The Handbook of © 1995, 1996 by Blackwell Publishing Ltd

Chapter 15 © 1995, 1996 by Lila Gleitman and Jane Gillette

BLACKWELL PUBLISHING

Child Language 350 Main Street, Malden, MA 02148-5020, USA

108 Cowley Road, Oxford OX4 lJF, UK
550 Swanston Street, Carlton, Victoria 3053, Australia

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system,
or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording
or otherwise, except as permitted by the UK Copyright, Designs, and Patents Act 1988,
without the prior permission of the publisher. ·

First published 1995

Edited by First published in paperback 1996
Reprinted 2000, 2004 (twice)

Paul Fletcher and Library of Congress Cataloging-in-Publication Data

The Handbook of Child Language/ edited by Paul Fletcher and Brian MacWhinney
Brian MacWhinney p. cm.
Includes bibliographical references and index.
ISBN ~1-18405-8 - ISBN ~1-20312-5 (pbk)
1. Language acquisition. 2. Language disorders in children.
I. Fletcher, Paul. II. MacWhinney, Brian
Pl18.H347 1995
401'.93-dc20 94-15066
CIP

A catalogue record for this title is available from the British Library.

Set in 10 on 12 pt Palatino
by Graphicraft Typesetters Ltd, Hong Kong

The publisher's policy is to use permanent paper from mills that operate a sustainable forestry
policy, and which has been manufactured from pulp processed using acid-free and elementary
chlorine-free practices. Furthermore, the publisher ensures that the text paper and cover board
used have met acceptable environmental accreditation standards.

For further information on

Blackwell Publishing, visit our website:
www.blackwellpublishing.com

Blackwell
Publishing
 vi Contents

9 Socialization across Contexts 251

RICHARD ELY AND JEAN BERKO GLEASON
Contents
Part II The Emergence and Consolidation of Linguistic
Abilities 271
Introduction 272
The Spoken Language: Early Speech Development 277
10 Development of the Capacity for Spoken Language 278
JOHN L. LOCKE
11 Phonetic Abilities in the First Year of Life 303
RAY D. KENT AND GIULIANA MIOLO
12 Phonological Development 335
LISE MENN AND CAROL STOEL-GAMMON

List of Contributors viii Leaming Words 361

Acknowledgments xi 13 Early Lexical Development 362
MARTYN BARRETT
Part I Theory, Method, and Context 1 14 Later Lexical Development and Word Formation 393
Introduction 2 EVE V. CLARK
Theoretical Approaches 9 15 The Role of Syntax in Verb Learning 413
LILA R. GLEITMAN AND JANE GILLETTE
1 Parameters in Acquisition 10
JtjRGEN M. MEISEL Leaming Grammar 429
2 Connectionist Approaches to Language Acquisition 36 16 Reinterpreting Children's Sentence Comprehension: Toward a
KIM PLUNKETT
New Framework 430
3 The Impact of Language Socialization on Grammatical ROBERTA MICHNICK GOLINKOFF AND KATHY HIRSH-PASEK
Development 73 17 Strategies in the Acquisition of Syntax 462
ELINOR OCHS AND BAMBI SCHIEFFELIN ANN M . PETERS
Methods 95 18 Phrase Structure and Functional Categories 483
4 Individual Differences and their Implications for Theories of ANDREW RADFORD

Language Development 96 19 Empty Categories and Complex Sentences: The Case of

ELIZABETH BATES, PHILIP S. DALE, AND DONNA THAL wh-questions 508
5 Computational Analysis of Interactions 152 ]ILL DE VILLIERS
BRIAN MAcWHINNEY
179
Part III Nonnormal Language Development 541
Social and Contextual Influences
6 Issues in the Study of Input: Finetuning, Universality, Individual Introduction 542
and Developmental Differences, and Necessary Causes 180 20 Computational Approaches to the Analysis of Language
CATHERINE E. SNOW Impairment 545
]ON F. MILLER AND THOMAS KLEE
7 Discourse Organization and the Development of Reference to
Person, Space, and Time 194 21 Phonological Impairment 573
MAYA HICKMANN LAURENCE B. LEONARD
8 Bilingual Language Acquisition 219 22 Grammatical Impairment 603
ANNICK DE HouwER p AUL FLETCHER AND RICHARD INGHAM
 Parameters in Acquisition 11

structures of language. The kind of knowledge involved seems to be largely

underdeterrnined by the data, i.e. humans apparently have access to a sub-
1 Parameters in Acquisition stantial body of knowledge about language which cannot be "learned," since
the relevant information is not present in the empirical basis for such learning,
neither in the primary linguistic data nor in the context. Because of this dis-
crepancy between experience and knowledge, termed "Plato' s problem," the
JURGEN M. MEISEL implicit knowledge constituting the initial state of the language faculty is
claimed to be genetically transmitted. The theory of UG formulates this a priori
knowledge in terms of principles and parameters which determine the set of
possible human languages.
This approach represents an important difference, when compared to earlier
versions of generative grammar. In the "Standard Theory" (Chomsky, 1965),
for example, the initial state had been assigned the task of defining possible
rules, as well as providing for an evaluation procedure selecting the optimal
rule system. In the "Principles and Parameters Theory," as developed by
1 Introduction Chomsky (1981) and others, however, UG is not conceived of as a set of rules.
In fact, the notion that independent "rules" can be stipulated has been entirely
eliminated. The reason is that it became increasingly obvious, as pointed out
Parameter theory, a version of a theory of universal grammar (UG), has been by Williams (1987), that even if one were to succeed in writing descriptively
developed to account for universal as well as particular aspects of grammars. adequate grammars consisting of sets of rules, no progress would be made
The latter point, the fact that it is designed to explain language variation, towards explanatory adequacy unless the evaluation metric was discovered
makes it particularly attractive for research concerned with the development which enables the child to choose between the (possibly very large number)
of grammar in children, for it offers theoretical concepts and mechanisms of grammars which account equally well for the given data. Yet little progress
designed to deal not only with diversity among languages but also with vari- had been made with regard to insights concerning this evaluation metric.
ation across the range of possible interim grammars in language acquisition. Rather, the more elaborate the rules - transformations and others - became,
The present chapter introduces and discusses crucial notions of parameter the less obvious it was how children could acquire such rule systems and how
theory and outlines current developments. It thus focuses on variable proper- they would evaluate them in comparison to de~criptively equally adequate
ties of child grammars. ones. Reflections of this kind, in addition to empirical observations indicating
It should be noted that, in spite of the fact that this theory has been received that language development proceeds quite uniformly even under very differ-
very positively by many researchers, a fair number of the issues addressed in ent input conditions (e.g. multilingual environment, blindness, deafness) and
this chapter are far from being settled. In fact, even a number of rather fun- with severe neurological deficits, led to the conclusion that UG is part of the
damental questions about the epistemological status of parameters, as well as genetically endowed knowledge of humans.
about how exactly one should imagine them actually operating, remain unan- The well-foundedness of this assumption cannot be discussed here. What
swered. In what follows, I will summarize, in each case, what is generally matters for the present purpose is that the principles of UG are not rules
agreed upon and what I believe to be the crucial arguments in the ongoing comparable to relatively unconstrained rewriting rules of phrase structure
debate; whenever possible, I will suggest solutions for controversial issues, grammar or to transformations, as familiar from earlier versions of generative
even though, in more than one case, these may not represent the most widely theory. Phrase structure rules of the type S ➔ NP + VP, for example, have
accepted opinion. been replaced by the X-bar principle. It states that the internal structures of
sentential constituents (phrases) is identical in many ways. There is always a
head (X°, or "zero projection" ), and one finds in all phrases the same number
2 Principles and Parameters of Universal of structural layers 1 (projections), probably two, i.e. X1 and X2; the latter is,
thus, the maximal projection (XP). X° combines with a complement, itself a
Grammar maximal projection, to form an X1, and X1 combines with a specifier, again an
XP, to form the maximal projection of the head. X-bar theory claims that not
According to Chomsky (1986a, b ), the goal of the theory of UG is to model the only do all categories of a particular grammar conform to this schema, but it
specific mental capacity which enables humans to process the highly complex applies universally, across the languages of the world.

The Handbook of Child Language, First Edition. Edited by Paul Fletcher and Brian MacWhinney.
© 1995 Blackwell Publishing Ltd. Chapter 15 © 1995 Lila Gleitman and Jane Gillette.
Published 1995 by Blackwell Publishing Ltd.
12 Jurgen M. Meisel Parameters in Acquisition 13

In certain respects, however, some principles of UG do not account exhaus- secondary importance whether the various properties just mentioned are in-
tively for properties of grammars; they are "under-specified" (Williams, 1987), deed all related to the null-subject parameter; in fact, "free inversion" has been
offering several options, i.e. parameters are left unspecified. These parameters called into question. Only if one had to conclude that, in general, parameters
not only account for the obvious differences between grammars of different each determine individual properties of grammars, would the entire concept
languages; they also allow for variation in a more general fashion, including lose much of its theoretical attractiveness, and parameters would merely be
variation in language development. Note that instead of opening an infinite descriptive devices. Recent developments, indeed, give reason for apprehen-
space of variation, parameter theory restricts variation across grammars in a sion of this kind since one finds a certain inflation of new parameters which
rather tight way: the principle in question is given by UG, and the parameterized are proposed whenever a generalization about grammar does not hold univer-
options are limited. Take the example of phrase structure, again. X-bar theory sally. In view of this, it should be stated unequivocally that parameter theory
provides the general format of a syntactic phrase, the internal structure of as an explanatory framework for language acquisition depends to a large extent
which, however, may vary crosslinguistically. One such point of variability is on the assumption that parameterized options are defined on an abstract level
the position of the head of an XP; it may appear to the left or to the right of of grammar, typically triggering a number of apparently unrelated surface
its complement, e.g. as VO or OV in a VP. Note that, in this case, the para- effects.
meter offers just two options, head-initial or head-final. But this is not a theore- These as well as a number of other issues concerning parameter theory need
tically necessary restriction. It is, instead, quite likely that, in other instances, to be further investigated in order to reach broader consensus than is currently
a choice may be given between more tharr two possible parameter settings; the case. I will address some of them during the ensuing presentation, in as
see, for example, Manzini and Wexler (1987), who discuss a multivalued para- far as they relate to acquisition problems. This discussion will show that, most
meter concerning the binding domain for anaphors in a given language. Most importantly, parameter theory needs to be more strictly constrained. Various
parameters suggested so far, however, offer a choice between only two values. constraints have, indeed, been suggested as a result of analyses of developing
One important aspect of parameter theory which does not become apparent grammars, thus demonstrating that language acquisition studies can contrib-
by our example of head direction is that the setting of a parameter value may ute to grammatical theory.
cause a cluster of superficially unrelated grammatical properties to appear in
the language; see Chomsky (1981: 6). This can be illustrated by means of the
"null-subject" or "pro-drop" parameter, perhaps the most extensively studied
case. Whereas some languages allow empty subjects in tensed clauses, e.g.
3 Parameters of UG in Language Development
Spanish, others normally require this position to be filled lexically, e.g. Eng-
lish. The null-subject pararfl.eter specifies the grammatical conditions which Since UG, in generative theory, is conceived of as a set of principles and
must be met for this empty category "pro" to be allowed to occur. What parameters representing the innately specified initial state of the language
exactly these conditions are is the object of quite some controversy; see Jaeggli faculty, it is also understood as a crucial component of the language acquisi-
and Safir (1989) for a discussion of such problems. All authors seem to agree, tion device. This implies, most importantly, that a developing grammatical
however, that the nature of the inflectional system of the verb agreeing with system, at each stage of the process of change, only contains structures and
the subject plays a decisive role. Now, with respect to the issue discussed here, mechanisms which do not violate the principles of UG. In other words, the
the clustering of superficially unrelated phenomena, it has been suggested relevance of grammatical theory for language acquisition studies - and vice
that this parameter relates not only to null-subjects but to a number of other versa - depends on the continuity assumption. This is to say that "the child's
syntactic properties as well. One is that null-subject languages do not exhibit grammatical rules should be drawn from the same basic rule types, and be
expletive elements, such as English it and there in "it seems" or "there is"; on composed of primitive symbols from the same class, as the grammatical rules
the other hand, these languages allow "free inversion" of subjects in simple attributed to adults in standard linguistic investigations" (Pinker, 1984: 7).
sentences, as in Italian "ha mangiato Giovanni," "Giovanni ate." Other prop- The continuity assumption, however, leads to an apparent paradox: chil-
erties which have been claimed to be related to the null-subject parameter are dren's grammars differ from adult grammars but they are of the same type as
(a) "long wh-movement" of subjects, (b) empty resumptive pronouns in em- mature grammars. This is where parameter theory becomes relevant. It offers
bedded clauses, and (c) apparent violations of the [that-trace] filter (Chomsky, a solution for this puzzle and it helps to explain the incomplete knowledge
1981: 240ff). A more comprehensive account of these issues is not necessary about language one must attribute to children. Recall that grammatical theory
for the present purpose. The crucial point is that, according to parameter distinguishes between phenomena related to nonparameterized universal prin-
theory, a number of surface phenomena may depend on the setting of a single ciples and those that depend on parameters of UG or on language-specific
parameter, thus lending this concept considerable explanatory force. It is of properties of grammar. Only in the case of nonparameterized principles will
14 Jurgen M. Meisel Parameters in Acquisition 15

the continuity assumption predict that they are invariably present in child as time. The null-subject parameter can serve as an example, again. The discov-
well as in adult grammars. Concerning phenomena that depend on language- ery of one of its properties, e.g. the existence of expletives, is predicted to
specific properties, UG and the continuity assumption quite obviously have suffice to set the parameter to the correct value - in this case to [-null-subject];
nothing to say. The most interesting properties of grammar, from an acquisi- all other phenomena should emerge more or less simultaneously.
tional perspective, are the parameterized principles of UG, for children need
to find out how the values of the parameters are set in the language(s) they
are acquiring. Since UG allows for different solutions in these cases, one can, 4 The Initial State of Parameters
in fact, expect children to explore 2 the range of variation defined by para-
meterized options of UG, precisely because these choices are not determined
An essential, in fact the defining property of a grammatical parameter, is that
a priori but have to be made on the basis of information available in the input.
it needs to be specified for the adequate value required by the particular
A number of consequences follow from these assumptions and make it
language. This raises the question about its initial state. Asked differently:
possible to test them empirically. One claim, implicit in these hypotheses, is
what does a grammar look like before the parameter values have been set
that nonparameterized properties of grammar should exhibit no variation,
correctly? There seem to be two possibilities, and both have indeed been sug-
neither in the course of development nor interindividually, for they are not
gested in the literature:
learned, in the usual sense of the term; see Atkinson (1992) for an in-depth
discussion. With respect to the X-bar principle, for example, the claim is that 1 The parameter is set on <\ default value provided by UG; this need not
the child does not have to learn about the internal structure of the various be the correct option in terms of the target system but it can be found
syntactic phrases; since this principle is part of UG, it belongs to the child's in some adult grammar; in this case, parameters will have to be reset
implicit knowledge right from the start. Variation comes in with language- in some languages.
specific learning or as the result of parametric choice. The latter determines, 2 Both (or all) values are accessible simultaneously; this can be inter-
for example, the particular set of categories which are instantiated in a specific preted as meaning that the parameter is not set, at first, but the vari-
grammar, e.g. whether or not there are prepositions. Once this has been de- ous options are, for a limited time, simultaneously present in the
cided, the internal structure of a PP is given. Language-specific learning, in developing grammar.
this case, involves, among other things, putting together the list of lexical
items representing prepositions in the language, e.g. adding the equivalent Let us briefly look at both of these possibilities. Option (1) is probably the
of English "for," French "de," or German "fur" to it and/ or to the list of best known and perhaps also the most widely accepted one. Hyams (1986), for
complementizers. Thus, although parameterized properties of grammar, just example, argued that the null-subject parameter is initially set on the [+null-
like other principles of UG, are part of the knowledge present prior to expe- subject] (or [+pro-drop]) value as the default specification, early grammars
rience, the appropriate setting of parameters needs to be triggered by experi- thus generating null-subject languages (the "early English as Italian" hypoth-
ence, and it is conceivable that a child chooses the wrong option. esis). It is a well-known fact that children, during early phases of language
The distinction between learning and triggering of knowledge is, neverthe- acquisition, tend to omit elements which are obligatory in adult grammar.
less, an essential one; see Carroll (1989). Leaming requires more frequent According to this analysis, at least some of these cases of subject omission are
exposure to the input, possibly also over a longer period of time, and it prob- grammatically licensed rather than being motivated by the context or by per-
ably needs salient and unambiguous input data; but it will nevertheless ex- formance limitations. This version of the default assumption raises, however,
hibit more intra- and interindividual variation due to trial and error procedures, a number of questions. Leaving aside specific problems of this particular analy-
and it may still lead to inadequate preliminary hypotheses. Triggering differs sis,3 there still remain some more general, theoretical and empirical issues. The
from learning in each of these points; it is predicted to happen faster, requiring most far-reaching one concerns the problem of whether and how the para-
less frequent and less simple input data, and the developmental pattern is meter can be reset, for it has, indeed, been argued (Clahsen, 1991b; Muller,
expected to be much more uniform across individuals. 1994) that parameters cannot be reset; this will be discussed in more detail in
Another consequence of parameter theory follows from what has been said section 8, below.
about the clustering effect of different surface properties of a single parameter Another problematic issue is how to determine the default value among the
setting. The obvious implication is that the child needs to discover only one various options offered. Note that there exist at least two types of parameters,
of the properties indicating the correct setting of the parameter for the parti- one with nested values, the other with disjoint ones. As for the former (nested),
cular language, and all others will be automatically available. The empirical fixing the parameter on one value yields a grammar which generates a larger
prediction is that the various phenomena will appear within a short period of set of structures, as compared to the grammar which results from setting the
16 Jurgen M. Meisel Parameters in Acquisition 17

parameter on the other value. The option [+pro-drop], for example, allows for
Figure 1.1
sentences with or without overt subjects, since, in null-subject languages, sub-
jects may also be lexically realized. The [-pro-drop) option, on the other hand,
does not, normally, tolerate subjects to be lexically empty. The set of structures
generated by the latter option thus seems to be a subset of the one generated
by the former. The "subset principle," as first suggested by Berwick (1985) for
language acquisition, states that the learner, as a first choice, must select the This is not to say that multiple settings are possible; rather, as suggested by
narrowest possible language consistent with the data. In fact, ordering "para- Lebeaux (1988: 179ff; 1990), the initial grammar, Go, allows access to the de-
meter values according to the subset relations of the languages that the values fault as well as to alternative values provided by UG before the parc1II\eter is
generate" (Manzini and Wexler, 1987: 414) has been claimed to be "the one actually set on a specific value. This amounts to saying that G0 preserves the
fundamental principle of learnability" (ibid.: 425). The argument contained initial state in which a value (X) is ordered before the default value Y.4 In the
in the subset principle, therefore, is that the default value of the parameter setting G 1, option X is erased, leaving only the default value Y. In the setting
should be the one generating the smaller set of structures. This will enable the G2, the brackets indicating optionality are erased, X applies and overrides Y;
child, on the basis of positive evidence alone, to realize that the parameter see figure 1.1.
needs to be reset. To illustrate this point, if the pro-drop parameter is set on In other words, parameter setting leads to the elimination of alternative op-
the negative option as the default value, a child acquiring Spanish should tions. But as long as the child has not yet set the parameter, the child "falls
merely have to realize that sentences lacking overt subjects are used in his or into" the default grammar, i.e. value Y is active as a default, rather than G0
her environment to be able to reset the parameter. In the reverse case, i.e. with being specified for Y, and alternative options remain available. Note that, even
the parameter set on [+pro-drop], a child learning English cannot easily arrive though UG principles are not violated, the early system G0 is not a possible
at the same conclusion. Only negative evidence, i.e. information indicating final grammar because the parameter has not been set; this has been pointed
that lexically empty subjects are not tolerated, could enable the learner to do out by Lebeaux (1988) himself and also by Roeper (1992). What is attractive
so. Yet negative evidence is normally not available, for children's grammatical about this approach, however, is that it can do without the rather problematic
errors are usually not corrected (see Brown and Hanlon, 1970), and correc- notion of parameter "resetting." It furthermore predicts that initial grammars
tions, even if given, are not easily understood and used (see Pinker, 1989). make available all values of a parameter, a solution which seems to be re-
In the absence of negative evidence and if positive evidence is not sufficient, quired for other reasons as well.
the child would have to become aware of the fact that one of the options This becomes evident from work by Valian (1988: 12ff; 1990a: 122ff), al-
offered by the grammar actually never appears in the input. This kind of though her proposal is not identical to that of Lebeaux. Expanding on a sug-
evidence has been referred to as "indirect negative evidence" (Chomsky, 1981). gestion by Berwick and Weinberg (1984), she argues that all parametric options
As Valian (1988, 1990a) points out, however, the problem with indirect neg- need to be available in early grammar for parsing considerations. Her argu-
ative evidence is that there is no principled way of knowing how long one ments are based on the idea that the child, right from the earliest point of
should wait before concluding that a specific structure will never emerge in grammatical development, has to be able to parse and assign a structure to a
the data. Therefore, the learner can never be sure that the corresponding sentence he or she hears. Since the parser is fed by the grammar it is also
parameter option may indeed be ruled out as a possibility for the grammar limited by the grammar. Thus, if parameters are initially set on one of the
being acquired. possible values, the parser cannot process structures generated by grammars
As for parameters with disjoint values, like the head-direction parameter, set on the other value. If, for example, a child acquiring Spanish has first set
similar problems do not arise, for no subset relationship holds. The two pos- the null-subject parameter on the [-null-subject] value, potentially disconfirming
sible grammars generate two disjoint sets of structures, e.g. with head-initial data cannot actually disconfirm the option to be changed, since the child's
or head-final categories. This, however, leads to the consequence that one does current grammar does not provide the parser with the necessary information.
not have principled reasons for deciding which value represents the default "Only if we give the parser access to the missing, correct, value, can the datum
setting. Whether or not a default option for parameters with disjoint values is serve the function of contradicting the incorrect value" (Valian, 1988: 14).
at all necessary is indeed not obvious, since learnability considerations of the In sum, the assumption that default options exist for at least some para-
sort discussed for nested parameters do not apply. In order to achieve a uni- meters appears to be plausible, but it is problematic to conclude that para-
form treatment of all parameters, one might nevertheless opt for this solution meters are already set on the default value at the initial state of language
pending further empirical scrutiny. development. Rather, first grammars need to allow for the possibility of ac-
Let us now consider option (2), where all values are accessible simultaneously. cessing all options before fixing one value.
18 Jiirgen M. Meisel Parameters in Acquisition 19

may not be acceptable to simply ignore the nongrammatical utterances as

5 Triggering Parameter Setting irrelevant. In the present context, these facts are of utmost importance. The
very notion of triggering implies that, in principle, an extremely small number
The triggering of parameter setting is perhaps the most difficult issue to be of examples should suffice to set the parameter on a specific value; see section
dealt with here, for it requires an assessment of the role of the input - a 3, above. Consequently, 20 percent or even 4 percent of ungrammatical sen-
delicate problem for a theory of language acquisition which holds as one of tences are likely to make the child's task impossible; he or she has to find out
its fundamental hypotheses that much of the relevant information is not whether a given sentence to which he or she cannot assign a grammatical
contained in the primary linguistic data. Remember the distinction made in structure is indeed ungrammatical or whether the grammar or the parser is
sectiort 3, above, between different acquisitional mechanisms: instantiation of inadequate. Valian (1988), discussing this issue in some detail, concludes that
UG principles, learning of language-specific features, and triggering of para- parameter setting theory, if it cannot handle this problem, will be "O percent
meters. Only in the latter do we find the subtle interplay between know- accurate."
ledge available prior to experience and information contained in the data. Yet Valian (1988; 1990a) argues that the child needs to be equipped with a
many accounts attribute properties of inductive learning to triggering (see mechanism evaluating the significance of parser failure or parser success. The
Atkinson, 1992, chapters 8-10)- probably a residue of trial and error learning, solution she advocates consists in attributing to the child the ability to (uncon-
as pointed out by Haider (1993). But it is quite likely that, in contrast to this, sciously) "run experiments" by "floating an utterance" and subsequently com-
triggering involves parametric choices leading to consequences from which paring it with the adult's reply. This hypothesis gains in plausibility by the
the learner may not be able to retreat. Importantly, the standard assumption observation that adults tend to reply to children using an utterance which is
is that triggering this knowledge, as opposed to learning, requires less robust lexically and structurally similar to the child's sentence, and they seem to do
data and less exposure to the input. This raises the question of what exactly so more frequently than when responding to other adults. Valian (1990a: 135)
is needed to serve as triggering information. found that, in addition to 5 percent verbatim replies, approximately 30 percent
One point has already been raised in section 4, namely that it may plausibly of the adults' replies are similar to the child's previous utterances, and about
be assumed that primary linguistic data only contain positive evidence. Neg- two thirds of these included an implicit correction of the child's error. The
ative evidence and indirect negative evidence, to the extent that they are at idea is that the comparison process will enable the child to identify
all present in a child's linguistic environment, are apparently not used in a ungrammatical input, and it avoids the problem normally tied to the use of
systematic fashion. indirect negative input, i.e. that there is no way to tell when to stop searching
A second point concerns the reliability of the input data. There are, in fact, for evidence. Assuming a comparison process, the child may reasonably
reasons to suspect that their reliability is flawed. First of all, it has been argued expect the phenomenon under consideration to emerge within a short time
that surface properties of sentences can be ambiguous in terms of parameter period.
values and therefore misleading with respect to the correct setting. This is the Another solution to the problem of how to deal with ambiguous and partly
case when specific lexical items give rise to exceptions to syntactic generaliza- ungrammatical data is to assume that parameter setting is triggered only by
tions; see Truscott and Wexler (1989). Another example is that of the omission narrowly defined properties of the data, not by all features which are possibly
of subjects in some languages. English and German, for example, tolerate null relatable to a given parameter value. Roeper and Weissenborn (1990: 151), for
subjects in certain contexts although they do not exhibit grammatical proper- example, make the following claim: "If a given parameter is marked by sev-
ties of pro-drop languages. This can be interpreted as ambiguous evidence for eral features, then there will be one unique trigger specified in UG"; see also
the null-subject parameter; see Weissenborn (1992: 286). Secondly, and more Roeper and de Villiers (1992), who discuss the existence of such a unique
seriously perhaps, adult language even when used with children is not always trigger (or "telltale"; Valian, 1990b). According to this hypothesis, the presence
grammatical. It is not only that performance errors occur; adults use utter- of a specific feature in the data indicates unambiguously that the parameter
ances which they themselves judge as nongrammatical. Furthermore, some has to be set on a given value. The pro-drop parameter can serve as an exam-
types of ungrammatical sentences are perfectly acceptable in specific contexts, ple again. We have seen that overtly missing subjects do not unambiguously
just as certain types of grammatical sentences are not always acceptable for point to the [+pro-drop] value. The use of expletive subjects, on the other
native speakers. Valian (1990a: 120ff), for example, found in a survey of her hand, seems to provide more reliable information to the effect that the lan-
data from American English children and parents that 4 percent of parental guage is not a null-subject language; see Hyams (1986, 1989). It remains to be
replies to children were ungrammatical, and 16 percent were acceptable but seen whether a unique trigger can indeed be identified for each parameter.5
not fully grammatical, i.e. parents say things like "Want your lunch now?," I will return below to the question of what the structural properties of the
"Raining hard," etc. Now, for linguistic analyses of adult language it may or trigger should be expected to be.
20 Jurgen M. Meisel Parameters in Acquisition 21

How, then, can we decide between the two solutions? The use of indirect criteria at all. In view of this, I find it preferable to stick to my suggestion to
negative evidence, as advocated by Valian, implies that there is hypothesis first attempt to solve the problem in a way consistent with grammatical theory.
testing involved, after all. Although this is certainly a possibility, it is one This amounts to saying that further grammatical constraints should be im-
which weakens the theoretical position taken by parameter theory, shifting posed on what may act as a trigger. Adopting the "unique trigger" proposal
from a grammatical explanation to one based on strategies of language use. has been a first step in this direction; eliminating the possibility of parameter
The unique trigger solution is not only more in tune with the spirit of para- resetting (Clahsen, 1991b) leads us further down this way. In what follows, a
meter theory as a grammar-based explanation; it also seems to be the more number of suggestions will be presented which drastically limit the range of
parsimonious approach, not having to rely on the existence of poorly under- possible triggers in terms of their grammatical properties; see section 8. Fur-
stood additional procedures of hypothesis testing. Ultimately, however, we ther restrictions are derived from the assumption that a particular feature can
may hope to be able to decide on empirical grounds between the two possib- only act as a trigger if certain prerequisites are already met by other parts of
ilities, for they make rather different empirical predictions. If hypothesis test- the developing grammar; see section 6.
ing is involved, one should expect to find interindividual variation; children Yet the better we succeed in more strictly defining possible triggers, the
may also go wrong, initially, before setting a parameter correctly, and this more acute becomes the second concern, i.e. the question whether these items
leads to intraindividual variation. The unique trigger solution does not predict will be accessible to all learners. Truscott and Wexler (1989: 160), for example,
variation of this sort, even though one cannot exclude, at this point, the pos- object that if one were to assume that a specific phenomenon reliably indicates
sibility that due to data ambiguity a feature might be identified erroneously as the correct setting of a parameter, "this unknown phenomenon would have to
the "unique trigger." Yet even in this case, one does not expect to find oscil- occur with a fairly high frequency (high enough that all learners of the lan-
lation between different parameter values but rather than the learner retreats guage would encounter it)." The latter point seems to be well taken: triggering
via lexical learning; see section 8 on parameter "resetting." data must be of a sort which may plausibly be expected to be accessible to
The third point to be raised here concerns the frequency of occurrence of every child. But this has little to do with frequency. Rather, what matters is
a feature and the length of exposure of the learner to the relevant data. As that triggering elements should not appear only in what Lightfoot (1991) calls
mentioned above, the theory claims that minimal exposure to the data should "exotic" contexts. In this sense it is reasonable to ask for "robust" data; fre-
be sufficient for parameter setting. Ideally, a single example encountered in quency of occurrence, however, does not appear to be relevant. The question
the input could indeed suffice. While it is obvious that this strong idealization thus remains, what may contribute to "robustness."
cannot be maintained under realistic conditions, not much is known as to how This leads to the fourth and last point. Summarizing what has been dis-
much exactly is necessary. Clahsen (1991b), for example, simply observes that cussed so far, one can say that triggering data must materialize as positive
the child "seems to require a certain amount of positive evidence to fix a evidence easily accessible to the child. In the current theoretical framework,
parameter at a particular value." Randall (1992: 100) goes one step further and their deficiency with regard to reliability cannot be amended in a satisfactory
postulates the existence of a "trigger threshold," i.e. that a "sufficient number way by frequency requirements. A more promising approach is to define ac-
of tokens" need to occur, admitting, however, that "we know surprisingly cessibility in structural terms. Structural properties of the data have only been
little about it." Lightfoot (1991) even goes so far as to claim that data must be mentioned in passing, up to now. It should be obvious, however, that this is
salient and frequent in order to be able to act as triggers. Two sorts of concerns what this discussion is all about. Only to the extent that the child is capable
apparently lead to this or similar conclusions. The first is related to what has of assigning a structure to an element encountered in the input can this feature
been said above about ambiguous and ungrammatical data. As Lightfoot (1991: be integrated into the grammar of the learner. Yet it is most likely that during
14) phrases it, the trigger should be "something less than the total linguistic the period of linguistic development when parameter values are set, the child's
experience," since occasional exposure to divergent forms, e.g. a different dialect capacity to (implicitly) perform such an analysis is restricted to a certain type
spoken by a houseguest, or the language use of a second language learner, of data, not ori.ly in terms of reliability, etc., but also with regard to grammati-
"normally has no noticeable effect on a child's linguistic development." cal complexity. This is suggested by Lightfoot (1989, 1991); he proposes that
Note, however, that these are merely empirical observations to the effect degree-0 learnability is sufficient for parameter setting, picking up on the
that not every form present in the input automatically triggers the parameter learnability proof by Wexler and Culicover (1980), who demonstrated that it
for which it is apparently relevant, that there is no evidence for constant is possible, in principle, for a child to identify the grammar of a language if
switching of parameter settings, and that parameterized acquisition, although it is exposed to sentences with at most two levels of embedding (degree-2
faster than inductive learning, does take some time, after all. Notice further learnability). Morgan (1986), based on different assumptions about what con-
that no empirical or-theoretical criteria are presented identifying the threshold stitutes the child's input, succeeded in giving proof for degree-1 learnability.
level. In fact, nothing obliges us to accept a solution based on quantitative Lightfoot does not aim at an improvement of the learnability proof (see
22 Jurgen M. Meisel Parameters in Acquisition 23

Williams, 1989). Rather, he attempts to show that in order to discover the

positive evidence necessary to decide on the settings of a number of para- 6 A Developmental Schedule: Maturation or
meters, it is sufficient to have access to degree-0 "plus a little," i.e. only main Ordering of Parameters?
clause structures plus the front of the lower clause need to be accessible. The
reason why Lightfoot (1989, 1991) arrives at this somewhat peculiar "zero-
plus" formula is that he defines structural range not in terms of clauses but The preceding section dealt with difficulties in determining what kind of
of binding domains. For the present purposes, however, it is possible to retain evidence is required for a parameter to be set to the correct value. Surpris-
a definition based on clauses. It has, in fact, been argued by several authors, ingly, perhaps, an equally puzzling problem emerges when one adopts the
e.g. Baker (1989), Wasow (1989), Wilkins (1989), that degree-1 is more ad- opposite perspective. Assuming we are able to identify the elements which
equate, and that adopting it does not, in fact, weaken or strengthen Lightfoot's can act as triggers, the question arises why the triggering does not happen
hypothesis. immediately, as soon as the relevant piece of evidence appears in the child's
This brings us to another proposal concerning the structural range within input. This is Borer and Wexler's (1987: 128) "triggering problem." The crucial
which triggers are to be found, the "subordinate clause strategy" suggested by issue here is to account for the fact that the data required to trigger the setting
Roeper (1973), according to which children pay particular attention to subor- of parameter values are, presumably, present in the child's linguistic environ-
dinate clauses. The idea was inspired by Emonds' (1970) finding that structure ment from early on; yet many structural properties appear in a sequential
preservation is strictly adhered to in subordinate clauses. Note that this strat- order which is uniform across individuals and, in many respects, even
egy represents a more radical version of the Degree-1 Hypothesis, for it claims crosslinguistically. Why, then, do triggers operate at a given point, why not
that certain clues about how to set a parameter are to be encountered in earlier or later? The obvious answer seems to be that language development
subordinate clauses, exclusively. Roeper and Weissenborn (1990) elaborate on follows a specific underlying logic. The nature of this logic, however, is quite
this; they advocate the Unique Trigger Hypothesis, but specify that it applies controversial. Whereas some authors believe that neurological maturation is
in subordinate clauses, at least as far as some parameters are concerned. An the underlying cause, others contend that it is a grammar-internal phenom-
example of this is the null-subject parameter, again. Arguing that all the po- enon by which parameters are ordered.
tential triggers proposed, including the presence of expletives, are not reliable, The Maturation Hypothesis was first developed by Felix (1984), but similar
they attempt to show that unambiguous information can indeed be detected ideas had already been proposed by Gleitman (1981) and White (1982). Felix
in embedded clauses where the occurrence of null-subjects is restricted to true (1984, 1987, 1992) claims that UG principles emerge according to an innately
pro-drop languages. Another example is the underlying clause-final position specified maturational schedule. Under his account, early grammars will, in
of verbs in languages like Dutch and German where finite verbs surface in contradiction to the continuity assumption (see section 3), exhibit properties
final position in subordinates, but in second position in root sentences. From which are in conflict with the as yet latent UG principles (Felix, 1987: 115)
these observations one might conclude that the language learner must be sen- before biological maturation makes the relevant principles accessible to the
sitive to the root/nonroot distinction right from the start of grammatical de- child. This, then, leads to a reorganization of the interim grammar in such a
velopment.6 Unfortunately, however, the evidence is not entirely convincing way as to make it conform to the newly developed principles. Another version
in either case. Lightfoot (1991: 52ff.) demonstrates that a number of unembed- of the Maturation Hypothesis has been suggested by Wexler and his asso-
ded indicators appear in so-called verb-second languages, enough for the child ciates. Borer and Wexler (1987: 124) state that "certain principles mature. The
to be able to discover underlying OV order. As for zero subjects, Rothweiler principles are not available at certain stages of the child's development, and
(1989) has found a number of lexically empty referential subjects in subordi- they are available at a later stage." The example they discuss is the formation
nate clauses of German children. Following Roeper and Weissenborn (1990: of A-chains; informally speaking, this refers to the mechanism necessary for
156), this should not be possible, since the null-subject parameter must have assigning a thematic role to a moved NP. It should, therefore, be impossible
been set correctly once the children are able to use embedded clauses. for the child to assign thematic roles nonlocally if A-chain formation is not
In sum, then, information contained in subordinate clauses is clearly neces- available; as a consequence, child grammars at this point of development should
sary for at least some parameters to be set on the correct value. It is, however, not be able to analyze constructions which depend on this mechanism, e.g.
doubtful whether children are able to exploit the matrix. v. embedded clause passives. Note that it is not clear whether this example really illustrates a case
distinction from very early on, and it also remains a controversial issue whether of nonavai_lability of UG principles. Subsequent work clarifies, in fact, that it
the child really focuses on subordinate clauses in cases of ambiguous data should not be interpreted as meaning that UG principles themselves are sub-
with respect to potential triggers. Information beyond the degree-1 complex- ject to maturation. Instead, Wexler (1990: 105) suggests a "VG-Constrained
ity, on the other hand, is certainly not required for parameter setting. Maturation" hypothesis according to which "UG constrains every stage of
24 Jurgen M. Meisel Parameters in Acquisition 25

child development, but certain constructs may not be available at certain ages." parameter, and the subset principle specifies the markedness hierarchy for
Under this hypothesis, learning constraints are responsible for restricting the these values. They further propose the "Subset Condition" (see also Manzini
availability of UG principles; these restrictions are successively removed as a and Wexler, 1987), which ensures that the languages generated by two or
result of physical maturation, thus giving the child access to more principles. more values of a parameter are ordered by proper inclusion. Yet although this,
The major difference between this weak version of the Maturation Hypothesis together with the subset principle, imposes a strict order on the choice of
and its stronger counterpart is that the former predicts intermediate grammars values of a specific parameter, they strongly reject the idea of ordered para-
to be consistent with UG at every point of development. meters, i.e. of interdependence of parameters.
Nonmaturational accounts of the developmental schedule look for grammar- At this point, in may be useful to pause briefly and to attempt a contrast-
internal ordering principles rather than referring to physical maturation. In ing evaluation of maturation and grammatical ordering hypotheses. The first
this vein, a number of authors, notably Lebeaux (1987), Roeper and Weissenborn point to observe is that no one is able to offer independent evidence for the
(1990), Roeper and de Villiers (1992), and Weissenborn (1992), have suggested respective mechanisms invoked as explanatory factors. In other words, we are
that parameter settings are interdependent, i.e. the value on which one para- left where we started from : grammatical development is observed to follow
meter is set depends on the setting of another one. Ordering here means that a uniform sequential pattern which asks for an explanation. The second step
"A decision in Parameter B is not executed until Parameter A is fixed . .. In could thus be to ask whether the various approaches attain descriptive ad-
effect, this constitutes an input filter: a sentence that involves Parameter B will equacy. Leaving details aside, they do seem to succeed here equally well.
simply not receive a full grammatical analysis until the acquisition system is Critics of maturational approaches, for example, generally concede that matu-
ready for revision (after Parameter A has been set)" (Roeper and de Villiers, ration might explain the properties of child grammar discussed by the authors
1992: 193). This can be illustrated by an example discussed by Roeper and under review - objecting at the same time that maturation is not needed as an
Weissenborn (1990); they claim that the correct setting of the pro-drop param- explanation; see Weinberg (1987), Guilfoyle and Noonan (1988), and Clahsen
eter depends on the parameter determining whether or not there is overt wh- (1991, 1992), among others. As far as I can see there is only one major point
movement. Weissenborn (1992) specifies that the crucial fact here is whether where empirically distinguishable claims are made, namely with respect to
COMP is filled lexically, either by a complementizer in the head position of CP possible violations of principles of UG. Under the strong Maturational Hy-
or by a wh-phrase in SpecCP. The line of reasoning then runs roughly as pothesis, they are expected to happen, whereas everybody else denies their
follows: the presence of complementizer phrases is itself parameterized; chil- existence. The aforementioned critics, for example, claim that the apparent
dren's initial sentence structures, like adult Chinese or Japanese, lack CPs; violations can all be accounted for in a UC-conforming fashion. For the time
overt complementizers or wh-movement trigger the implementation of a CP being, I suggest we accept this as a temporary solution although some doubts
system in the child's grammar, and only in structures of this kind does one remain, particularly with respect to child language at the so-called two word
find a unique trigger for the pro-drop parameter; see the preceding section. stage; see Felix (1987, 1992). Given this state of affairs, I suggest proceeding in
Note that in all these cases ordering is defined as intrinsic order/ and this the, by now, customary way, resorting to Occam's razor, i.e. to evaluate those
is not restricted to parameterized properties of grammars. New grammatical candidates most highly which can do without an additional apparatus. Ac-
features need to be integrated into the existing system, and unless this is cording to this logic, the least desirable approaches are those requiring extrin-
possible, i.e. the new feature fits into the system, the learner is "blind" towards sic ordering and those postulating learning principles whose sole function is
what the input offers. With respect to parameters again, the situation is quite to block access to grammatical knowledge which is supposedly present but
similar, for triggering depends on previously acquired grammatical know- cannot be used. The solution to be preferred, then, is the one which attempts
ledge, as well, for possible triggers cannot be recognized as such unless they to explain the observed sequences in terms of grammatical properties without
can be assigned a grammatical structure. Intrinsic ordering of this type ap- having recourse to extrinsic ordering, maturation schedules, or markedness
pears to be uncontroversial. hierarchies. Note that this verdict is tentative and will be withdrawn as soon
Strictly speaking, however, we have introduced at least one case of extrinsic as one can (a) give compelling empirical evidence to the contrary, and/ or (b)
ordering, namely default settings of parameters. The nondefault option can be independently motivate the additional mechanisms.
interpreted as the marked setting, yet markedness (or what counts as un- It should be pointed out, in this connection, that assuming maturation is, of
marked, i.e. the default case) is not defined by grammar-internal but by some course, by no means an implausible hypothesis, given that our theory empha-
external criteria. At best, it is j~stified by the subset principle (see section 4), sizes strongly the biological basis of language, and biological maturation is
but this too is a principle of learnability, not of grammar. Indeed, Wexler and quite uncontroversial in other areas for children of this age (Borer and Wexler,
Manzini (1987) view the theory of markedness which they outline as part of 1987; Felix, 1987). It is, nevertheless, necessary to indicate what the physiologi-
a "learning module." Markedness here applies to the possible values of a cal basis (neurological or other) for these claims might be. In fact, one such
26 Jurgen M. Meisel Parameters in Acquisition 27

piece of evidence may have been found by Greenfield (1991). Discussing find- below, in section 8.1. Let me merely add at this point that, in this framework,
ings from neurophysiological and neuroanatomical research, she concludes lexical learning is claimed to solve the triggering problem (see Clahsen, 1991b),
that the development of brain connections provides the foundation for struc- i.e. increases in the learner's lexicon (including morphological items) are said
tural development in language. A crucial point is that this research focuses on to induce restructuring of intermediate grammars.
neural circuits, not on brain areas alone. Specific circuits connecting various
cortical areas develop as a result of gradual differentiation by which "multiple
short-range connections are 'pruned' to fewer, more specific and longer-range 7 Setting Parameter Values
connections" (Greenfield, 1991: 544). In the case of Broca's area, dendritic growth
leads to a differentiation of two functionally distinct neighboring areas, and
this happens precisely during the period when complex grammatical struc- In the preceding sections, the initial state of parameters and the nature of
tures emerge in child language, beginning at around age two. Broca's area is triggering data have been discussed in some detail, as well as the question of
now starting to receive input from the anterior prefrontal area. "If this is why the presence of potential triggers in the input does not lead to instanta-
confirmed by further research, then language is not modular at birth or even neous fixing of all relevant parameters simultaneously. What remains to be
at the beginning of language development; it becomes increasingly modular asked is how triggering can be thought to actually operate. It has been argued
with age and neural differentiation" (Greenfield, 1991: 550). One way to inter- that parameters determine knowledge which cannot be learned inductively;
pret this is to say that UG becomes available at around age 2;0 as a result of several options are latently present, one of which must be chosen. This choice,
maturation. Such a view is in accordance with the theory developed by it has been claimed, is triggered when an analysis of the data reveals the
Bickerton (1990b), who distinguishes between protolanguage and language, presence of a triggering element. I furthermore argued that assuming the ex-
only the latter being the product of grammatical knowledge, with language istence of a unique trigger for each parameter appears to be the most pro-
proper emerging at about age two.8 Assuming, then, that the findings reported mising solution for the intricate triggering problem. But the role of the data
by Greenfield justify the claim that biological maturation makes UG accessible emerges once more as a potential problem, this time with respect to the ques-
to the child at approximately age 2;0, this provides independent evidence for tion of whether the relationship between the input data and the setting of the
the Maturation Hypothesis. Notice, however, that it only explains why early parameter value is deterministic in nature or not.
language uses are not constrained by UG, as claimed by Felix (1987) and Surprisingly enough, only a few studies address these issues explicitly,
Bickerton (1990b). In order to justify the hypothesis that principles within UG notably Valian (1988, 1990a), Haider (1993: 13), and Atkinson (1992: 206ff), and
mature successively, as suggested by Felix, evidence of a different kind is I will therefore only mention some basic points. Haider, for example, makes
required. some interesting observations which may help to clarify the problem under
Our discussion has thus led to the conclusion that the best candidate for an scrutiny, even though he wants to eliminate the notion of parameter fixing
adequate explanation of acquisitional sequences is an approach which relies altogether, for reasons beyond the scope of this chapter.9 He points out that
primarily on intrinsic ordering. As a consequence, one might opt for what fixing a parameter requires three distinct acts: "A particular property in the
Guilfoyle and Noonan (1988) termed the "Structure Building Hypothesis"; see input data must be identified, the property recognized in the data must be
also Radford (1986, 1990). It is based on the hypothesis that early grammars identified as relevant for setting the parameter of a specific principle, and,
only contain projections of referential categories, most importantly verbs and eventually, the parameter of the principle must be set to the value correspond-
nouns. Functional categories, on the other hand, are implemented in the course ing to the input data property" (Haider, 1993: 5). This amounts to saying that,
of further acquisition. This means, for example, that early grammars lack CP as a first step, the utterance will have to be parsed, and a structural represen-
and possibly also IP. Grammatical properties related to these categories, like tation must be constructed, independently of the fixing of parameters. It is this
certain word order phenomena resulting from movement to IP or CP, the structure to which the parameter setting mechanism can apply. Since the parser
finite/ infinitive distinction handled by INFL, case assignment depending on can be assumed to be fed by the grammar, it should be obvious that the first
properties of INFL, and so forth, will therefore also be absent during early step already requires grammatical knowledge. The more controversial point
phases, and parameters related to functional categories obviously cannot be here is whether steps two and three will indeed have to be distinguished.
set until these elements are instantiated in the developing grammar. More- One way to view this is to imagine that parameters function like switches,
over, since in most or all cases functional categories are interdependent in metaphorically speaking, tripped by incoming data which, thereby, set them
the way they are implemented in the grammar of a specific language, tying properly in a quasiautomatic fashion. If this is correct, the· child does not
parameter setting to the emergence of these elements provides us with an play an active part in this process. An alternative view would be to attribute
intrinsically motivated developmental schedule. I will return to this issue to the child a more active role. Valian (1988), for example, argues that simple
28 Jurgen M. Meisel . Parameters in Acquisition 29

exposure to the language is insufficient and that learning is required, where subgrammars can coexist simultaneously, and it is plausible to assume that
learning is understood as nonconscious hypothesis testing. this has also been the case within the grammatical competence of individuals,
In my understanding, there are a number of reasons, already alluded to in not only across the linguistic community, for this is what we find in bilinguals
section, to prefer the solution claiming that parameter setting is deductive as well, and even in bidialectal speakers. In fact, the individual's ability to set
learning, i.e. it happens in a quasiautomatic fashion. In fact, this seems to one parameter on contradictory values may very well be a necessary condition
follow logically from adopting the notion of a unique trigger. It also appears for becoming a multilingual. In other words, it is precisely this type of evi-
to be more in line with the theoretical framework of parameter theory, given dence for conflicting parameter settings which enables the bilingual child to
that it strictly reduces the role of inductive learning. Remember that the in- conclude that separate grammatical systems are underlying the incoming data.
sight that learning as hypothesis testing cannot account for crucial aspects of In conclusion, then, the view advocated here interprets parameter setting as
language development was one of the major reasons for adopting the Princi- a quasiautomatic process triggered by a telltale trigger detected in the gram-
ples and Parameters Theory. Consequently, a deterministic.approach to para- matical structure which has been assigned to the data. In exceptional cases,
meter setting is to be preferred, unless compelling evidence to the contrary is this does not exclude the possibility of incorrect decisions, and it predicts that
found. Note that it also seems to be the more commonly held view to assume conflicting evidence concerning the value to be chosen will serve as a trigger
that setting the values of a parameter is a causal effect of properly analyzed to differentiate two grammatical systems.
and recognized incoming data.
To offer another metaphor, one may think of each parameter value as a lock
into which only one key fits, i.e. the specific grammatical feature emerging
from an analysis of the data functioning as the telltale trigger. How can this 8 Constraints on Parameter Theory
be reconciled with the important role of variation, not only in language acqui-
sition but also in historical change? A theory of grammar failing to assign an It should have become apparent by now that parameter theory, in spite of its
adequate status to variability cannot hope to account satisfactorily for gram- success in explaining important aspects of grammatical development, still needs
matical development. The notion of parameter has been introduced to explain to be elaborated in more detail with respect to a number of crucial theoretical
crosslinguistic variation, and it is obviously also the locus where the possibil- assumptions. Most importantly, tighter constraints have to be imposed in order
ity of ontogenetic and historic development should be explained. Yet variabil- to make all of the major claims of the theory amenable to empirical scrutiny.
ity need not be a property of the parameter itself nor of the parameter setting In the remainder of this contribution, I will discuss briefly two important
mechanism. Instead, variability exists in language use and therefore also in the constraints proposed recently, namely that only functional heads should be
potential triggering data. As has been discussed at some length in section, this parameterized and that resetting should be excluded.
may even lead to ambiguity with respect to the correct choice of parameter
values. As opposed to most authors who believe that the notion of a "telltale"
or "unique" trigger is incompatible with the idea of incorrect settings, e.g. 8.1 Parameterization of (functional) heads
Roeper and Weissenborn (1990), I want to argue that telltale triggers do not
exclude the possibility of setting parameters incorrectly. The reason is that The first proposal concerns the issue of what parts of grammar can be para-
although a particular structural property may indicate unambiguously that a meterized. It appears to be desirable to restrict parameterization to a strictly
specific value must be chosen, the data may very well be structurally ambigu- limited domain of UG. Initially, it had been hypothesized that universal prin-
ous, and this may lead, in exceptional cases, to wrong analyses and conse- ciples themselves are parameterized; see Chomsky (1981, 1986a). Take the
quently also to wrong settings. I will return to this point in section 8.2. This principle of subjacency 10 as an example; informally speaking, it states that an
is not all: we even have to allow for the possibility that the input data may element, e.g. a wh-phrase, may not be moved over more than one bounding
contain conflicting evidence, i.e. that they can be analyzed by the child in a node. Whereas the principle as such seemed to hold universally, what would
way leading to the conclusion that there exists a telltale trigger for both pa- count as a bounding node in the grammar of a specific language had been
rameter values. What, at first sight, appears to be selfcontradicting is in fact argued to be subject to parameterization. In English grammar, for example,
a welcome result, I believe. To illustrate this point, let me mention French as NP and IP are bounding nodes, whereas in Italian CP, but not IP, functions as
an example. Old French was a null-subject language, but Modem Standard such. Yet in a number of studies, Wexler and others, e.g. Wexler and Manzini
French has lost this syntactic property. During an extended period of time, (1987), Truscott and Wexler (1989), demonstrated that the value of a parameter
however, French exhibited characteristics of both [+pro-drop] and [-pro-drop] "cannot be associated with grammars as a whole but rather must be associated
languages. In other words, language change shows that different grammars or with single lexical items" (Manzini and Wexler, 1987: 424). It was Borer (1984)
30 Jiirgen M. Meisel Parameters in Acquisition 31

who first suggested moving the burden of parametric choice from the compu- that grammars of various languages differ with respect to whether AGR ap-
tational component to the lexicon. Picking up on this, Manzini and Wexler pears above TNS or vice versa. This contradicts the widely held belief in
(1987) proposed their "Lexical Parameterization Hypothesis." Note, however, studies on UG that clause structure is identical for all languages at an abstract
that although this means that the value of a parameter is associated with level. The implications for language acquisition are obvious. According to the
properties of individual lexical items, rather than being defined over the en- Structure Building Hypothesis, the child has to discover which functional
tire grammar, it still implies that the principle itself is subject to parametric categories, out of a limited set offered by UG, are instantiated in the particular
variation. grammar to be acquired. If Ouhalla's claim can be maintained, it follows that
A more radical change has been introduced by Chomsky (1989), also taking the child faces the additional task of determining the hierarchical order of
up the suggestions by Borer (1984). He proposes that parameters of UG should these elements in the clause structure of the language he or she is acquiring
only relate to the lexicon, not to the computational system, thus restricting the using information contained in the data, most importantly by making infer-
class of possible parameters to categories of lexical items or to properties of ences based on the surface order of morphemes realizing these functional
lexical items (e.g. canonical government). This excludes, for example, categories. It remains to be seen whether this is theoretically viable and whether
parameterization of subjacency. Chomsky (1989) goes, in fact, a step further, it represents indeed a feasible task for the learner.
stating "If substantive elements (verbs, nouns, etc.) are drawn from an invari-
ant universal vocabulary, then only functional elements will be parameterized."
This amounts to saying that parameterization should be restricted to a closed 8.2 Can parameters be reset?
class of functional heads like COMP, INFL, and DET. Interestingly, this Func-
tional Parameterization Hypothesis is in line with hypotheses based on acqui- The second proposal concerns the possibility of parameter resetting, until re-
sition data like the Structure Building Hypothesis by Guilfoyle and Noonan cently an uncontroversial assumption. Under several scenarios this issue be-
(1988), mentioned above in section 6, and similar approaches by Radford (1986, comes crucial. One example is the above mentioned case (section 4) of default
1990) and Lebeaux (1988); see the contributions in Meisel (1992), for some settings of parameters (see Hyams, 1986, 1989). If this is correct, it is necessary
recent discussions of this issue. to allow the possibility of switching from one setting to another, and this again
The most consistent attempt to pursue the research program outlined by means that the respective other option must remain accessible to the learner.
Chomsky (1989), according to which functional heads are the prime locus of One way to avoid the necessity of resetting is to follow the suggestion by
parameterization, can be found in the work by Ouhalla (1990, 1991). He claims Lebeaux (1988) according to which the child "falls into" the default grammar
that grammatical properties of a language are all due to the set of functional without actively setting the parameter on one of its values. But the problem
categories implemented in its grammar, and all parameters reflect properties remains in view of the expected ambiguity of the data, discussed in section 5,
of functional categories. One of the merits of Ouhalla's proposal is that it above; if one cannot rule out the possibility that the child initially chooses the
offers an explicit definition of functional categories which distinguishes them wrong option, one will have to explain how the correct choice can eventually
from their substantive counterparts, resulting in a set comprising some catego- be made and, perhaps even more difficult, make sure that the parameter is not
ries normally not classified as functional elements, e.g. pronouns and certain constantly set and reset.
prepositions. Ouhalla (1991) succeeds in giving a unified treatment to a variety An explicit account of this "pendulum problem" and of how it might per-
of phenomena. Word order patterns, for example, can be related to properties haps be solved is given by Randall (1990, 1992). She proposes the Catapult
of functional categories. Take the position of the verb. Rather than having the Hypothesis, based on the assumption that there exists, indeed, a unique trig-
head-direction parameter apply to the VP, verb placement depends on the ger for each parameter. The "catapult" is a mechanism enabling the child to
internal structure and on the position of the functional category to which retreat from a false choice. It consists of a principle of grammar and a piece
the verb is moved; in fact, the order of V in VP is not fixed at all. This has of primary data. The logic underlying the catapult is the modus ponendo tollens,
interesting consequences for language development. Assuming that first gram- i.e. "Either A or B. A, therefore: not B." A is a phenomenon encountered in the
mars lack functional categories, early child utterances are correctly predicted data which can serve as a trigger, B is a particular setting of the parameter,
to exhibit variable verb placement patterns. It is only when AGRP or some e.g. A = there exist expletives, B = [+pro-drop]. The either/ or logic of the
other functional phrase of this type has been implemented that the verb will catapult forces the learner to eliminate possibility B if fact A is found in the
assume a fixed position corresponding to the position of the functional head. data. Yet the second question remains: what prevents the child from resetting
Note that under this approach word order is partly accounted for as resulting the parameter at every instance when it encounters data which appear to
from variable order of functional categories in clause structure. In other words, be in conflict with an earlier decision? Randall believes that because of unique
at least for some of these categories, namely AGR, TNS, and NEG, it is claimed triggers the pendulum problem will not arise in syntax but only in the lexicon,
32 Jurgen M. Meisel Parameters in Acquisition 33

the case discussed by her (1990, 1992). Here, the child can assume a conserva- to _reset parameters. For one. child, however, Miiller finds that he has indeed
tive strategy, i.e. retreat item by item. set the finiteness parameter on the wrong value, in the fashion described above.
It is, however, by no means obvious that the pendulum effect should not be As a consequence, once subordinate clauses appear, he uses verb-second pat-
a problem for syntax, given the repeatedly mentioned ambiguity of the data; terns and verb-third constructions with the finite verb immediately following
see section 7. And yet, one first important observation (Clahsen, 1992) is that the complementizer as well as the subject or the topicalized nonsubject. The
one does not find evidence for the pendulum effect; children do not switch reason for the erroneous setting apparently is that for this child complementizers
back and forth between different parameter values. Clahsen (1991b}, therefore, evolve out of prepositions instantiated initially by far, "for." Given patterns of
proposes the Parameter Setting Constraint, according to which parameters historic change of a similar sort, this is by no means an implausible hypoth-
cannot be reset. He then proceeds to show that certain word order phenomena esis, but it results in a misanalysis of COMP as a substantive rather than a
in the acquisition of German can be accounted for without resorting to para- functional category. The most fascinating point, however, is what happens
meter setting; see below. But if this is correct, does it mean that the settings subsequently. Once the boy starts using items representing the functional
must be correct in the first place, or are there other means to retreat from the category COMP, this does not entail adultlike use of word order in subordi-
error? nate clauses. In other words, no clustering of syntactic phenomena related to
An interesting case is discussed by Miiller (1993, 1994). As has been men- one parameter can be observed. Rather, he learns the correct ordering, sepa-
tioned before in connection with the Structure Building Hypothesis, a number rately for each complementizer, in an item-by-item process extending over a
of studies indicate that early grammars lack CP. This creates a problem in considerable period of time (approximately two years).
explaining certain verb placement phenomena with children acquiring Ger- To conclude then, not only has Clahsen (1991b) demonstrated that certain
man, since they do use SVO, during this period, and occasionally also appar- phenomena for which parameter resetting had been suggested as an explana-
ent V2 structures, with some element preceding the finite verb and the subject tion can be accounted for adhering to the Parameter Setting Constraint, Muller
following it (e.g. OVS, AdvVS). For adult German, a verb-second language, (1993) offers empirical evidence indicating that erroneous settings need to be
the usual analysis argues that finite verbs move to the head of CP, and a remedied by a long drawn-out process of inductive learning, thus lending
maximal projection (e.g. the subject or some other topicalized constituent) strong support to the proposed constraint.
appears in the specifier position of CP. Both these positions are, however, not
available as long as the child does not have access to CP; assuming that VP is
head-final, the child should therefore only use verb-final constructions. One
solution could be that, in these cases, the verb appears in a head-initial func- 9 Conclusions
tional category other than CP, i.e. IP or some variant thereof (e.g. FP, AGRP);
see Clahsen (1991), Meisel and Miiller (1992). This is in accordance with the The objective of this chapter ahs been to present a critical discussion of the
verb-second parameter proposed by Platzack and Holmberg (1989), according contribution of parameter theory to explanations of language development.
to which the finiteness operator is placed either in IP (Romance languages, Focusing on two examples, the null-subject and the head-direction parameter,
English, etc.) or in CP (V2 languages). Eventually, however, the child needs to its fundamental theoretical concepts were introduced and a number of em-
revise his or her grammar in order to make it conform to the adult version. pirical consequences were outlined. I hope to have shown that this theory is
Meisel and Miiller (1992) suggest that the verb-second parameter is reset, once potentially of great interest for language acquisition studies. At the same time,
CP has been implemented in the grammar, triggered by the acquisition of however, it should also have become clear that the theory still needs to be
complernentizers. A functional category thus has to be added. Yet if resetting elaborated in more detail. Most importantly, it is too powerful in some re-
of parameters is excluded, this solution has to be ruled out. As a consequence, spects and needs to be constrained more tightly. Several such proposals have
one would have to predict that if the parameter has indeed been set on an been discussed here. In the current state of our knowledge, it seems to be
incorrect value, the corresponding properties of the target grammar which the possible to restrict parameterization to lexical heads, probably even to func-
correct parameter value would have yielded automatically will have to be tional heads. It furthermore appears to be preferable to exclude the possibility
acquired by inductive learning. Analyzing the language development of of resetting parameters, once they have been set to a specific value; this also
bilingual (French and German) children, Muller (1993, 1994) demonstrates that constrains the use of default options.
this is indeed what happens. Leaving details aside, one can summarize her The specific interest of parameter theory is in great deal due to the fact that
findings by stating that functional categories in early grammars tend to be it defines the role of the input data for grammatical development, a rather
underspecified in terms of their featural composition, as suggested by Clahsen sensitive issue within a theoretical framework which has virtually eliminated
(1991b); these intermediate grammars can thus be restructured without having inductive learning as a means to acquire abstract grammatical knowledge. It
34 Jurgen M. Meisel Parameters in Acquisition 35

is, nevertheless, crucial not to confound issues related to parameters as ele- unambiguous information can only terms of semantic and pragmatic
ments of a theory of UG and those concerning the way they are put to use. The be found in subordinate clauses; see functions. If these distinctions are
subset principle, to mention one example, defines an important learnability below. kept in mind (see Meisel, 1994), it is
6 I want to thank Jurgen Weissenborn of only marginal interest which
condition, but it is not part of parameter theory proper, i.e. of UG. A number terminology is used.
(pers. comm.) for clarifications on
of requirements concerning the data have been dealt with at some length this point. 9 Haider (1993) argues that fixing
during this chapter, e.g. that parameters should not require anything other 7 Roeper and de Villiers (1992), parameters presupposes the
than positive evidence, probably of a structurally simple· nature. The claim, however, also discuss the possibility existence of an interpreting interface
however, that for each parameter the:-e should be a telltale or unique trigger, of extrinsic ordering. between UG and the general
defines the nature of parameters and is therefore part of the theory, notwith- 8 Although protolanguage is not cognitive mechanisms for which
standing the fact that the learner may go wrong in identifying this trigger. It constrained by UG, it does not there is no evidence. In his view,
must be kept in mind that triggers do not appear labeled as such in the data; seem to be adequate to talk of "The UC-potential is a cognitive co-
rather, they need to be identified by means of a grammatical analysis. violations of UG principles, in this processor. It is activated whenever
I believe that parameter theory has become popular because the notion of case, for UG has not yet been there is a data structure that suits
parameter has been understood as offering a particularly useful metaphor activated. Strictly speaking, this is the capacity .. . Under this
not even in contradiction with the perspective, the processing potential
for an as yet poorly understood mechanism of the language-specific mental
continuity assumption, although amounts to a filtering and
capacity. In my view, it is quite legitimate to proceed in this way and to oper- reinforcement device. It will channel
Bickerton argues for discontinuity.
ate temporarily with a vaguely defined concept, for this enables one to reflect Discontinuity, as defined by Pinker the processing of the language data
upon issues of which one only has a vague understanding. It should be obvi- (1984), refers to the claim that adult into UC-conformable channels, and
ous, however, that during the course of further research one must make every grammars evolve out of earlier it will do so inevitably" (Haider,
effort to arrive at more precise definitions. At the same time, the epistemologi- systems consisting of categories and 1993: 13).
cal implications should be spelled out, and the neurobiological basis for these rules which are not defined in 10 For technical details, see any of
assumptions will have to be clarified. In this way, one may hope to gain pro- terms of abstract grammatical the recent introductions to
found insights into the human linguistic capacities. Although the enterprise principles, as mature grammars, but generative syntax, e.g. Freidin
may still fail, I believe that it is worthwhile pursuing. are motivated differently, e.g. in (1992: ll0ff).

NOTES

I am grateful to Susanne E. Carroll and intentionality on behalf of the child

Jurgen Weissenborn, as well as to or that this is a case of hypothesis
Andolin Eguzkitza, Lynn Eubank, testing.
Georg Kaiser, Helen Leuninger, 3 See Meisel (1990c}, Valian (1990b},
Natascha Muller, Zvi Penner, and and Weissenborn (1992), among
Achim Stenzel, for reading and others, who argue against this
commenting on an earlier version of hypothesis.
this chapter. 4 Lebeaux (1988: 184££) actually
1 There has been some discussion, revises this approach in order to
recently, as to whether all maximal make it compatible. with his General
projections have the same internal Congruence Principle.
structure, e.g. whether they all 5 Roeper and Weissenborn (1990), for
require a Spec position; see Speas example, mention a number of
(1990). studies which apparently provide
2 Verbs like "explore," "choose," etc. evidence against each telltale
are used metaphorically in this suggested for the null-subject
context. I do not mean to imply parameter so far. They claim that