquisition of Implicit Arguments and the Distinction Between Theory, Process, and Mechanism." - In: Brian MacWhinney (ed.): Mechanisms of Language Acquisition (Hillsdale, N.J.: Erlbaum) 309-344. Toman, Jindfich (1986): "A (Word-)Syntax for Participles." - In: Linguistische Berichte 105, 367-408. Tracy, Rosemarie (1986): "The Acquisition of Case Morphology in German." - In: Linguistics 24, 47- 78. Verrips, Maike (this volume): "Learnability Meets Development: The Case of Pro-Drop." Wasow, Tom (1977): "Transformations and the Lexicon." - In: Peter W. Culicover, Thomas Wasow, Henk van der Hulst & Michael Moortgat (eds.): Lexical Grammar (Dordrecht: Foris) 327-360. Wilder, Chris (1989): "Passive and the German Infinitive." - In: Frankfurter Linguistische Forschungen 9, 1-31. Williams, Edwin (1981): "Argument Structure and Morphology." - In: The Linguistic Review 1,81-114.

Beatrice Primus & Katrin Lindner

Variation in Grammar and First Language Acquisition: A New Concept of Parameter in Universal Grammar

1.

Introduction

The present article discusses evidence against approaches that reduce the descriptive inventory of linguistic models to a single type of relational concepts. The relational concepts at issue are the thematic relations such as agent and patient, the structural relations as conceived in recent Generative Grammar, and the morpho-syntactic case relations such as nominative argument or accusative argument of a verbal head. Most of the approaches to grammar have tried to limit their descriptive inventory to just one of the relational systems mentioned above. The other relational systems have been neglected, or considered to be derived from the chosen set of relations. As is well known, Functional Grammars (cf. Dik 1989, Givon 1984, Foley & Van Valin 1984) have concentrated on semantic relations, notably on the above-mentioned thematic relations, and disregarded structural relations and case relations. Generative Grammars, by contrast, have formulated rules of grammar solely in terms of structural relations. The data we will present in this article clearly suggest that these approaches, which we will call Reductionistic Approaches, are not appropriate. We will present an alternative approach that incorporates more than one relational system. We call our approach the Generalized Hierarchy Approach. The simplicity of the model to be presented in this article is not achieved by reducing the descriptive inventory, but rather by organizing the relational concepts on hierarchies in such a way as to permit very simple and general rule formulations. The generalizations and observations to be presented here are based on studies that have been designed and carried out independently by the authors of this paper. Due to this fact, the findings of the language acquisition study are particularly valuable as independent evidence for the Generalized Hierarchy Approach. The order of presentation is as follows. Section 2 introduces the relevant relational concepts and the main hypotheses of our approach. Section 3 illustrates the advantages of our approach to grammar for basic order facts. Section 4 discusses language acquisition data which are incompatible with approaches that reduce their inventory to a single set of relational concepts. Idiolectal and other variation in adult grammar and language acquisition will be shown to be problematic for the criticized approaches.

2.

The relational concepts and their hierarchies

We start our discussion by introducing the major relational concepts that are relevant in universal grammar as well as in the grammar of many individual languages including German. We do not claim that our list of relational concepts is exhaustive. The addition of new relations would not diminish the simplicity and the restrictive power of the rules to be presented below.

186

Β. Primus & Κ. Lindner

T h e first type of relational concepts w e will discuss are thematic relations (or roles) as introduced by various well-known thematic role theories. It is one of our main assumptions that, in the unmarked case, thematic roles do not f o r m an unordered set but are ranked along a hierarchy. T h e relevant hierarchy for German and very many other languages is given in (l).1 (1)

agent/experiencer > θ recipient > θ benefactive > θ patient/stimulus > θ other roles

Hierarchy (1) was proposed in slightly differing terminology by Jackendoff (1972) and Givon (1984) as a universal hierarchy that explains various p h e n o m e n a across languages. A s demonstrated elsewhere (cf. Primus 1987), agent and experiencer (which also includes the possessor and all first participants of Stative situations) have the same status within their respective role systems (cf. Dowty (1991) for a semantic explanation). There are approaches that try to explain linguistic phenomena on the basis of thematic roles and their hierarchy to the exclusion of other relational concepts. One way of reducing the descriptive inventory to thematic roles and their hierarchy is to define grammatical relations in terms of thematic roles in such a way as to m a p the thematic hierarchy (1) onto the hierarchy of grammatical relations introduced by Keenan & C o m r i e (1977) and Johnson (1977) as in (2): (2)

subject > g r direct object > g r indirect object >gr other oblique arguments or modifiers

T h e definition of grammatical relations in terms of thematic roles will explain, for example, subject properties as agent properties. Such approaches have s o m e motivation. There are rules of grammar that are clearly determined by thematic roles. W e will see that basic order in German and other languages is sensitive to thematic relations. Recent work within Functional G r a m m a r (cf. Dik 1989), Role and Reference G r a m m a r (cf. Foley & Van Valin 1984) and Lexical Functional G r a m m a r (cf. Bresnan & Kanerva 1989) supplies further evidence for the relevance of thematic roles in grammar. T h e attempt at defining (2) exclusively in terms of (1) fails in view of the fact that there is n o one-to-one relation between a particular grammatical relation and a particular thematic relation but only a m o r e general, hierarchy-based constraint of linking grammatical relations to thematic relations (cf. Dik 1978). But even if a definition of grammatical relations in terms of thematic roles were possible for some language, it would not yield the desired result for universal g r a m m a r . T h e reason is that thematic relations explain only some of the properties languages have. Other properties are best described in terms of other relational systems.

'

The value "Θ" for the ranking of thematic relations is an instance of the hierarchy relation ">h" (notational variant "n *>n dddddd des sin des BI:ns/ sin BI:n\ a these are 9 3 a a this/there these are these bees are bee(s) 'These are bees - are bees.'

Paul09 (3;07.11)

Like the other children Paul produces utterances where he replaces -en with -e for masculine nouns, cf. (113) and (114). Here we have the same interpretation difficulty as with Max's Note that the appropriate inflectional ending is missing on the adjective; adults would say die ganzen Punkte. See also the discussion of Max's deviant forms. 64

Note that in spoken German the "full" plural form [BI:nan] is very often reduced to [BI:p] with a syllabic ending, and this is very difficult to distinguish from [BI:n], so the second attempt might be adequate.

257

German Plurals and Compounding

feuerwehrmanne (cf. (92) above): Only one lexical item is affected, but note that the two examples are not taken from the same sample. So the use of this inflectional ending cannot be dismissed as a unique slip. (113)

ich hol mal zwei BÄre\ I get PRT two bears

(114)

zweibäre two bears

(adult: Bär-en)

Paull6 (3;10.30)

Paull8 (3;11.20)

In Paul's data we find evidence once more that both plural markers -(e)η and -s are overgeneralized. This does not seem too problematic for the level model and confirms Clahsen et ai's findings. Examples such as (113) and (114), however, cannot be explained within the level model in its present form. This is especially interesting as Lisa and Max, too, produce a few of these forms, and Wegener (1992, this volume) reports similar forms in her data. We find considerably fewer deviant compounds in Paul's data. Examples (115) to (118) are taken from an early recording; they illustrate that Paul has productive command over the German compounding rules. All these forms would be possible in adult German. (115)

einesitzbank a sit-bench 'a bench'

Paul04 (3;05.15)

(116)

sister is imitating P's "gell":65 Ρ: nein julia is η gellsager\ gettsager\ gelll no julia is a TAG-sayer TAG-sayeA TAG/ 'No, Julia is the one who always says "right", right?'

(117)

nein keinefußgarage no no foot-garage 'no, (this is) no foot-garage'

(118)

nein eine Hochgarage mag ich no a high-garage like I 'No, I'd like to have a multi-story garage'

3x

Deviant compounds offer conflicting evidence. In (119) to (122) Paul leaves out linking elements where adults would use them. (119)

blumkohl flowercabbage 'cauliflower'

(adult: Blume-n-kohl)

3x Paul08 (3;06.27)

gell is a South German tag word. It can be roughly translated as 'right' or 'don't you agree.' This is one of the few Swabian features Paul adopted either from the investigator, who is Swabian, or from kindergarten.

I. Gawlitzek-Maiwald

258

Paul04 (3;05.15)

(120)

Ρ pointing at someone on a skateboard in a book: S: Und da hinten drauf? And at the back? P: da steht ein menschfahrrad\ (adult: Mensch-en-fahrrad) there stands a human-being bicycle 'There is a bicycle for human beings'

(121)

bauerhof farmer-yard 'farm'

(adult: Bauer-n-hof)

Pauli3 (3;09.12)

(122)

jungeauto boy-car 'boy's car'

(adult: Junge-n-auto)

Paul 18 (3; 11.20)

(123)

ah nich auf meine klötzestraße ah not on my blocks-road (adult: Klotz-straße) 'Not on my road build of blocks!'

Paul04 (3;05.15)

(124)

meerenstrand sea-beach 'sea shore'

(adult: Meer-es-strand)

Paul29 (4;05.00)

(125)

einfischennetz a fish-net or fisher-net 'fishing net' or 'fisher's net'

(adult: Fisch-er-netz or: Fisch-netz)

Paul29 (4;05.00)

In all these cases -(e)n is affected, as in Clahsen et al.'s (1990, 1992) data. As this is one of Paul's overgeneralized plural markers, it would be expected according to the level model, but in (123) to (125) Paul inserts -(e)n and -e where adults would not use them.

259

German Plurals and Compounding

5.5

Benny

Table 7: BENNY Utterances

02 292

03 287

04 359

05 414

06 348

07 176

08 402

09 390

10 443

Plural

+

?

0 0 0

7 12 0

1 4 1

0 10 2

4 5 2

1 3 1

0 7 2

0 1 0

5 15 0

Comp.

t ?

0 17 0

1 14 1

4 13 1

0 6 0

6 18 2

0 7 0

0 32 0

1 6 0

0 12 0 66

After the first two recordings (BennyOl and Benny02), in which no plural contexts occurred, the number of adult-like plural forms varies between 46% and 70%. After a few months the proportion of appropriate forms rises to 75% (Benny08) and more. As with Max and Paul, we find some variation in the marking of DPs in Benny's data; within one recording some forms are deviant, cf. (126), some are not, cf. (127). (126)

talking about a glove puppet without teeth: 2x Benny03 (2;06.25) hat kein zähne (adult: hat keine Zähne) has no (sg) teeth

(127)

hat keine zähne

8x

has no (pi) teeth Examples such as (128) and (129) illustrate that UL is not something that Benny gets for nothing. In the first example he uses UL in a singular form where adults would not use it, and in the second one he does not use UL where it is necessary as part of the plural marker. This example is especially interesting as the plural jnarker -er causes UL in all vowels that can undergo the change. One would therefore expect that in the case of -er UL should be easily acquired. (128)

eine nüß daHIN ein nüß ein NÜß dahin a nut there a nut a nut there (adult: eine Nuß)

Benny05 (2;10.29)

(129)

talking about a piece of paper with holes punched in it: I: Was is denn in dem Papier? 'What's in this piece of paper?'

Benny07 (3;00.03)

The following have been classified as questionable: {ho:rxa}Jtain listening stone (adult: stethoscope) placing jigsaw-puzzle pieces: {lisigaga) (?) (adult: ?) polizistamter (adult: Polizeibeamter)

Benny03 Benny04 2xBenny06

1. Gawlitzek-Maiwald

260

Β: äh locher ehm holes

(adult: Löch-er)

Like the other children Benny overgeneralizes -(e)n and -e, but he never uses -s where adults would not do so. Again, we find a few examples where he uses -e instead of -(e)η in a masculine or neuter noun, cf. (130): (130)

talking about a physician, Β pointing to his ears: hier die ohre here the ears

Benny04 (2;09.19) (adult: Ohr-en)

In the majority of cases, however, he replaces -e with -(e)n for masculine and neuter nouns, cf. (131) to (133): (131)

Β & I are playing with cars and smurfs: I: Wer is da? 'Who's there?' Β points at a car Β -.deine FREUNden\ (adult: Freund-e) your friends

(132)

talking about rays of sunshine in a picture: hat haaren (adult: Haar-e) has hairs

(133)

at a paddock: I: Was sieht denn so ähnlich aus wie Esele? 'What is similar to donkeys (diminutive)?' B: p/erden\ (adult: Pferd-e) horses

Benny06 (2;11.21)

4x BennylO (3;01.13)

BennylO (3;01.13)

Later on, in Bennyll (3;03), he does what Max and Paul also do: He marks masculine and neuter plural forms for plural, not only "on top of" -0 plurals but on top of -er plurals too, cf. (134) and (135). (134)

(135)

during a walk in autumn: I: Des grüne, was da liegt. 'These green things lying there' B: blättem\ leaves describing trees on wrapping paper: B: da sin immer so vögeln drauf\ there are always such birds on-it 'They are all full of birds'

Bennyll (3;03.00)

(adult: Blätt-er)

Bennyll (3;03.00) (adult: Vögel-0)

261

German Plurals and Compounding

An example recorded much later shows that the overgeneralization of the so-called idiosyncratic allomorphs as illustrated by ohre in (130) are not a short-time phenomenon: (136)

Benny (6;00.06)

Β showing his (water) police station to K: (adult: Chef-s) B: des sind die zwei chefe\ (ai these are the two bosses K: Was sind des? 'What are they?' B: die zwei che/e\ the two bosses K: Heißt des nicht Chefs? 'Don't we say "chefs"?' B: chefs kama au sagn\ des is doch beides des gleiche\ bosses can-one also say that is after-all both the same 'you can say "chefs" as well\ they are both the sameY

Two aspects are important in connection with this example: The fact that Benny repeats the form twice and then gives an explanation shows a kind of metalinguistic awareness, and that this is not simply a slip of the tongue. Second, the fact that the overgeneralized form fits into the German system of plural formation points in the same direction. Benny's deviant compounds seem to be very consistent: Apart from (137), he omits the linking element which should occur and would be the equivalent of the plural morph -(e)η for the first constituent, cf. (138) to (141): (137)

Wäschenkorb laundr-ies-basket 'clothes-basket'

(adult: Wäsche-korb)

Benny09 (3;01.04)

(138)

sonnedach sun-roof

(adult: Sonne-n-dach)

Benny03 (2;06.25)

(139)

taschetuch pocket-cloth 'handkerchief

(adult: Tasche-n-tuch)

Benny04 (2;09.19)

(140)

krankewagen

(adult: Kranke-n-wagen)

Benny04 (2;09.19)

ill-person(s)-car 'ambulance' (141)

schraubezieher screw-puller 'screwdriver'

&2xBenny06 (2;11.21)

3x Benny06 (2;11.21) (adult: Schraube-n-zieher)

If Benny really treated -(e)n as the default plural marker, forms such as (138) to (141) would be expected under the assumption of a level model (cf. implication (I)). But there are four

I. Gawlitzek-Maiwald

262

reasons for being very careful with this interpretation: First, -(e)η is not the only plural marker that Benny overgeneralizes. Second, we find a deviant example such as (137), where the added -n- cannot function as a plural marker. But what is its function? It links the first and second elements of the compound. Third, many non-deviant forms such as Traube-nzucker ('dextrose') and Kranke-n-wagen ('ambulance') can be found, where Benny uses -(e)η within compounds. Fourth - and this is the most important argument - the forms in (138) to (141) are not deviant at all in the dialect of which Benny is a native speaker. Each adult speaker of Swabian produces these forms. 67 This means that we find almost no deviant compounds in Benny's utterances. So, what has been said about all of the children is particularly clear for Benny: There is a quantitative difference between deviant forms in plural formation and compounding; deviant forms in the latter are very few, and there are no real correlations in the sense of implication (I) between plural inflection and word formation.

6.

Similarities and variation in the language acquisition data

While we could find some inter-individual variation with respect to a)

the consistency of plural marking,

b)

whether or not number marking within DPs is adult-like, and

c)

the preference for a particular plural marker in overgeneralizations,

similarities across the four children could also be found. These can be summarized as follows: a)

they all produce fewer deviant compounds than plurals,

b)

they all overgeneralize the plural marker -(e)n,

c)

in the data of all the children there were some examples of -e where adults would not use it as a plural marker, and

d)

three of the four children overuse -s as a plural marker.

How can these findings be explained within the model of a level-ordered lexicon? The (in-)consistency of number marking probably has to be accounted for on independent possibly syntactic - grounds; it might not fall within the domain of the lexicon. It would be very interesting to find reasons for particular children preferring one or the other plural marker, but to do this we would have to look at other morphological phenomena to perhaps find correlations with the preference for particular markers. At the present stage nothing further can be said about this observation. As all the children produce fewer, if any, deviant compounds than plural forms, the first assumption about a strict correlation between plural marking and compounding (impliΛ 7 Ό/

The same is true for his father's dialect, Hessian. In other words, Benny receives very little input where linking elements in nominal compounds are present.

German Plurals and Compounding

263

cation I) is ruled out. Neither do we find the same type and number of overgeneralizations and deviations within compounds as in plural forms. Nor do we observe a strict nonoccurrence of overgeneralized plural markers within compounds. A possible explanation is that children do not treat the elements occurring between the nominal parts of a compound as plural markers, but as linking elements. In other words, for children (and adults) these are elements with a phonological function, and they are probably inserted by a phonological rule. 6 8 The overgeneralization of -(e)n has been observed before. 6 9 Clahsen et al. (1990, 1992) suggested as an explanation that some children wrongly attach -(e)n to level 3 first and then recategorize it as belonging to level 2. It does not seem to be a small number of children who overgeneralize -(e)n, but a large proportion, perhaps the majority of children. Two questions immediately arise: What leads so many children astray? And why, how, and when do they realize that they have to attach -(e)n at a different level? As we saw in the discussion of adult compounds, it is not likely that these can function as the stimulus for reanalyzing -(e)n and eventually moving this marker to the second level. Evidence from compounding is just not clear and reliable enough as far as the level model is concerned. The overgeneralization of -e cannot be accounted for within the level model as it stands. As a level 1 marker there is no way it could be used in non-idiosyncratic contexts; nevertheless children use it in that way. The overgeneralization of -s is, of course, expected within the level model and does not cause any problems. However, why do not all children treat -5 as a default marker? And for those - like Benny - who do not, the problem arises that they have to reanalyze the marker eventually. One can agree with Clahsen (1992:9) that "in the noun-plurals of German-speaking children the default affix occurs in errors whilst the irregulars do not." I think, however, that what the default affixes of German are must be redefined within the level model. Children overgeneralize several plural markers, i.e. -(e)n, -e, and -s, and they clearly observe the gender distinction in their spontaneous plural forms and in the inflection of nonce items. Therefore the level model will have to include gender-based plural rules for -(e)n and -e and a basically phonological rule for -s.70 In the above sections I have presented evidence from child language which showed no correlation between plural marking and the occurrence of linking elements within nominal compounds. A careful analysis of adult German also highlighted problems with the level model for German as it stands. Thus, I propose to treat all elements occurring between the nominal constituents within compounds, not only -s-, as linking elements and introduce them into compounds through special, probably phonological, rules. 71 The case of German plural and compounding has, again, illustrated how important and fruitful it is to look at a wider data base and how much insight we can gain from child language acquisition data for the evaluation of linguistic theories.

68 69

Cf. Wiese (1992) for the insertion of -s- as a linking element. Cf. among others Clahsen et al. Π990, 1992) and Wegener (1992). This is in addition to rules dealing with the plural inflection of clippings, abbreviations, and names, which also form the plural with -s. The generalizations about linking elements in Ortner et al. A991) could probably serve as a starting point for formulating these rules.

264

I. Gawlitzek-Maiwald

Bibliography Augst, Gerhard (1975): Untersuchungen

zum Morpheminventar

der deutschen

Gegenwartssprache.

-

Tübingen: Narr. Bech, Gunnar (1963): "Zur Morphologie der deutschen Substantive." - In: Lingua 12,177-189. Berko, Jean (1958): "The Child's Learning of English Morphology." - In: Word 14, 150-177. Bittner, Dagmar (1991): "Strukturprinzipien der neuhochdeutschen Substantivflexion: Vorschlag eines Modells." - In: Bochum - Essener Beiträge zur Sprachwissenschaft 9, 207-225. Chomsky, Noam (1970): "Remarks on Nominalization." - In: Roderick A. Jacobsen & Peter S. Rosenbaum (eds.): Readings in English Transformational Grammar (Walham, Mass.: Ginn & Co.) 184-221. Clahsen, Harald (1992): "Overregularization in the Acquisition of Inflectional Morphology." - In: Theorie des Lexikons 16. - University of Düsseldorf. Clahsen, Harald, Monika Rothweiler & Andreas Woest (1990): "Lexikalische Ebenen und morphologische Entwicklung: Eine Untersuchung zum Erwerb des deutschen Pluralsystems im Rahmen der Lexikalischen Morphologie."

-

In:

Monika

Rothweiler

(ed.):

Spracherwerb

und

Grammatik:

Linguistische

Untersuchungen zum Erwerb von Syntax und Morphologie (Opladen: Westdeutscher Verlag) 105-126. Clahsen, Harald, Gary Marcus, Monika Rothweiler & Andreas Woest (1992): "Regular and Irregular Inflection in the Acquisition of German Noun Plurals." - In: Cognition 45, 225-255. Fleischer, Wolfgang (1982): Wortbildung der deutschen Gegenwartssprache. - Tübingen: Niemeyer. Fritzenschaft, Agnes (this volume): "Activiating Passives in Child Grammar." Fritzenschaft, Agnes (in prep.): Erwerb des Passivs im Deutschen. - Ph.D. Diss. University of Tübingen. Fritzenschaft, Agnes, Ira Gawlitzek-Maiwald, Rosemarie Tracy & Susanne Winkler (1990): "Wege zur komplexen Syntax." - In: Zeitschriftfür Sprachwissenschaft 9, 52-134. Gawlitzek-Maiwald, Ira, Rosemarie Tracy & Agnes Fritzenschaft (1992): "Language Acquisition and Competing Linguistic Representations: The Child as Arbiter." - In: Jürgen Μ. Meisel (ed.): The Acquisition of Verb Placement: Functional Categories and V2 Phenomena in Language Acquisition

(Dordrecht:

Kluwer) 139-179. Giegerich, Heinz J. (1985): Metrical Phonology and Phonological

Structure: German and English.

-

Cambridge: Cambridge University Press. Gordon, Peter (1985): "Level-Ordering in Lexical Development." - In: Cognition 21, 73-93. Grimm, Hannelore & Hermann Schöler (1978a): Der Heidelberger Sprachentwicklungstest

H-S-E-T. -

Göttingen: Hogrefe. Grimm, Hannelore & Hermann Schöler (1978b): H-S-E-T Durchführungsanweisung. - Göttingen: Hogrefe. Höhle, Tilman Ν. (1982): "Über Komposition und Derivation: Zur Konstituentenstruktur von Wortbildungsprodukten im Deutschen." - In: Zeitschrift für Sprachwissenschaft

1, 76-112.

Höhle, Tilman N. (1985): "On Composition and Derivation: The Constituent Structure of Secondary Words." In: Jindrich Toman (ed.): Studies in German Grammar (Dordrecht: Foris) 319-376.

265

German Plurals and Compounding

Kiparsky, Paul (1982): "From Cyclic Phonology to Lexical Phonology." - In: Harry van der Hulst & Norval Smith (eds.): The Structures of Phonological Representations. Part I. (Dordrecht: Foris) 131-175. Kiparsky, Paul (1983): "Word Formation and the Lexicon." - In: Francis Ingemann (ed.): 1982 Mid-America Linguistics Conference Papers (Kansas: Lawrence) 3-29. Kopeke, Klaus-Michael (1987): "Die Beherrschung der deutschen Pluralmorphologie durch muttersprachliche Sprecher und L2-Lemer mit englischer Muttersprache." - In: Linguistische Berichte 107, 23-43. Kopeke, Klaus-Michael (1988): "Schemas in German Plural Formation." - In: Lingua 74, 303-335. Körte, Barbara (1986): "Die Pluralbildung als Paradigma linguistischer Theorien." - In: Der Deutschunterricht 38, 15- 36. Mohanan, Karuvannur P. (1982): Lexical Phonology. - Austin, Texas: Linguistics Club. Mugdan, Joachim (1977): Flexionsmorphologie und Psycholinguistik. - Tübingen: Narr. Olsen, Susan (1990a): "Zum Begriff des morphologischen Heads." - In: Deutsche Sprache 18, 126-147. Olsen, Susan (1990b): "Konversion als ein kombinatorischer Wortbildungsprozeß." - In: Linguistische Berichte 127, 185-216. Ortner, Lorelies, Elgin Müller-Bollhagen, Hanspeter Ortner, Hans Wellmann, Maria Pümpel-Mader & Hildegard Gärtner (1991): Deutsche Wortbildung: Typen und Tendenzen in der Gegenwartssprache 4: Substantiv-komposita. - Berlin: de Gruyter. Plank, Frans (1976): "Morphological Aspects of Nominal Compounding in German and Certain Other Languages: What to Acquire in Language Acquisition in Case the Rule Fails?" - In: Gaberell Drachmann (ed.): Akten des 1. Salzburger Kolloquiums über Kindersprache (Tübingen: Narr) 201-219. Plank, Frans (1981): Morphologische (Irregularitäten.

- Tübingen: Narr.

Reis, Marga (1983): "Gegen die Kompositionstheorie der Affigierung." - In: Zeitschriftfur Sprachwissenschaft 2, 110-131. Reis, Marga (1988): "Word Structure and Argument Inheritance: How Much Is Semantics?" - In: Linguistische Studien A 179, 53-67. Toman, JindFich (1987): Wortsyntax: Eine Diskussion ausgewählter Probleme deutscher Wortbildung. Tübingen: Niemeyer. Wegener, Heide (1992): "Pluralregeln und mentale Grammatik." - In: Anatoli Strigin, Ilse Zimmermann (eds.): Fügungspotenzen: Festschrift für Manfred Bierwisch (Berlin: Akademie Verlag) (=Studia grammatica 34) 225-249. Wegener, Heide (this volume): "Variation in the Acquisition of German Plural Morphology." Wellmann, Hans, Nikolaus Reindl & Annemarie Fahrmaier (1974): "Zur morphologischen Regelung der Substantivkomposition im heutigen Deutsch." - In: Zeitschriftfür deutsche Philologie 93, 358-378. Werner, Otmar (1969): "Das deutsche Pluralsystem: Strukturelle Diachronie." - In: Hugo Moser (ed.): Sprache, Gegenwart und Geschichte (Düsseldorf: Schwann) 92-128. Wiese, Richard (1988): Silbische und lexikalische Phonologie: Studien zum Chinesischen und Deutschen. Tübingen: Niemeyer.

266

I.

Gawlitzek-Maiwald

Wiese, Richard (1992): Phonology of German. - Ms. University of Düsseldorf. Wunderlich, Dieter (1986): "Probleme der Wortstruktur." - In: Zeilschrift für Sprachwissenschaft

5, 209-252.

Wurzel, Wolfgang U. (1970): "Flexion und Wortbildung." - In: Wolfgang U. Wurzel (ed.): Studien deutschen Lautstruktur

Wurzel, Wolfgang U. (1983): "Thesen zur morphologischen Natürlichkeit." - In: Zeitschrift für Germanistik 196-208.

zur

(Berlin: Akademie Verlag) 202-213. 4,

Heide Wegener

Variation in the Acquisition of German Plural Morphology by Second Language Learners

1.

The German plural system

Before going into the acquisition paths and strategies of the children in the study,1 I feel it is necessary to give a brief presentation of the formation of nominal plurals in German and to explain their systematicity. This is essential because it is a widely held belief among psycholinguists that there are no systematic rules for the allocation of the plural markers. 2 It comes as no surprise that language acquisition researchers who are convinced of the lack of system in German plurals conclude from the study of plural acquisition that for the most part the plural in German is learned by memorization (cf. for example Park 1978:237; SchanerWolles 1989:173) or that the acquisition of the plural must be explained in a totally different way (cf. Clahsen et al. 1990). It is often concluded that the German plural lacks systematicity: "Not only is the learning of plurals difficult, but also no systematic approach is possible" (Park 1978:247). This seems circular to me. As a result one finds in psycholinguistic presentations the claim that in German eight or even nine plural markers have to be learned (cf. Kopeke 1987:24; Clahsen et al. 1990:107):3

Engel, Balken, Ruder

Brüder, Mütter Kinder, Felder Wälder, Hühner Hunde, Jahre, Tische Hände, Höfe, Bänke Ecken, Nasen, Löwen Banken, Hemden Autos, Wessis, LKWs

angels, beams, oars

brothers, mothers children, fields forests, chickens dogs, years, tables hands, yards, benches corners, noses, lions banks, shirts cars, West-Germans, trucks

1

For details on the children cf. 2.

2

For example Wode (1988:219) simply denies their existence: "Für jedes Wort muß der Plural gesondert gelernt werden." According to Kopeke (1987:40) the system of plural marking appears in German "als eine komplexe Abfolge unmotivierter Regeln und Listen von Ausnahmen." According to Schaner-Wolles (1989:172) the German plural formation is "kaum regelgeleitet," according to Clahsen et al. (1990:110) all plural forms, apart from those with -s, are irregular; according to Park (1978:237) the German plural has a morphological complexity "which cannot be subsumed under any general rule." In contrast see Bittner (1988) and Mugdan as early as (1977:114), who asks the question whether the language alone is responsible for the fact that the descriptions of noun inflection in many grammars are in fact more or less without rules or system.

3

UL = umlaut; UL+er = umlaut+ -er ending.

Η. Wegener

268

This claim is, however, an exaggeration. The German plural system is by no means as confusing as it at first seems. Fortunately for the learner the problem of which plural marker goes with which noun is not solved in such an arbitrary and irregular way. I wish to show in the following that there are indeed rules for the selection of the plural endings in German and I want to demonstrate the basis for these rules. To this end it is necessary to state:

1.1

1.

that the number of plural markers can be drastically reduced by one single phonological rule.

2.

that there are two relevant criteria for the distribution of the five remaining plural markers, that is, the gender classes and the markedness of the nouns.

The phonological rule

Those who claim that in German allocation rules must be learned for eight or nine plural markers have overlooked the fact that six plural allomorphs can be complementarity grouped into pairs by one phonological rule, thus halving their number (as already in Bech 1963:179). This rule combines the plurals -en and -n, -e and -0, as well as UL+€ and UL+0 by dropping the schwa sound of the plural marker if there is already a /a/ present in the final syllable of the noun, independently of whether this /a/ is in final position or is followed by a consonant. As a result of the schwa-drop rule (more detail in Wegener 1992a), the many nouns ending in the pseudosuffixes -e, -el, -er, -en cannot form their plural with one of the syllabic markers -e, -en or -er, but only with the null variants, that is -0, UL+0, or -/ι.4 As soon as the child has discovered this phonological rule, the number of possible errors is greatly reduced, because the number of plural markers is reduced from nine to five:

PL

1) -(e)n

Bank | -en benches Amp -el \ -n traffic lights

2) -(e)

Jahr \ -e Balk -en \ -0

3) UL+(e) Bärt \ -e Gärt -en \ -0

beards gardens

4) UL+er Wäld | -er

forests

5) -s

4

years beams

Oma

\ -s

grandmas

The null-variants are in strict complementary distribution with the schwa variants: *Banbt, *Ampelen *Hund, *Rudere.

Second Language Acquisition: the German Plural

1.2

269

Criteria for the formation of plural rules

It is not possible for me here to discuss in detail the individual arguments by Augst, Mugdan, and Bittner, who generally assume two criteria as relevant for the selection of the plural marker, that is, word ending and gender. The authors differ as to which of these criteria they consider to be of primary importance, the ending (Augst 1975:7 and Mugdan 1977:84) or the gender (Bittner 1988:41). While I agree that gender is critical for the selection of plural markers, I would like to suggest that the markedness of nouns also plays a decisive role. The categorization into gender classes is familiar, but for the allocation of the plural only the difference between the feminine on the one hand and masculine and neuter, that is nonfeminine, on the other hand is relevant. A separation into only two gender classes follows from the fact that masculine and neuter generally take the same plural marker, -(e) according to the main rule, UL+(e), -(e)n, or -er according to the subsidiary rules; cf. Table 1 in 1.5 below for the statistical distribution. The second decisive factor is whether the noun is marked or unmarked, that is, whether it has the normal structure of German nouns or whether it is phonologically or lexically conspicuous. According to Bornschein & Butt (1987:140), we can talk about the normal structure of German nouns when a multi-syllabic noun ends in /d/ or /a/ + consonant, or one of the derivational suffixes. Consequently all nouns are classed as conspicuous or marked which •

end with any vowel other than schwa (+C), for example unstressed full vowels, like Auto, Mobile, Mutti-,



are truncations like Urti (for Universität 'university') and abbreviations like LKW (Lastkraftwagen 'truck'), since phonetically these are not different from nouns ending in full vowels: ['elkave];



are foreign nouns ending in consonants, but only if the final syllable is pronounced in a "foreign" way (cf. Clahsen's unassimilated borrowings).5



We must include here also all nouns where there are special reasons for not changing the phonetic sound in forming the plural - as in cases of onomatopoeia (Kuckucks 'cuckoos'); citation forms, that is, words used metalinguistically (die Warums 'the whys'A· and finally proper names.

All these marked nouns, or nouns with marked endings, take the -s plural and as such behave in a regular way. This claim can be justified as follows: The normal plural marker, -en or -e, can only be attached to stressed vowels at the end of a word: Theorien ['i:dn] 'theories,' Ideen [i'de:3n] 'ideas,' Schuhe [ju:a] 'shoes.'6 The addition of a plural -en to unstressed vowels at the word ending would cause a lengthening which would be unnatural for German speakers. Instead the language opts for a solution, namely the -s plural, which always occurs whenever a native German plural cannot be used for phonetic reasons. This is generally the case for words ending in unstressed vowels: die 5

Compare, for example, Balkons -Balkone [ös] - [o:na], Vibrations - Vibrationen [Jans] - [o.ndnj.

6

This also applies for abbreviations where the final vowel is stressed: LPGen (Landwirtschaftliche Produktionsgenossenschaften 'farming cooperatives in the GDR'), NPen against LPGs, NPs.

270

Η. Wegener

Muttis 'the Moms,' die Autos 'the cars,' die Omas 'the grandmas.' The same rule yields the -s plural for abbreviations: die LKWs 'the trucks,' die PKWs (Personenkraftwagen) 'the cars.' It is after unstressedfiill vowels that the -s plural stands with absolute regularity. Hence it is not attached to Computer or Gangster, which, although they come from English, a language with an -s plural marker, do not have marked endings in German. Nouns ending in a consonant which take the -5 plural, e.g. T-shirts, Jobs, are not "exceptions" (Mugdan 1977:89) but rather "phonologically conspicuous" or marked, as long as they are pronounced in a foreign way. German speakers obviously have reservations about adding one of the syllabic plural endings to a foreign-sounding or lexically marked noun, since this changes the structure of the word much more than the -s plural would. This explains why in cases of interjections {die Achs 'the alases') and onomatopoeia (die Wauwaus 'the bowwows'), as well as proper names, even if they are not foreign sounding, preference is given to the -s plural. Thus we have Bäche und Seen 'streams and lakes,' but Achs und Wehs 'weepings and wailings,1 and not *Äche und Wehen; we have die Brauns 'the Browns,' not die *Bräune analogous to Zäune 'fences;' die Loks, not die *Loken; we have die Länder 'the countries,' but die zwei Deutschlands 'the two Germanys'; Zeppeline 'Zeppelins' and Hermeline 'ermines,' but die beiden Berlins 'the two Berlins.'7 In contrast to the syllabic plural endings, the -s plural causes no shift of the syllable boundary of the singular stem, compare Termin, Termi\ne vs. Berlin, Berlins|. As we can see, there is a great variety of nouns that are classified as "marked nouns": They are all phonologically or lexically conspicuous and therefore select the -s plural. 1.3

The main rules

According to these criteria three main rules can be formulated for the selection of plural markers.

Rl: R2: R3:

1.

Unmarked feminine nouns select the plural marker -(e)n.

2.

Unmarked masculine and neuter nouns select the plural marker -(e).

3.

Marked nouns select the gender-independent, plural marker -s.

[N,+Fem]-> - -(e) [N, m]-> -s

Examples for Rl:

Uhren 'clocks,' Kugeln 'balls, spheres,' Ecken 'corners,' Zeitungen 'newspapers,' Garagen 'garages,' Krankheiten 'illnesses.'

Examples for R2:

Hunde 'dogs,' Jahre 'years,' Balken 'beams,' Ruder 'rudders,' Lehrlinge 'apprentices,' Bäcker 'bakers,' Ergebnisse 'results.'

7

In an utterance by Momper, mayor of Berlin.

Second Language Acquisition: the German Plural

Examples for R3:

1.4

271

Uhus 'eagle-owls,' Autos, Omas 'grandmas,' Jobs, Chips, Bars, Wessis 'West(erner)s,' Kuwaitis, Duos.

The subsidiary rules

To the learners' dismay, not all nouns follow the main rules. And the so-called "exceptions" are very frequent nouns in the core lexicon. In contrast to unique phenomena like Atlanten 'atlases,' Lexika 'lexicons,' Kakteen 'cactuses,' which constitute genuine exceptions and clearly need to be learned by rote, it is not clear whether the marked-regular cases are processed via rote learning or rule formation. I suggest we adopt subsidiary rules for these "regular exceptions." These subsidiary rules come into play when, in contrast to the unmarked regular case (default value), which follows the main rule, the noun is assigned to a class with markedregular plural formation (cf. Wurzel 1990:205, Wegener 1992a) because of a particular feature. The learning task for the child or the foreigner, then, is not a question of storing single entries but of recognizing that it is not the main rule but a subsidiary rule which has to be applied. Three instances must be differentiated here: 1. The largest number of marked-regular plural forms results from the nouns behaving in a way which violates gender rules; that is to say, certain masculine and neuter nouns take a "feminine" plural -(e)n, and certain feminine nouns take the plural -(e), usually reserved for non-feminine nouns. On non-derivational words in the latter category, however, the umlaut always occurs as well. Examples for the first case are Staaten 'states,' Bären 'bears;1 Betten 'beds,' Vettern 'cousins,' Muskeln 'muscles,' and for the second case Hände 'hands,' Bänke 'benches,' Nässe 'nuts,' Mütter 'mothers' and Töchter 'daughters.' If we regard plural formation according to the gender-determined rules 1 and 2 as unmarked, then we can consider plural formation which violates gender rules to be marked. What we have then is a case of reversal of markedness in the morphology: Hund 'dog' can be unmarked in the lexicon as far as plural formation is concerned, since the presence of the feature [-Fem] will yield the plural marker -e; correspondingly, Uhr 'clock,' is unmarked in the lexicon, since the feature [+Fem] will yield the plural marker -en. In contrast, Bär 'bear,' Bett 'bed,' Staat 'state' must be marked for plural formation in the lexicon, so that the featureinterpretation rules will produce the plural marker -en in the presence of the feature [-Fem]. Correspondingly, Hand 'hand' and Tochter 'daughter' must be marked for plural formation in the lexicon, so that UL+(e) is selected in the presence of the feature [+Fem]. The umlaut in the marked plural formation for Hand and Tochter comes from the assumption that all marked plural formations with -e or -er contain front vowels, that is, they follow the umlaut rule: UR:

back V - > front V / mPL -e or -er

This gives us the following marked subsidiary plural rules: 1.

Marked-regular feminine nouns select UL+(e) as plural marker;

2.

Marked-regular masculine and neuter nouns select -(e)n as plural marker.

Η. Wegener

272

S R I : IN, +Fem, m PL] - > VL+fe) SR2: |N, -Fem, m PL] - > -(e)η Examples for SRI: Hände 'hands,' Töchter 'daughters,' Kenntnisse 'knowledge' [no pi. in English]. Examples for SR2: Bären 'bears,' Betten 'beds,' Vettern 'cousins,' Studenten Geologen 'geologists.'

'students,'

2. SRI and the UR account for the umlaut in marked feminine nouns, where it is regular. For 41% of the masculine nouns and the two neuter nouns (Flöße 'rafts' and Klöster 'convents, monasteries')8 which contain a vowel able to take an umlaut and undergo unmarked plural formation with -(e), the umlaut must be recorded in the dictionary entry, which gives us a third subsidiary rule: 3.

Several unmarked masculine and two neuter nouns form the regular plural with an umlaut.

SR2': [N, -Fem, +UL PL] -> VL+(e) Examples for SR21: Wölfe 'wolves,' Flöße 'rafts,' Väter 'fathers,' Klöster convents.'

'monasteries,

3. Finally there remain those masculine and neuter nouns 9 with -er plural whose umlauts result from the UR. The group of nouns which select the -er plural cannot be determined with any precision by ending, by number of syllables or by gender: some end in consonants, some in vowels; some are mono-, some multi-syllabic (Bücher 'books,' Eier 'eggs,' Reichtümer 'riches'). They can be delimited only as [-Fem], The subsidiary rule for them is as follows: 4.

Certain masculine and neuter nouns form the plural with UL+er.

SR2": [N, -Fem, +er PL] - > UL+er Examples for SR2": Männer 'men,' Kinder 'children,' Skier 'skiers,' Irrtümer 'misconceptions.' As the examples show, all plural markers apart from the pure umlaut appear in derived as well as in simple words. That is, the derivational suffixes select the same plural markers as

8

And in the single feminine exception Werkstatt, PI Werkstätten 'workshops.'

9

According to Mugdan (1977:88) this applies to about 70 neuter and 20 masculine nouns.

Second Language Acquisition: the German Plural

273

the simple words in accordance with gender and markedness.10 Suffixes like -ling, -er, -nis, which form masculine or neuter nouns, select the -(e) plural; -tum selects the -er plural in accordance with the subsidiary rule; suffixes like -and, -ent and -e, which form marked, in this case, animate masculine nouns, select the -(e)n plural. Feminine noun-forming suffixes, like -heit, -ung, -age, select the -(e)n plural; and the suffixes which create marked nouns (-/ and -o) select the -s plural. The classification of plural endings according to gender and markedness can be seen below in Figure 1. The heavy black lines on the outside right and left indicate the default values for the individual noun classes.

1.5

The relevance of the plural rules

According to the absolute numbers and the percentages of plural markers (of the nouns) in the basic lexicon (cf. Mugdan 1977:101f.), the following statistical relevance can be determined for the plural rules assumed here. The relevance value derives from the number of possible cases of application of a rule in the basic lexicon (type values, column 3) in relation to the rule-determined criteria of markedness (m) in rule 3 (column 4), gender in rules 1 and 2, and the corresponding subsidiary rules (columns 5-7). At most, therefore, one could assign the umlaut plural to level 1, but not any of the plural suffixes, as Wiese (1988) and Clahsen et al (1990), following Kiparsky, assume. Cf. critically also Wurzel (1990:214 Note 6), and in more detail Wegener (1992b:Chap. 2).

Η. Wegener

274

The table shows the high frequency with which the masculine and neuter nouns form the plural ending with -(e), and the feminine nouns with -(en), and consequently the high relevance of rules 1-2: 33,9% for the -(e) plural according to R2, 35,6% for the -(e)η plural according to R l . The three main rules allow the competent speaker who knows the phonological and lexical features of the nouns to predict the plural form with certainty for 71,5% of all nouns in the basic lexicon! Only 29,5% of the plurals of nouns in the basic vocabulary have to be lexically mastered. Table 1:

Application domain of plural rules

Rule

Plural Marker

% Basic Vocabulary

R1

-(e)n

35.6

R2

-(e)

33.9

R3

-s

2.0

SR1

-UL+(e)

3.1

SR2

-(e)n

SR2' SR2"

% Marked

% Fem.

% Masc.

% Neut.

91.0 51.8

62.5

4.6

9.6

4.8

-UL+e

12.9

32.8

1.0

-er

6.3

1.8

25.0

94.1 8.0

For the child who does not yet know the phonological and lexical features of the noun (markedness and gender), the learning task presents quite a different problem, however. When analyzing language acquisition data, one should therefore pay attention to the following points, among others: Do children realize 1.

that only one plural marker is applicable for a word, or do they regard the different plural markers as free variants?

2.

that there is a schwa rule, or do they form ungrammatical plurals which are impossible in German like *Vogele, *Eimere, *Ampelenl

3.

that Auto, Uhu and T-Shirt are marked nouns and as such require the -s plural, or, put differently, do they realize that the -s plural appears only with certain nouns and that *Traktors is just as impossible as a plural form in German as *Uhue and

* Auto el 4.

that the plural rules are dependent on gender, and can they as a consequence form the main rules for feminine and non-feminine nouns?

5.

that there are subsidiary rules a.

for plural formations which violate gender rules, like Hände 'hands,' Mütter 'mothers,' Menschen 'people,' Löwen 'lions,' and

Second Language Acquisition: the German Plural

b.

2.

275

for the umlaut plural and the -er plural? Or do they have to learn these forms as lexical entries? In other words: Which plural markers are acquired and overgeneralized, which are acquired but not overgeneralized, and which are acquired particularly late or not at all?

The plural and child second language acquisition

Variation can be understood to mean not only the appearance of different courses of acquisition in a variety of children (inter-individual variation) but also the use of several different acquisition strategies by the same child (intra-individual variation). My claim is that both sorts of variation are here determined by the complexity of the learning task. Plural marking in German cannot be accomplished with one single uniform learning strategy, whatever it may be. Instead it requires the parallel application of several competing acquisition strategies that complement one other. Kaltenbacher (1990) differentiates between a cognitive-analytical approach and a more strongly imitative, formulaic, holistic approach. This differentiation can also be applied to the acquisition of morphology, where it means that the forms encountered in the input can be adopted holistically and initially used without being analyzed. Mistakes occur here because of inappropriate use of the holistically stored forms, that is: Plural forms are used in singular contexts and singular forms in plural contexts, e.g. ein Eier 'one eggs,' zwei Baum 'two tree.' In contrast, the cognitive acquisition strategy segments and analyzes the forms in the input, identifies and classifies the morphemes and assigns to them a "grammaticizable notion" (following Slobin 1973), e.g. the semantic concept of "muchness," or number. Errors that occur here are due to wrongly classified morphemes, failure to take restrictions into consideration, and the wrong assignment of allomorphs to nouns, thus leading to an overgeneralization of certain plural markers, which results in the formation of regularized (possible, but in fact nonexisting) plural forms. The use of correct plural forms does not say anything about whether children have taken over the form holistically and stored it as such in their lexicon and now need only to retrieve it, or whether they construct the plural form actively according to a rule. The nature and quantity of the two types of error mentioned above reveal to what degree the two acquisition strategies are used by the children. The formation of plural rules and storage of plural forms will be illustrated with data from an investigation of German second language acquisition in four children with Turkish, three with Polish, and one with Russian as a first language. They were recorded regularly by me from the outset of their German acquisition over a period of two years. At the time of the study the children are between seven and nine years old and differ in one very important way: The Turkish children go to so-called bilingual classes. 1 1 The Aussiedler children (children of readmitted former German nationals from Eastern European countries) go to transition or regular classes, that is to say, they have all their lessons in German. Plural marking was, however, never dealt with in their lessons. Furthermore the out-of-school contact varies quite a lot, since several of the Aussiedler children, but none of

11

In the first and second year they have six hours of classes of or in German, in the third and fourth years sixteen and nineteen hours respectively.

Η. Wegener

276

the Turkish children, go to a German day-care center in the afternoons. However, several of the Turkish children also play with German children. The part played by rule formation in language acquisition is probably not only more clearly observable but in actual fact greater in second language acquisition than in first language acquisition. This is not only because the learners' cognitive capabilities are more highly developed but also because they - this applies to the Turkish children here - often do not have sufficient input at their disposal to acquire the plural forms as lexical entries and are therefore forced to apply their capabilites of construction to a greater extent. The Aussiedler children, whose second language acquisition occurs under almost natural conditions, have relatively fewer regularized plural forms. 1 2 Apparently they not only have sufficient input to acquire plural forms, at first unanalyzed, directly, but also enough to anchor these in longterm memory and then later analyze and classify them as plural forms, assigning the function of plurality marking to the endings only then.

2.1

Intra-individual variation

2.1.1 Application of the acquisition strategies is dependent on the speech situation How frequently the analytical-constructive acquisition strategy is used cannot, as I mentioned above, be determined from the occurrence of correct plural forms - they could have been either regularly formed or taken over holistically and without analysis. The figures in Table 2 show that, as expected, the analytical-constructive strategy is more productive in test situations than in spontaneous conversation in interviews - cf. the statistical distribution of plural forms in the example of a Turkish child (Ne). Table 2:

Occurring plural forms for Ne

Total Interviews

(tokens / types)

Tests (tokens / types)

CP

AP

609 / 298

73.2%/65%

120/111

66.6% / 63.9% 8 . 3 % / 9 %

RP

14.6%/17.1% 12.1%/17.7% 25.5% / 27%

CP = correct plural, AP = absent plural marking (singular instead of plural), RP = regularized plural forms (e.g. Frosche, Fröschen 'frogs'). Table 2 shows that the same child constructs regularized plural forms to a much greater extent in the test situation 1 3 (where she uses twice as many regularized plurals as in the interview situation) than in spontaneous conversation, where she refers back much more frequently to a form stored in memory, even when it is a singular form. The use of the two acquisition strategies also varies during the acquisition process: The formulaic-holistic strategy is used 12

5.4% vs. 9.1% in the whole period covered by the study, cf. Wegener 1992b.

13

In a game with memory cards, plural forms are elicited for real words.

277

Second Language Acquisition: the German Plural

by the child especially at the beginning, which is quite trivially due to the fact that she is initially familiar with neither plural forms nor plural markers, not to mention plural rules. In contrast, in the third test, during the 27th month of contact, only the cognitive-analytical strategy was applied. The implementation of the individual acquisition strategies for plural formation does not always show a continuing increase or decrease. Their frequencies can also differ considerably from one test to another. Figure 7 below shows for the Turkish girl Me neither a continual increase of RP, as was the case with Mu, nor a regular decrease, as was the case with Ne (see below for explanation), but great variation. Only in test 2 does Me apply the rule-formation strategy frequently - in tests 1 and 3 she hardly uses it at all. While the test 1 results can be explained by the fact that the plural rules have not yet been formed at this point, Me's lack of the rule-formation strategy in test 3 (in the 19th contact month) cannot be explained by saying that the plural forms have been mastered and that therefore analytical formation of the plural is no longer needed, as is the case with the Turkish boy Mt. It can rather be explained by the fact that Me is willing to take risks in the second test, actively producing forms, whereas in the third test six months later, she is more form-conscious and is careful to avoid making errors. She is thus less willing to take risks and, as a result, less creative.

2.1.2 The application of acquisition strategies is dependent on the complexity of the learning task Finally the application of the different acquisition strategies varies with the properties of the plural forms and plural markers themselves. It is therefore determined by target-language conditions and as such is also a language-specific phenomenon. For the most part regularized plurals are formed only with -(e), -(e)η and -s, but, with the exception of single instances in the test, practically never with -er and the umlaut. 14 In other words, only the main rules are formed, even though the necessary restrictions are missing, and not the subsidiary rules. Table 3 shows what proportion the individual plural markers have in the RP (types) - at the top for Ne, below that for the group of Turkish children as a whole. 1 5 It can be seen that the Turkish children form over half of the regularized plural forms with the particularly iconic and valid plural marker -(e)n (for discussion see Wegener 1992b), about one third with -(e), and only about an eighth with -5. There is only one instance of -er, and the umlaut is under-rather than overgeneralized.

14

Examples for RP are: Eie, Baume, Frosche, Kugel, Kartoffel, Uhue, TShirte, Elefante; Eien, Haaren, Schuhen, Freunden, Blätten, Zöpfen, Kinden; Traktors, Mädchens; Bäller, Hünde.

15

The difference is due, among other things, to the fact that Ne has a particularly differentiated -s plural phase.

Kleiden,

278

Table 3:

Η. Wegener

Percentage of the individual plural markers in the RP (types):

Ne

Turk. Child.

-e

-0

-(e)n

-s

Interviews

28.3%

7.5%

56.6%

7.5%

Tests

30.0%

33%

30.0%

33.0%

Interviews

27.3%

6.5%

55.8%

10.4%

Tests

26.7%

4.7%

50.0%

17.4%

-er

+ UL

-UL

3.3%

6.6%

3.3%

5.2%

6.5%

3.5%

26.7%

1.2%

Further variation comes about because regularized plurals are not formed at the same pace and the same time; rather, each of them has different peaks. Figure 2 shows at which point which plural markers form the basis for the plural rules and overgeneralizations. It can be clearly seen that first -(e), then -(e)n, and finally -s are overgeneralized. For the first two plural markers there are parallel results in the LI acquisition data, cf. Schaner-Wolles (1989:166), MacWhinney (1978:58), Park (1978:242). The data show that for the most part only the two main rules for unmarked nouns are constructed in language acquisition. Thus, not only LI data but also L2-acquisition data refute the claim in Clahsen et al. (1990) that the -s plural represents the default value in German. Figure 2:

Occurrence of the various plural markers in RP for Ne 100 ι 90 BO 70

'////////ft.

60 50 40

vNNNN^C

30

20 10 r - · ? X-9

T1

10-12

T2

I

16-18 19-21

T2

22-27

T3

Time

The question is why it is these plural markers and not others which form the basis for the plural rules and which are then also overgeneralized. In contrast to Slobin's study of the acquisition of Russian case morphology, in which it is the not particularly frequent but particularly valid instrumental ending which is overgeneralized, neither validity nor salience of the -e plural can be called on to explain its early overgeneralization, since both show minimal values (cf. Kopeke 1988:315). The determining factor seems to be its frequency in

279

Second Language Acquisition: the German Plural

the input: The formation of the plural rules in the sequence -(e) before -(e)η and finally -s 1 6 follows - with the exception only of -er - almost exactly the distribution of correct plural forms in Ne's data, cf. Table 4. Table 4:

Occurrence of plural markers on the correct plural forms in Ne (Tokens/Types):

UL-(e)

-0

-(e) η

UL+er

-s

With UL

1Θ1 / as

65/33

137/82

111 / 41

26/14

80/49

The -e plural has the highest frequency not only in the basic lexicon (cf. Table 1) but also in the child's lexicon, where, together with the 0 plural, it constitutes almost half of all plural forms. The reason why the -(e)n plural is the most frequently overgeneralized plural (though not the earliest) may be that it has a higher validity and a low type-token index. The higher the type frequency of a marker, the greater the probability that it will become productive and be overgeneralized (Bybee 1985:133): "Productivity of morphological rules must be connected to high type frequency, or as MacWhinney 1978 phrases it, high applicability." Children then overgeneralize first the plural marker that is most frequent in the input and later, after they compare the plural allomorphs with one another, the most valid plural marker. Now a choice is made among plural markers according to their iconicity and validity, a phenomenon known in language acquisition research as "inflectional imperialism" (Slobin 1973:203). This also provides answers to some of the questions posed in section 1.5: The children do not take gender restrictions into account, which is trivially due to the fact that at the time of acquisition of plural marking they have not yet mastered gender. As a result they use "feminine" markers for masculine and neuter nouns too (Tagen 'days,' Freunden 'friends,' Jahren 'years,' Haaren 'hair'), and also non-feminine for feminine nouns (Türe 'doors,1 Kugel 'balls, spheres'). They use regular plural markers also for marked nouns (T-Shirte, Uhue), the -s plural, if they use it productively at all (which does not happen with the Aussiedler children), also for unmarked nouns (Mädchens, Traktors), which can also be easily explained: Children cannot recognize the markedness of certain nouns, since lexical markedness is only recognizable against the background of the whole lexicon. Since it is precisely the "normal" structure of the German noun that language learners are not yet familiar with, they do not know that Uhu and T-Shirt are phonologically conspicuous. Because of the small number of -s plural forms, they have to rely on lexical learning until they have classified the nouns in question as marked, which happens first for those ending in a full vowel. As a test with nonce words shows, however (cf. Wegener 1992b), even the nouns recognizable as being marked because of their endings, e.g. Kafti, Ziro, were pluralized with -s by only 18.7% of the Aussiedler children and 47% of the Turkish children; 17 33% and 32%, respectively, pluralized them with -(e)n. Also, German children of the same age, i.e. 9 years old, use the -5 plural for only 40% of these nouns and the more valid -n plural for 40%! The -s plural is only overgeneralized by several children. This sort of plural formation can not be observed in the Aussiedler children, and it is also more rare in first language acquisition, cf. Schaner-Wolles (1989:166), MacWhinney (1978:58), Park (1978:242). Here I attribute the Turkish children's better results to the longer time they have lived in Germany.

280

Η. Wegener

The umlaut and the -er ending are special cases. In contrast to -e, -en, and -s, they are almost never overgeneralized; and in contrast to the -s plural, the -er plural is especially easy to acquire, particularly for the Aussiedler children, and it is also always formed with an umlaut. Obviously -er plurals tend to be holistically acquired more than all the others. The claim in Korte (1986:26) that children seem to recognize very quickly that the umlaut plural and the -er plural are exceptions to be learned on an individual basis credits children with a knowledge that they cannot possibly possess nor acquire, and it does not explain how children gain this "insight." How should children, without any negative evidence, recognize that certain plural forms, and moreover those that they come across often and use without difficulty, are irregular? The fact that no productive rules are formed in any phase for the plural markers -er and UL, apart from marginal cases,18 can be explained precisely by their low type frequency and their high token frequency. The type-token ratio (see Table 4) is approximately 2:1 for the plural forms with -e, -0, -s, and the umlaut, lower for -(e)n, but higher, almost 3:1, for the -er plural. It shows that this plural marker is used very frequently in the child's lexicon and as such is greatly overrepresented. The same result is also found in Augst's LI data (1979:230). While high token frequency makes formulaic acquisition easier, a relatively low type frequency hinders the use of an operating principle that Peters (1985:1033) calls "Compare," i.e. comparison with other forms in the input. This comparison is, however, a necessary precondition for the segmentation and identification of endings. As a consequence, children are for a long time not aware that the -er ending is a plural marker and they learn these forms by entering them as units in the lexicon. A further explanation for the low productivity of the -er plural lies in the low validity of the -er ending as a plural marker. More than any other ending, -er has additional functions to fulfill besides number marking, namely the signalling of several cases (nominative, genitive, dative), of gender (masculine), the comparative, the agent in verb derivations, and, finally, it also occurs as a pseudosuffix: Functions of -er·.

-er

NomMSg Gen PL GenFSg Dat F Sg Comparative AGENT Plural Pseudosuffix

ein kleiner Mann Kind reicher Eltern trotz großer Mühe mit bester Laune größer als du Sprecher, Hörer Kinder Fenster

a little man child of rich parents despite great effort in a good mood bigger than you speaker, hearer children window

It is therefore harder for children to recognize and classify -er as a plural marker than the other endings.

The eight children in the study form only one -er plural, namely Bäller (two instances), and this was only in the test; overgeneralization of the umlaut was also rare: Hunde, Traktor.

281

Second Language Acquisition: the German Plural

The undergeneralization of the umlaut can be attributed to its poor salience and low validity: The umlaut is not a segmentable element; it does not appear at the end of a word; and it occurs not only as a plural marker, but also in diminutives and derived forms, compare Hündchen, Hündin (doggie, bitch) and in simple words: Löwe (lion). The plural endings thus differ in regard to validity as well as in token frequency and/or type-token index. As a result, the rankings for easiest/most difficult to learn and most/least overgeneralized plural endings are not identical, but differ clearly. If we examine the figures on which plural forms were correctly realized most frequently during the period of the study, we find the following acquisition sequences: Acquisition sequence of the plural markers: TC:

-0

>

-er

>

-e

>

-n

>

-en

>

-s

AC:

-0

>

-er

>

-e

>

-en

>

-s

>

-n

If we examine the frequency of overgeneralization of the plural endings, the two groups of children exhibit the following rankings (for the token and type figures these rankings are based on Wegener 1992b, Chap. 2.3): Frequency rankings of the overgeneralizations: TC:

-(e) η >

-s

>

-e

>

UL

>

-er

AC:

-(e)n >

-e

>

-s

>

UL

>

-er

Apart from the -0 plural, whose easy acquisition is trivial as there is no marker that needs to be learned, we see that both groups of children learn the -er plural most easily because of its high token frequency, but overgeneralize the -(e)n plural most because of its high validity and iconicity. The polyfunctionality of the pseudosuffixes is also responsible for the children's formulating and later rejecting certain false hypotheses. During the phase in which they form the iconic -(e)η plural forms with particular frequency, "double plurals"19 like Männern are also produced. Here children must recognize that the -er is already a plural marker and they must rehabilitate it accordingly, that is, they must break down the double plural forms and go back a step. Consistency in the learner's behavior would in this case only be detrimental. Conversely, children must recognize that the ending -el is not a plural marker and that consequently feminine nouns ending in -el require additional plural marking. They must recognize that their null hypothesis for these nouns, which for months made it possible for them to say "zwei Kugel, " is wrong. The fact that learners apparently stick to a null hypothesis for a long time here shows not only that they have difficulty in differentiating between pseudosuffixes and plural markers, " Forms such as *Männere, *Männeren, *Männerer, which would mean that children were violating the schwa deletion rule, never appear, neither do forms such as *Vogele, apart from in the test. This indicates that the children know the schwa deletion rule before they begin to construct productive plural forms. From the point of view of the child's grammar these are, of course, not double markings for the function of plural! Also the so-called double plural forms are, in contradiction to Park (1978), possible plural forms in German, cf. Bauern.

282

Η. Wegener

which is understandable with -er, -e, and -en, since these appear both as plural markers (Eier, Hunde, Banken) and as pseudosuffixes (Eimer, Bande, Balken). The null overgeneralizations also confirm the tendency to avoid identical double marking commented on in recent work on morphology and psycholinguistics (cf. Kopeke 1987:34 for references). The children in this study not only avoid identical double marking - forms like *Sponkenen appear only in the nonce word test and make up less than 1% of the data - but they generally shy away from attaching a plural suffix to a pseudosuffix, even to -el, which several of the children obviously classify as a plural marker as well. 20 2.1.3 The significance of the two acquisition strategies for successful learning The phenomena leading to intra-individual variation should not be seen as different phases of acquisition, since they tend to exist side by side rather than emerge in sequence. They can therefore be much more appropriately characterized as reflexes of two different acquisition strategies. The holistic-formulaic acquisition strategy takes over pluralized forms from the input, uses them in an appropriate, or at the beginning not so appropriate way, and finally stores them as lexical entries. This acquisition strategy lasts for the whole of the acquisition process in the case of the umlaut and the -er plural forms. It requires a subsequent analysis of word forms which are at first used monomorphemically. The significance of this acquisition strategy lies, among other things, in the fact that it can explain phenomena in language acquisition whose interpretation is controversial in the literature, as e.g. the claims that: 1.

Children never violate the schwa deletion rule.

2.

Children never attach invented endings to nouns for plural marking; even the form zwei Knirfel (for ein Knirf) represents an analogy to zwei Ärmel on the basis of false segmentation of the pseudosuffix.

3.

Children never attach the plural endings to wrong word classes.

The explanation for this is that children store sequences of morphemes in the form of whole words or segments of speech. Then they unequivocally construct the rules, as the generalizations show, and their speech productions on the basis of such stored forms, which serve as a model, cf. also Bybee (1985:114). The cognitive-analytical acquisition strategy identifies and classifies endings as plural markers, develops rules for the formation of plurals and uses them likewise in an appropriate or inappropriate way, that is to say, it leads to overgeneralizations and redundant forms. As soon as the restrictions for these rules have been recognized, a great number of new nouns can be pluralized with their help. The simultaneous application of both acquisition strategies is shown by the fact that during the course of acquisition the children exhibit great variation in their choice of plural forms for In the nonce-word tests several children form plural forms like zwei Scheitel, zwei Knirfel, for the singular Schett, Knirf, although these are the only cases of invented plural forms which appear in my data. This phenomenon also appeared in the German control children (nine years old).

283

Second Language Acquisition: the German Plural

a particular lexeme. For example, in the same interview, the plural forms Kugel,

Kugeln,

Kugen, Kugels were used by N e ( T ) or Ka(P), 2 1 although the children were not corrected, and likewise Frösche, Frosche, Fröschen.

The child thus violates the unique-entry principle and

also goes against the claim in Pinker (1984:177f.), and following him Clahsen (1988:91), that this principle is not a linguistic but rather a psycholinguistic principle, which prevents double occupancy of the cells of the paradigm (Clahsen 1988:91).

Clahsen, like Pinker starts out

from the impossibility of irregular and regular forms existing side by side, since he assumes, as Pinker does for English, that there is only one regular class of endings. The claim does not hold true for the acquisition of German morphology for at least two reasons:

First, unlike

English, German has several regular and marked-regular plural classes at its disposal. same applies to verb classes, cf. Wurzel (1990:205).

The

It is therefore not surprising that

children overgeneralize several plural markers at the same time, forming and trying out several plural rules side by side.

Secondly, the children vacillate considerably between

holistically stored marked-regular forms, which they have, like irregular forms, adopted on the basis of positive evidence from the input, and forms developed and regularized later. According to my data they do not make a distinction between those forms actually found in the input and their own made-up forms; the two sorts of plural formation are for a long time equivalent for them. Some children stick resolutely to RP forms (Elefante, Kugel), they receive counterevidence from the input.

although

Such examples are familiar from

LI

acquisition, cf. Ramge (1975:71), Wode (1988:242f.), Bowerman (1987:451), GawlitzekMaiwald (this volume). Even one of the German control children (nine years old!) used the plural forms Lutschers/Lutscher, T-ShirtslT-Shirte

as free variants.

These violations of the rule can be explained in either of two ways: 1.

2.

The children vacillate because of their uncertainty about which of the five plural markers to select for a particular lexeme. It is assumed that the children do in fact "know" that only one plural marker is appropriate, but they are not certain which one it is. It is precisely this that the children do not as yet "know":

They have not yet

realized that in German (in natural language in general?) there is a unique-entry principle 22 for grammatical morphemes.

They assume that the various plural

markers are free variants - just as they can sometimes choose not only between two different words to express a semantic concept but also between different structures. In both cases the children need positive evidence from the input to discover, a)

that free variation in plural marking is not tolerated

b)

which of the plural markers is required for each concrete case.

21 τ _ Turkish child, Ρ = comes from Poland. 22

The unique entry principle is to be viewed as strictly morphological, otherwise the appearance of, for example, the semantic relation "possessor of X," which can in German be realized in various ways, would contradict it: Das Auto meines Vaters, Das Auto von meinem Vater, Meinem Vater sein Auto ('my father's car').

Η. Wegener

284

A unique-entry principle as a psycholinguistic principle and acquisition mechanism would without a doubt be desirable and would prevent such vacillations, but this assumption does not stand up to the data. The conditions are not like that. The same applies to the assumption that children mark regularized forms as provisional (with "?"; cf. Clahsen 1988:91, Pinker 1984), and not those forms for which they have positive evidence in the input. According to my data, not only are the regularized forms substituted by marked-regular forms, but also the other way round. So, also this certainly attractive idea does not stand up to the data, neither in L2 nor in LI acquisition.23 Pinker and Clahsen fail to recognize that in the acquisition of the plural several acquisition strategies compete, complement each other, but at times also cancel each other out. The acquisition of a relatively complicated system, like that of German plural marking, is evidently more complex and thus also needs a more complex theory of acquisition. 2.2

Inter-individual variation

The biggest difference between the individual children and the two groups of children can be seen in the success with which they acquire plural marking. Within two years the Aussiedler children attained over 90% correct plural forms (=CP) in the tests and correspondingly low marks for AP (absent plural marking, i.e. singular forms) and RP (regularized plural forms). For the Turkish children, however, development stagnates in their third year of learning - we find fossilization and backsliding of a sort not seen in any of the Aussiedler children. Already mastered marked-regular plurals are regularized again (instead of Bücher Buche is formed, instead of Häuser Hausen appears again, etc.). Figures 3 and 4, containing the results of three tests (with real words) for each of four children, clearly show the continual increase in learning in a continually rising line in the Aussiedler children and the stagnation with a levelling off of the line for the Turkish children. The explanation for the interruption in learning growth can be found in the different learning conditions of the two groups of children and is basically trivial: A linguistic element which relies substantially on evidence from the input cannot be adequately and successfully acquired with only little input and reduced contact with the target language. The Turkish children do in actual fact form the main rules for German plural marking too, but do not have sufficient opportunity to check their own forms against the input, and compare and revise them accordingly. Nor do they have sufficient access to the L2 to anchor the marked-regular forms in their memory.

23

Cf. Park (1978:244): "The children had in their lexicon two or three types of plural for the same singular noun: e.g. Kinder, Kindern 'children,' Stühle, Stühl, Stühlen 'chairs.'"

Second Language Acquisition: the German Plural

Figure 3:

Development of the plural forms in the Aussiedler children (AC) (spontaneous data)

Time

Figure 4:

Development of the plural forms in the Turkish children (spontaneous data) ioo

TC01-6 TC07-12

TC13-18 TC19-24 TC25-32 Time

285

286

2.2.1

Η. Wegener

Observing the gender restrictions

A less trivial reason for the differences in learning success lies in the fact that the Aussiedler children are more successful than the Turkish children in discovering gender restrictions and forming plural rules which are gender-appropriate. T o answer the question of whether the children's plural rules were gender-restricted, a test with nonce words was used 24 (similar to Kopeke 1987, but with additional nonce words, see Appendix) in which the gender of the items which had the same endings or the same pseudosuffixes, but different suffixes, was given.

W e recorded whether or not a gender-

appropriate plural marker was chosen - that is, -(e)n

for feminine, -(e) and/or -er for non-

feminine. Table

Plural formation of nonce words by Turkish children and Aussiedler

children in

the third grade of primary school

Turkish C h i l d r e n Ending & G e n d e r Features

-(e)n

-(e),

-er

Aussiedler -(e)n

Children -(e),

-er

C, + F

64.7

27.4

44.4

44.4

C, - F

54.4

39.7

45.8

52.0

Pseudosuffix + F

58.8

20.5

62.5

20.8

Pseudosuffix -F

60.0

20.5

47.9

39.5

schwa

92.0

1.9

94.4

2.7

84.3

1.9

83.3

11.1

82.3

10.2

81.0

4.1

56.8

29.4

6.2

62.0

+F

schwa -F -heit etc.

+F

-ling e t c . - F

( K e y : C = Consonant, F = Feminine) The Turkish children show for all items independent of gender a clear preference for the more iconic -(e)η plural, which they also choose for non-feminine nouns. Consequently they attain low marks for the plural markers -(e) or -er required by the non-feminine plural rule. Even in suffixed, and therefore indirectly gender-marked nouns (e.g. ending in -heit or -ling),

the

17 Turkish children and 12 Aussiedler children took part in the test. At the time of the test the Aussiedler children had been in Germany for only 1-2 years, the Turkish children for 2-3 years, if not already from birth. Each noun class was represented by 2-4 items. The missing percentage values are for -s plural formations and "no answer" reactions. Column 1 shows the ending and gender features for the items. Columns 2 and 3 show the percentage values for the items from the Turkish children, columns 4 and 5 the plural markers chosen by the Aussiedler children. The boldface type indicates where the plural markers were formed as expected according to the rules.

Second Language Acquisition: the German Plural

287

Turkish children tend to follow an iconic strategy rather than to pick a plural marker based on gender. The Aussiedler children, however, choose rather the "masculine" plural markers for nonce words ending in a consonant and differentiate between nouns ending in the pseudosuffixes -el, -er, or -e more clearly by gender than the Turkish children. Both groups also apply gender differentiation to nouns ending in /a/, where it is not needed, but here again the Aussiedler children do so more clearly than the Turkish children. This hypercorrect gender differentiation is even more pronounced in German children and students (cf. Wegener 1992b). The explanation for the clearly better observance of the gender criterion by the Aussiedler children lies, on the one hand, again quite simply in the fact that they have a better knowledge of the language, especially in an area which relies to an even greater extent than the acquisition of plural marking on rote learning and thus on input, namely the acquisition of gender. On the other hand their earlier compliance with the gender criterion may just be due to the Aussiedler children's greater sensitivity for the category of gender, passed on from their LI, that is, they are familiar with this phenomenon from their first language, Polish or Russian. The fact that the Aussiedler children prefer the "masculine" gender marker for nonce words ending in consonants could be interference: In Polish and Russian, nouns ending in hard consonants are generally masculine, see Corbett (1991:36,83). 2.2.2 Building up a system of rules The Aussiedler children also differ from the Turkish children in that they manage at an earlier stage to recognize, or rather assume phonological restrictions for the individual plural markers. When the children form regular plurals, the -e and - 0 overgeneralizations are at first predominant, as we have shown - the latter particularly for nouns with pseudosuffixes, which children evidently interpret as plural markers. In a second sub-phase these null forms are replaced by more iconic and also redundant markings ("double plural forms"). The children's need here for overt marking of a grammatical concept can be seen quite clearly. The children's plural rules are at this stage in no way restricted and apply equally 1. for all genders, 2. for nouns denoting animate and inanimate objects (*Tischen, *Freunden), 3. for all sorts of word ending (*T-Shirte, *Uhue). They always observe the schwa deletion rule, however. Violations of this phonological rule do not appear at all in the spontaneous data and lie under 1% even in the nonce-word test. From this one can deduce that creative plural formation begins only after the children have internalized the phonological rules of German. For the Aussiedler children, a third phase reveals the first signs of the formation of a plural system which contains several rules. The rules of this interim grammar are not based on gender but on word ending, and according to the observable overgeneralizations they are as follows:

288

Η. Wegener

1.

Nouns ending in /θ/ choose the plural marker -n (Jungen 'boys').26

2.

Nouns ending in a pseudosuffix choose the plural markers - 0 or -n (Kugel (n) 'balls, spheres').

3.

Nouns ending in a consonant choose the plural marker -e (Türe 'doors,' Elefante 'elephants').

The effectiveness of these rules can be seen from the pluralization of nonce words in a test; cf. the results in Table 5. Only about 45% of items which end in consonants were pluralized with -en by the Aussiedler children, even if they were feminine (in comparison with 65% and 55% for the Turkish children). In contrast, however, over 80% of items ending in schwa were pluralized with -en in both gender classes. Masculine and neuter items ending in a consonant were pluralized with the expected -(e), -er plural by 52% of the Aussiedler children, but by only about 40% of the Turkish children. And the corresponding items ending in a pseudosuffix were formed with the expected - 0 plural by almost 40% of the Aussiedler children but only by 20% of the Turkish children, since the latter show an even greater preference here for the more iconic plural marker -n. Also, the fact that the Aussiedler children form plural rules based on word ending more than the Turkish children may be due to a greater sensitivity for phonological rules, influenced by their LI, since the division into inflectional classes, and thus plural formation as well, is extensively determined by noun endings in the Slavic languages. There is no evidence in my data for further instances of interference, e.g. the direct transfer of plural markers from the LI into German, the preference for plural markers which sound similar to those in the LI (if this were the case, the Turkish children should prefer the er plural and the Aussiedler children the -e plural), or omission of the plural markers after quantifiers, which is usually the case in Turkish. The LI does not influence the children's language acquisition through direct transfer and consequently does not appear as interference, but, at most, in superordinate predispositions or in a general sensitivity for particular language phenomena and categories. 2.2.3 Preference for a particular strategy of acquisition Inter-individual variation is further demonstrated in the different willingness of the children to use either the cognitive-analytical or the formulaic-holistic acquisition strategy. Variation occurs group internally too, and is especially pronounced among the Turkish children, who must rely more on rule formation. Figures 5 and 6 show clear differences in plural formation for the Turkish children Ne and Mu in the test.27 The proportion of CP continually increases in Ne, but in Mu, a boy with little contact, who uses limited analytical strategies, it falls again so that the AP (singular forms), which were completely avoided by Ne in the third test, actually increased for Mu.

26

This rule captures some of the cases in subsidiary rule 2, but not the consonant-final masculine nouns or the neuter nouns which take the -en plural, as Bären, Betten.

27

The tests for real words were done with memory cards which show objects and beings typically associated with children to ensure familiarity.

289

Second Language Acquisition: the German Plural

Figure 5:

Development of plural forms in Ne TOO

Ne09

Ne21

Ne27

Time

Figure 6:

Development of plural forms in Mu

RP

Mu09

Mul9

Mu26

Time

Most interesting is the comparison of RPs, which show the application of analytical strategies of acquisition (cf. Figure 7). In Ne the RP attain relatively high values in all three tests. Ne is a child who will 'regularize away' if she does not know the plural form. In the second test this strategy causes Ne to come off worse than Mt, although in the spontaneous data her performance is superior to his. In contrast, Me and Mu are cautious (Me varies his strategy of plural formation from test to test - see above); they prefer to use the singular forms which

Η. Wegener

290

they are already familiar with rather than take the risk of trying to form an unknown plural. Only in the third test does Mu use RP forms to about the same extent as Ne in the first test, i.e. 17 months previously! Mt's proportion of RP is low, since he, a child with especially extensive play contact with German children, can reproduce the plural forms from memory (see the high scores for CP in Figure 8) and, as a consequence, is not at all dependent on use of the analytical strategy. Figure 7:

Proportion of RP in the plural forms of the 4 Turkish children (test data)

Me

Mt

Mu

Time

Ne

Second Language Acquisition: the German Plural

291

Mt is, thus, more like the Aussiedler children, who construct RP only at the beginning, but who have the CP - even the marked-regular forms - stored in their mental lexicon and consequently at their disposal much earlier than the Turkish children. Which of the two basic strategies of acquisition is given priority is therefore also directly dependent on the conditions of acquisition. Thus, the course of acquisition in the Turkish children shows that for the acquisition of a language as morphologically complex as German, a basic minimum of input and contact with L2 speakers must be present. This is not the case in the schools operating under the 'Bavarian model,' or is the case only if the children have extensive contact with German peers outside of school. 2.3

Concluding remarks

The intra-individual and inter-individual variation observed in German plural acquisition makes it clear that empirical language acquisition studies must take into account both a holistic and an analytical acquisition strategy for complex language phenomena. In fact, they must consider the possibility that children use several strategies side by side for language processing and that they differ in the strategies they give priority to, these being determined in part also by the linguistic elements to be learned. The high number of vacillations among different plural forms, and the fact that especially the Turkish children give up marked-regular plural forms, which they had already used fluently, in favor of regularized plural forms28 also shows that holistically stored sequences cannot be regarded as acquired as long as the individual morphemes have not been assigned the right functions, that is, as long as they have not been cognitively analyzed. Consequently the cognitive-analytical approach is the decisive strategy in language processing and language acquisition, and the holistic-formulaic strategy is only a prerequisite, albeit an absolutely essential prerequisite (cf. also Kaltenbacher 1990:4).

28

E.g. Bücher is replaced again by Buche, Eier by Eien, etc.

Η. Wegener

292

Appendix Nonce words for the plural test:

1.

-C: die Schett, die Pucht, die Liihr das Grett, das Kett der Troch, der Trunt, der Knirf

2.

-e: die Tilte, die Muhre, die Limpe das Priere der Knumpe, der Olke -el·. dieToftel das Trilchel der Knafel -er: dieBachter das Sierer der Rnauker -en: der Sponken

3)

full V: die Kafti die Kaftu das Ziro der Traika

4)

-ung etc.: die Schergung die Trauschaft die Borchheit die Tonität

5)

-ling etc.: der Knauling das Verknöpfnis das Quettchen das Schwirklein

Bibliography Augst, Gerhard (1975): Untersuchungen Tübingen: Narr.

zum Morpheminventar

der deutschen

Gegenwartssprache.

-

Augst, Gerhard (1979): "Neuere Forschungen zur Substantivflexion." - In: Zeitschrift flir Germanistische Linguistik 7, 220-232.

293

Second Language Acquisition: the German Plural

Bech, Gunnar (1963): "Zur Morphologie der deutschen Substantive." - In: Lingua 12, 177-189. Berko, Jean (1958): "The Child's Learning of English Morphology." - In: Word 14, 150-177. Bierwisch, Manfred (1975): "Syntaktische Merkmale in der Morphologie: Generelle Probleme der sogenannten pronominalen Flexion im Deutschen." - In: Ferenc Kiefer (ed.): Morphologie und generative

Grammatik

(Frankfurt: Athenäum) 1-55. Bittner, Dagmar (1988): "Motivationsstrukturen im Flexionsverhalten der neuhochdeutschen Substantive Vorschlag eines Modells." - In: Linguistische Studien A 188, 36-52. Bornschein, Matthias & Matthias Butt (1987): "Zum Status des -s Plurals im gegenwärtigen Deutsch." - In: Werner Abraham & Rita Arhammar (eds.): Linguistik in Deutschland (Tübingen: Niemeyer) 135-153. Bowerman, Melissa (1987): "Commentary. Mechanisms of Language Acquisition" - In: Brian MacWhinney (ed.): Mechanisms of Language Acquisition (Hillsdale, N.J.: Erlbaum) 443-466. Bybee, Joan L. (1985): Morphology: A Study of the Relation Between Meaning and Form. - Amsterdam: Benjamins. Clahsen, Harald (1988): Normale und gestörte Kindersprache: Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie. - Amsterdam: Benjamins. Clahsen, Harald, Monika Rothweiler & Andreas Woest (1990): "Lexikalische Ebenen und morphologische Entwicklung: Eine Untersuchung zum Erwerb des deutschen Pluralsystems im Rahmen der Lexikalischen Morphologie."

-

In:

Monika

Rothweiler

(ed.):

Spracherwerb

und

Grammatik·

Linguistische

Untersuchungen zum Erwerb von Syntax und Morphologie (»Linguistische Berichte, Special issue 3) 105126.

Corbett, Greville G. (1991): Gender. - New York: Cambridge University Press. Kaltenbacher, Erika (1990): Strategien beim frühkindlichen Syntaxenverb. - Tübingen: Narr. Kopeke, Klaus-Michael (1987): "Die Beherrschung der deutschen Pluralmorphologie durch muttersprachliche Sprecher und L2-Lerner mit englischer Muttersprache." - In: Linguistische Berichte 107, 23-43. Kopeke, Klaus-Michael (1988): "Schemas in German Plural Formation." - In: Lingua 74, 303-335. Körte, Barbara (1986): "Die Pluralbildung als Paradigma linguistischer Theorien." - In: Der Deutschunterricht 38, 15-30. Mater, Erich (1965): Rückläufiges

Wörterbuch der deutschen

Gegenwartssprache.

- Leipzig: VEB

Bibliographisches Institut. MacWhinney, Brian (1978): The Acquisition of Morphophonology. Monographs of the Society for Research in Child Development 43:1. Mills, Anne E. (1985): "The Acquisition of German." - In: Dan I. Slobin (ed.), 141-254. Mills, Anne E. (1986): The Acquisition of Gender. - Berlin: Springer. Mugdan, Joachim (1977): Flexionsmorphologie und Psycholinguistik. - Tübingen: Narr. Park, Tschang-Zin (1978): "Plurals in Child Speech." - In. Journal of Child Language 5, 237:150.

294

Η. Wegener

Peters, Ann Μ. (1985): "Language Segmentation: Operating Principles for the Perception and Analysis of Language." - In: Dan I. Slobin (ed.): The Crosslinguistic Study of Language Acquisition, Vol. 1 (Hillsdale, NJ.: Erlbaum) 1029-1068. Phillips, Betty S. & Lowell Bouma (1980): "The Acquisition of German Plurals in Native Children and NonNative Adults." - In: International Review of Applied Linguistics 18,21-29. Pinker, Steven (1984): Language Learnability and Language Development. - Cambridge/London: Harvard UP. Plank, Frans (1977): "Markiertheitsumkehrung in der Syntax." - In: Papiere zur Linguistik 17/18, 6-66. Ramge, Hans (1975): Spracherwerb: Grundzüge der Sprachentwicklung des Kindes. - Tübingen: Niemeyer. Rothweiler, Monika (ed.) (1990): Spracherwerb und Grammatik: Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie. Opladen: Westdeutscher Verlag (^Linguistische Berichte, Special issue 3). Russ, Charles (1989): "Die Pluralbildung im Deutschen." - In: Zeitschrift fir Germanistische Linguistik 17, 5867. Schaner-Wolles, Chris (1989): "Plural- vs. Komparativ-Erwerb im Deutschen: Von der Diskrepanz zwischen konzeptueller und morphologischer Entwicklung." - In: Hartmut Günther (ed.): Experimentelle Studien zur deutschen Flexionsmorphologie (Hamburg: Buske) 155-186. Slobin, Dan I. (1973): "Cognitive Prerequisites for the Development of Grammar." - In: Charles A. Ferguson & Dan I. Slobin (eds.): Studies of Child Language Development (New York: Holt, Rinehart, and Winston) 175-208. Slobin, Dan I. (ed.) (1985): The Crosslinguistic Study of Language Acquisition, Vol. 1. - Hillsdale, N.J.: Erlbaum. Wegener, Heide (1992a): "Pluralregeln und mentale Grammatik." - In: Anatoli Strigin & Ilse Zimmermann (eds.): Fügungspotenzen: Festschrift für Manfred Bierwisch (Berlin: Akademie Verlag) (=Studia Grammatica 34) 225-249. Wegener, Heide (1992b): Kindlicher Zweitspracherwerb: Untersuchungen zur Morphologie des Deutschen und ihrem Erwerb durch Kinder mit LI Polnisch, Russisch und Türkisch. - Habilitation Thesis University of Augsburg. Werner, Otmar (1969): "Das deutsche Pluralsystem: Strukturelle Diachronie." - In: Hugo Moser (ed.): Sprache, Gegenwart und Geschichte (Düsseldorf: Schwann) 92-128. Wiese, Richard (1988): Silbische und lexikalische Phonologie: Studien zum Chinesischen und Deutschen. Tübingen: Niemeyer. Wode, Henning (1988): Einführung in die Psycholinguistik. - München: Hueber. Wurzel, Wolfgang U. (1984): Flexionsmorphologie und Natürlichkeit. - Berlin: Akademie Verlag. Wurzel, Wolfgang U. (1990): "The Mechanism of Inflection: Lexicon Representations, Rules, and Irregularities." - In: Wolfgang U. Dressler, Hans C. Luschützky, Oskar Ε. Pfeiffer & John R. Rennison (eds.): Contemporary Morphology (Berlin: de Gruyter) 203-216. Zwicky, Arnold M. (1967): "Umlaut and Noun Plurals in German." - In: Studio Grammatica 6 (Berlin: Akademie Verlag) 35-45.

Elsa Lattey

Inference and Learnability in Second Language Acquisition: Universals vs. Language-Specific Phenomena in the Domain of Idiomatic Expression1

1.

Introduction

One of the most important maxims in language pedagogy in this century was "Do not let errors stand uncorrected - especially not in writing - for students might learn the mistakes instead of the correct structures." This is the problem faced by every teacher when students make mistakes: Should you correct them immediately, make note of them to correct them later, ignore them in the hope that they will go away after the students have been exposed to more correct data, or what? Mistakes are something we wish did not happen and when they do, we hope they will disappear of their own accord as fast as possible. One of my aims in this article, however, is to accord to mistakes a different, more useful role: I want to demonstrate that mistakes are extremely useful phenomena, not only for the language learner but for the language teacher as well. In language analysis, in contrast to language pedagogy, errors were acknowledged and given a certain due in that researchers asked the question: To what extent can these errors be explained by interference, i.e., what is the role of LI (the student's mother tongue) in the production of erroneous structures in L2 (the language being acquired)? 2 Often this led to contrastive analyses of the two languages and an evaluation of the systems to see what points of potential interference there might be, and how these are reflected in the intermediary systems developed by the learners. 3 Parallel to this discussion, there was and is a concern with learnability, a search for the conditions necessary in principle for learning to take place. What is relevant here is whether the data that would enable the child (in first language acquisition) or the second-language learner (child or adult) to actually learn some linguistic fact ever occur in the input. Where this is not the case, the appeal is made to "universal grammar" (UG) as an explanation for the occurrence of the linguistic phenomenon in the output. The logic of this argument is that whatever is part of UG does not have to be learned, because it is innate. The body of linguistic facts the learner can draw on consists, on this line of reasoning, of UG plus what becomes available via input. One of the problems, however, is "negative input" and the question of whether it plays a role in language acquisition or not. "Negative input" (or 1

Special thanks go lo my students for revealing facts about English idioms that I would otherwise have had no access to. I also thank Rosemarie Tracy and the Tiibingen acquisition project for encouraging me to pose my problem to first language researchers as well as for their helpful suggestions.

2

I simplify here to discussing only LI and L2. The problems are of course extendable to L3,... Ln.

3

Ellis (1985:32-33) points out that "Contrastive analysis was predicated on the need to avoid error, but if error is seen as a positive aspect - evidence of continued hypothesis testing... - then the importance of devising a teaching programme geared to its prevention becomes less obvious."

296

Ε Lattey

"negative data," cf. Lasnik 1989) refers to structures or linguistic sequences that do not occur in a particular language, and the question that puts itself is how does one learn that these structures or sequences are not part of the language, i.e., how does one obtain negative evidence? 4 Specifically, how can a learner in a natural language situation come to the valid conclusion that the language being learned does not contain, for example, complex prenominal modifiers. This is a very productive structure of German but not so of English (at least, not yet - English does seem to be moving in this direction, though). In German one can say things like the equivalent of (1)

the by-me-selected bouquet of flowers

but one can also say, as in English, (2)

the bouquet of flowers which I selected.

How does the German speaker learning English learn that utterances like (1) are not part of the structure of English? 5 Of course, one may say that foreign learners can simply hypothesize a gap in the input, try out the structure, wait for a reaction, and if the listeners laugh or do not understand, then the conclusion is: The structure does not work in the target language. But the question is even more basic than that. We must ask: How does the child growing up in a native English-speaking environment learn that there is no such structure in the language? Given that it is possible to have such a structure - and German is evidence that it is possible - then universal grammar needs to account for it, and speakers of languages that do not have this structure must have some way of learning that their languages do not. That is, we are faced with the very basic question of "how it is that we are able to come to know what we know on the basis of limited experience" (Matthews 1989:2). Lasnik (1989) discusses certain substitutes for negative data, including "indirect negative evidence," ruling out structures that fail to turn up where they are expected, and "semantic bootstrapping," drawing on semantic data to help solve problems in the acquisition of phrase structure. Of these, only the former is potentially relevant to the problem at hand. However, the difficulty Lasnik points out of defining "where they are expected" applies here as well. Beyond that,

4

The terms "negative input," "negative data" and "negative evidence" are often used interchangeably in the literature. I use the first two this way; "negative evidence," however, needs to be distinguished from the others. Evidence can only refer to the linguistic conclusions the speaker is able to draw on the basis - in this case - of the absence of certain structures in the language s/he hears. Note that "negative evidence" can be direct (correction) or indirect (based on non-occurrence).

5

A similar example is discussed by Ellis (1985:192). Ellis cites: (i) *We explained the girl the answer, as an ungrammatical structure of English and states "there must be some innate principle which prevents the child producing sentences like [i]." However, this structure does exist in German: (ii) Wir erklärten dem Mädchen die Antwort. Do the German- and the English-speaking children have different innate principles? Certainly the German-speaking learner of English tries on occasion to apply the German principle to English, producing utterances like (iii) *... when you recommended us that movie. Isn't this rather a feature to be included in the individual language lexicon, i.e. a feature of explain or recommend^. And is that perhaps also where the more specifically linguistic constraints on idioms (e.g. those in the structuraldependence category, cf. below) are to be found?

Second Language Idioms

297

we have lack of awareness to deal with, making it difficult in the absence of error to even define our problem.6

2.

The practical problem

2.1

Data source

I would like to touch on each of the areas I have just introduced and discuss them in relation to second-language acquisition. I will do this by characterizing a practical problem, namely the second-language acquisition of idioms and their strategies of use, and exploring it in some detail.7 My examples are taken primarily from dialogues written by German students of English in a course on "Phraseology and Vocabulary," where the task was to use the idioms being learned in concrete dialogue contexts. In writing these dialogues, the students made mistakes that revealed facts about language usage that are normally not accessible, since native speakers just "avoid" the idioms in the relevant contexts (though "avoid" is too strong a term here, suggesting, as it does, a conscious act, i.e., we simply have cases of natural nonoccurrence here). The contexts in question do not belong to the set of possible contexts for the respective idioms, and so the learners' errors here constitute instances of making explicit what is under normal acquisition conditions simply part of the domain of "negative input" for both first- and second-language learners of English. My pragmatic concern is with the second-language learner, but the theoretical question applies to the first-language learner as well, and needs to be answered, I think, in that context. 2.2

A sketch of the problem

The negative input, or missing information, does not confine itself, as one might expect, to the pragmatic domain, although much of what is at issue has to do with language usage. I will present some examples, on the basis of which I will introduce some of the relevant categories, categories such as whether an idiom is used punctually or generically, who can qualify as a referent of a particular phrase, etc. In doing so, I will demonstrate the usefulness of the occurring error for the analysis. The analysis proceeds by means of a step-by-step approach to the subconscious (unconscious?) constraints available to the native speaker (how will remain an open question) but never spelled out for the language learner, as they are below the level of awareness for native speakers.® The only way the language learner comes to know some of these linguistic facts is via negative feedback, in a classroom situation or in

6

Klein (1992:178, note 11) suggests that learnability theories are pretty far removed from the actual learning of a natural language because they were developed in the context of formal languages and what conditions one needs to fulfill for the generating grammar to produce the occurring strings. Nevertheless, in my opinion the question of whether "positive input" suffices or whether one needs "negative input" statements in order to satisfactorily deal with learnability is fully applicable to natural language situations.

7

Already Weinreich (1969:23) pointed to the interesting nature of idiomaticity, "because it ... yields certain distinct subpatterns - some perhaps universal, others specific to each language."

® As Cook (1988:177) puts it, "Teachers or native speakers can only explain what they are consciously aware of."

298

Ε. Lattey

communicative contexts (e.g. via giggling responses or misunderstanding, as mentioned above). For the native speakers of the language, these constraints, if they are not part of the "meaning" of the relevant idioms, certainly are anchored in the inventory of communicative strategies for using them. The question is: How do they become part of the communicativestrategy inventory of the native speaker, when the information they contain is subconscious and the violations - the errors made by the students - never occur for them, the native speakers?

2.3

A pragmatic classification of idioms for pedagogical purposes

In the Phraseology and Vocabulary class mentioned above, idioms were introduced to the students in a pragmatic approach. The idioms were grouped into four pragmatic categories (cf. Hieke & Lattey 1983 and Lattey & Hieke 1990): 9 Focus on the World: the tables are turned it's six of one half a dozen of the other it goes without saying that takes the cake! Focus on the Individual: be an unknown quantity wear one's heart on one's sleeve be slow on the uptake not be able to put two and two together The Individual and the World: be out of the woods have a long dry stretch ahead something is not my bag strike while the iron is hot The Interaction of Individuals: be of one heart and mind put a bug in someone's ear pull someone's leg lend someone a helping hand This approach makes it possible to discuss together idioms that need to be distinguished because they are close in meaning and function and yet make important distinctions. So, in the "Focus on the Individual" category, we need to distinguish between being slow on the

'

Cf. Lattey (1986) for a theoretical elaboration.

Second Language idioms

299

uptake, i.e. not very quick to understand but able to get the point eventually, from not able to put two and two together, which suggests a permanent inability to comprehend even the simplest things. 10 Some of the idioms have syntactic peculiarities, such as have a long dry stretch ahead lacking a progressive form, or something is not my bag usually occurring in the present tense. These are facts about the usage of these idioms that are also not fully in the native speaker's awareness, but they are the sorts of information that a trained teacher will come up with if asked to think about what one would have to tell a language learner about this idiom. 2.4

Errors as a road to linguistic insights

The student dialogues, of course, also contain errors that derive from a lack of knowledge about such syntactic peculiarities. But the more interesting errors were such as helped me, the teacher, to discover characteristics of the idioms and their usage that I had not realized before, despite extensive work with the idioms in question (my colleague Al Hieke and I had published two workbooks on them and I had written an academic paper on our system of classification as well). Cook (1988:177) points out, citing Chomsky, that "one does not learn the grammatical structure of a second language through 'explanation and instruction' beyond the most rudimentary level [because] no one has enough explicit knowledge about [it] [emphasis added]." Some explicit knowledge that I had been lacking before I now obtained from the errors of my students. What I learned was the following: First, that certain specific contexts were inappropriate for certain idioms. This is an observation about the distribution of an idiom (an observational linguistic statement). The second stage is a recognition of what grammatical or pragmatic convention is being violated by the student usage. (A descriptive linguistic statement, this amounts to a linguistic or psycholinguistic hypothesis.) The third step is to come up with an explanation based on the fact that language is a device of communication and that the users of this device are human beings. (This is what I consider the level of an explanatory statement.)11

3.

Categories relevant for the learning problem

I will now discuss some of the data, after which I will return to the theoretical issue of learnability and universal grammar. The relevant categories (at least for the examples I discuss below - they are a representative but not an exhaustive listing) are: 10

I claim that this distinction exists in the meaning of the idioms, though it belongs to that domain of knowledge that is not accessible even to the native speaker - under normal conditions (i.e., when not faced with the error). Evidence for this inaccessibility can be seen in that even a dictionary like COLLINS COBUILD offers the following definition (in the typical COBUILD direct-address style): "You say that someone is slow on the uptake when they have difficulty understanding simple or obvious things."

11

The reader will no doubt recognize the "levels of adequacy" introduced by Chomsky (1965), though with a modification of the "explanatory" one. To my mind, explanation must come from outside the theory that Chomsky delineates; consistency within the theory and capturing a linguistic generalization are part of the descriptive level as I see it, and explanation must relate that description to behavioral and cognitive facts about human beings.

Ε Lattey

300

1. 2. 3. 4. 5. 6.

Potential referents Context scope (generic vs. punctual) Structural dependence Considerations of intent Similarities and conflicts between LI and L2 Semantic and pragmatic context limitations

Let us look at these one by one.

3.1

Potential referents

The question of potential referents for an idiom has to do with a "feel" for what one is likely to say about oneself, about an addressee, and about a third person. In the data at hand, the greatest restrictions appear to be on what one would or would not say about oneself. Consider the idiom sit on the fence, for example, 'be undecided about which course of action to take.' The student dialogue that raised the issue was: (3)

A: Β:

Are you coming to London with us? I don't know yet. There's a party in Tom's flat. I am sitting on the fence.

It seems that one hardly uses sitting on the fence of oneself, at least not in a specific event, although one can say things like "I wish I didn't always sit on the fence so long," describing a general character trait. A similar case is look like death warmed over (that is American, the British say look like death warmed up), 'look or feel very tired or very ill.' Here is the relevant dialogue: (4)

A: B: A:

Yesterday I found a spider sitting on my bed. I died a thousand deaths. So what did you do? I called my brother Eric. I forgot that he is all thumbs, and so the spider was able to escape and disappeared. I was on pins and needles the whole night. I look like death warmed over today.

This again appears to be something one would not say of oneself unless it is qualified by epistemic must - "I must look like death wanned over today" (and then one hopes the other person says "no, you don't") - or if one happened to be looking in a mirror at the moment of speaking. A similar restriction holds of eat out of someone's hand, 'obey or believe someone without question; have no independent will': (5)

When I first saw Tom, I immediately took a shine to him. Now I know him better, and we are really of one heart and mind. My friends say that I am eating out of his hand now, but that's not true at all! He's just a perfect guy!

Although it is perfectly o.k. for the writer's friends to say "She's eating out of his hand now," the writer herself would better have described the situation with: My friends say that "he's got me eating out of his hand," i.e. using the reverse direction on the idiom.

Second Language Idioms

301

In (6) we have the final example of the "things you wouldn't say about yourself" 12 category, be slow on the uptake, 'have difficulty comprehending what is being said': (6)

A: Would you like to help me repair this old radio? It's very difficult. B: What did you say? Am I slow on the uptake? You want me to help you with this old radio? Don't you know that I'm all thumbs?

Here we have the added dimension of the question form. We can see that question formation is not the problem, however, for it is perfectly natural to say "Is he slow on the uptake, or what?" if you observe some third person not comprehending something that you and the addressee understand. So again the problem appears to be the first person. Here speakers might also say something like "I must be slow on the uptake today," but only if they haven't yet understood, or "I was really slow on the uptake," when they have finally understood (after a long time of not doing so). What the four contexts (3)-(6) have in common is that the idioms make a negative comment about the referent - with respect to behavior, appearance, control of one's own life, or intellectual ability. Given human nature, it is not surprising that as speakers of English we should be reluctant to speak that way about ourselves unless we mitigate the statement by means of a modal expression such as must be or attribute the responsibility for the situation to someone else, as in he has me eating out of his hand. I have said "as speakers of English" in the preceding paragraph because English is my point of departure, but one could easily imagine the argumentation being transferred to any other (western?) society, at least to any society in which the speakers are not routinely selfdeprecating in their behavior. The data discussed here thus fall into a potentially "universal" domain, a question I will return to below. 3.2

Context scope (generic vs. punctual)

This category deals more directly with the context of situation than with the referents involved. Here the issue is whether the idiom has generic characteristics or not, that is, whether it is a generally valid statement or whether is makes reference to a particular occurrence or event. An example with have bats in one's belfry, 'be eccentric, peculiar, crazy (literally or figuratively),' will make the distinction clear: (7)

I guess_you had bats in your belfry when you recommended that movie to us.

Here the instinctive teacher comment was: "Not usually used punctually, i.e. together with a wAe/j-clause."13 We can see the difference if we consider the similar idiom have a screw loose. This can be used punctually, i.e. it could be substituted in (7) to produce an acceptable utterance, cf. (8).

When enclosed in quotation marks - i.e., treated as a single modifier, as here - English can use constructions like the German complex prenominal modifiers mentioned above. Used literally the phrase can occur with a when-clause: (i) When they lived in the old abandoned church in Laredo, they had bats in their belfry.

F.. Lattey

302

(8)

I g u e s s h a d a screw loose when you recommended that movie to us.

Apparently speakers of English have a different scope understanding of these two idioms. Again, we are dealing with a domain not readily accessible until one stumbles over the awkward student construction. Three dictionaries that I consulted - Barron's Dictionary of American Idioms, National Textbook Company's American Idioms Dictionary (NTC) and Collins COBUILD - make no mention of the generic character of have bats in one's belfiy. The interesting thing about these two examples is that one can explain the difference - in my sense of explain - by thinking about the real world that gave rise to the idiomatic expressions: When bats occupy a belfry or other tower or cave, they tend to move in permanently, unless there is a real disruption that drives them out. The looseness of a screw, in contrast, is a relatively impermanent state, for the application of a screwdriver will remedy the situation. The interpretation of the idiom not have a penny to one's name is related to whether it occurs with a time reference or not, but the nature of the time reference seems to be rather constrained. So, for example, in (9), the student's use of this month appears inappropriate, while just now (if added at the end of the sentence) seems to work perfectly well, though surely the constraint is not to the moment of speaking. (9)

You needn't ask Peter for his rent. This month he hasn't a penny to his name. But I reckon he's trying hard to get it by the end of this week, since he's been out working every night.

The general meaning of this idiom, when there is no mentioned time reference, is generic, i.e., as Collins COBUILD defines it: "If someone doesn't have a penny to their name, they are very poor indeed." It describes the overall financial state of the individual, not any momentary lack of funds. This distinction between generic and punctual usage of an idiom became clear to my colleague Al Hieke and me as we were trying to find an English equivalent for the German idiom: Glück im Unglück haben. The closest we could come in English was: Every cloud has a silver lining. The difference between the two idioms, however, is that they are not applicable in the same situational contexts. The German idiom is punctual and is usually used after the fact, when someone has experienced a misfortune that turned out to be not as bad as it might have. Contextually, an English speaker might say something like, "Cheer up, it could have been worse." The English idiom, on the other hand, says that no misfortune is entirely dismal, there's always something positive about it. The distinction between generic and punctual seems to be a candidate for a universal feature, especially when we appeal to zoological or physiological knowledge in the explanation (as we did with the bats and the screws). Which particular idioms behave in which way may, however, be a language-specific fact that needs to be learned. Certainly related structural facts will be language specific (cf. 3.3). 3.3

Structural dependence

What is at issue in this category is the co-occurrence or non-co-occurrence of a particular idiom with a particular grammatical construction.

Second Language Idioms

303

We have seen, for example, that have bats in one's belfry (Ex. 7) should not co-occur with a when clause. Here there is an interaction between two factors: context scope and structural dependence. The context-scope factor is a potential universal, as mentioned above. In contrast, the structural dependence that follows from it - i.e. in this case the non-cooccurrence of the idiom with a when clause - is, most probably, language specific. In (10), we have an idiom, something suits one to a T, 'one is satisfied with every detail of something, it is just right,' a portion of which, namely to a T, is not usually used in an if clause:14 (10)

A: Excuse me, can you please show me the way to the White House on my map? B: Sure, I know Washington inside out. I'm just on my way to see the president. If it suits you to a T, you can accompany me right now. A: That'll be wonderful, thank you.

We would have expected, in this dialogue, for speaker Β to say, "If it suits you, you can accompany me," to which speaker A would have had the option of replying (in lieu of the response given) "That suits me to a T," expressing the relevant phrase in an assertive utterance. My final example for the structural-dependence category is the idiom pull someone's leg (that is the American version; pull the other one (it's got bells on) is the British counterpart).15 The inappropriate student context is given in (11): (11)

I just got the brush-off from Μ. I even suppose that she only pulled my leg when she told me she loved me.

Here the correction would be: "she was only pulling my leg when she told me she loved me," i.e. this idiom is usually used in the progressive, with be + -ing. We see from the examples discussed in this section that this category (structural dependence) is likely to be a reflection of language-specific characteristics of the idioms. 3.4

Considerations of intent

These reflect what speakers do with language in communication. So, for example, the idiom just discussed,/»«//someone's leg, describes an intentional speech act on the part of the actor. The idiom can be defined as 'fool someone, often teasingly,' and one cannot do that accidentally. But the student who wrote (12) didn't realize that.

14

One can in discourse say things like "If it suits you to a T, get on with it." but only after the other person has made the comment that something suits them to a T, i.e., this usage is like a quotation of something the other person said.

15

We have a difference here in situational appropriateness, the British version being a form of direct address (a metaphorical command) issued after the speaker (S) recognizes that the addressee (A) is teasing, and communicating to A that S does not believe what has just been said but takes it as a joke. While the American version can also be phrased as a command (Stop pulling my leg!), it needn't be and can thus also describe an observed event.

304

(12)

Ε. Lattey

A: Joe promised me the moon. But soon I realized that he was lying through his teeth.

B: Perhaps he was only pulling your leg! A: No, he did it on purpose. Until we get to A's second turn, everything is fine. But A's response indicates that there is something missing in the student's understanding of the pragmatic import of this idiom. S/he has not realized the volitional character of the event described by the idiom. Example (13) reflects a lack of understanding that is not surprising at all when one considers the usual paraphrase of get out of the frying pan into the fire, which is something along the lines of 'be worse off than before' or 'go from bad to worse.' (13)

A: Have you already talked to the doctor? B: Yes, and he says John is out of the woods now. A: I heard Betty say that if John had gotten an infection, he would have gotten from

thefryingpan into the fire. It seems to me - and I entertain the possibility that the reader may not agree with me here that the action that gets one from the frying pan into the fire has to be a volitional one, at least in part. That is, the person who is worse off than before is at least partially responsible for that, even if the negative outcome was unintentional. An example may help to make this clear: (14)

Sandra moved to the city in the hope of finding work there. But she didn't get a job and without her family there to fall back on, she had gotten from the frying

pan into the fire. The action (moving to the city) that resulted in the negative outcome (being worse off than before) was intentional. The context is therefore appropriate for this idiom. The "getting an infection" in example (13), because it does not meet this condition (being intentional), is an inappropriate context for the use of the idiom. Getting an infection is not something you have control over as a hospital patient. (15a) and (15b) are two student productions, both reflecting the same misunderstanding of the idiom sleep like a log. (15a) A: Well, did you get all of the exam papers corrected? B: Yes, but I was up all night and this afternoon I'll sleep like a log. (15b)

I have to go now. I'm so hungry I could eat a horse. And tonight / think I'll sleep

like a log. These examples reflect how sensitive idioms are to discourse enhancers, let me call them that for the moment, for want of a better term. By discourse enhancers I mean words (I suppose they could sometimes be phrases also) that add a comment and give the utterance a modal character. Both of these examples would be greatly improved by the use of the discourse enhancer probably : I'll probably sleep like a log. In (15b) the I think should be deleted (because it suggests intention, which does not work with this idiom, although it would be fine

Second Language Idioms

305

with its paraphrase 'sleep long and soundly'),16 or replaced by I bet. In (15a) the absence of a probably also suggests interpretation as a volitional act. The usage phenomena discussed in this section all have to do with whether an action is intentional or not: In the case of pulling someone's leg, the event described by the idiom is an intentional act, in the case of sleep like a log it cannot be; in the third case, what is at issue is whether the event that leads up to getting from the flying pan into the fire involves volition on the part of the actor or not. Here again, the category - i.e. the taking account of intent - may be a universal, whereas the individual languages may have different conditions for similar idioms.17 3.5

Similarities and conflicts between LI and L2

This category raises some interesting questions about language learners' ability to access information about their mother tongue and transfer it to a subsequently learned language. In (16) and (17) we have English idioms that have close counterparts in German: be the spitting image of: wie aus dem Gesicht geschnitten (literal paraphrase of the German: 'as though cut out of someone's face') and have a bone to pick with someone : mit jemandem ein Hühnchen zu rupfen haben ('to have to pluck a chicken with someone'). Be the spitting image of someone, like the German equivalent, refers to physical similarity only, and yet several students produced dialogues in which they used it to describe a similarity in behavior. (16a) A: Well, I think Tom has always had a soft spot for her. But who would have expected this? B: Why do you doubt it? They're both real sports. He's the spitting image of his father, who's always laughing, and she can be like her mother, soft and gentle. (16b)

Fred seems to be the spitting image of George. Once he has discovered a new hobby, he puts all his time and money into it.

Here, a fact learned in the acquisition of German, that there is an idiom that refers to physical likeness only, is not transferred to English, even when the English idiom be the spitting image of someone is learned in the context of the corresponding German one (i.e., in introducing the English idiom and its paraphrase "look exactly like someone," the teacher also asked whether students could come up with the German equivalent, and they did, so jemandem wie aus dem Gesicht geschnitten was raised to the level of awareness at the same time as the English idiom was introduced and was presumably available while the English idiom was being learned). The question is, however: How salient is the constraint on the idiom even for the native speaker? Dictionary definitions do not highlight it; they simply offer a definition in terms of sieht X sehr ähnlich, 'looks very much like X,1 or of "perfect likeness or counterpart," the second element of which does not on the face of it limit the class of referents to those with physical likeness only.

16

One can intend to sleep for a long time (and help it along by not setting an alarm clock), especially when one is very tired. Whether one can intend to sleep "soundly" might be open to discussion.

17

In German, for example, the equivalent of (15) would be Ich werde schlafen wie ein Murmeltier, and that construction (though future, just like the English example) does not imply intent.

Ε. Lattey

306

In the second case I want to discuss, we have a similar lack of transference. Consider (17): (17a)

I think I'll have a bone to pick with this guy.

(17b)

Next time I see him I'll have a bone to pick with him.

This idiom means that one has something to complain about to someone relating to that person's behavior or actions. The complaint is valid at the time of speaking. The fact that there will be only a future encounter with the offending party (s/he is not around at the moment) does not postpone the condition (state) of having the complaint. We could describe these facts as a structural dependence (cf. 3.3 above) and say that this idiom usually has no future form. 18 But it seems more relevant in the context of universale and second language acquisition to view them in the light of a similarity between LI and L2 that is not being capitalized on by the language learner. The question is why not? And this question seems all the more justified when we encounter (18), where it is apparent that transfer - in this case negative transfer - is taking place. (18)

For the last few weeks his attitude towards me has somehow changed: he's irritated; he always has to have the final say; he is a pain in the neck to me!

In this English idiom the annoyance is usually not attributed to anyone - it is assumed that it is a reflection of the speaker's feelings - but even when the experiencer is made explicit, it is done by saying something like "I find her to be a pain in the neck," rather than with the prepositional phrase, as in (18). What is being transferred here? The German close equivalents in (19) (19)

(a) jemandem auf den Wecker gehen to-someone on the alarm-clock get 'get on someone's nerves' (b) jemandem auf die Nerven fallen to-someone on the nerves fall 'get on someone's nerves' (c) jemand ist mir ein Dorn im Auge someone is to-me a thorn in-the eye 'someone annoys me intensely'

do attribute the annoyance, i.e. the person who is annoyed is mentioned explicitly. Although there are other German near equivalents without an expressed affected individual (so, for example, eine Nervensäge sein [be a nerve-saw] or ein Quälgeist sein [be a torturing spirit]), it does seem that German structures like those in (19) are the source for the prepositional phrase to we in (18). Example (18), of course, is of a different type than (16) or (17). The phenomenon that was transferred in (18) could be described as either a structural one (having to do with the

Except after an if-clause: If you do that, I'll have a bone to pick with you, but this is an unusual usage.

Second Language Idioms

307

subcategorization frame of the idiom)19 or a lexical one (deriving from the idiom entry itself). 20 As such, it may have a low salience for the learner - akin to the awareness of whether the idiom contains a definite or indefinite article, a feature the students often disregarded - and be characterizable as incomplete learning. There is another example which may fit into the considerations-of-intent category, but I am not entirely sure. It involves the idiom come to a standstill, which is paraphraseable as 'unable to go on; be at a point where neither party will yield any further or is able to make a move.' The German equivalent is festgefahren sein, which literally means 'be stuck,' as in the mud (said of a vehicle). Collins COBUILD gives both a literal and a figurative example for the English idiom: a) b)

The traffic has come to a standstill, (literal) The negotiations are at a standstill, (figurative)

The problematic student dialogue is as follows: (20)

A: I think there's no use staying together any longer ... our relationship has come to a standstill·. B: Do you really think so? In my opinion we should try again.

This is unusual, though I cannot fully say why. E.g., it would be natural to say "Our relationship isn't going anywhere," so it is not that relationships do not move. Perhaps it is because the idiom usually refers to such events as negotiations (cf. the Collins COBUILD figurative example), where a certain result or outcome is expected. German, on the other hand, does seem to allow the figurative collocation of Beziehung 'relationship' and festgefahren sein 'come to a standstill', and may thus be the source of the awkward collocation in English. This section, then, highlighted examples which are, if not candidates for universal categories, at least to be found in both of these languages (English and German). Yet the potential positive transfer is not activated by the second-language learners while the negative transfer is. 3.6

Semantic and pragmatic context limitations

The final category of examples I want to discuss is that involving semantic and pragmatic context limitations. The idiom be all thumbs, for example, means to be awkward or clumsy (with one's hands - a crucial constraint, determined by the semantic content of the idiom: the thumb stands in opposition to the fingers, and what makes us able to grasp things is the ability to bring thumb and fingers together. If our hands were all thumbs, we would not be able to We cannot, of course, speak of the subcategorization frame of the verb here, so perhaps this is not the right concept. What I would include in this notion, along with mention of particular participants (such as the traditional indirect object in (18)) are such distinctions as between having a bee in one's bonnet and treading on someone's toes, where in the first case the indefinite reference is to the "subject" of the idiom and in the second to the "object." 20

The idiom entry must, of course, contain selectional (or subselectional) restrictions as well. Cf. Weinreich 1969:36-39, Höhle 1978:26-29 for a discussion of idiom entries in the lexicon.

308

Ε Lattey

grasp, as least not with the finesse that the combination of fingers and thumbs gives us. Ergo, being all thumbs reduces our tactile skill. But the thumb being a part of the hand, the idiom is constrained to such activities as can be done with the hands). Cf. (21). (21)

A: B: A: B:

How are you? Well, today it seems as if I'm all thumbs. How come? First of all I broke two dishes this morning, then I almost fell down the stairs and last but not least I missed my bus.

"Breaking the dishes" is good, but "falling down the stairs" and "missing the bus" are not the kind of activity you associate with "being all thumbs" because they are not done with the hands. How does the native speaker leam this? Common sense? Why not the secondlanguage learner? The German equivalent is zwei linke Hände haben, 'to have two left hands' and is restricted in the same way. That was a semantic constraint. The following is a pragmatic one: To have a skeleton in the closet is to have some shameful secret that one tries to keep concealed from others. Because it is the sort of thing one conceals from others, it is difficult for others to find out about. And that is what makes the following utterance odd: (22)

I wonder why she didn't want to tell me why she had to move out of her flat in Downing St. She sure has a skeleton in the closet.

Pragmatically, someone else's "having a skeleton in the closet" is something you have to make guesses about. You do not usually have the certainty to say "she sure has." Here, again, native speakers would employ the epistemic must·, "she must have a skeleton in the closet," at least until such time as the secret has been revealed. 21 To know which way the wind blows is 'to be clever, experienced, able to assess a situation accurately,' or as Collins COBUILD puts it: "If you want to find out which way the wind is blowing, you want to find out what is likely to happen." Neither of these paraphrases suggests that it is people and their attitudes that are at issue. That is, "what is likely to happen" does not refer to such natural events as whether it is going to rain or not, or to such physical effects as whether an experiment in the laboratory is likely to produce positive or negative results. The wind that is blowing is, in effect, being blown by people, and it reflects their attitudes and likely intentions. The following dialogue appears to be odd in just this respect, having excluded the individuals that might be affecting the "wind." (23)

A: B:

Mom, can you help me with these math exercises? We have a test tomorrow. Sure, let me have a look at them. When I was in school I knew these math problems cold. In three h o u r s ' / / know which way the wind blows. The test will be no problem for you.

It may be that we have interference from German sicher, 'certainly,' here, which is used in conversational implicative also in the sense of I bet that ...' or Ί presume that ...' and so comes close to the must I am suggesting as a replacement in English.

Second Language Idioms

309

Here, a more appropriate idiom to use would have been: be able to do something blindfolded. Apparently knowing which way the wind is blowing is knowing what other people are thinking, how they are likely to interact with you, etc. Again: How does the native speaker learn this? In the use of the following idiom there is a connotative element that needs to be learned for appropriate deployment. To have someone/something on one's hands means to have someone/something as one's responsibility, but it conveys more than that, namely that this responsibility is a burden. So when one reads (24) one feels that something is missing in the context that would justify use of this idiom, given its inherent implication that the situation being described is considered to be a burden. (24)

By the way, C was with her cousin D, and I couldn't believe it: that girl is the spitting image of her. D will have C's pets on her hands when she's on holiday.

The teacher's comment on the student paper will, I believe, explain: "In order to justify the comment inherent in have... on her hands (namely that it is a burden), you need to set up the dialogue with something like: Ί feel sorry for D. When C's on holiday, D will have C's pets on her hands.' Otherwise in your dialogue as it is you'd be more likely to hear 'D will be taking care of C's pets ..."' The notion of a burden must be retrievable from the context to motivate the idiom's use. I would like to include one final example in this category, the idiom we began our discussion of potential referents with: sit on the fence. Here is the dialogue: (25)

A: Β: A:

Would you like to go to lunch with me, Mary? I just had a long talk with my boss. I'm not sure whether I'm hungry anymore after that. Oh, don't just sit there staring into space. Come on, get off the fence and come along.

Apart from the poetic license taken with the idiom here (which is fully within the acceptable flexibility of such phraseological idioms), 22 it seems to me that the context is not really indecisive enough and that there is really only one alternative to sitting on the fence. The difficulty should be as to which side to come down on, i.e. there should be two potential courses of action. This lies in the semantic content of the idiom again. We can visualize a fence and see that it has two sides. The paraphrases or definitions we find in dictionaries, however, do not pick up on this. They often do not go beyond "undecided" or "not able or wanting to choose," without a mention of choosing from among how many alternatives. Perhaps this distinction is not as clear for others as it is for me. NTC even gives the following example dialogue:

22

The dictionaries are not of one mind about the form of this idiom. COLLINS COBUILD gives sit on the fence. National Textbook Company offers on the fence (about something), while Barron's just gives on the fence, adding that it is "often used with sit." Their example makes the alternative between two courses of action or two goals clear, however: "Jack sat on the fence for a week last spring before he finally joined the track team instead of the baseball team."

Κ Lattey

310

"Ann is on the fence about going to Mexico." "I wouldn't be on the fence. I'd love to go." Here the choice is obviously between going and not going, as in (25), between doing something, or not doing something, or, seen from the point of view of the fence, between jumping off and not jumping off. This accords with a definition of "undecided." For me, this is not enough for the successful use of this idiom, however, and I have found agreement from other native speakers. One might improve the context, to make staying at home a clear alternative "course of action," and increase the acceptability of the idiom, but as it stands, I find its use odd. Depending on how one views these data (the idiom, its paraphrases, the "erroneous use"), we might classify this last context limitation as semantic or pragmatic. If I am right and sit on the fence requires clear alternatives of action about which the actor is undecided, then I would tend to say it is a semantic question. If we accept different interpretations and different viewpoints, then we have surely moved over into the pragmatic domain.

4.

Summary and concluding remarks

I have sketched and exemplified in section 3. six categories which I find relevant to determining what is going on in the production of what I see as second-language errors in the use of these phraseological idioms. I would like now to come back to the question of how the language learner copes with data such as these. Stein (1991:242-243) points out that learners get little help in acquiring a sureness about idiom use, adding that it is precisely with phraseological idioms that the cognitive, conceptual, factual, communicative and situational relationships are very complex. We have seen that this is indeed the case. 2 3 Stein (1991:255) also makes the claim, however, that learners reproduce without any problem those phraseological idioms that have equivalent lexicon, structure and meaning in LI and L2 and those that derive from borrowing or from common historical and cultural sources. We have seen that this is not always the case (cf., e.g., be the spitting image of or

have a bone to pick with someone in 3.5). The learning of idioms and the strategies of their use is a complicated, as yet little studied task. The individual concrete linguistic realization of a phraseological lexeme by the learner must be identified in a whole series of applications and in an ample number of meaningful speech-act contexts (Stein 1991:256). But how do the learners (first- and second-language learners) do this? That is the basic question, given the fact that normal discourse does not provide examples of inappropriate contexts, does not provide negative data or evidence. That is, we do not expect the errors that I have discussed - insofar as my observations and explanations are correct - to ever occur in the discourse of native speakers of English who have acquired their language fully. If that is so, then how does the learner - how did I as a native speaker - ever get to the point of avoiding these contexts? Did some sort of unconscious learning take place? Can we appeal to universal grammar? That would seem to

Precisely how the six categories I have described map onto the "cognitive, conceptual, factual, communicative and situational relationships" addressed by Stein will not be explored here.

Second Language Idioms

311

be the theoretical course one must take if there is no chance for learning from the environment. But we are dealing with idioms! What is more language-specific than idioms? For some of the categories I have discussed, appealing to universale may indeed be attractive, but it seems that they would be universale of cognitive behavior rather than of grammar. So, the constraints on possible referents (3.1), for example, appear to follow from general observations about human behavior. They could thus be cognitive universale or the products of cognitive development and applied intuitive faculties. The question, then, is why does the second-language learner make mistakes, if we are dealing with a cognitive universal and the students in question are adults, i.e. have completed their cognitive development? The same question applies to the cases of LI and L2 similarity (3.5), where learners did not succeed in transferring certain facts acquired along with their native German to their use of English. One possible answer here is that the artificiality of the classroom learning situation, possibly considering the potential of stress in situations where performance is being evaluated (however mildly), may lead to the suspension of "known" cognitive universals. 2 4 We are back in trouble, though, when we look at cases of negative transfer. Why should positive transfer be avoided and negative transfer practiced? It may be that we cannot explain this. 2 5 Ellis (1985:35) suggests that "any particular error may be the result of one factor on one occasion and another factor on another. There is no logical or psycholinguistic reason why a given error should have a single, invariable cause." 1 agree as far as "single" and "invariable" are concerned. But this should not keep us from looking for possible sources. Especially since, as we have seen, errors can be extremely useful. We should appeal with Hatch (1983) to "naturalness" factors: how salient, how clear is the relationship of form and meaning - and how can we heighten the salience and put it to use to achieve successful and to avoid unsuccessful transfer? The qualities of genericness and punctuality (3.2) appear to be useful in determining the scope of applicability of an idiom and can be directly correlated with facts about the realworld phenomena that gave rise to the idioms. They thus have universal character. Nevertheless, there are undoubtedly idioms whose generic or punctual characteristics are dependent on social facts, an issue not explored in this article. This would suggest that although the qualities of genericness or punctuality could be drawn on in UG, how an individual language makes use of them may be, in part at least, language specific. The most clearly language-specific features observable in the idioms in question are the phenomena discussed in 3.3, 3.4 and 3.6: structure dependence, the role of volition and individual semantic and pragmatic limitations of the contexts. Yet even here, the potential for universal features to look out for is given. For example, the notion that it can be important in determining the range of applicability of an idiom whether the individual to whom that idiom is to apply contributed intentional behavior to the development of the context or not could be worth considering in any language.

This explanation is more attractive than a possible avoidance of structures like those in LI on the basis of "it's a different language, so chances are that there are more differences than similarities," an avoidance strategy used, for example, by the learner who avoids using an idiom in L2 that is like the familiar LI idiom, concluding that s/he must be translating and that cannot be right. 25

It may also be that we need more examples of lack of positive and presence of negative transfer. There may well be instances of the other alternatives as well, the presence of positive and the absence of negative transfer, as more data is collected and analyzed.

312

Ε, Lattey

It seems that we know far too little about how anyone learns idioms to answer the questions that I have raised here. We need research on first language acquisition of meanings and strategies of idiom usage. We need to know about how acquisition proceeds in the face of negative input. As Bowerman (1988:96) concludes, "the 'no negative evidence' problem is not a myth, but a very real and serious challenge for the construction of an adequate theory of language acquisition." And we need to know about the role of cognitive universals and universal strategies (White 1988). In the light of so many open questions, the recognition that errors, rather than being problems to be avoided at all costs, can contribute to the development of linguistic understanding and the achievement of precision in defining meanings and communicative strategies far beyond what even trained linguistic intuitions can provide is indeed refreshing.

Postscript·. Let me return in the context of this discussion of idioms to address the title question of this volume: How tolerant is Universal Grammar? When I consider the restrictions on language production I have formulated for the domain of idiomatic expression on the preceding pages - based on my native-speaker view of what the individual idioms mean and the conventions I have internalized in the course of my life - and envisage the consequence, my imparting what I take to be facts about proper English usage to my students, I cannot help thinking that a grammar, any natural language grammar (and UG, if it turns out there is such a thing, so much the more), must be far more tolerant than any individual user could ever be. This thought takes me back to Ferdinand de Saussure and his characterization of a langue as an entity of which each individual speaker has only an incomplete realization (de Sausssure 1916/1959). Universal Grammar, like Saussure's langue, must be very tolerant indeed.

313

Second Language Idioms

Bibliography Bowerman, Melissa (1988): "The 'No Negative Evidence' Problem." - In: John A. Hawkins (ed.): Explaining Language Universals (Oxford: Basil Blackwell) 73-101. Chomsky, Noam (1965): Aspects of the Theory of Syntax. - Cambridge, Mass.: MIT Press. Cook, Vivian J. (1988): Chomsky's Universal Grammar: An Introduction. - Oxford: Blackwell. Ellis, Rod (1985): Understanding Second Language Acquisition. - Oxford: Oxford University Press. Hatch, Evelyn (1983): Psycholinguistics: A Second Language Perspective. - Rowley, Mass.: Newbury House. Hieke, A. E. & Elsa Lattey (1983): Using Idioms: Situationsbezogene Redensarten. - Tübingen: Niemeyer. Höhle, Tilman Ν. (1978): Lexikalistische Syntax: Die Aktiv-Passiv-Relation

und andere

Infinitkonstruktionen

im Deutschen. - Tübingen: Niemeyer. Kean, Maiy-Louise (1988): "The Relation Between Linguistic Theory and Second Language Acquisition: A Biological Perspective. - In: Pankhurst et al., 61-70. Klein, Wolfgang (1992): Zweitspracherwerb: Eine Einführung. - Frankfurt a. Main: Hain. Lasnik, Howard (1989): "On Certain Substitutes for Negative Data." - In: Robert J. Matthews & William Demopoulos (eds.): Learnability and Linguistic Theory (Dordrecht: Kluwer) 89-105. Lattey, Elsa (1986): "Pragmatic Classification of Idioms as an Aid for the Language Learner." - In: International Review of Applied Linguistics 24, 217-233. Lattey, Elsa & A. E. Hieke (1990): Using Idioms in Situational Contexts: A Workbook. - Tübingen: Francke (=UTB 1589). Matthews, Robert J. (1989): "Introduction." - In: Robert J. Matthews & William Demopoulos (eds.): Learnability and Linguistic Theory (Dordrecht: Kluwer) 1-17. Pankhurst, James, Michael Sharwood Smith & Paul Van Buren (eds.) (1988): Learnability and Second Languages. - Dordrecht: Foris. de Saussure, Ferdinand (1916/1959): Cours de linguistique generate [Course in General Linguistics]. C. Bally & A. Sechehaye (eds.), in collaboration with A. Riedlinger. Translated, with introduction an noted by W. Baskin. New York: McGraw-Hill Book Co. Selinker, Larry (1972): "Interlanguage." - In: International Review of Applied Linguistics 10, 209-231. Sharwood Smith, Michael (1988): "Second Language Acquisition: Logical Problems and Empirical Solutions." - In: Pankhurst et al., 9-35. Sorace, Antonella (1988): "Linguistic Intuitions in IL Development: The Problem of Indeterminacy." - In: Pankhurst et al., 167-190. Stein, Barbara (1991): "Probleme der Beschreibung von Phraseologismen (Beispiel Französisch)." - In: Die Neueren Sprachen 90:3, 242-258. Weinreich, Uriel (1969): "Problems in the Analysis of Idioms." - In: Jaan Puhvel (ed.): Substance and Structure of Language (Berkeley, California: University of California Press) 23-81. White, Lydia (1988): "Universal Grammar and Language Transfer." - In: Pankhurst et al., 36-60.

315

List of Contributors Franz-Josef d'Avis & Petra Gretsch Seminar für Englische Philologie und Sonderforschungsbereich 340 Universität Tübingen Wilhelmstr. 50 / Wilhelmstr. 113 72074 Tübingen Germany Agnes Fritzenschaft Seminar für Englische Philologie Universität Tübingen Wilhelmstr. 50 72074 Tübingen Germany Ira Gawlitzek-Maiwald Seminar für Englische Philologie Universität Tübingen Wilhelmstr. 50 72074 Tübingen Germany Mary Aizawa Kato Depto de Linguistica, IEL UNICAMP Cidade Universitaria Zeferino Vaz Barao Geraldo 13081 Campinas, Sao Paulo Brazil Elsa Lattey Seminar für Englische Philologie Universität Tübingen Wilhelmstr. 50 72074 Tübingen Germany Zvi Penner Institut für Sprachwissenschaft Universität Bem Länggassstr. 49 3000 Bern 9 Switzerland

Beatrice Primus & Katrin Lindner Institut für Deutsche Philologie Universität München Schellingstr. 3 80799 München Germany Chris Schaner-Wolles Institut für Sprachwissenschaft Universität Wien Berggasse 11 1090 Wien Austria Rosemarie Tracy Seminar für Englische Philologie Universität Tübingen Wilhelmstr. 50 72074 Tübingen Germany Maaike Verrips Department of General Linguistics Universiteit van Amsterdam Spuistraat 210 Amsterdam Holland Heide Wegener Lehrstuhl für Deutsche Philologie Deutsch als Zweitsprache Universität Augsburg Universitätsstr. 10 86159 Augsburg Germany